HK1234870B - Methods and apparatus to compensate impression data for misattribution and/or non-coverage by a database proprietor - Google Patents

Description

Method and apparatus for compensating database owners for misidentification and/or non-coverage of impression data

This patent claims priority to U.S. Provisional Patent Application Serial No. 61/952,726 filed on March 13, 2014, U.S. Provisional Patent Application Serial No. 61/979,391 filed on April 14, 2014, U.S. Provisional Patent Application Serial No. 61/986,784 filed on April 30, 2014, U.S. Provisional Patent Application Serial No. 61/991,286 filed on May 9, 2014, U.S. Provisional Patent Application Serial No. 62/014,659 filed on June 19, 2014, U.S. Provisional Patent Application Serial No. 62/023,675 filed on July 11, 2014, and U.S. Provisional Patent Application Serial No. 62/030,571 filed on July 29, 2014. The entire contents of U.S. Provisional Patent Application Serial No. 61/952,726, U.S. Provisional Patent Application Serial No. 61/979,391, U.S. Provisional Patent Application Serial No. 61/986,784, U.S. Provisional Patent Application Serial No. 61/991,286, U.S. Provisional Patent Application Serial No. 62/014,659, U.S. Provisional Patent Application Serial No. 62/023,675, and U.S. Provisional Patent Application Serial No. 62/030,571 are incorporated herein by reference.

Technical Field

The present invention relates generally to monitoring media and, more particularly, to methods and apparatus for compensating impression data by database owners for misidentification and/or non-coverage.

Background Art

Traditionally, audience measurement entities determine audience engagement with media programs based on registered panelists. Specifically, the audience measurement entity enrolls individuals who have consented to be monitored into a panel. The audience measurement entity then monitors those panelists to determine the media (e.g., television or radio programs, movies, DVDs, advertisements, etc.) to which those panelists were exposed. In this manner, the audience measurement entity can determine exposure measurements for different media based on the collected media measurement data.

The technology for monitoring user access to Internet resources (such as web pages, advertisements, and/or other media) has evolved significantly over the years. Some existing systems primarily perform this type of monitoring through server logs. In particular, an entity serving media on the Internet may use such existing systems to record the number of requests received for its media at its server.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 illustrates an exemplary system for collecting impressions of media presented at a mobile device and collecting user information from a distributed database owner for association with the collected impressions.

2 is an example impression data compensator that may be implemented in the example audience measurement server of FIG. 1 to compensate impression data for inaccuracies related to misidentification and non-coverage resulting from the database owner's impression collection techniques.

3A-3C collectively illustrate an exemplary data flow that may be executed by the exemplary impression data compensator of FIG. 2 to compensate impression data for inaccuracies related to misidentification and non-coverage resulting from the database owner's impression collection techniques.

4 is a flow diagram representative of example machine-readable instructions that may be executed to implement the example impression data compensator of FIG. 2 to compensate impression data for inaccuracies related to misidentification and non-coverage resulting from a database owner's impression collection techniques.

5 is a flow diagram representative of example machine readable instructions that may be executed to implement the example impression data compensator of FIG. 2 to calculate a misidentification calibration matrix and/or a co-viewing matrix.

6 is a flow chart representative of example machine readable instructions that may be executed to implement the example impression data compensator of FIG. 2 to determine the probability of a false positive for multiple pairs of demographic groups.

7 is a flow diagram representative of example machine readable instructions that may be executed to implement the example impression data compensator of FIG. 2 to generate a misidentification calibration matrix and/or a co-viewing matrix.

8 is a flow diagram of example machine readable instructions that may be executed to implement the example impression data compensator of FIG. 2 to generate a co-viewing matrix.

9 is a flow diagram representative of example machine-readable instructions that may be executed to implement the example impression data compensator of FIG. 2 to generate an alpha factor associated with a media audience not covered by a database owner.

10 is a flow diagram representative of example machine readable instructions that may be executed to implement the example impression data compensator of FIG. 2 to adjust impressions based on a misidentification calibration matrix.

11 is a flow diagram representative of example machine readable instructions that may be executed to implement the example impression data compensator of FIG. 2 to adjust impressions based on an uncoverage factor corresponding to a database owner's impression collection technique.

12 is a flow diagram representative of example machine-readable instructions that may be executed to implement the example impression data compensator of FIG. 2 to calculate demographics (or non-coverage factors) associated with media audiences not covered by a database owner.

13 is a flow diagram representative of example machine readable instructions that may be executed to implement the example impression data compensator of FIG. 2 to adjust impressions and/or unique audiences based on an uncovered factor for a database owner.

14 illustrates an exemplary system configured to determine the type of media impression data to collect in accordance with the teachings of the present invention.

FIG. 15 illustrates an example of the type predictor of FIG. 14 that determines a type for collected media impression data.

16 illustrates an example of the data classifier of FIG. 15 for classifying collected media impression data to be used by the genre analyzer of FIG. 15 to predict genre.

17 shows an example graph plotting example daytime portions and example daytime categories used by the type predictors of FIGS. 14 and 15 .

FIG. 18 illustrates an example of the type modeler of FIG. 14 that may be used to construct a type model.

19 is a flow diagram representative of example machine readable instructions that may be executed to implement the example type modeler of FIGS. 14 and 18 to construct a type model.

20 is a flow diagram representative of example machine readable instructions that may be executed to implement the example type predictor of FIGS. 14 and 15 to assign types to collected media impression data.

FIG. 21 illustrates an exemplary processor platform that may be used to execute the exemplary instructions of FIG. 4 through FIG. 13 to implement the exemplary apparatus and systems disclosed herein.

FIG. 22 illustrates another exemplary processor platform that may be used to execute the exemplary instructions of FIG. 19 and/or FIG. 20 to implement the exemplary apparatus and systems disclosed herein.

DETAILED DESCRIPTION

The technology used to monitor user access to Internet resources (such as web pages, advertisements, and/or other media) has evolved significantly over the years. At one time, such monitoring was primarily accomplished through server logs. In particular, entities serving media on the Internet would record at their servers the number of requests received for their media. Basing Internet usage studies on server logs is problematic for several reasons. For example, server logs can be tampered with directly or with the help of bots that repeatedly request media from the server to increase the server log count of the requested media. Second, media is sometimes retrieved once, cached locally, and then viewed repeatedly from the local cache without involving the server in the repeated viewings. Server logs cannot track these views of cached media because reproducing the locally cached media does not require further requests to the server for the media. Consequently, server logs are susceptible to both overcounting and undercounting errors.

The invention disclosed in Blumenau's U.S. Patent 6,108,637 fundamentally changes the way Internet monitoring is performed and overcomes the limitations of the server-side log monitoring techniques described above. For example, Blumenau discloses a technique in which Internet media to be tracked is tagged with beacon instructions. Specifically, the monitoring instructions are associated with the Hypertext Markup Language (HTML) of the media to be tracked. When a client requests the media, the media and the beacon instructions are both downloaded to the client. Thus, the beacon instructions are executed whenever the media is accessed from the server or from a cache.

The beacon instructions cause monitoring data reflecting information about access to the media to be sent from the client that downloaded the media to a monitoring entity. Typically, the monitoring entity is an Audience Measurement Entity (AME) (e.g., any entity interested in measuring or tracking audience reach for advertising, media, and/or any other media) that does not provide the media to the client and is a trusted third party (e.g., Nielsen Inc.) used to provide accurate usage statistics. Advantageously, because the beacon instructions are associated with the media and are executed by the client browser whenever the media is accessed, monitoring information is provided to the AME regardless of whether the client is a panel member of the AME.

However, it would be useful to link demographics and/or other user information to monitoring information. To address this problem, AME creates groups of users who have agreed to provide their demographic information and have their Internet browsing activity monitored. When individuals join the group, they provide AME with detailed information about their identity and demographics (e.g., gender, race, income, home address, occupation, etc.). AME sets up a cookie on the group member's computer that enables AME to identify the group member whenever the group member accesses tagged media, thereby sending monitoring information to AME.

Because most clients providing monitoring information from tagged pages are not panel members and are therefore unknown to AME, statistical methods are needed to attribute demographic information based on data collected for panel members to the larger population of users providing data for tagging media. However, compared to the total user population, the panel size of AME is small. Therefore, the problem arises of how to increase the panel size while ensuring accurate panel demographic data.

There are many database owners operating on the Internet. These database owners provide services (e.g., social networking services, email services, media access services, etc.) to a large number of subscribers. In exchange for providing such services, subscribers register with the owner. As part of this registration, subscribers provide detailed demographic information. Examples of such database owners include social networking providers such as Facebook, Myspace, Twitter, etc. These database owners place cookies on the computers of their subscribers, enabling the database owners to recognize registered users when they visit their website.

The examples disclosed herein can be used to determine media impressions, ad impressions, media exposure, and/or ad exposure using user information that is distributed across different databases on the Internet (e.g., different website owners, service providers, etc.). The example methods, apparatus, and articles of manufacture disclosed herein not only accurately correlate Internet media exposure with user information, but also effectively expand the panel size and composition beyond the panel of participating audience measurement entities and/or ratings entities to include persons registered in other Internet databases, such as those of wireless service operators, mobile software/service providers, social media websites (e.g., Facebook, Twitter, Google, etc.), and/or any other Internet website, such as Yahoo!, MSN, Apple iTunes, Experian, etc. This expansion effectively leverages AME's media impression tracking capabilities and the use of databases of non-AME entities (such as social media and other websites) to create a large, demographically accurate panel that forms an accurate and reliable measurement of exposure to Internet media (such as advertisements and/or programs). Examples of such media include websites, images presented on websites, and/or streaming media accessible via a computing device (Amazon Video, Netflix, Hulu, etc.).

Traditionally, AMEs (also referred to herein as "rating entities") determine demographic coverage for advertising and media programming based on registered panelists. That is, AMEs enroll individuals who consent to being monitored into a panel. During enrollment, AMEs receive demographic information from enrollees, enabling subsequent correlations between advertising/media exposure to those panelists and different demographic markets. Unlike conventional techniques in which AMEs rely solely on their own panelist data to gather demographic-based audience measurement, the exemplary methods, apparatus, and/or articles of manufacture disclosed herein enable AMEs to share demographic information with other entities that operate based on a user registration model. As used herein, a user registration model is one in which users subscribe to the services of those entities by creating an account and providing demographic-related information about themselves. Sharing demographic information associated with a database owner's registered users enables AMEs to expand or supplement their panel data with substantially reliable demographic information from external sources (e.g., the database owner), thereby expanding the coverage, accuracy, and/or completeness of their demographic-based audience measurement. This access also enables AMEs to monitor individuals who would not otherwise join an AME panel. Any entity that has a database that identifies demographic data of a group of individuals can partner with AME. Such entities may be referred to as "database owners" and include entities such as wireless service carriers, mobile software/service providers, social media sites (e.g., Facebook, Twitter, Google, etc.), and/or any other Internet site (such as Yahoo!, MSN, Apple iTunes, Experian, etc.) that collects demographic data on users in exchange for services.

The examples disclosed herein can be implemented by AME (e.g., any entity interested in measuring or tracking audience reach for advertisements, content, and/or any other media) in collaboration with any number of database owners, such as online web service providers that develop online media reach metrics. Such database owners/online web service providers can be wireless service carriers, mobile software/service providers, social networking sites (e.g., Facebook, Twitter, MySpace, etc.), multi-service sites (e.g., Yahoo!, Google, Experian, etc.), online retailer sites (e.g., Amazon.com, Buy.com, etc.), and/or any other web service sites that maintain user registration records.

The use of demographic information from heterogeneous data sources (e.g., high-quality demographic information from a panel of audience measurement entities and/or registered user data from network service providers) improves the efficiency of reporting metrics for both online and offline advertising campaigns. The exemplary techniques disclosed herein use online registration data to identify user demographics and/or other user information, and use server impression counting and/or other techniques to track the number of impressions generated by those users. Online network service providers (such as wireless service operators, mobile software/service providers, social networking sites (e.g., Facebook, Twitter, MySpace, etc.), multi-service sites (e.g., Yahoo!, Google, Experian, etc.), online retailer sites (e.g., Amazon.com, Buy.com, etc.)) (collectively and individually referred to herein as online database owners) maintain detailed demographic information (e.g., age, gender, geographic location, race, income level, education level, religious affiliation, etc.) collected through the user registration process. An impression corresponds to a household or individual that has been exposed to the corresponding media and/or advertisement. Thus, an impression represents a household or individual that has been exposed to an advertisement or media or a group of advertisements or media. In Internet advertising, the number of impressions or impression count is the total number of times an advertisement or advertising campaign has been visited by network users (e.g., including visits reduced by, for example, pop-up blockers and/or augmented by, for example, retrieval from a local cache).

Examples disclosed herein adjust impression information obtained from one or more client devices and one or more database owners to improve the accuracy of demographic data corresponding to recorded impressions. When using database owner data to provide demographic information for impressions, audience demographics and/or impression information obtained from one or more client devices and one or more database owners can be skewed due to errors, including: 1) misidentification errors due to device sharing, and/or 2) database owner non-coverage errors. In some cases, these two different sources of bias appear to result in similar error factors, but are actually different biases. Examples disclosed herein generate calibration factors and apply the calibration factors to audience data to correct for these errors.

Misidentification error refers to a measurement bias that occurs when a first person belonging to a first demographic group is considered to be the person associated with a media impression on a device when, in fact, a second person belonging to a second demographic group is the person for whom the media impression occurred. In some examples of such situations, a mobile device is shared among multiple people in a household. Initially, a first person in the household uses the mobile device to access a website associated with a database owner (e.g., via the mobile device's web browser, via an application installed on the mobile device, etc.), and the database owner may identify the first person as associated with the mobile device based on the access performed by the first person (e.g., a login event). Subsequently, the first person stops using the device but does not log out of the database owner's system, and/or the second person does not log into the database owner's system, allowing the database owner to identify the second person as a different user than the first person. Consequently, when the second person begins accessing media using the same mobile device, the database owner continues (in this case, incorrectly) to identify the use of the mobile device (e.g., media impression) as associated with the first person. Consequently, impressions that should be attributed to the second person and the second demographic group are incorrectly attributed to the first person and the first demographic group. The effect of large-scale misidentification errors can create measurement bias errors by misrepresenting the demographic distribution of media impressions among a large audience and thereby distorting the audience demographics of impressions collected for advertisements and/or other media whose exposure is monitored by the audience measurement entity.

Exemplary Technical Advantages of the Disclosed Examples

Existing techniques for correcting false positives include determining an adjustment factor by comparing: A) demographic information collected for a computation session using panelist metering software installed on the client computer, and B) demographic information determined for the same computation session using cookie-based impressions from the database owner. Examples of such techniques are disclosed in U.S. patent application Ser. No. 13/756,493, filed on January 31, 2013. The entire contents of U.S. patent application Ser. No. 13/756,493 are incorporated herein by reference. The examples disclosed in U.S. patent application Ser. No. 13/756,493 rely on panelist metering software installed locally on the client computer to accurately identify panelists registered with a panel of an audience measurement entity. The examples disclosed in U.S. patent application Ser. No. 13/756,493 also rely on cookie or cookie-like data to determine the adjustment factor. Such techniques are not suitable for correcting false positives on devices that do not have panelist metering software installed and/or do not provide a cookie that can be used as a client device identifier, such as some mobile devices (e.g., iOS devices). In other words, the prior art relies on locally installed panelist metering software and cookies to generate the misidentification adjustment factor. Without such locally installed panelist metering software and/or without such cookies, the prior art would not be able to successfully generate the misidentification adjustment factor.

Compared to existing systems and methods, the examples disclosed herein generate misidentification factors using a misidentification correction matrix based on responses to surveys conducted on randomly selected persons and/or households, and do not rely on or use web trackers to generate the misidentification correction factors and/or misidentification correction matrices. As used herein, the term "database owner registration status" as used with reference to a person refers to whether the person is registered with one or more specific database owners. Existing techniques that rely on locally installed panelist metering software to collect database owner registration status (e.g., whether each family member is registered with a specific database owner) and to collect media access data performed by family members may not accurately correct impression information for misidentification bias error and/or non-coverage bias error, which are errors for impressions that occur on device types that are not monitored by the panelist metering software. The examples disclosed herein generate misidentification correction factors without relying on panelist metering software installed locally at a client computer to collect database owner registration status data. The examples disclosed herein also generate misidentification correction factors without relying on such locally installed panelist metering software to collect data about the media access of family members. Thus, the examples disclosed herein determine misidentification correction factors for any device type, including device types for which database owner registration status data, database owner login data, and/or data regarding family member media access is not collected (and/or is not collectable) using locally installed panelist metering software. Such devices are referred to herein as "non-locally metered devices."

In some of these disclosed examples, panelist data is collected and used differently than in the prior art. In such examples, panelist data is used to adjust the misidentification correction factor to more accurately determine the incidence of people in demographic groups living together. As used herein, the "incidence of people in demographic groups living together" refers to the relative frequency with which people in a first demographic group live with people in a second demographic group (e.g., the percentage of people in demographic group A who live with someone in demographic group B). However, an aggregated audience distribution corresponding to the collected impressions is still generated without reference to the panelist data. In the examples disclosed herein, the panelist data is used only to adjust the generated aggregated audience distribution. The adjusted aggregated audience distribution is used to generate and/or adjust the misidentification correction factor. For example, panelist data that can be used to adjust the incidence of people in demographic groups living together can include the corresponding number of people in the first demographic group who live with people in the second demographic group (e.g., the same demographic group as the first demographic group or a different demographic group). In some examples, panelist data used to adjust the incidence of people in demographic groups living together does not indicate whether the panelist is a registered user of the database owner (e.g., does not include cookies) and does not include information indicating media access using a computing device.

The disclosed examples determine the probability of misidentification by determining the probability that, when a database owner identifies a person in demographic group (i) as a viewer of the media, a person in demographic group (j) is the actual viewer of the media. In some examples, the probability of misidentification is calculated by redistributing the audience and/or impressions observed in association with households using a survey calibration data source. In such examples, the survey calibration data source is used to aggregate and adjust the redistributed audience and/or impressions for multiple households for oversampling and/or undersampling. In some such examples, the resulting probabilities are normalized to reflect the probability of misidentification for each impression observed by the database owner and associated with the demographic group.

Other disclosed examples determine the probability of misidentification due to device sharing as a combination of three separate probabilities: a) the probability of living in the same household, b) the probability of having access to a mobile device (of any type) in the household, and c) the probability of sharing a mobile device for a particular content type. Examples disclosed herein apply the resulting probability of misidentification to impression data as a factor to compensate for errors in the unique audience represented by the collected impression data. In some examples, providing such error compensation involves constructing a misidentification correction matrix to reflect the probability that an impression identified by the database owner as associated with a first demographic group should actually be associated with a second demographic group.

As used herein, non-coverage error is defined as measurement bias that occurs because the database owner is unable to identify a portion of the audience (e.g., their demographics) who use mobile devices to view media. In some instances, when a request is sent from a mobile device to the database owner, as in the examples disclosed above, the database owner is unable to match the data in the request to a person. The database owner may not be able to identify the person associated with a given impression because: 1) the person accessing the media that gave rise to the impression has not provided information about themselves to the database owner (e.g., the person is not registered with the database owner (e.g., Facebook) and therefore has no record of the person with the database owner, the registration profile for the person is incomplete, the registration profile for the person has been flagged as suspected of containing inaccurate information, etc.), 2) the person is registered with the database owner but does not access the database owner using the specific mobile device on which the impression occurred (e.g., the database owner is accessed only from a computer and/or a mobile device other than the mobile device associated with the current request, and/or a user identifier for the person is not available on the mobile device on which the impression occurred), and/or 3) the person is registered with the database owner and accesses the database owner using the mobile device on which the impression occurred (e.g., the person has previously logged into the database owner from the mobile device), but has taken other active or passive measures (e.g., blocking or deleting cookies) that prevent the database owner from associating the mobile device with the person. In some examples, a user identifier for a person is not available on the mobile device on which the impression occurred because the mobile device and/or application/software on the mobile device is not a cookie-based device and/or application.

The examples disclosed herein generate device sharing matrices and/or non-coverage factors for different combinations of media categories and/or mobile device types. Examples of media categories (for which separate device sharing matrices and/or non-coverage factors may be generated) include advertising, children's programming, comedies, dramas, feature films, information and/or news programming, sports, variety shows (e.g., game shows, reality shows, talk shows), and/or other categories. Examples of device types (for which separate device sharing matrices and/or non-coverage factors may be generated) include smartphones (e.g., iPhones, smartphones based on the Android operating system, Blackberry smartphones, smartphones based on Windows Mobile, etc.), tablets (e.g., iPads, tablets based on the Android operating system, etc.), portable media players (e.g., iPods, etc.), and/or other device types. Such device types may be cookie-based devices (e.g., devices running cookie-based applications/software) and/or non-cookie-based devices (e.g., devices running applications/software that do not employ cookies, such as Apple iOS devices).

The disclosed exemplary methods and apparatus for compensating impression information for misidentification and/or non-coverage errors solve the technical problem of accurately determining demographic information associated with media impressions that are transmitted and monitored via a network such as the Internet. Such computing devices are used to collect media impressions that occur at the computing device and data indicating the occurrence of media impressions. Calibration of such data can include collecting, processing, and/or analyzing hundreds, thousands, millions, or more impressions. Therefore, calibration and/or correction of such large amounts of data presents a significant technical challenge. The disclosed examples can be applied to a set of media impressions collected via a computing device to generate accurate demographic information for a large number of media impressions (e.g., 100,000 impressions per week) in an efficient and timely manner. This provides a significant improvement in the technical field of audience measurement.

The examples disclosed herein can be used with an online activity rating (OCR) system developed by Nielsen Inc. (USA). The OCR system is an efficient system for collecting and analyzing large amounts of data. The OCR system does not require panelist software to obtain the data to be processed. The technology disclosed herein is able to generate adjustment factors without the need to introduce panelist software. This reduces the amount of software required and eliminates the need for end-user computers. These are technical advantages that reduce the cost and use of computing resources. The technology disclosed herein is also used to effectively calibrate OCR automatic systems to correct misidentification errors that may occur in their operation. Therefore, the disclosed technology solves the technical problem of calibrating OCR systems to accurately reflect real-world conditions, and therefore, calibration is achieved by eliminating the need for panelist software distributed throughout the system.

The disclosed exemplary method includes: receiving a first request from a first type of computing device at a first Internet domain. In the exemplary method, the first request indicates access to media at the computing device. The exemplary method also includes: sending a request for demographic information, the demographic information corresponding to a plurality of requests received from the first type of computing device at the first Internet domain. The plurality of requests includes the first request. The exemplary method also includes: obtaining a count of media impressions that occurred on the first type of computing device, a first portion of the media impressions corresponding to persons whose demographic information is identifiable by a database owner and a second portion of the media impressions corresponding to persons whose demographic information is not available to the database owner; and determining demographic information for the second portion of the media impressions based on a first probability that the media corresponding to the media impression was accessed on the first type of computing device and based on a second probability that the media was accessed on the second type of computing device.

In some exemplary methods, determining the demographic information includes multiplying a ratio of a first probability to a second probability by the number of media impressions attributed to a first demographic group. In some exemplary methods, the first probability is a probability that a person in the first demographic group accesses the media on a first type of computing device, and the second probability is a probability that a person in the first demographic group accesses the media on a second type of computing device. Some exemplary methods further include adjusting the media impressions to compensate for incorrectly attributing a subset of the media impressions to a second person in a second demographic group, and determining the number of media impressions attributed to the first demographic group from the subset of media impressions adjusted to compensate for the incorrectly attributed media impressions.

In some exemplary methods, the first type of computing device includes a mobile device and the second type of computing device includes a television. In some exemplary methods, the first type of computing device includes at least one of the following: a smartphone, a tablet computer, or a portable media player. In some exemplary methods, the first probability and the second probability correspond to a media category of the media. In some exemplary methods, the media category is at least one of comedy, drama, politics, reality, or a combination of media categories.

Some example methods further include, before determining the demographic information, adjusting the media impression for a situation in which the media impression is incorrectly attributed to a person who did not cause the media impression. Some example methods further include calculating a first probability and a second probability based on survey responses to a survey of the person, calculating the first probability comprising determining, from the survey responses, a weight associated with at least one of a demographic group, a media category, a type of computing device, or a geographic region, the weight indicating a corresponding probability of accessing the media associated with the media impression on the device type of interest.

In some exemplary methods, the survey is conducted on at least one of a random panel or a panel of audience members maintained by an audience measurement entity. In some exemplary methods, determining demographic information for the portion of the media impression includes determining proportions of media impressions attributed to different demographic groups corresponding to the person, and scaling the proportions of the media impressions to the portion of the media impression.

Some exemplary methods also include providing instructions to an issuer, wherein the instructions are to be provided by the issuer to a computing device. When executed by the computing device, the instructions provided by the issuer cause the computing device to send a first request. Some exemplary methods also include saving computer processing resources by not communicating with individual online users about their online media access activities and by not requesting survey responses from online users to determine a first probability of accessing media corresponding to a media impression on a first type of computing device or to determine a second probability of accessing the media on a second type of computing device. Some exemplary methods also include saving network communication bandwidth by not communicating with individual online users about their online media access activities and by not requesting survey responses from online users to determine a first probability of accessing media corresponding to a media impression on a first type of computing device or to determine a second probability of accessing the media on a second type of computing device.

The disclosed exemplary apparatus includes an impression collector configured to receive a first request from a computing device at a first internet domain and to send a request for demographic information corresponding to a plurality of requests received at the first internet domain from computing devices of a first type, the plurality of requests including the first request. In the disclosed exemplary apparatus, the first request indicates access to media at the computing device. The disclosed exemplary apparatus also includes an impression information collector configured to access counts of media impressions occurring on computing devices of the first type, a first portion of the media impressions corresponding to persons whose demographic information is identifiable by a database owner, and a second portion of the media impressions corresponding to persons whose demographic information is not available to the database owner. The disclosed exemplary apparatus also includes an uncovered detector configured to determine demographic information for the portion of the media impressions based on a first probability of accessing the media corresponding to the media impression on the computing device of the first type and a second probability of accessing the media on the computing device of the second type, at least one of the impression information collector or the uncovered detector being implemented by logic circuitry.

Some example apparatuses further include an uncovered calculator configured to multiply a ratio of the first probability to the second probability by the number of media impressions attributed to the first demographic group. In some example apparatuses, the first probability is a probability that a person in the first demographic group accesses the media on a first type of computing device, and the second probability is a probability that a person in the first demographic group accesses the media on a second type of computing device.

Some exemplary apparatuses further include a misattribution corrector for adjusting media impressions to compensate for incorrectly attributing a subset of media impressions to a second person in a second demographic group, determining a number of media impressions attributed to the first demographic group from the subset of media impressions adjusted to compensate for the incorrect attribution. In some exemplary apparatuses, the first type of computing device comprises a mobile device and the second type of computing device comprises a television. In some exemplary apparatuses, the first type of computing device comprises at least one of the following: a smartphone, a tablet computer, or a portable media player. In some exemplary apparatuses, the non-coverage calculator is for calculating the first probability and the second probability based on survey responses to a survey of persons, and the non-coverage calculator is for calculating the first probability by determining, from the survey responses, a weight associated with at least one of a demographic group, a media category, a type of computing device, or a geographic region, the weight indicating a corresponding probability of accessing the media associated with the media impression on the device type of interest.

In some exemplary arrangements, the survey is conducted on at least one of a random panel or a panel of audience members maintained by an audience measurement entity. In some exemplary arrangements, the first probability and the second probability correspond to a media category of the media. In some exemplary arrangements, the media category is at least one of comedy, drama, politics, reality, or a combination of media categories. In some exemplary arrangements, the uncovering corrector is configured to determine demographic information for the portion of the media impression by determining a proportion of the media impression that is attributed to different demographic groups corresponding to the person, and scaling the proportion of the media impression to the portion of the media impression.

An additional disclosed exemplary method includes receiving a first request from a computing device of a first type at a first internet domain and sending a request for demographic information corresponding to a plurality of requests received from the computing device of the first type at the first internet domain, the plurality of requests including the first request. In the disclosed exemplary method, the first request indicates access to media at the computing device. The disclosed exemplary method also includes generating an aggregated audience distribution comprising a first audience distribution for a first household aggregated with a second audience distribution for a second household based on survey responses of the first household, the first audience distribution comprising a distribution of first family members of the first household across demographic groups of the first family members registered with a database owner, the first audience distribution based on access to first media by the first family members; normalizing the aggregated audience distribution to generate a false positive correction matrix comprising a probability of attributing an impression of the media to the first demographic group when the database owner determines that the impression corresponds to a person in the second demographic group; and compensating for false positive errors in the impression by reallocating the impression from the second demographic group to the first demographic group using the false positive correction matrix.

Some exemplary methods also include: generating a correction metric to correct the aggregate audience distribution for at least one of oversampling or undersampling associated with a survey calibration data source that indicates access to media by a first family member and a registration status of the first family member with a database owner. In some exemplary methods, the correction metric is based on a second probability that a first person in a first demographic group lives with a second person in a second demographic group. In some exemplary methods, generating the correction metric includes: determining a first number of first person pairs based on the first survey calibration data, representing first person pairs in the first person pair including a first person in the first demographic group and a second person in the second demographic group living together; determining a second number of second person pairs based on the second survey calibration data, representing second person pairs in the second person pair including a third person in the first demographic group and a fourth person in the second demographic group living together, the second survey calibration data having a higher accuracy than the first survey calibration data; and determining a ratio of the first number to the second number.

In some exemplary methods, for each first demographic group, the aggregate audience distribution describes the number of persons in the second demographic group who would be attributed to the first demographic group, and normalizing the aggregate audience distribution includes scaling the number of persons in the second demographic group so that the total number of persons for the second demographic group is a specified value. Some exemplary methods also include: determining a sharing pattern for the first household, the sharing pattern indicating first members of the first household who access the media category and other members of the first household who do not access the media category; and determining a probability density function of device sharing probabilities for first members of the first household in the first household based on the sharing pattern for the media category, the probability density function indicating a corresponding probability that the first member of the household accesses media in the media category, the first audience distribution being based on the device sharing probabilities.

Some exemplary methods further include aggregating the device sharing probabilities for the first family members when the first household has two or more registered first family members of the first family members who are in the same second demographic group. Some exemplary methods further include generating a first audience distribution by distributing a portion of the numbers into the demographic groups, the portion of the numbers totaling a count of those first family members of the first family members that are registered with the database owner. In some exemplary methods, generating the first audience distribution does not use a cookie. In some exemplary methods, generating the first audience distribution and generating the aggregated audience distribution does not use metering software to collect the registration status of the family members with the database owner or to collect media access data. In some exemplary methods, reallocating impressions includes determining the product of an error correction matrix and an impression matrix, the impression matrix indicating the number of impressions corresponding to the respective demographic groups determined by the database owner. In some exemplary methods, reallocating the impressions results in a total number of impressions being the same as the total number of impressions in the impression matrix.

Some example methods also include providing instructions to an issuer, which are provided by the issuer to a computing device and, when executed by the computing device, cause the computing device to send a first request. Some example methods also include conserving computer processing resources by not communicating with individual online users about their online media access activities and not requesting survey responses from online users to generate an aggregate audience distribution. Some example methods also include conserving network communication bandwidth by not communicating with individual online users about their online media access activities and not requesting survey responses from online users to generate an aggregate audience distribution.

An additional disclosed exemplary apparatus includes an impression collector configured to receive a first request from a computing device of a first type at a first internet domain and to send a request for demographic information corresponding to a plurality of requests received from the computing device of the first type at the first internet domain, the plurality of requests including the first request. In the disclosed exemplary apparatus, the first request indicates access to media at the computing device. The disclosed exemplary apparatus also includes an aggregated distribution generator configured to generate an aggregated audience distribution comprising a first audience distribution for a first household aggregated with a second audience distribution for a second household, based on survey responses from the first household, the first audience distribution comprising a distribution of first family members of the first household across demographic groups of the first family members registered with a database owner, the first audience distribution based on access to the first media by the first family members. The exemplary apparatus also includes a matrix normalizer configured to normalize the aggregated audience distribution to generate a false positive correction matrix comprising a probability that an impression of the media is attributed to a person in the first demographic group when the database owner determines that the impression corresponds to a person in the second demographic group. The disclosed exemplary apparatus also includes a misidentification corrector for compensating for misidentification errors in impressions by reallocating impressions from the second demographic group to the first demographic group using a misidentification correction matrix, at least one of the aggregate distribution generator, the matrix normalizer, or the misidentification corrector being implemented by logic circuitry.

Some exemplary devices also include a matrix corrector for generating a correction metric to correct the aggregate audience distribution for at least one of oversampling or undersampling associated with a survey calibration data source indicating access to media by a first family member and a registration status of the first family member with a database owner. In some exemplary devices, the correction metric is based on a second probability that a first person in a first demographic group lives with a second person in a second demographic group. In some exemplary devices, the matrix corrector is configured to generate the correction metric by: determining a first number of first person pairs based on first survey calibration data, i.e., the number of first person pairs that respectively include a first person in the first demographic group and a second person in the second demographic group living together; determining a second number of second person pairs based on second survey calibration data, i.e., the number of second person pairs that respectively include a third person in the first demographic group and a fourth person in the second demographic group living together, the second survey calibration data having a higher accuracy than the first survey calibration data; and determining a ratio of the first number to the second number.

In some exemplary apparatuses, for each first demographic group, an aggregate audience distribution describes the number of persons in a second demographic group that would be attributed to the first demographic group, and a matrix normalizer is used to normalize the aggregate audience distribution by scaling the corresponding number of persons in the second demographic group so that the number of persons for the second demographic group is a specified value.

Some exemplary devices also include a household distribution generator for: determining a sharing pattern for a first household, the sharing pattern indicating first household members among the first household members who access a media category and first household members among the first household members who do not access the media category; and determining a probability density function of device sharing probabilities for first household members in the first household based on the sharing pattern for the media category, the probability density function indicating corresponding probabilities of first household members accessing media in the media category, the first audience distribution being based on the device sharing probabilities.

In some exemplary apparatuses, an aggregate distribution generator is configured to aggregate device sharing probabilities for first family members when the first household has two or more registered first family members of the first family members who are in the same second demographic group. Some exemplary apparatuses also include a household distribution generator configured to generate a first audience distribution by distributing portion numbers into demographic groups, the portion numbers totaling a count of those first family members of the first family members that are registered with a database owner. In some exemplary apparatuses, a false positive corrector is configured to reallocate impressions by determining the product of a false positive correction matrix and an impression matrix, the impression matrix indicating the number of impressions corresponding to the respective demographic groups as determined by the database owner. In some exemplary apparatuses, the false positive corrector is configured to reallocate impressions such that the total number of reallocated impressions is the same total number of impressions as the total number of impressions in the impression matrix.

An additional disclosed exemplary method includes receiving a first request from a computing device of a first type at a first internet domain and sending a request for demographic information corresponding to a plurality of requests received from the computing device of the first type at the first internet domain, the plurality of requests including the first request. In the disclosed exemplary method, the first request indicates access to media at the computing device. The disclosed exemplary method also includes generating an aggregated audience distribution comprising a first audience distribution for a first household aggregated with a second audience distribution for a second household, the first audience distribution comprising a distribution of first family members of the first household across a demographic group of those first family members registered with a database owner, the first audience distribution based on access to first media by the first family members, the aggregated audience distribution generated without using cookies; normalizing the aggregated audience distribution to generate a false positive correction matrix comprising a probability of attributing an impression of the media to the first demographic group when the database owner determines that the impression corresponds to a person in the second demographic group; and compensating for false positive errors in an impression by reallocating the impression from the second demographic group to the first demographic group using the false positive correction matrix, the impression indicating media accessed on a mobile device.

Some exemplary methods also include: generating a correction metric to correct the aggregate audience distribution for at least one of oversampling or undersampling associated with a survey calibration data source that indicates access to media by a first family member and a registration status of the first family member with a database owner. In some exemplary methods, the correction metric is based on a second probability that a first person in a first demographic group lives with a second person in a second demographic group. In some exemplary methods, generating the correction metric includes: determining a first number of first person pairs based on the first survey calibration data, representing the number of first person pairs that include a first person in the first demographic group and a second person in the second demographic group living together; determining a second number of second person pairs based on second survey calibration data, representing the number of second person pairs that include a third person in the first demographic group and a fourth person in the second demographic group living together, the second survey calibration data having a higher accuracy than the first survey calibration data; and determining a ratio of the first number to the second number.

In some exemplary methods, for each first demographic group, the aggregate audience distribution describes the number of persons in the second demographic group that would be classified as belonging to the first demographic group, and normalizing the aggregate audience distribution includes scaling the number of persons in the second demographic group so that the number of persons for the second demographic group is a specified value. Some exemplary methods also include: determining a sharing pattern for the first household, the sharing pattern indicating first members of the first household who access the media category and other members of the first household who do not access the media category; and determining a probability density function of device sharing probabilities for first members of the first household in the first household based on the sharing pattern for the media category, the probability density function indicating a corresponding probability that the first member of the household accesses media in the media category, the first audience distribution being based on the device sharing probabilities.

Some exemplary methods further include aggregating the device sharing probabilities for the first family members when the first household has two or more registered first family members of the first family members who are in the same second demographic group. Some exemplary methods further include generating a first audience distribution by distributing a portion of the numbers into the demographic groups, the portion of the numbers totaling a count of those first family members of the first family members that are registered with the database owner. In some exemplary methods, generating the first audience distribution does not use a cookie. In some exemplary methods, generating the first audience distribution and generating the aggregated audience distribution does not use metering software to collect the registration status of the family members with the database owner or to collect media access data. In some exemplary methods, reallocating impressions includes determining the product of an error correction matrix and an impression matrix, the impression matrix indicating the number of impressions corresponding to the respective demographic groups determined by the database owner. In some exemplary methods, reallocating the impressions results in a total number of impressions being the same as the total number of impressions in the impression matrix.

Some example methods also include providing instructions to an issuer, which are provided by the issuer to a computing device and, when executed by the computing device, cause the computing device to send a first request. Some example methods also include conserving computer processing resources by not communicating with individual online users about their online media access activities and not requesting survey responses from online users to generate an aggregate audience distribution. Some example methods also include conserving network communication bandwidth by not communicating with individual online users about their online media access activities and not requesting survey responses from online users to generate an aggregate audience distribution.

An additional disclosed exemplary apparatus includes an impression collector for receiving a first request from a first type of computing device at a first Internet domain and sending a request for demographic information, the demographic information corresponding to a plurality of requests received from the first type of computing device at the first Internet domain, the plurality of requests including the first request. In the disclosed exemplary apparatus, the first request indicates access to media at the computing device. The disclosed exemplary apparatus also includes an aggregated distribution generator for generating an aggregated audience distribution, the aggregated audience distribution including a first audience distribution for a first household aggregated with a second audience distribution for a second household, the first audience distribution including a distribution of first family members of the first household across demographic groups of those first family members registered with a database owner, the aggregated distribution generator generating the aggregated audience distribution without using web trackers, and the first audience distribution based on access to the first media by the first family member. The exemplary apparatus also includes a matrix normalizer configured to normalize the aggregated audience distribution to generate a misidentification correction matrix, the misidentification correction matrix including probabilities of attributing an impression of media to a first demographic group when a database owner determines that the impression corresponds to a person in a second demographic group. The exemplary apparatus also includes a misidentification corrector configured to compensate for misidentification errors in an impression by reallocating the impression to the first demographic group using the misidentification correction matrix, the impression indicating media accessed on a mobile device. At least one of the aggregated distribution generator, the matrix normalizer, or the misidentification corrector is implemented by logic circuitry.

Some exemplary devices also include a matrix corrector for generating a correction metric to correct the aggregate audience distribution for at least one of oversampling or undersampling associated with a survey calibration data source indicating access to media by a first family member and a registration status of the first family member with a database owner. In some exemplary devices, the correction metric is based on a second probability that a first person in a first demographic group lives with a second person in a second demographic group. In some exemplary devices, the matrix corrector is configured to generate the correction metric by: determining a first number of first person pairs based on first survey calibration data, i.e., the number of first person pairs that respectively include a first person in the first demographic group and a second person in the second demographic group living together; determining a second number of second person pairs based on second survey calibration data, i.e., the number of second person pairs that respectively include a third person in the first demographic group and a fourth person in the second demographic group living together, the second survey calibration data having a higher accuracy than the first survey calibration data; and determining a ratio of the first number to the second number.

In some exemplary apparatuses, for each first demographic group, an aggregate audience distribution describes the number of persons in a second demographic group that would be attributed to the first demographic group, and a matrix normalizer is used to normalize the aggregate audience distribution by scaling the number of persons in the second demographic group so that the number of persons for the second demographic group is a specified value.

Some exemplary devices also include a household distribution generator for: determining a sharing pattern for a first household, the sharing pattern indicating first household members among the first household members who access a media category and other first household members among the first household members who do not access the media category; and determining a probability density function of device sharing probability for first household members in the first household based on the sharing pattern for the media category, the probability density function indicating corresponding probabilities of first household members accessing media in the media category, the first audience distribution being based on the device sharing probability.

In some exemplary apparatuses, an aggregate distribution generator is configured to aggregate device sharing probabilities for first family members when the first household has two or more registered first family members of the first family members who are in the same second demographic group. Some exemplary apparatuses also include a household distribution generator configured to generate a first audience distribution by distributing portion numbers into demographic groups, the portion numbers totaling a count of those first family members of the first family members that are registered with a database owner. In some exemplary apparatuses, a false positive corrector is configured to reallocate impressions by determining the product of a false positive correction matrix and an impression matrix, the impression matrix indicating the number of impressions corresponding to the respective demographic groups as determined by the database owner. In some exemplary apparatuses, the false positive corrector is configured to reallocate impressions such that the total number of reallocated impressions is the same total number of impressions as the total number of impressions in the impression matrix.

An additional disclosed exemplary method includes: collecting media impressions from a first type of computing device; requesting demographic information for the media impressions from a database owner, a first portion of the media impressions corresponding to persons for whom the database owner stores demographic information, and a second portion of the media impressions corresponding to persons for whom the database owner does not have demographic information available; receiving demographic information corresponding to the first portion of the media impressions from the database owner; determining, using a processor, a number of media impressions in the second portion of the media impressions; and determining, using the processor, demographic information for the second portion of the media impressions based on a first probability of accessing media corresponding to the media impressions on the first type of computing device and based on a second probability of accessing the media on a second type of device.

An additional disclosed exemplary method includes: receiving a first request from a first type of computing device at a first Internet domain and receiving a second request from the computing device at the first Internet domain, the first request indicating access to media at the computing device, the second request indicating access to duration units of the media at the computing device. The exemplary method also includes: sending a request for demographic information corresponding to a plurality of requests received from the first type of computing device at the first Internet domain, the plurality of requests including the second request. The disclosed exemplary method also includes: obtaining a count of duration units that occurred on the first type of computing device, a first portion of the duration units corresponding to persons whose demographic information is identifiable by a database owner, and a second portion of the duration units corresponding to persons whose demographic information is not available to the database owner; and determining demographic information for the second portion of the duration units based on a first probability of accessing the media corresponding to the duration units on the first type of computing device and based on a second probability of accessing the media on the second type of computing device.

An additional disclosed exemplary apparatus includes an impression collector. In the disclosed exemplary apparatus, the impression collector is configured to receive a first request from a computing device of a first type at a first Internet domain, the first request indicating access to media at the computing device. In the disclosed exemplary apparatus, the impression collector is further configured to receive a second request from the computing device at the first Internet domain, the second request indicating access to duration units of media at the computing device. In the disclosed exemplary apparatus, the impression collector is further configured to send a request for demographic information corresponding to a plurality of requests received from the computing device of the first type at the first Internet domain, the plurality of requests including the second request. The disclosed exemplary apparatus also includes an impression information collector configured to access a count of duration units occurring on the computing device of the first type, a first portion of the duration units corresponding to persons for whom demographic information is identifiable by a database owner, and a second portion of the duration units corresponding to persons for whom demographic information is not available to the database owner. The disclosed exemplary apparatus also includes an uncovered corrector for determining demographic information for a second portion of the duration unit based on a first probability of accessing the media corresponding to the duration unit on a first type of computing device and based on a second probability of accessing the media on a second type of computing device, at least one of the impression information collector or the uncovered corrector being implemented by a logic circuit.

Although the examples disclosed herein are described with reference to compensating or adjusting impression information obtained from mobile devices, these examples are also applicable to non-mobile devices, such as desktop computers, televisions, video game consoles, set-top boxes, and/or other devices.

Impression and demographic information collection

FIG1 illustrates an exemplary system 100 for collecting user information (e.g., user information 102a, 102b) from distributed database owners 104a, 104b for use in association with an impression of media presented at a client device 106. In the illustrated example, user information 102a, 102b, or user data, includes one or more of the following: demographic data, purchase data, and/or other data indicating user activity, behavior, and/or preferences regarding information accessed via the Internet, purchases, media accessed on electronic devices, physical locations visited by users (e.g., retail or commercial establishments, restaurants, venues, etc.). The examples disclosed herein are described in conjunction with mobile devices, which may be mobile phones, mobile communication devices, tablet computers, gaming devices, portable media presentation devices, etc. However, the examples disclosed herein may be implemented in conjunction with non-mobile devices, such as Internet appliances, smart televisions, Internet terminals, computers, or any other device capable of presenting media received via network communications.

In the illustrated example of FIG1 , to track media impressions on a client device 106 , an audience measurement entity (AME) 108 cooperates or collaborates with an application publisher 110 to download and install a data collector 112 on the client device 106 . The application publisher 110 of the illustrated example can be a software application developer that develops and distributes applications to mobile devices, and/or a distributor that receives applications from a software application developer and distributes them to mobile devices. The data collector 112 can be included in other software loaded onto the client device 106 , such as an operating system 114 , an application (or app) 116 , a web browser 117 , and/or any other software. The exemplary client device 106 of FIG1 is a non-natively metered device. That is, the client device 106 does not support and/or is not provided with metering software (e.g., metering software provided by AME 108 ).

Any of the exemplary software 114-117 can present media 118 received from a media publisher 120. The media 118 can be an advertisement, video, audio, text, graphics, a web page, news, educational media, entertainment media, or any other type of media. In the illustrated example, a media ID 122 is provided in the media 118 to identify the media 118 so that when the media 118 is presented on the client device 106 or any other device monitored by the AME 108, the AME 108 can trust that the media 118 has a media impression.

The data collector 112 of the illustrated example includes instructions (e.g., Java, Java script, or any other computer language or script) that, when executed by the client device 106, causes the client device 106 to collect media IDs 122 of media 118 presented by the application 116 and/or the client device 106, and to collect one or more device/user identifiers 124 stored on the client device 106. The one or more device/user identifiers 124 of the illustrated example include identifiers that can be used by corresponding ones of the collaborating database owners 104a-b to identify one or more users of the client device 106 and to locate user information 102a-b corresponding to the one or more users. For example, the one or more device/user identifiers 124 may include software identifiers (e.g., International Mobile Equipment Identity (IMEI), Mobile Equipment Identifier (MEID), Media Access Control (MAC) address, etc.), application storage identifiers (e.g., Google Android ID, Apple ID, Amazon ID, etc.), Open Unique Device Identifier (OpenUDID), Open Device Identification Number (ODIN), login identifiers (e.g., username), email address, user agent data (e.g., application type, operating system, software vendor, software version, etc.), third-party service identifiers (e.g., advertising service identifier, device usage analysis service identifier, demographic profile collection service identifier), network storage data, Document Object Model (DOM) storage data, locally shared objects (also known as "Flash Cookies"), etc. In some examples, fewer or more device/user identifiers 124 may be used. Furthermore, although only two cooperating database owners 104a-b are shown in FIG1 , AME 108 may collaborate with any number of cooperating database owners to collect distributed user information (e.g., user information 102a-b).

In some examples, the client device 106 may not allow access to identifying information stored on the client device 106. For such instances, the disclosed examples enable the AME 108 to store an AME-provided identifier (e.g., an identifier managed and tracked by the AME 108) on the client device 106 to track media impressions on the client device 106. For example, the AME 108 may provide instructions in the data collector 112 to set the AME-provided identifier in a storage space accessible to and/or allocated to the application 116, and the data collector 112 uses the identifier as the device/user identifier 124. In such examples, the AME-provided identifier set by the data collector 112 persists in the storage space even when the application 116 and the data collector 112 are not running. In this manner, the same AME-provided identifier can remain associated with the client device 106 for an extended period of time. In some examples where the data collector 112 sets an identifier in the client device 106, the AME 108 may recruit the user of the client device 106 as a panelist and may store user information collected from the user during the panelist registration process and/or collected by monitoring user activity/behavior via the client device 106 and/or any other device used by the user and monitored by the AME 108. In this manner, the AME 108 may associate the user information of the user (from the panelist data stored by the AME 108) with media impressions attributed to the user on the client device 106.

In the illustrated example, the data collector 112 sends the media ID 122 and one or more device/user identifiers 124 to the application publisher 110 as collected data 126. Alternatively, the data collector 112 can be configured to send the collected data 126 to another collection entity (other than the application publisher 110) that has been contracted by or is partnering with AME 108 to collect media IDs (e.g., media ID 122) and device/user identifiers (e.g., one or more device/user identifiers 124) from mobile devices (e.g., client devices 106). In the illustrated example, the application publisher 110 (or collecting entity) sends the media ID 122 and one or more device/user identifiers 124 as impression data 130 to an impression collector 132 at AME 108. The impression data 130 of the illustrated example may include one media ID 122 and one or more device/user identifiers 124 for a single impression of the reported media 118, or the impression data 130 may include multiple media IDs 122 and one or more device/user identifiers 124 for multiple impressions of the reported media based on multiple instances of collected data (e.g., collected data 126) received from the client device 106 and/or other mobile devices.

In the illustrated example, the impression collector 132 stores the impression data 130 in an AME media impression storage 134 (e.g., a database or other data structure). Subsequently, the AME 108 sends the one or more device/user identifiers 124 to corresponding collaborative database owners (e.g., collaborative database owners 104a-b) to receive user information (e.g., user information 102a-b) corresponding to the one or more device/user identifiers 124 from the collaborative database owners 104a-b, so that the AME 108 can associate the user information with the corresponding media impression of the media (media 118) presented at the mobile device (e.g., client device 106).

In some examples, to protect the privacy of the user of the client device 106, the media identifier 122 and/or the one or more device/user identifiers 124 are encrypted before being sent to the AME 108 and/or the collaborating database owners 104a-b. In other examples, the media identifier 122 and/or the one or more device/user identifiers 124 are not encrypted.

After the AME 108 receives one or more device/user identifiers 124, the AME 108 sends a device/user identifier log 136a-b to a corresponding collaborative database owner (e.g., collaborative database owners 104a-b). In some examples, each of the device/user identifier logs 136a-b may include a single device/user identifier, or each of the device/user identifier logs 136a-b may include multiple aggregated device/user identifiers received over time from one or more mobile devices. After receiving the device/user identifier logs 136a-b, each of the collaborative database owners 104a-b looks up the user corresponding to the device/user identifier 124 in its respective log 136a-b. In this manner, each of the collaborative database owners 104a-b collects user information 102a-b corresponding to the user identified in the device/user identifier log 136a-b for transmission to the AME 108. For example, if the partner database owner 104a is a wireless service provider and the device/user identifier log 136a includes an IMEI code that can be recognized by the wireless service provider, the wireless service provider accesses its subscriber records to find a user with an IMEI code that matches the IMEI code received in the device/user identifier log 136a. When the user is identified, the wireless service provider copies the user's user information to the user information 102a for transmission to the AME 108.

In some other examples, data collector 112 is configured to collect one or more device/user identifiers 124 from client device 106. Example data collector 112 sends one or more device/user identifiers 124 to application publisher 110 in collected data 126, and example data collector 112 also sends one or more device/user identifiers 124 to media publisher 120. In such other examples, data collector 112 does not collect media ID 122 from media 118 at client device 106 because data collector 112 is not within example system 100 of FIG. 1 . Instead, media publisher 120 that publishes media 118 to client device 106 retrieves media ID 122 from the media 118 it publishes. The media publisher 120 then associates the media ID 122 with the one or more device/user identifiers 124 received from the data collector 112 executing in the client device 106, and sends the collected data 138 to the application publisher 110, the collected data 138 including the media ID 122 and the associated one or more device/user identifiers 124 of the client device 106. For example, when the media publisher 120 sends the media 118 to the client device 106, this is accomplished by using one or more of the one or more device/user identifiers 124 received from the client device 106 to identify the client device 106 as the target device for the media 118. In this manner, the media publisher 120 can associate the media ID 122 of the media 118 with the one or more device/user identifiers 124 of the client device 106, indicating that the media 118 is being sent to a specific client device 106 for presentation (e.g., to generate an impression of the media 118).

In some other examples where the data collector 112 is configured to send the one or more device/user identifiers 124 to the media publisher 120, the data collector 112 does not collect the media ID 122 from the media 118 at the client device 106. Instead, the media publisher 120 that issues the media 118 to the client device 106 also retrieves the media ID 122 from the media 118 it issues. The media publisher 120 then associates the media ID 122 with the one or more device/user identifiers 124 of the client device 106. The media publisher 120 then sends the media impression data 130 (including the media ID 122 and the one or more device/user identifiers 124) to the AME 108. For example, when the media publisher 120 sends the media 118 to the client device 106, it does so by identifying the client device 106 as the target device for the media 118 using one or more of the one or more device/user identifiers 124. In this manner, the media publisher 120 can associate the media ID 122 of the media 118 with one or more device/user identifiers 124 of the client device 106, indicating that the media 118 was sent to a particular client device 106 for presentation (e.g., to generate an impression of the media 118). In the illustrated example, after the AME 108 receives the impression data 130 from the media publisher 120, the AME 108 can then send the device/user identifier logs 136a-b to the collaborating database owners 104a-b to request the user information 102a-b described above in conjunction with FIG. 1 .

1 , application publisher 110 may implement at least some of the operations of media publisher 120 to send media 118 to client device 106 for presentation. For example, an advertisement provider, media provider, or other information provider may send media (media 118) to application publisher 110, which then distributes the media to client device 106 via, for example, application 116 when executed on client device 106. In such examples, application publisher 110 implements the operations described above as performed by media publisher 120.

Additionally or alternatively, compared to the above examples in which the client device 106 sends identifiers to the audience measurement entity 108 (e.g., via the application publisher 110, the media publisher 120, and/or other entities), in other examples, the client device 106 (e.g., the data collector 112 installed on the client device 106) sends identifiers (e.g., one or more device/user identifiers 124) directly to the corresponding database owners 104a, 104b (e.g., without the AME 108). In such examples, the exemplary client device 106 sends the media identifier 122 to the audience measurement entity 108 (e.g., directly or through an intermediary, such as via the application publisher 110), but does not send the media identifier 122 to the database owners 104a-b.

As described above, the exemplary collaborative database owners 104a-b provide user information 102a-b to the exemplary AME 108 for matching with the media identifier 122 to form media impression information. As also described above, the database owners 104a-b are not provided with a copy of the media identifier 122. Instead, the client provides the database owners 104a-b with an impression identifier 140. The impression identifier uniquely identifies the impression event relative to other impression events of the client device 106, so that the occurrence of the impression at the client device 106 can be distinguished from other occurrences of the impression. However, the impression identifier 140 itself does not identify the media associated with the impression event. In such examples, the impression data 130 from the client device 106 to the AME 108 also includes the impression identifier 140 and the corresponding media identifier 122. To match the user information 102a-b with the media identifier 122, the exemplary collaborative database owners 104a-b provide the user information 102a-b to the AME 108 associated with the impression identifier 140 that was used to trigger the impression event that collected the user information 102a-b. In this manner, the AME 108 can match the impression identifier 140 received from the client device 106 to the corresponding impression identifier 140 received from the collaborative database owners 104a-b to associate the media identifier 122 received from the client device 106 with the demographic information in the user information 102a-b received from the database owners 104a-b. The impression identifier 140 can additionally be used to reduce or avoid duplication of demographic information. For example, the exemplary collaborative database owner 104a-b may provide the user information 102a-b and the impression identifier 140 to the AME 108 on a per-impression basis (e.g., whenever a client device 106 sends a request including the encrypted identifier 208a-b and the impression identifier 140 to the collaborative database owner 104a-b) and/or on an aggregate basis (e.g., sending a set of user information 102a-b to the AME 108 presented at the client device 106, where the set of user information 102a-b may include indications of multiple impressions (e.g., multiple impression identifiers 140) at the mobile device 102a-b).

The impression identifier 140 provided to the AME 108 enables the AME 108 to distinguish unique impressions and avoid overcounting the number of unique users and/or devices viewing the media. For example, the relationship between the user information 102a from the Collaborative A database owner 104a and the user information 102b from the Collaborative B database owner 104b for the client device 106 may not be readily apparent to the AME 108. By including the impression identifier 140 (or any similar identifier), the example AME 108 can associate user information corresponding to the same user between the user information 102a-b based on matching impression identifiers 140 stored in both user information 102a-b. The example AME 108 can use such matching impression identifiers 140 on the user information 102a-b to avoid overcounting mobile devices and/or users (e.g., by only counting unique users rather than counting the same user multiple times).

For example, if an impression causes a client device 106 to send multiple device/user identifiers to multiple different database owners 104a-b without an impression identifier (e.g., impression identifier 140), the same user may be counted multiple times. For example, a first of the database owners 104a sends first user information 102a to AME 108, which signals that the impression occurred. Additionally, a second of the database owners 104b sends second user information 102b to AME 108, which signals (separately) that the impression occurred. Additionally, separately, client device 106 sends an indication of the impression to AME 108. Without knowing that user information 102a-b is from the same impression, AME 108 has an indication of a single impression from client device 106 and indications of multiple impressions from database owners 104a-b.

To avoid overcounting impressions, the AME 108 can use the impression identifier 140. For example, after looking up the user information 102a-b, the exemplary collaborating database owners 104a-b transmit the impression identifier 140 to the AME 108 having the corresponding user information 102a-b. The AME 108 matches the impression identifier 140 obtained directly from the client device 106 to the impression identifier 140 received from the database owner 104a-b having the user information 102a-b, thereby associating the user information 102a-b with the media identifier 122 and generating the impression information. This is possible because the AME 108 receives the media identifier 122 associated with the impression identifier 140 directly from the client device 106. Thus, the AME 108 can map user data from two or more database owners 104a-b to the same media contact event, thereby avoiding double counting.

Each unique impression identifier 140 in the illustrated example is associated with a particular impression of media on the client device 106. The collaborative database owners 104a-b receive the respective device/user identifiers 124 and generate the user information 102a-b independently (e.g., without regard to the other of the collaborative database owners 104a-b) and without knowledge of the media identifiers 122 involved in the impressions. Without indicating that particular user demographics in user information 102a (received from collaborative database owner 104a) are associated with (e.g., are a result of) the same impression at client device 106 as particular user demographics in user information 102b (received from collaborative database owner 104b independently of user information 102a being received from collaborative database owner 104a), and without reference to impression identifier 140, AME 108 may be unable to associate user information 102a with user information 102b and/or determine that different pieces of user information 102a-b are associated with the same impression, and therefore, would count user information 102a and user information 102b as corresponding to two different users/devices and/or two different impressions.

The above examples illustrate methods and apparatus for collecting data at an audience measurement entity (or other entity). The examples discussed above can be used to collect impression information for any type of media, including static media (e.g., advertising images), streaming media (e.g., streaming video and/or audio, including content, advertisements, and/or other types of media), and/or other types of media. For static media (e.g., media without a temporal component, such as images, text, web pages, etc.), the exemplary AME 108 records an impression for each occurrence of the media being presented to, transmitted to, or otherwise provided to the client device 106. For streaming media (e.g., video, audio, etc.), the exemplary AME 108 measures demographic information for media that appears over a period of time. For example, when the media is located in a client application/software 114-117, the AME 108 (e.g., with the aid of an application publisher 110 and/or a media publisher 120) provides beacon instructions to the client application or client software (e.g., OS 114, web browser 117, application 116, etc.) executing on the client device 106. In some examples, the beacon instructions cause the client applications/software 114-117 to send requests (e.g., broadcast messages) to the impression collector 132 at regular intervals and/or irregular intervals (e.g., every minute, every 30 seconds, every 2 minutes, etc.). By monitoring and/or counting the requests that occur at intervals, the exemplary AME 108 monitors the duration of individual impressions of duration-based media (e.g., video, audio, etc.). The exemplary AME 108 can determine the number of impressions (e.g., initial loads) of duration-based media, the unique audience for duration-based media, and/or the total duration (in units such as seconds or minutes) of duration-based media viewed across multiple impressions. As used herein, the term "impression information" may include impressions and/or duration units. The exemplary impression collector 132 identifies requests from the web browser 117 and, in conjunction with one or more database owners, matches the impression information for the media to the demographics of the user of the web browser 117.

In some examples, a user loads (e.g., via browser 117) a web page from a website publisher, where the web page corresponds to a particular 60-minute video. As part of the exemplary web page or in addition to the exemplary web page, the website publisher causes data collector 112 to send a broadcast message (e.g., a beacon request) to beacon server 142, for example, by providing beacon instructions to browser 117. For example, when the beacon instructions are executed by exemplary browser 117, the beacon instructions cause data collector 112 to send a broadcast message (e.g., a beacon request, an HTTP request, a sound pulse) to impression collector 132 at a specified interval (e.g., every minute or any other suitable interval). The exemplary beacon instructions (or, for example, a redirect message from impression collector 132 or database owners 104a-b) also cause data collector 112 to send the broadcast message or beacon request to one or more database owners 104a-b, which collect and/or maintain demographic information about the user. The database owners 104a-b transmit demographic information about the users associated with the data collector 112 for combination or association with impressions determined by the impression collector 132. If the user closes the webpage containing the video before the video ends, the beaconing instruction is stopped and the data collector 112 stops sending broadcast messages to the impression collector 132. In some examples, the broadcast messages include timestamps and/or other information indicating the locations in the video to which the multiple broadcast messages correspond. By determining the number and/or content of broadcast messages received at the impression collector 132 from the client device 106, the example impression collector 132 can determine that the user watched a particular length of the video (e.g., the portion of the video for which the broadcast message was received at the impression collector 132).

The client device 106 of the illustrated example executes client applications/software 114-117 directed to a host website (e.g., www.acme.com), from which media 118 (e.g., audio, video, interactive media, streaming media, etc.) is obtained for presentation via the client device 106. In the illustrated example, the media 118 (e.g., advertisements and/or content) is tagged with identifier information (e.g., media ID 122, creation type ID, layout ID, publisher source URL, etc.) and beacon instructions. The exemplary beacon instructions cause the client applications/software 114-117 to request additional beacon instructions from a beacon server 142, which instructs the client applications/software 114-117 on how and where to send beacon requests to report impressions of the media 118. For example, the exemplary client applications/software 114-117 send a request including an identification of the media 118 (e.g., media identifier 122) to the beacon server 142. The beacon server 142 then generates beacon instructions 144 and returns them to the exemplary client device 106. Although beacon server 142 and impression collector 132 are shown separately, in some examples, beacon server 142 and impression collector 132 are combined. In the illustrated example, beacon instructions 144 include the URL of one or more database owners (e.g., one or more of collaborating database owners 104a-b) or any other server to which client device 106 should send a beacon request (e.g., an impression request). In some examples, the broadcast message or beacon request can be implemented as an HTTP request. However, while the HTTP request sent identifies a web page or other resource to be downloaded, the broadcast message or beacon request includes audience measurement information (e.g., an ad campaign identifier, a content identifier, and/or device/user identification information) as its payload. The server to which the broadcast message or beacon request is directed is programmed to record the audience measurement data of the broadcast message or beacon request as an impression (e.g., an ad and/or content impression depending on the nature of the media tagged with the beacon instruction). In some examples, the beacon instructions received with the tagged media 118 include beacon instructions 144. In such examples, the client applications/software 114 - 117 do not need to request the beaconing instructions 144 from the beaconing server 142 because the beaconing instructions 144 are already provided in the tagged media 118 .

When beaconing instructions 144 are executed by client device 106, beaconing instructions 144 cause client device 106 to send beaconing requests (e.g., repeatedly at specified intervals) to a remote server (e.g., impression collector 132, media publisher 120, database owners 104a-b, or other server) specified in beaconing instructions 144. In the illustrated example, the specified server is a server of audience measurement entity 108, i.e., at impression collector 132. Beaconing instructions 144 may be implemented using Java script or any other type of instructions or script (e.g., including Java, HTML, etc.) that can be executed by a client application (e.g., a web browser).

Examples of systems that may be used to implement FIG. 1 are disclosed in U.S. patent application serial number 14/127,414, filed on August 28, 2013; U.S. patent application serial number 14/261,085, filed on April 24, 2014; U.S. provisional patent application serial number 61/952,726, filed on March 13, 2014; U.S. provisional patent application serial number 61/979,391, filed on April 14, 2014; U.S. provisional patent application serial number 61/986,784, filed on April 30, 2014; U.S. provisional patent application serial number 61/991,286, filed on May 9, 2014; and U.S. provisional patent application serial number 62/014,659, filed on June 19, 2014. The entire contents of U.S. patent application serial number 14/127,414, U.S. patent application serial number 14/261,085, U.S. provisional patent application serial number 61/952,726, U.S. provisional patent application serial number 61/979,391, U.S. provisional patent application serial number 61/986,784, U.S. provisional patent application serial number 61/991,286, and U.S. provisional patent application serial number 62/014,659 are incorporated herein by reference.

The examples of Figures 2-11 can be used to compensate for misidentification and/or non-coverage errors in impression information collected from a client device through which a user accesses media. Such impression information can be collected using any suitable technique, including the exemplary techniques discussed above. For example, the impression information collected from database owners 104a-b can be aggregate impression information that describes a number of impressions for a media item of interest (e.g., an advertisement, streaming media, website, etc.), a number of duration units (e.g., minutes, seconds, etc.) during which the duration-based media was presented, and/or a count of audience members corresponding to the impressions. The aggregate impression information obtained from database owners 104a-b may be subject to misidentification errors (e.g., errors caused by a database owner mistakenly associating an impression with a first person in a first demographic group when it should be associated with a second person in a second demographic group) and/or non-coverage errors (e.g., errors caused by a database owner being unable to associate an impression with a person). Because impressions and/or duration units that cannot be associated with demographic information by database owners 104a-b may not be included in the aggregated demographic information, in some disclosed examples, non-coverage errors in the aggregated demographic information can be detected using, for example, counting impressions at AME 108 and comparing the counted impressions to a number of impressions for which database owners 104a-b identified demographic information. In some other examples, database owners 104a-b return a number of impressions and/or duration units for which database owners 104a-b were unable to determine demographic information. The number of impressions and/or duration units for which database owners 104a-b were unable to determine demographic information can be used as the number of non-coverage impressions.

Examples disclosed herein use survey calibration data to estimate respective errors and generate compensated impression information that is adjusted to correct for misidentification errors and/or non-coverage errors. Examples disclosed herein can be used for impressions and/or duration units collected from client devices (e.g., mobile devices and/or non-mobile devices), can be applied to impressions and/or duration units collected only from mobile devices, can be applied to impressions and/or duration units collected from mobile devices (separate from impressions and/or duration units collected from non-mobile devices), and/or can be applied to impressions and/or duration units collected from a first type of mobile device (separate from impressions collected from a second type of mobile device). In some examples, compensated impression information is presented or reported separately for mobile devices and non-mobile devices, and/or reported as aggregated data corresponding to both mobile devices and non-mobile devices.

The examples disclosed herein can be applied to input data in real time or substantially real time (e.g., within seconds or minutes of receiving the data), and can be used to compensate impression information (e.g., impressions, duration units) over any desired time period (e.g., hourly, daily, weekly, monthly, etc.) and/or cumulatively (e.g., applied to impressions and/or duration units collected over multiple time periods). Thus, the examples disclosed herein can provide advertisers and/or media publishers with accurate demographic information to enable faster adjustments to media campaign strategies to align with measured demographics than known methods.

FIG2 is a block diagram of an exemplary impression data compensator 200 that can be used to implement the exemplary impression collector 132 of FIG1 to compensate impression information for misidentification and/or non-coverage errors. The exemplary impression data compensator 200 of FIG2 compensates or adjusts impression information obtained from a client device (e.g., the client device 106 of FIG1) and/or from database owners 104a-b to reduce (e.g., avoid) errors such as those mentioned above.

The exemplary impression data compensator 200 of FIG. 2 includes a calibration data collector 202 , a shared matrix generator 204 , a false positive corrector 206 , an impression information collector 208 , a non-coverage calculator 210 , a non-coverage corrector 212 , and an impression information adjuster 214 .

The example calibration data collector 202 of FIG. 2 collects or obtains survey calibration data that describes characteristics of mobile device usage by an audience. For example, the survey calibration data may include and/or be based on responses to a survey of randomly selected households. In some examples, the calibration survey obtains information including the number of people in the household, demographic characteristics of the household (e.g., age and gender, race, ethnicity, language characteristics, household income, geographic location, etc.), the number and/or type of mobile devices (e.g., smartphones, tablets, portable media players, etc.) present in the household, and/or the registrations of the people in the household with a designated database owner (e.g., partner database owners 104a-b of FIG. 1 ). In some examples, for each person in the household, the calibration survey obtains usage characteristics of each mobile device and/or the type of mobile devices present in the household; the media categories typically viewed by the person; the media categories typically viewed by the person on each mobile device and/or the type of mobile devices in the household; the applications used by the person on each mobile device and/or the type of mobile devices in the household; and/or interaction characteristics on each mobile device with the designated database owner and/or the type of mobile devices in the household. The example calibration data collector 202 obtains survey calibration data from at least a threshold number of households and, if appropriate, weights the results to reflect a general population or audience.

In some other examples, the survey calibration data source includes a survey of one or more established respondent panels, such as the Nielsen National People's Monitor (NPM) Panel used for ratings. Established panel surveys often provide higher quality survey calibration data. In some examples, data from multiple surveys are used to calculate different compensation factors and/or are combined to calculate the compensation factor.

Error identification correction

The example sharing matrix generator 204 of Figure 2 calculates a device sharing matrix based on the survey calibration data. In the example of Figure 2, the sharing matrix generator 204 creates a separate false positive correction matrix for each combination of device type and media category represented in the survey calibration data.

To generate the misidentification correction matrix for the device types and media categories of interest, the example shared matrix generator 204 includes a household distribution generator 216 , an aggregate distribution generator 218 , a matrix corrector 220 , and a matrix normalizer 222 .

The exemplary household distribution generator 216 of FIG. 2 generates an audience distribution for a household based on the survey calibration data. For example, the household distribution generator 216 determines the corresponding likelihood that the person represented in the survey calibration data uses the device type of interest to view media of the media type of interest. To illustrate, consider the following example. The exemplary household from which the survey calibration data was collected includes four members: 1) a male aged 45-54, 2) a female aged 35-44, 3) a female aged 18-24, and 4) a male aged 12-17. The female aged 18-24 and the male aged 12-17 have registered with the exemplary database owner 104a of FIG. 1 (e.g., Facebook) (e.g., can be identified as registered users by the exemplary database owner 104a of FIG. 1) and access the database owner 104a using a tablet (e.g., the client device 106 of FIG. 1) (although not simultaneously). The male aged 45-54 and the female aged 35-44 are not identifiable by the database owner 104a on the tablet. Table 1 below shows exemplary sharing patterns for tablets by media category. In Table 1, cells marked with an "X" indicate that the person identified at the demographic group label of the cell watches media having the category identified in the content type label. Conversely, a blank cell in Table 1 indicates that the person identified at the demographic group label of the cell does not watch media having the category identified in the content type label. Media categories can be defined based on, for example, media categories used in survey calibration data and/or media categories used in television and/or other media evaluations.

Example sharing patterns for tablets by media category based on survey calibration data in an example first household

Table 1

As shown in Table 1, males aged 45-54 use tablets to view media categorized as political media (e.g., websites, streaming media, etc.), females aged 35-44 use tablets to view media categorized as drama, comedy, and/or reality (e.g., websites, streaming media, etc.), and females aged 18-24 use tablets to view media categorized as drama and comedy (e.g., websites, streaming media, etc.). When males aged 12-17 use tablets to log in to database owner 104a, they do not view media monitored by audience measurement entity 108 on their tablets. Based on the sharing patterns in Table 1, the exemplary household distribution generator 216 calculates device sharing probabilities for each media category in Table 1, as shown in Table 2 below. Table 2 shows the device sharing probabilities as probability density functions (PDFs) of the content types (e.g., media categories) viewed on devices by people identified in the demographic group annotations.

Example device sharing probabilities by media category for a first example household

Table 2

In this example, if a male aged 12-17 logs into the database owner on a tablet (via a browser and/or an application) and does not log out of the database owner, and a female aged 35-44 subsequently uses the same tablet (while the male aged 12-17 is still logged into database owner 104a) and does not log in to database owner 104a with her own credentials, when the female aged 35-44 views media on the tablet, database owner 104a will misattribute impressions and/or duration units that were correctly attributed to the female aged 35-44 to the male aged 12-17. Thus, in such examples, using database owner information to associate impressions and/or duration units with demographic information results in impressions and/or duration units being not attributed (or underattributed) to the male aged 45-54 and the female aged 35-44, and impressions and/or duration units being overattributed to the female aged 18-24 and/or the male aged 12-17.

To determine the misidentification correction matrix for households for the tablet and 'comedy' categories, the example household distribution generator 216 converts the 'comedy' probabilities in Table 2 above into the example redistributed audience matrix shown in Table 3 below. In Table 3, the columns (Identified Demographic Group i) represent the demographic groups identified by the database owner 104a as being associated with an impression, and the rows (Actual Viewer Demographic Group j) represent the demographic groups that actually viewed (e.g., actual viewers) the media corresponding to the impression. Thus, Table 3 includes a PDF for the event that, when the database owner identifies a person as being in Identified Demographic Group i, the actual or real viewer is a person in Actual Viewer Demographic Group j. The value in each cell is the probability γ _{ij that} , when the database owner 104a associates an impression of media with Identified Demographic Group i in the column, Actual Viewer Demographic Group j in that row is viewing the media.

j\i M45-54 F35-44 M12-17 F18-24 M45-54 0 0 0 0 F35-44 0 0 0.5 0.5 M12-17 0 0 0 0 F18-24 0 0 0.5 0.5 total 0 0 1 1

Example Redistribution Audience Matrix for the "Comedy" Media Category for a First Example Household

Table 3

As an example of determining impressions from the redistributed audience matrix for a household of Table 3, for 10 impressions in the ‘comedy’ media category identified by the database owner 104a as viewed by 12-17 year old males in the identified demographic group i, 5 impressions should be attributed to 35-44 year old females in the actual viewer demographic group j (e.g., 10 impressions multiplied by the probability of 0.5 in Table 3), 5 impressions should be attributed to 18-24 year old females in the actual viewer demographic group j (e.g., 10 impressions multiplied by the probability of 0.5 in Table 3), and no impressions should be attributed to 12-17 year old males or 45-54 year old males in the actual viewer demographic group j (e.g., 10 impressions multiplied by the probability of 0 in Table 3).

The exemplary household distribution generator 216 may use the PDFs in Table 2 and/or Table 3, wherein the PDFs have different probabilities labeled for different demographic groups (Table 2) and/or different probabilities for different actual viewer demographic groups (Table 3). For example, when one member of a household views a media category of significant interest more frequently than another member of the household, the different probabilities in the PDFs may be determined from the survey calibration data. For example, if females aged 35-44 from Table 2 report viewing media in the category of 'comedy' as 'frequently', while females aged 18-24 from Table 2 report viewing media in the category of 'comedy' as 'rarely', the PDF for the 'comedy' category in Table 2 may be (0, 0.75, 0, 0.25) to reflect the different viewing frequencies. Additionally or alternatively, the exemplary household distribution generator 216 may use PDFs with different probabilities in the exemplary sharing matrix of Table 2 based on the presence of multiple persons in the same demographic group. For example, a family with two females aged 12-17 and one female aged 35-44 may have a PDF where the probability of the F12-17 demographic group is twice the probability of the F35-44 demographic group.

The example aggregate distribution generator 218 of FIG. 2 generates an aggregate redistributed audience matrix for each combination of device type and media category based on all households in the example survey calibration data.

In some examples, household distribution generator 216 generates individual device share matrices from the survey responses of individual households, and aggregate distribution generator 218 aggregates the individual redistributed audience matrices into an aggregated redistributed audience matrix. For example, household distribution generator 216 may redistribute individuals within identified demographic group i that are identifiable by database owner 104a within households. Thus, household distribution generator 216 also redistributes impressions associated with those individuals to actual viewer demographic group j based on the survey calibration data.

In another example of a redistributed audience matrix, Table 4 below shows exemplary device sharing probabilities by media category for a household having two females of the same identified demographic group i (e.g., identified by database owner 104a as viewers of the media to which the impression corresponds), shown as females aged 18-24 (F18-24). In the illustrated example, the two females in the F18-24 demographic group are identifiable registered users of database owner 104a belonging to the F18-24 demographic group.

Content Type M45-54 F35-44 F18-24 F18-24 all 0.33 0.33 0 0.33 politics 1 0 0 0 drama 0 0.5 0 0.5 comedy 0 0.5 0 0.5 Reality 0 1 0 0

Example Device Sharing Probabilities by Media Category for a Second Example Household

Table 4

The cells of Table 4 above include a probability density function (PDF) indicating the probability of sharing a media device between persons belonging to a specified identified demographic group i (e.g., one person in the M45-54 demographic group, one person in the F35-44 demographic group, and two persons in the F18-24 demographic group) for a specified media category (e.g., all, politics, drama, comedy, reality). For example, for each of a person in demographic group i identified by M45-54, a person in demographic group i identified by F35-44, and one of persons in demographic group i identified by F18-24, the PDF for viewing media in the "all" media category is 0.33. In the illustrated example, for another person in demographic group i identified by F18-24, the PDF is 0. The data in Table 4 is based on a survey-calibrated data source (e.g., a survey of randomly selected persons and/or households) that provides information about the media viewed by persons in a household.

In the illustrated example, two women in demographic group i, identified by F18-24 of the exemplary household represented in Table 4 above, are registered users of database owner 104a (e.g., a social networking service). Based on the data in Table 4 above, household distribution generator 216 redistributes the audience (and therefore associated impressions) of each registered database owner user (e.g., viewers in the F18-24 demographic group) among the actual viewer demographic groups j, M45-54, F35-44, and F18-24, who watch the media of interest category in the household based on the device sharing patterns and viewing patterns collected from the households in the exemplary survey calibration data source. An exemplary redistributed audience for the "all" media category and the second exemplary household of Table 4 are shown in Table 5 below.

j\i M45-54 F35-44 F18-24 F18-24 M45-54 0 0 0.33 0.33 F35-44 0 0 0.33 0.33 F18-24 0 0 0 0 F18-24 0 0 0.33 0.33 total 0 0 1 1

Example Redistribution Audience Matrix for the "All" Category

Table 5

In Table 5 above, the household distribution generator 216 applies the same PDF (e.g., 0.33) to each registered database owner user in the same identified demographic group i (e.g., two family members of demographic groups F18-24). In the illustrated example of Table 5, the cell value indicates the corresponding probability that when a person in the household is identified by the database owner as a person in identified demographic group i (i.e., the person and/or demographic group indicated in the column), the actual or real viewer is a person in actual viewer demographic group j (i.e., the person and/or demographic group indicated in the row). For example, the probability is 0.33 that when the database owner identifies the first person in demographic group i identified by F18-24, the real viewer of the media in the "All" category is a person in actual viewer demographic group j M45-54. In this example, the probability of the following events is the same (e.g., 0.33): when the database owner identifies a second person in demographic group i identified by F18-24, the actual or real viewer of the media in the "All" category is a person in actual viewer demographic group j by M45-54.

In the illustrated example, it does not matter whether the female viewers in demographic group i identified by F18-24 indicate that they watch a particular media category represented by the exemplary redistributed audience matrix of Table 5 on a particular device. As long as the database owner 104a captures impressions and/or duration units for that particular device, the household distribution generator 216 can redistribute the impressions and/or duration units equally among all actual viewer demographic groups j, M45-54, F35-44, and F18-24, for that household. Thus, the two columns labeled F18-24 (e.g., corresponding to the two daughters of the household) are filled with equal redistributions of 0.33 (for M45-54 actual viewer demographic group j), 0.33 (for F35-44 actual viewer demographic group j), and 0.33 (for F18-24 actual viewer demographic group j). The probability values in the same row for actual viewer demographic group j, F18-24, are however summed across columns F18-24 because these values correspond to the same identified demographic group i, F18-24. The cells in the "Total" row indicate the number of audience members within the corresponding identified demographic group i (M45-54, F35-44, F18-24) who are registered users of the database owner 104a.

Table 6 below shows an exemplary redistributed audience matrix of type "all," where the summed values for demographic group i identified by F18-24 are shown.

j\i M45-54 F35-44 F18-24 M45-54 0 0 0.66 F35-44 0 0 0.66 M12-17 0 0 0 F18-24 0 0 0.66 total 0 0 2

Example Redistributed Audience Matrix for the "All" Media Category for a Household with Two Persons in the F18-24 Demographic Group

Table 6

In the exemplary upper Table 6, each column (identified demographic group i M45-54, F35-44, and F18-24) corresponds to the total number of registered users of the database owner 104a in the identified demographic group i in the household. The cells in the "Total" row indicate the number of audience members within the corresponding identified demographic group i (e.g., M45-54, F35-44, F18-24) who are registered users of the database owner 104a. In some examples, the actual viewer demographic group j (e.g., row) of the redistributed audience matrix of Table 6 is expanded to include all identified demographic groups i used by the audience measurement entity 108 and/or database owner 104a to achieve aggregation of the matrix. Thus, although the household represented by the upper Table 6 does not have any family members in the actual viewer demographic group M12-17, the exemplary upper Table 6 includes the M12-17 rows to achieve aggregation of the PDF of Table 6 with the PDF of Table 7 below.

After the device sharing probabilities shown in Table 4 are used to generate the redistributed audience matrix for the second household described above in conjunction with Tables 5 and 6, the redistributed audience matrices of Tables 6 and 7 are aggregated (e.g., summed) across households to combine registered users of database owner 104a in identified demographic group i and sum the redistributed audiences for actual viewer demographic group j. For example, the redistributed audiences of Tables 6 and 7 above are aggregated to generate the aggregated redistributed audience across households shown in Table 8 below. In the illustrated example, Tables 6 and 7 correspond to two different households.

j\i M45-54 F35-44 M12-17 F18-24 M45-54 0 0 0.33 0.33 F35-44 0 0 0.33 0.33 M12-17 0 0 0 0 F18-24 0 0 0.33 0.33 total 0 0 1 1

Example Redistributed Audience Matrix for the "All" Media Category for a Household with One M12-17 Person and One F18-24 Person

Table 7

Example Aggregate Redistribution Audience Matrix for "All" Media Categories for Multiple Households

Table 8

The probabilities in Table 8 above reflect the redistributed audience for registered users of database owner 104a within the two households represented in the corresponding tables in Tables 6 and 7. The sum of the columns for demographic group i identified by M12-17 and F18-24 reflects the total number of registered users of database owner 104a within the identified demographic group i. The cells in the "Total" row indicate the number of audience members who are registered users of database owner 104a within the corresponding identified demographic group i (M45-54, F35-44, M12-17, F18-24).

In some examples, the matrix corrector 220 generates a Nielsen National People Meter (NPM) metric to account for the probability of demographic pairs i, j (e.g., a person from identified demographic group i and a person from actual viewer demographic group j) living together in the same household. For example, P(L) _ij is the probability that the first person in identified demographic group i and a person from actual viewer demographic group j live in the same household. In the illustrated example, the NPM metric is another source of calibration data. In the examples disclosed herein, NPM data is collected from a household where a meter installed by AME 108 ( FIG. 1 ) collects identification of household members as they view/listen to media via media devices associated with the household. Because the NPM data is collected in the household using a locally installed meter that facilitates household members identifying themselves (or otherwise accurately collects identification of household members viewing/listening to media), the NPM data has a high degree of accuracy in identifying which household member is actually viewing the media presented by the media device (e.g., is the actual viewer of the media).

The NPM metrics described below can be used to interpret the P(L) _ij probabilities. In these examples, the matrix corrector 220 applies the NPM metrics to the corresponding i, j cells in the table of registered users of the redistribution database owner (e.g., the exemplary redistribution audience matrix for the "all" category shown in Table 5 above). Table 9 below shows estimates for demographic groups i, j pairs of people living together based on survey responses collected from households (in these examples, this also serves as a survey calibration data source from which the redistribution audience table is determined). Table 10 below shows estimates for demographic groups i, j pairs of people living together based on the collected NPM data. Table 11 below shows the NPM metrics generated by the matrix corrector 220 based on the estimates in Tables 9 and 10.

j\i M45-54 F35-44 M12-17 F18-24 M45-54 100 60 10 55 F35-44 30 100 10 80 M12-17 20 50 80 15 F18-24 50 20 20 90

Exemplary estimators for demographic groups i, j pairs of people living together based on survey responses

Table 9

j\i M45-54 F35-44 M12-17 F18-24 M45-54 105 50 15 48 F35-44 35 102 12 80 M12-17 25 40 70 20 F18-24 40 15 18 98

Exemplary estimators for demographic groups i, j pairs of people living together based on NPM data

Table 10

j\i M45-54 F35-44 M12-17 F18-24 M45-54 1.05 0.83 1.50 0.87 F35-44 1.17 1.02 1.20 1.00 M12-17 1.25 0.80 0.88 1.33 F18-24 0.80 0.75 0.90 1.09

Example NPM Metrics

Table 11

The exemplary NPM metric of Table 11 above is calculated by dividing the value of Table 10 (the estimated number of people living together for demographic group i, j pairs based on panelist data) by the corresponding value of Table 9 (the estimated number of people living together for demographic group i, j pairs based on survey responses). In the illustrated example, Table 11 is used to account for oversampling/undersampling of any viewers from different demographic groups i, j living together in the same household. For example, Table 9 estimates that 100 people in demographic group i, j for M45-54 live together based on survey responses. Based on more accurate NPM data, Table 10 estimates that 105 people in demographic group i, j for the same M45-54 live together in the same household. Therefore, the exemplary NPM metric of Table 11 for this demographic group M45-54 is 1.05, which is greater than 1 to compensate for the fact that Table 9 undersamples the number of people living together from demographic group i, j for M45-54 (e.g., 100) relative to the corresponding value of 105 in Table 10. For instances where the estimates in Table 9 are oversampled relative to the corresponding amounts in Table 10, the corresponding NPM metric in Table 11 will be less than 1 (e.g., for actual viewer demographic group j of F18-24 and identified demographic group i of F35-44, NPM metric in Table 11 = 0.75).

In some examples, the estimates for some demographic group pairs i, j living together from Table 9 (e.g., determined based on survey responses) align with the estimates based on NPM data from Table 10 because NPM data is a higher quality data source than the demographic data from database owner 104a. Tables 12 and 13 illustrate exemplary data in which the estimates for demographic group pairs i, j of people living together are aligned closer to the NPM data. In Table 12, matrix corrector 220 adjusts the redistributed audience values from Table 8 above for the probability P(L) _ij that the first person in identified demographic group i lives in the same household as a person in actual viewer demographic group j. In the illustrated example, matrix corrector 220 determines each i, j cell value in Table 12 by multiplying the redistributed audience value in the i, j cell of Table 8 above by the corresponding NPM indicator in the i, j cell value of Table 11 above. In this way, the sharing matrix corrector 220 applies the NPM metric to the redistributed impressions collected by the database owner 104a across different households to account for the probability P(L) _ij that the first person in the identified demographic group i lives in the same household as a person in the actual viewer demographic group j.

Example database owner redistribution audience adjusted for P(L) _ij

Table 12

j\i M45-54 F35-44 M12-17 F18-24 M45-54 0 0 0.417 0.295 F35-44 0 0 0.333 0.338 M12-17 0 0 0 0 F18-24 0 0 0.250 0.368 total 0 0 1 1

Example Normalized Redistributed Audience

Table 13

In Table 13 above, the matrix normalizer 222 normalizes the adjusted impression redistribution from Table 12 above so that the columns for identified demographic group i (e.g., M12-17 and F18-24 identified demographic group i) sum to 1. In this manner, each column (e.g., identified demographic group i M12-17 and F18-24) represents a probability density function (PDF) of what the AME 108 determines to be the actual viewer demographic group j of the real viewers of the media when the database owner 104a detects a particular identified demographic group i. In the illustrated example, the normalized redistributed audience of Table 13 is the misidentification correction factor, or the probability γ _ij for each pair of demographic groups i, j, which is the probability that when a person in identified demographic group i is identified as a media viewer by the database owner 104a, a person in actual viewer demographic group j is the actual viewer (e.g., γ _ij = _0.417 , 0.295, 0.333, 0.338, 0.250, and 0.368).

In some examples, the probability γ _ij for each pair of identified demographic group i detected by the database owner and actual viewer demographic group j assigned as a real or actual viewer can be weighted and/or averaged across all individual household matrices to determine an aggregate probability. In the illustrated example, a real viewer or actual viewer refers to a person in a particular demographic group who is considered an actual audience member who engages with (e.g., watches, listens to, consumes, etc.) media on a device. For example, an actual viewer can be a viewer determined by AME 108 to be a person who actually views or engages with the monitored media. The determination that a person is an actual viewer can be based on statistical probability, which indicates the likelihood of an actual viewer based on responses to a randomly selected person and/or household survey discussed above. The determination of an actual viewer as perceived by AME 108 can also be based on observational or other collected data (e.g., NPM data) representing actual viewers in a household. In any case, an actual viewer is a strong inference by AME 108 as to who is an actual viewer, but as used herein, an actual viewer is not necessarily an absolute determination of the identity of an actual viewer. However, the inference strength of actual viewers as used in conjunction with the examples disclosed herein is sufficient to accurately be used in conjunction with the examples disclosed herein to provide corrected impression and/or duration units with a high degree of accuracy. In some examples, the sharing matrix generator 204 generates device sharing matrices based on additional and/or alternative distinctions, such as different demographic markets, different stations, and/or different time periods.

As an alternative to the exemplary method for calculating _γij using the redistributed audience described above, in some examples, the aggregated distribution generator 218 calculates the aggregated redistributed audience matrix by calculating compound probabilities, as shown in Equation 1:

γ _ij ＝P(L) _ij ×P(D|L) _ij ×P(S _x |D) _ij (Equation 1)

In Equation 1 above, P(L) _ij is the probability that the first person in identified demographic group i lives in the same household as a person in actual viewer demographic group j; P(D|L) _ij is the probability that the first person in identified demographic group i and a person in actual viewer demographic group j have access to a mobile device (of any type) in the same household, assuming that the two people live in the same household; and P(S _x |D) _ij is the probability that the first person in identified demographic group i has access to the same selected type of mobile device as a person in actual viewer demographic group j, and the two people share the mobile device for media of a selected category. As used herein, "selected category" refers to the category of interest that is subject to analysis. Therefore, "selected" means "selected for analysis," as used in this context. The same meaning applies to "selected device type,""selected demographic group," and "pair of selected demographic groups."

In some examples, the aggregate distribution generator 218 uses data from one or more calibration data sources (e.g., survey calibration data) to determine one or more of the probabilities. For example, in some examples, the shared matrix calculator determines the probabilities P(L) _ij and P(D|L) _ij from a survey of a panel, such as the panel used to determine the NPM indicator data discussed above, and P(S _x |D) _ij from another survey of random households.

Referring to the examples described above with reference to Tables 1-13, the exemplary aggregate distribution generator 218 uses the redistributed aggregate audience matrix to calculate, for each pair of demographic groups i, j, the probability γ _ij that (1) a person in the identified demographic group i is identified by the database owner 104a as a viewer of the media, and (2) a person in the actual viewer demographic group j is an actual viewer. The exemplary aggregate distribution generator 218 generates a false positive correction matrix that includes each of the calculated probabilities γ _ij . The exemplary aggregate distribution generator 218 can use the exemplary equation (1) above and/or can use the NPM indicator method described above to calculate the probabilities γ _ij for the false positive correction matrix. An exemplary aggregate redistributed audience matrix for the selected tablet device type and the selected comedy media type is shown below in Table 14 as an exemplary false positive correction matrix. The subsequent sections of the exemplary Table 14 extend to the right so that the table has an equal number of rows and columns.

Exemplary False Positive Correction Matrix

Table 14

As shown in the exemplary misidentification correction matrix of Table 14 above, the values of each column total 100%. Therefore, when redistributing impressions, duration units, and/or audiences based on the exemplary misidentification correction matrix of Table 14 above, the impressions, duration units, and/or audiences attributed to the identified demographic group i are accounted for. As described in more detail below, the misidentification corrector 206 of FIG. 2 applies the exemplary misidentification correction matrix of Table 14 above to compensate for misidentification errors in the impression information.

Co-viewing Matrix

In some examples, the sharing matrix generator 204 also includes a co-viewing matrix generator 224 that generates a co-viewing matrix (e.g., for each combination of media category and device type) in addition to the exemplary misidentification correction matrix of Table 14 above. A co-viewing matrix is a matrix that provides the PDF of the probability of simultaneous viewing between two demographic groups. For example, the co-viewing probability κ _ij is the probability that, when the database owner identifies a person in the identified demographic group i as being associated with a media impression, a person in the actual viewer demographic group j is also viewing (e.g., co-viewing) the media with the identified person. Therefore, for situations where an impression can be correctly associated with multiple people, the probability κ _ij in the co-viewing matrix can be used to compensate for the impression information.

Like the exemplary false positive correction matrix of Table 14 above, the co-viewing matrix generated by the co-viewing matrix generator 224 has an equal number of rows and columns. However, unlike the exemplary false positive correction matrix of Table 14 above, the columns of the co-viewing matrix do not necessarily sum to any particular number (e.g., 100%). In the illustrated example, the exemplary co-viewing matrix generator 224 uses the probabilities P(L) _ij and P(D|L) _ij discussed above and also uses the probability P( _Cx |D) _ij to calculate the co-viewing probability κ _ij , _which describes the probability that a person in the identified demographic group _i , who lives in the same household and has access to a mobile device of a selected device type, and a person in the actual viewer demographic group j, simultaneously uses the same mobile device of device type x to view media of the selected media category. In some examples, the co-viewing matrix generator 224 replaces the P( _Sx |D) _ij term in equation (1) above with P( _Cx |D) _ij to calculate the co-viewing probability κ _ij .

In some examples, the false positive corrector 206 applies the co-viewing matrix to the impressions and/or duration units adjusted for device-sharing false positives. In some other examples, because co-viewing can be considered to represent additional impressions and/or duration units that were not accounted for in the data collection, the false positive corrector 206 applies the probabilities κ _ij in the co-viewing matrix to the impressions and/or duration units adjusted for device-sharing false positives and for lack of coverage by the database owner 104 a, thereby determining co-viewing based on the compensated (e.g., corrected) impressions and/or duration units.

Misidentification Correction Example 1

After generating the example false positive correction matrix of Table 14 above, the example false positive corrector 206 applies the false positive correction matrix to a set of impressions obtained from the impression information collector 208 of FIG. 2 . In a first example of false positive correction, the false positive correction matrix is applied to correct impressions for attribution to demographic groups without affecting audience counts. This type of example can, for example, be applied to correct impressions of Internet-based streaming media (e.g., streaming video and/or streaming audio). For example, corrected impression information for Internet-based streaming media can be combined with impression information for another type of device on which the media can be accessed in addition or alternatively. For example, after an episode is presented (e.g., broadcast) on television, an audience member can access the television series via a computing device for delayed or time-shifted streaming playback. Impressions of the streaming media accessed via the computing device can be added to the Live+7 ratings metric, which measures the sum of viewing impressions (or corresponding audience size) of the initial presentation and impressions (or corresponding audience size) occurring on the day of the initially scheduled broadcast presentation and within 7 days after the initially scheduled broadcast presentation.

The example impression information collector 208 of Figure 2 collects impression information from the database owner 104a and collects impression volume information obtained from client devices, such as the client device 106 of Figure 1. The example impression information obtained from the database owner 104a includes an aggregate number of impressions associated with each demographic group from the database owner 104a.

The example database owner 104a may provide impression information (e.g., the number of impressions identified by the database owner 104a as being associated with an identified person) for each media item of interest (e.g., a media item being monitored by the audience measurement entity 108). Additionally or alternatively, the example database owner 104a provides impression information for various types of devices to the impression information collector 208. Table 15 below illustrates example impression information for a tablet device obtained from the example database owner 104a by the impression information collector 208. The example impression information collector 208 may collect similar data for other types of mobile devices (e.g., smartphones, portable media players, etc.).

Example impression information for a tablet device

Table 15

In some examples, the impressions are processed through a decision tree before being provided to the impression information collector 208. In some examples, the decision tree is used to determine the delineation between the identified demographic group i and/or the actual viewer demographic group j. Examples of processing impressions are disclosed in U.S. patent application serial number 13/209,292 filed on August 12, 2011 and in U.S. provisional patent application serial number 61/923,959 filed on January 6, 2014. The entire contents of U.S. patent application serial number 13/209,292 and U.S. provisional patent application serial number 61/923,959 are incorporated herein by reference.

The example false positive corrector 206 of FIG2 adjusts or compensates for false positive errors in impression information obtained from the database owner 104a. FIG3A illustrates an example calculation performed by the false positive corrector 206 to adjust the impression information. In the examples of FIG2 and FIG3A, the false positive corrector 206 uses the device sharing matrix generated by the example sharing matrix generator 204 to adjust demographic information provided by the database owner 104a (e.g., impression counts per demographic group per device type and/or media category, etc.).

Using the database owner data (e.g., impression information 102a, 102b of FIG. 1 ) obtained by the impression information collector 208, the misidentification corrector 206 of FIG. 2 calculates the dot product of an n×n misidentification correction matrix 302 for device types and/or media categories and an n×1 database owner data 304 (e.g., an impression count matrix) for device types and/or media categories. The result of the dot product is an n×1 misidentification adjusted data matrix 306 having an adjusted number of impressions for the device types and/or media categories.

Table 16 below shows an exemplary misidentification-adjusted impression calculated by misidentification corrector 206 using the exemplary misidentification correction matrix of Table 14 and the impression information of Table 15. Table 16 includes the adjusted impression calculated based on the dot product discussed above with reference to FIG. For ease of comparison, the unadjusted impression from Table 15 above is also shown in Table 16.

Misidentification of adjusted and unadjusted impressions

Table 16

Thus, as shown in Table 16 above, the adjusted impressions compensate for misidentification errors in the impression information 102a, 102b received from the exemplary database owner 104a.

Misidentification Correction Example 2

In a second example of misidentification correction, the misidentification correction matrix of Table 14 above is applied to correct misidentification of impressions and audiences to demographic groups. As described above, misidentification of an impression to an incorrect demographic group can occur when the database owner 104a attributes an impression to a person in a first demographic group when, in fact, the impression is correctly attributed to a second person in a second demographic group (e.g., because a person in the first demographic group was logged into the database owner's device during the time the second person in the second demographic group viewed the media that gave rise to the impression). The exemplary misidentification correction of impressions and audiences can be used, for example, when media can be accessed from different types of computing devices and the audiences recorded or measured for those computing devices can be deduplicated. For example, duplication of recorded audience members occurs when an audience member accesses the same media multiple times from the same device and/or different devices. Multiple visits to the same media by the same audience member result in audience duplication because the same audience member is counted twice or more in the audience size count based on the multiple impressions recorded for the audience member's multiple visits to the same media. This type of audience duplication can result in an inflated representation of the true size of the audience that was exposed to or visited a particular media. Therefore, deduplication can be used to more accurately count the audience size to which impressions of a medium are attributable.

2 collects impression information from the database owner 104a and collects volume information for impressions occurring at the client device 106. Exemplary impression information obtained from the database owner 104a includes an aggregate number of impressions by demographic group and/or the size of the audience from each demographic group generated by the database owner 104a.

The exemplary database owner 104a may provide impression information (e.g., impression counts, impression counts by demographic group, etc.) and/or audience information (e.g., audience size, audience size by demographic group, etc.) for each media item of interest (e.g., media items being monitored by the audience measurement entity 108). Additionally or alternatively, the exemplary database owner 104a provides impression and/or audience information for various types of devices to the impression information collector 208. In some examples, the impression information collector 208 also collects impressions and/or audience information for media impressions that occur on computer platforms (e.g., non-mobile device platforms, such as desktop computers and/or laptop computers). Table 17 below shows exemplary impression and audience information (e.g., uncorrected impression counts and audience size) for tablet devices obtained by the impression information collector 208 from the exemplary database owner 104a. The exemplary impression information collector 208 may collect similar data for other types of mobile devices (e.g., smartphones, portable media players, etc.) and/or computer platforms. Exemplary lower table 17 is similar to exemplary upper table 15, except that lower table 17 also includes audience size and frequency information (eg, from database owner 104a).

Example impression and audience information for tablet devices obtained from an example database owner

Table 17

In some examples, the database owner's total impressions and audience size are processed through a decision tree before being provided to the impression information collector 208. Examples of processing impressions and specific audiences are disclosed in U.S. Non-Provisional Patent Application No. 13/209,292, filed on August 12, 2011, and U.S. Provisional Patent Application No. 61/923,959, filed on January 6, 2014. The entire contents of U.S. Non-Provisional Patent Application No. 13/209,292 and U.S. Provisional Patent Application No. 61/923,959 are incorporated herein by reference.

The example false positive corrector 206 of FIG2 adjusts or compensates for the impression information obtained from the database owner 104a for the false positive error. FIG3A discussed above also illustrates an example calculation performed by the false positive corrector 206 to adjust the impression and/or audience information. In this example, the false positive corrector 206 uses the device sharing matrix generated by the example sharing matrix generator 204 to adjust the demographic information provided by the database owner 104a (e.g., impression counts per demographic group per device type and/or media category, audience size per demographic group per device type and/or media category, etc.).

Using the database owner data (e.g., impression counts and/or audience size information) obtained by the impression information collector 208, the false positive corrector 206 of FIG2 calculates the dot product of the n×n false positive correction matrix 302 and/or the n×1 database owner data 304 (e.g., impression count matrix, audience size matrix) for device type and/or media category. The result of the dot product is an n×1 false positive adjusted data matrix 306 having adjusted impression counts or adjusted audience size.

Table 18 below shows an exemplary misidentification-adjusted data matrix calculated by misidentification corrector 206 using the misidentification correction matrix of Table 14 above and the exemplary impression count and/or audience size data of Table 17 above. Table 18 below includes the audience calculated based on the dot product discussed above, as well as the impression count information calculated by misidentification corrector 206 using the adjusted audience size. In this example, misidentification corrector 206 determines the misidentification-adjusted impressions (e.g., 118,492 for the F45-49 demographic group) by dividing the misidentification-adjusted audience size (e.g., 12,216 for the F45-49 demographic group) by the frequency corresponding to that demographic group from Table 17 above (e.g., 9.7 for the F45-49 demographic group). The unadjusted audience size and unadjusted impression counts are also shown in Table 18 for comparison purposes.

Misidentification of adjusted impression counts and audience sizes, and unadjusted impression counts and audience sizes for comparison

Table 18

Thus, as shown in Table 18 above, the misidentification-adjusted impression counts and misidentification-adjusted audience size compensate for misidentification errors in the impression count and audience size information received from the example database owner 104a. The total misidentification-adjusted audience size is substantially equal to (e.g., equal to, except for rounding errors) the total unadjusted audience size reported by the database owner 104a. The example misidentification-adjusted audience size and/or misidentification-adjusted impression counts of this example are corrected for uncovered errors as described in the examples below.

Alpha factor for uncovered

In some examples, such as to compensate for errors in impression count information corresponding to media associated with viewership (e.g., episodes of television programs available for streaming), the non-coverage calculator 210 of FIG. 2 uses an "alpha factor" to calculate the non-coverage factor for an audience. As used herein, the term "alpha factor" refers to the ratio B/A, where B is defined as the probability that a person (e.g., a person in a demographic group of interest) accesses a media item of interest (e.g., an episode of a television program via streaming video) using a device type of interest that is not covered by the database owner (e.g., on a mobile device and/or on a specific type of mobile device, such as a tablet, smartphone, or portable media player). For example, if the database owner does not have access to any identifiers or information on the device type of interest (e.g., one or more device/user identifiers 124 of FIG. 1 ), the database owner may not cover the device type, which the database owner can use to associate registered user information (e.g., demographic profile). As used herein, in the alpha factor ratio B/A, A is defined as the probability that a person accesses the media item of interest on a device other than a mobile device (e.g., a standard device used to access the media item of interest, such as a television in the case of a television program). For example, for episodes of a television program in a particular media category, the designated type may be a television and the first device type may be a computing device (e.g., a mobile device and/or a more specific type of mobile device, such as a smartphone, tablet, and/or portable media player), on which episodes of the television program may be accessed via streaming video. Such television programs are often made available via streaming video after an initial or premiere presentation on the television. Thus, a person may access episodes of the television program on the television and/or on a computing device via streaming media.

The example uncovered calculator 210 of FIG. 2 can generate different alpha factors for different demographic groups, different media categories, different mobile device types, mobile and non-mobile devices, different geographic regions, different stations, different time periods, and/or any other factors identified based on a survey calibration data source.

2 , the example non-coverage calculator 210 calculates probability B (e.g., the probability that a person views the media item of interest using a device type of interest) as the proportion of the selected demographic group (e.g., based on responses from a survey or another calibration data source) that accessed the media item of interest on the selected device type to the total number of people in the selected demographic group (e.g., based on responses from the survey or another calibration data source). For example, if 40 people in the M18-24 demographic group responded to a survey that they accessed media in the 'comedy' media category on a tablet (out of 100 people in the M18-24 demographic group represented in the survey), probability B is 40% or 0.40. Similarly, the example non-coverage calculator 210 calculates probability A (e.g., the probability that a person views the media item of interest using a selected other device type) as the proportion of the selected demographic group (e.g., based on responses from a survey or another calibration data source) that accessed the media item of interest on the other device type to the total number of people in the selected demographic group (e.g., based on responses from a survey or another calibration data source). For example, if 20 people in the M18-24 demographic group responded to a survey that they accessed media in the 'comedy' media category on television (out of the 100 people in the M18-24 demographic group represented in the survey), then Probability A is 20% or 0.20. Example models for calculating Probability A and Probability B are shown in Equations 2 and 3 below, respectively. The alpha factor for the M18-24 demographic group, the 'comedy' media category on tablets, is 0.40/0.20 = 2.

Equation 2A = (number of people in age and gender group X who watch media category Y on TV) / (total number of people in age and gender group X)

Equation 3B = (Number of people in age and gender group X who watch media category Y on the device type of interest) / (Total number of people in age and gender group X)

In an example where the non-coverage calculator 210 determines an alpha factor for determining non-coverage by the database owner 104a, the exemplary non-coverage corrector 212 corrects the impression information by multiplying the alpha factor for the demographic group by the distribution percentage of the demographic group for the media item of interest. For example, if the 35-39 year old female group (e.g., F35-39 in Table 19 below) represents 2.9% of the impressions for a particular episode of a television program, and the alpha factor for the 35-39 year old female group is 3.8 for the media category that categorizes the television program, the newly calculated percentage is approximately 11.2%. However, the resulting percentage (i.e., 11.2%) is normalized so that the percentages for all demographic groups for the episodes of the television program total 100%. The exemplary non-coverage corrector 212 multiplies the normalized percentage by the number of impressions that were not associated to the user by the database owner 104a to determine the number of impressions attributed to the 35-39 year old female group (F35-39). In some examples, the number of impressions that the database owner 104a cannot associate with a user is determined from the difference between: 1) the number of impressions identified by the AME 108 and 2) the number of impressions that the database owner 104a associates with a user. Additionally or alternatively, the database owner 104a monitors and reports the number of impressions that the database owner 104a cannot associate with a user, while also monitoring the number of impressions corresponding to a demographic group (e.g., impressions that the database owner can associate with a user).

Table 19 below shows exemplary data generated by the uncovered corrector 212 to correct impression information using an alpha factor. In the example of Table 19, AME 108 counts 2,000 media impressions that the database owner 104a is unable to associate to a demographic group (e.g., labeled "Demos" in the first column of Table 19 below).

Uncovered impressions

Table 19

As shown in the example of Table 19 above, for each demographic group (Demos), the non-coverage corrector 212 calculates the adjusted percentage (Adj%) as the product of the alpha factor (α=B/A) and the measured percentage. The non-coverage corrector 212 normalizes the adjusted percentage (Adj%) to a sum of 100% (e.g., dividing the adjusted percentage (Adj%) by the total adjusted percentage (e.g., 160.9%)) to obtain a normalized percentage (Norm%). The non-coverage corrector 212 multiplies the normalized percentage (Norm%) by the number of impressions that were not associated with the demographic group (Demos) by the database owner 104a to obtain the number of uncovered impressions (Non-covered Impressions) attributed to each demographic group (Demos). The exemplary impressions (Non-covered Impressions) determined by the non-coverage corrector 212 can be added to the misidentification-adjusted impressions to determine misidentification- and non-coverage-adjusted impressions.

Non-coverage factor for non-coverage correction

As an alternative to using the exemplary α factor (α=B/A) disclosed above to compensate impression information for non-coverage error, the exemplary non-coverage calculator 210 of FIG2 can calculate a non-coverage factor for each exemplary demographic group. The non-coverage factor reflects the portion of the impression that is not attributed to a person by the database owner 104a.

To calculate the uncovered factor for a demographic group and a particular device type, the example uncovered calculator 210 determines from a survey calibration data source a significant number or portion of persons in the demographic group (e.g., respondents to a survey) who indicate that they would not be recognized by the database owner 104a when using a particular device type as being a portion of the persons in the demographic group who own and access the particular device type. For example, the uncovered calculator 210 may determine that a respondent would not be recognized by the database owner 104a when using a particular device type if the respondent indicates that no one in the respondent's home uses the particular device type to access the database owner 104a, that the respondent is not registered with the database owner 104a, and/or that the respondent has taken any other action or inaction that prevents the database owner 104a from recognizing the respondent when accessing media on the particular device type.

The example non-coverage calculator 210 of FIG2 creates a table of non-coverage factors for each demographic group and device type, where the non-coverage factor is calculated as: non-coverage factor = (uncovered portion of respondents for the device type in the demographic group) / (total number of respondents in the demographic group who accessed devices of the device type). An example non-coverage factor table for a tablet is shown in Table 20 below. In some examples, the example non-coverage calculator 210 creates similar tables for other device types. As shown in Table 20, 20% of the 21-24 year old female demographic group is not covered (e.g., not identifiable) on the tablet by the database owner 104a. In other words, 10% of the 21-24 year old females who accessed the tablet would not be recognized by the database owner 104a. Similarly, 10% of the 18-20 year old male demographic group is not covered on the tablet by the database owner 104a.

Exemplary non-coverage factors for tablets

Table 20

The non-coverage factor of Table 20 above can be used as an alternative to the α factor (α = B/A) for performing non-coverage adjustments on impression information. For example, the frequency observed by the database owner 104a (e.g., the average impression per audience member during a time period) can be used to calculate the audience corresponding to the impressions observed by the database owner 104a. The example non-coverage calculator 210 then adjusts the audience by the non-coverage factor (e.g., adjusted audience = audience / (1-non-coverage factor)) and uses the frequency to convert the adjusted audience into non-coverage adjusted impressions.

Although the above example describes a non-coverage factor for one database owner, the example non-coverage calculator 210 can additionally or alternatively calculate non-coverage factors for multiple database owners. For example, if two database owners are used, the example survey calibration data source can include data designated to determine whether a respondent accessed any of the database owners 104a-b via one or more device types. The example non-coverage calculator 210 of FIG2 then determines that the non-coverage error is limited to those persons and/or impressions that are not identifiable by any of the database owners 104a-b. In the example of FIG2, the example person is considered covered if at least one of the database owners 104a-b can identify the person on a mobile device.

Uncovered Error Compensation and Scaling Example 1

A first example of compensating for uncovered errors includes multiplying an alpha factor (α = B/A) for a demographic group by the misidentification-adjusted impression count corresponding to the same demographic group. For example, the alpha factor can be used to calculate the uncovered impression count based on the misidentification-adjusted impression count of the first misidentification compensation example described above with reference to Table 16 and/or Table 18.

FIG3B illustrates exemplary calculations that may be performed by the non-coverage corrector 212 of FIG2 to compensate for non-coverage errors. In the example of FIG3B , the non-coverage corrector 212 obtains the misidentification-adjusted data 306 calculated by the misidentification corrector 206 in the example of FIG3A . The non-coverage corrector 212 applies non-coverage factors 308 (e.g., a set of alpha factors, a set of scalars) corresponding to the same device type (and/or media category) to the misidentification-adjusted data 306 for the device type (and/or media category) to determine misidentification- and non-coverage-adjusted data 310 for the device type (and/or media category).

For example, Table 21 below shows an exemplary calculation for determining the amount of uncovered impressions per demographic group using the total number of corresponding misidentification-adjusted impression counts from Table 18 above. The uncovered corrector 212 adjusts the misidentification-adjusted impression counts (e.g., 210,945 for the M25-29 demographic group from Table 18 above) using an alpha factor (α=B/A) for the M25-29 demographic group (e.g., 1.68 from Table 19 above). Table 21 below shows an example of misidentification- and uncovered-adjusted data 310 (e.g., misidentification- and uncovered-adjusted impression counts). In the following example, AME 108 has identified 1,126,462 total impressions that the database owner 104 has not associated with a demographic group.

Example Misidentification and Non-coverage Adjusted Impression Data

Table 21

In Table 21 above, exemplary misidentification-adjusted impressions (Misatt.-adjusted Imp. Count) are obtained from the misidentification correction example described above with reference to Table 18. The exemplary measured percentages (Meas %) of Table 21 are determined based on the misidentification-adjusted impression count (Misatt.-adjusted Imp. Count) for each demographic group by determining the percentage of the misidentification-adjusted impression count (e.g., 182,026 for the F30-34 demographic group) relative to the total misidentification-adjusted impression count (e.g., 3,034,551).

Table 21 includes a set of exemplary alpha factors (α=B/A) for various exemplary demographic groups, which can be calculated as described above with reference to exemplary Table 19. The alpha factors are determined, for example, using the results of a calibration survey (such as the survey discussed above) and equations 2 and 3 disclosed above. Using the alpha factors, the exemplary non-coverage corrector 212 calculates an adjusted percentage (Adj.%) for each exemplary demographic group of Table 21 by multiplying the measured percentage (Meas.%) by the corresponding alpha factor. The exemplary non-coverage corrector 212 also normalizes the adjusted percentage (Adj.%) for each exemplary demographic group of Table 21 to obtain a normalized percentage (Norm.%).

The example non-coverage corrector 212 determines the non-covered impression count (Non-Covered Imp. Count) by multiplying the normalized percentage (Norm. %) for each example demographic group of Table 21 by the total non-covered impression count (e.g., 1,126,426). For example, the non-covered impression count (Non-Covered Imp. Count) for the F30-34 demographic group is calculated as 3.85%*1,126,426=43,388. The example non-coverage corrector 212 can then add the non-covered impression count (Non-Covered Imp. Count) to the misidentification-adjusted impression count (Misatt.-adjusted Imp. Count) for each example demographic group of Table 21 to determine the misidentification and non-coverage adjusted impression count (Misatt. and Non-Cov.-Adj. Imp. Count).

Uncovered Error Compensation and Scaling Example 2

A second example of compensating for uncovered errors includes using the uncovered factors of Table 20. For example, when the alpha factor is not available for a particular type of media and/or a particular device type (e.g., the probability of viewing the media and/or the probability of viewing on the device type is not available), the exemplary uncovered method disclosed above in conjunction with the uncovered factor can be used in place of the alpha factor of the uncovered error compensation example disclosed above in conjunction with the alpha factor.

2 uses the non-coverage factor to correct the impression information obtained from the database owner 104a. For example, the non-coverage corrector 212 can determine the adjusted number of impressions as: (reported impression count)/(1-(non-coverage factor for the demographic group)).

Using the example of FIG3B , the non-coverage corrector 212 obtains the misidentification-adjusted data 306 calculated by the misidentification corrector 206 in the example of FIG3A . The non-coverage corrector 212 applies a non-coverage factor 308 corresponding to the same device type (e.g., a set of scalars rather than a set of alpha factors) to the misidentification-adjusted data 306 for that device type to determine the misidentification- and non-coverage-adjusted data 310 for that device type. For example, the non-coverage corrector 212 adjusts the misidentification-adjusted impression count (e.g., 210,945 for the M25-29 demographic group from Table 18 above) using the 10.0% non-coverage factor from Table 20 to determine the misidentification- and non-coverage-adjusted impression count as 19,046. The example non-coverage corrector 212 also determines the misidentification- and non-coverage-adjusted audience size as the quotient of the misidentification- and non-coverage-adjusted impression count divided by the frequency calculated or obtained from the database owner (e.g., the frequency from Table 17 above). Table 22 shows an example of misidentification and non-reach adjusted data 310 (eg, misidentification and non-reach adjusted impressions and audience) for a tablet device type.

Example misidentification and non-reach adjustment impression counts and audience size for the tablet device type

Table 22

In the example of Table 22 above, the example non-coverage corrector 212 determines the misidentified and non-coverage adjusted audience size for each demographic group by dividing the corresponding misidentified and non-coverage adjusted impression count by the corresponding frequency of Table 17 (e.g., from the database owner 104a).

The example impression information adjuster 214 of Figure 2 adjusts compensated impression counts and audience size to align with the number of impressions observed by the audience measurement entity 108. Figure 3C illustrates an example process for adjusting compensated (e.g., misidentification and non-coverage adjusted) impression counts and/or audience size 310 based on observed census data (e.g., impression volume counts).

While the examples of compensating for misidentification and/or non-coverage described above with reference to Tables 15-22 describe compensating for impressions and specific audience sizes, the example misidentification corrector 206 and/or the example non-coverage corrector 212 of FIG2 can additionally or alternatively use the same techniques to compensate for duration unit misidentification and/or non-coverage. Table 23 below shows an example of applying the misidentification matrix of Table 14 above to duration units associated with impressions and specific audience sizes of the example of Table 18 above.

Example misidentification and non-coverage adjusted impression counts, duration units, and special audience sizes for the tablet device type, and unadjusted impression counts, duration units, and special audience sizes

Table 23

As shown in Table 23 above, applying the misidentification correction matrix of Table 14 to the unadjusted duration units (Unadj.DurationUnits) of Table 23 results in a redistribution of duration units among the demographic groups (Demo.Group). The adjustments for misidentification and special audience size are the same as in the example described above with reference to Table 18. The example uncover corrector 212 can then correct the misidentified adjusted duration units using the uncover correction techniques described above with reference to the impressions of Tables 19-21.

Examples of impression scaling, duration unit scaling, and/or audience scaling

The example impression information adjuster 214 of Figure 2 adjusts compensated impressions to align with the number of impressions observed by the audience measurement entity 108. Figure 3C illustrates an example process for adjusting compensated (e.g., misidentification and non-coverage adjusted) impressions based on observed census data (e.g., impression volume counts).

Using the example misidentified and uncovered adjusted impression counts of the example of Table 21 above, the impression information adjuster 214 scales 312 the misidentified and uncovered adjusted impression counts 310 to match (e.g., be equal to) the number of impressions observed from the tablet (e.g., as observed by the AME 108 of FIG. 1 ), which in this example is 6,385,686 impressions. The example impression information adjuster 214 can scale the impression counts for the primary website (e.g., the website on which the advertisement or other media is hosted) and/or can scale the impressions of the media (advertisements or other media) placed on the primary website.

To scale 312 the exemplary compensated impression information for the demographic group, the exemplary impression information adjuster 214 multiplies the total number of impressions (e.g., recognized and unrecognized) observed from the database owner (e.g., 6,385,686 in this example) by the compensated impression count attributable to the M21-24 demographic group (e.g., 158,067 in Table 21 above) as a fraction (e.g., a percentage) of the total compensated impressions (e.g., 4,161,011 in Table 21 above). For example, the impression information adjuster 214 determines the scaled compensated impressions for the 21-24 male demographic group as (6,385,686)*(123,750/4,161,011)=189,913. Table 24 below shows exemplary scaled compensated impression counts for a tablet.

Exemplary scaled compensated impression count for tablets

Table 24

In some other examples, the impression information adjuster 214 calculates impression information for the media based on the PDF scaled impressions of the site hosting the media. For example, to scale impressions for media delivered on a primary website, the exemplary impression information adjuster 214 determines, for the demographic group of interest and the devices of interest (or all devices), the difference between: a) the number of impressions for the device type identified by the audience measurement entity 108 for the primary site (the census count of impressions) and b) the impressions adjusted for misidentification and non-coverage for the device type. The difference is the number of impressions explained by the scaling. The exemplary impression information adjuster 214 determines the product of this difference and the fraction of impressions attributable to the demographic group of interest for the device type of interest. In other words, the impression information adjuster 214 applies the probability distribution function for the demographic group of the primary site to the number of impressions needed to add to scale the misidentification and non-coverage adjusted impressions. This product (e.g., the fraction of impressions for the demographic group) is added to the misidentification and non-coverage adjusted impressions for the demographic group. Therefore, in this example, the impression information adjuster 214 determines the scaled impressions as: scaled impressions = (misidentification and non-coverage adjusted impressions for the selected demographic group and the selected device type) + (census impressions for media on the primary site for the selected device type - total misidentification and non-coverage adjusted impressions for all demographic groups for the selected device type) * (scaled impressions for the primary site for the selected demographic group and the selected device type / total scaled impressions for the primary site for all demographic groups and the selected device type).

Table 25 below shows an exemplary scaling of impressions for scaled media impressions (rather than primary site impressions) using the probability distribution function described above and the exemplary misidentification and non-coverage adjusted impressions of Table 19 above. Table 25 is based on 6,385,687 total census impressions for tablets on the primary site for the exemplary media.

Example Scaling

Table 25

In the example of Table 25 above, the example impression information adjuster 214 scales impressions for the F21-24 demographic groups and tablet device type to scaled impressions = (misidentification and non-coverage adjusted impression count for device type and demographic group) + (total observed tablet impression count for all demographic groups - total misidentification and non-coverage adjusted tablet impression count for all demographic groups) * (scaled primary site impression count for demographic group / total scaled primary site impression count for all demographic groups) = 123,750 + (6,385,686 - 4,161,015) * (1,132,301 / 37,873,074) = 190,262. The example scaling of the above example can be performed to scale impressions for different demographic groups to other census impression counts, such as the primary site's impression count (e.g., if the primary site exclusively presents media of interest).

Although the examples disclosed herein are described with reference to viewing and video media and/or combined audio/video media, the examples disclosed herein can also be used to measure listeners of audio-only media. For example, the media categories, survey calibration data, and/or second device types used to calculate the alpha factor can be tailored for audio media. For example, the second device type used to calculate the "A" term (e.g., Equation 2 above) can be modified to refer to (number of people in age and gender group X who listen to media category Y on the radio)/(total number of people in age and gender group X).

The exemplary scaling techniques described above with reference to Tables 24 and 25 may be used to scale misidentification and/or non-coverage adjusted duration units to duration unit counts observed for the primary site (eg, census duration counts).

Although the above examples disclose performing both misidentification correction and non-coverage correction, misidentification correction may be performed on impression counts and/or audience size without performing non-coverage correction. Alternatively, non-coverage correction may be performed on impression counts and/or audience size without performing misidentification correction.

Although an exemplary manner of implementing the exemplary impression data compensator 200 is shown in FIG2 , one or more of the elements, processes, and/or devices shown in FIG2 may be combined, divided, rearranged, omitted, eliminated, and/or implemented in any other manner. Additionally, the exemplary calibration data collector 202 , the exemplary sharing matrix generator 204 , the exemplary false positive corrector 206 , the exemplary impression information collector 208 , the exemplary non-coverage calculator 210 , the exemplary non-coverage corrector 212 , the exemplary impression information adjuster 214 , the exemplary household distribution generator 216 , the exemplary aggregate distribution generator 218 , the exemplary matrix corrector 220 , the exemplary matrix normalizer 222 , the exemplary co-viewing matrix generator 224 , and/or more generally, the exemplary impression data compensator 200 may be implemented using hardware, software, firmware, and/or any combination of hardware, software, and/or firmware. Thus, for example, any of the exemplary calibration data collector 202, the exemplary sharing matrix generator 204, the exemplary false positive corrector 206, the exemplary impression information collector 208, the exemplary non-coverage calculator 210, the exemplary non-coverage corrector 212, the exemplary impression information adjuster 214, the exemplary household distribution generator 216, the exemplary aggregate distribution generator 218, the exemplary matrix corrector 220, the exemplary matrix normalizer 222, the exemplary co-viewing matrix generator 224 and/or more generally, the exemplary impression data compensator 200 may be implemented using one or more analog or digital circuits, one or more logic circuits, one or more programmable processors, one or more application specific integrated circuits (ASICs), one or more programmable logic devices (PLDs) and/or one or more field programmable devices (FPLDs), etc. When reading any of the device or system claims of this patent covering a purely software and/or firmware implementation, at least one of the exemplary calibration data collector 202, exemplary sharing matrix generator 204, exemplary false positive corrector 206, exemplary impression information collector 208, exemplary non-coverage calculator 210, exemplary non-coverage corrector 212, exemplary impression information adjuster 214, exemplary household distribution generator 216, exemplary aggregate distribution generator 218, exemplary matrix corrector 220, exemplary matrix normalizer 222, and/or exemplary co-viewing matrix generator 224 is hereby expressly defined to include a tangible computer-readable storage device or storage disk, such as a memory, digital versatile disk (DVD), compact disk (CD), Blu-ray disk, etc., storing software and/or firmware. In addition, the exemplary impression data compensator 200 of FIG. 2 may include one or more elements, processes, and/or devices in addition to or in place of those shown in FIG. 2 , and/or may include more than one of any or all of the illustrated elements, processes, and devices.

Flowcharts representing exemplary machine-readable instructions for implementing the impression data compensator 200 of FIG. 2 are shown in FIG. 4 through FIG. 13 . In this example, the machine-readable instructions comprise programs executed by a processor, such as the processor 1412 shown in the exemplary processor platform 1400 discussed below in conjunction with FIG. 14 . These programs may be embedded in software stored on a tangible, computer-readable storage medium, such as a CD-ROM, floppy disk, hard drive, digital versatile disk (DVD), Blu-ray Disc, or memory associated with the processor 1412, but all programs and/or portions thereof may alternatively be executed by a device other than the processor 1412 and/or embodied in firmware or dedicated hardware. Furthermore, while the exemplary programs are described with reference to the flowcharts shown in FIG. 4 through FIG. 13 , many other methods of implementing the exemplary impression data compensator 200 may alternatively be used. For example, the order of execution of the blocks may be changed, and/or some of the blocks described may be changed, eliminated, or combined.

As described above, the exemplary processes of Figures 4 to 13 can be implemented using coded instructions (e.g., computer and/or machine-readable instructions) stored on a tangible computer-readable storage medium, such as a hard drive, flash memory, read-only memory (ROM), compact disc (CD), digital versatile disc (DVD), cache, random access memory (RAM), and/or any other storage device or storage disk in which information is stored for any duration (e.g., for an extended period, permanently, temporarily, temporarily buffered, and/or cached information). As used herein, the term "tangible computer-readable storage medium" is expressly defined to include any type of computer-readable storage medium and/or storage disk and to exclude propagating signals and transmission media. As used herein, "tangible computer-readable storage medium" and "tangible machine-readable storage medium" are used interchangeably. Additionally or alternatively, the exemplary processes of Figures 4 to 13 can be implemented using coded instructions (e.g., computer and/or machine readable instructions) stored on a non-volatile computer and/or machine readable medium, such as a hard drive, flash memory, read-only memory, optical disc, digital versatile disc, cache, random access memory, and/or any other storage device or storage disk in which information is stored for any duration (e.g., an extended period, permanently, temporarily, temporarily buffered, and/or cached information). As used herein, the term "non-volatile computer readable medium" is expressly defined to include any type of computer readable storage medium and/or storage disk and to exclude propagation signals and transmission media. As used herein, when the phrase "at least" is used as a transitional word in the preamble of a claim, it is open ended, as is the term "comprising."

FIG. 4 is a flow diagram representative of example machine readable instructions 400 that may be executed to implement the example impression data compensator 200 of FIG. 2 for compensating impression data.

The example sharing matrix generator 204 of FIG. 2 generates a false positive correction matrix and/or a co-viewing matrix (block 402). For example, the sharing matrix generator 204 calculates a device sharing matrix for each device type represented in the impression information (e.g., impressions) obtained by the calibration data collector 202 ( FIG. 2 ) from the database owner 104a ( FIG. 1 ). The example sharing matrix generator 204 may calculate the false positive correction matrix for the device type by calculating the probabilities described above in Equation 1 and/or by calculating and aggregating household sharing matrices. Additionally or alternatively, the example sharing matrix generator 204 may calculate a co-viewing matrix for each device type and/or media category that includes a co-viewing probability κ _ij . The co-viewing probability κ _ij may also be calculated using an example survey calibration data source (e.g., a survey of randomly selected individuals and/or households) to determine, for example, the incidence of co-viewing within a representative household. Example instructions that may be used to implement block 402 are described below with reference to FIG. 5 .

The example coverage calculator 210 of FIG2 determines the probability that the uncovered audience accessed the media on the device type of interest (block 404). The example coverage calculator 210 also determines a coverage factor (block 405). For example, the coverage calculator 210 can determine an alpha factor (e.g., the example alpha factor α=B/A described above with reference to Table 17) that is used to calculate the demographic distribution for impressions that are not associated with demographic information by the database owner 104a. Example instructions that can be executed to implement blocks 404 and 405 are described below with reference to FIG9.

The example impression information collector 208 of FIG2 obtains impression counts from a volumetric (e.g., census) measurement of impressions (block 406). For example, the impression information collector 208 determines the number of impressions identified at the audience measurement entity 108 ( FIG1 ) for each media item of interest (e.g., monitored media) and/or for each device type. Example volumetric data is shown in Table 16 above. The example impression information collector 208 also obtains demographic information corresponding to the impressions (and/or a subset of the impressions) (block 408). For example, the impression information collector 208 receives a count of impressions associated with each demographic group by the database owner 104a for each device type and/or for all device types.

The example misidentification corrector 206 of FIG. 2 selects a media category (e.g., comedy, drama, feature film, etc.) (block 410). The example misidentification corrector 206 also selects a device type (e.g., smartphone, tablet, portable media player) (block 412). The example misidentification corrector 206 adjusts impressions, duration units, and/or special audience size (obtained from the database owner 104a) based on a misidentification correction matrix for the selected media category and the selected device type (block 414). For example, the misidentification corrector 206 may determine a dot product of the misidentification correction matrix 302 and the database owner impression data 304, as shown in FIG. 3A. In some examples, the misidentification corrector 206 calculates a misidentification corrected special audience size based on the misidentification corrected impressions and the frequency determined by the database owner 104a. Example instructions that may be executed to implement block 414 are described below with reference to FIG. 7.

The example non-coverage corrector 212 of FIG. 2 also adjusts the impressions, duration units, and/or special audience sizes based on the non-coverage factor for the selected media category and the selected device type (block 416). For example, the non-coverage corrector 212 may divide the misidentified adjusted impressions 306 of FIG. 3B by the coverage percentage (e.g., 1-non-coverage percentage) to obtain the misidentified and non-coverage adjusted impression information 310. In some examples, the misidentified corrector 206 calculates the misidentified and non-coverage corrected audience based on the misidentified and non-coverage corrected impressions and the frequency determined by the database owner 104a. Example instructions that may be executed to implement block 416 are described below with reference to FIG. 8.

The example error correction module 206 determines whether there are additional device types for which impression information is to be compensated (block 418). If there are additional device types for which impression information is to be compensated (block 418), control returns to block 412 to select another device type. When there are no more device types for which impression information is to be compensated (block 418), the example error correction module 206 determines whether there are additional media categories for which impression information is to be compensated (block 420). If there are additional media categories for which impression information is to be compensated (block 420), control returns to block 410 to select another media category.

When there are no more media categories whose impression information is to be compensated (block 420), the example impression information adjuster 214 of FIG. 2 calculates media ratings based on the adjusted impression information (block 422). For example, the impression information adjuster 214 may scale the misidentified and uncovered adjusted impressions to match the impressions identified by the audience measurement entity 108. Additionally or alternatively, the example impression information adjuster 214 may combine device types to determine impression information for multiple or all mobile device types. The example instructions 400 of FIG. 4 then end.

5 is a flow diagram representing example machine-readable instructions 500 that may be executed to implement the example impression data compensator 200 of FIG 2 for calculating a misidentification correction matrix and/or a co-viewing matrix. The example instructions 500 may be executed by the example sharing matrix generator 204 of FIG 2 to implement block 402 of FIG 4 .

The example sharing matrix generator 204 of FIG. 2 obtains survey calibration data from a survey calibration data source (e.g., from the calibration data collector 202 of FIG. 2 ) (block 502). For example, the sharing matrix generator 204 may obtain information indicating the devices accessible to persons in a respondent's household and/or the media categories viewed by the respondent on different device types. The example aggregate distribution generator 218 selects a media category (block 504). The example aggregate distribution generator 218 selects a device type (block 506). In the example of FIG. 5 , the media category and/or device type are selected from the media categories and/or device types represented in the survey calibration data. For all demographic groups i and j, the example aggregate distribution generator 218 determines the probability γ _ij of the following events: (1) a person in the identified demographic group i is identified as a viewer of the media by a database owner (e.g., the database owner 104 a of FIG. 1 ), and (2) a person in the actual viewer demographic group j is an actual viewer (block 508). For example, the aggregated distribution generator 218 may use the exemplary Tables 1-13 above to generate the false positive correction matrix shown in the above Table 14. Example instructions that may be executed to implement block 508 are described below with reference to FIG.

For all pairs of demographic groups i and j, the example aggregate distribution generator 218 determines the probability κ _ij that: (1) a person in the identified demographic group i is identified by the database owner as a viewer of the media, and (2) there is also an impression for a person in demographic group j who is an actual viewer (block 510). The set of probabilities κ _ij for pairs of demographic groups i and j can be aggregated into a co-viewing matrix. Example instructions that can be executed to implement block 510 are described below with reference to FIG8.

The example aggregated distribution generator 218 determines whether there are any additional device types for which the survey calibration data from the survey calibration data source should be processed (block 512). If there are additional device types (block 512), control returns to block 506. If there are no additional device types (block 512), the example aggregated distribution generator 218 determines whether there are additional media categories for which the survey calibration data from the survey calibration data source should be processed (block 514). If there are additional media categories, control returns to block 504. Otherwise, if there are no additional media categories (block 514), the example instructions 500 end, and control returns to, for example, block 404 of FIG. 4.

FIG6 is a flow diagram of exemplary machine-readable instructions 600 that may be executed to determine a probability of false positives γ _ij . The exemplary instructions 600 may be executed to implement block 508 of FIG5 , which determines the probability γ _ij of the following events: (1) a person in an identified demographic group i is identified by a database owner (e.g., database owner 104a of FIG1 ) as a viewer of the media, and (2) a person in an actual viewer demographic group j is an actual viewer. Initially in the exemplary instructions 600 , the exemplary household distribution generator 216 redistributes the audience of registered users of the database owner 104a who are members of the household (block 602 ). For example, the exemplary household distribution generator 216 may generate an exemplary redistributed audience matrix for the selected media category, as described above in connection with Tables 5 and/or 6. The exemplary aggregated distribution generator 218 aggregates the redistributed audiences in the household for the identified demographic group i and the actual viewer demographic group j (block 604 ). For example, the aggregate distribution generator 218 may generate an exemplary redistributed audience matrix for a selected media category for a plurality of households, as described above in conjunction with Table 8.

The example matrix corrector 220 generates an NPM metric that accounts for the probability P(L) _ij that the first person in the identified demographic group i lives in the same household as a person in the actual viewer demographic group j (block 606). For example, the example matrix corrector 220 may apply NPM data to the survey response data to generate the NPM metric, as described above in conjunction with Tables 9-11. The example matrix corrector 220 applies the NPM metric to the database owner impressions redistributed across households to account for the probability P(L) _ij that the first person in the identified demographic group i lives in the same household as a person in the actual viewer demographic group j (block 608). For example, as described above in conjunction with Table 12, the matrix corrector 220 may determine each i, j cell value in Table 12 by multiplying the redistributed audience value in the i, j cell of Table 8 above by the corresponding NPM metric in the i, j cell value in Table 11 above. In this way, the matrix corrector 220 applies the NPM metric to the redistributed impressions collected by the database owner 104a across different households to account for the probability P(L) _ij that the first person in the identified demographic group i lives in the same household as a person in the actual viewer demographic group j.

The example matrix normalizer 222 normalizes the probabilities for each identified demographic group i (block 610). For example, the matrix normalizer 222 normalizes each column in the false positive matrix of Table 14 (e.g., each demographic group identified by the database owner 104a) so that the totals for each column equal the same number (e.g., 1.00 or 100%). The columns are normalized so that the number of impressions after correcting the false positives equals the total number of impressions detected by the database owner 104a, and the proportion of impressions associated with each identified demographic group i by the database owner 104a is maintained. The example instructions 600 of FIG. 6 end, and control returns to, for example, block 510 of FIG. 5.

7 is a flow diagram representing example machine-readable instructions 700 that may be executed to implement the example impression data compensator 200 of FIG2 for calculating a misidentification correction matrix and/or a co-viewing matrix. The example instructions 700 are an alternative example of the example instructions implementing block 402 of FIG4 described above with reference to FIG5.

The example sharing matrix generator 204 of FIG. 2 obtains survey calibration data from a survey calibration data source (e.g., from the calibration data collector 202 of FIG. 2 ) (block 702). For example, the sharing matrix generator 204 may obtain information indicating the devices accessible to persons in a respondent's household and/or the media categories viewed by the respondent on different device types. The example aggregated distribution generator 218 selects a media category (block 704). The example aggregated distribution generator 218 selects a device type (block 706). In the example of FIG. 7 , the media category and/or device type are selected from the media categories and/or device types represented in the survey calibration data. For all pairs of demographic groups i and j, the example aggregated distribution generator 218 calculates the probability P(L) _ij (e.g., based on the survey calibration data source) that a person in identified demographic group i lives in the same household as a person in actual viewer demographic group j (block 708). For example, the aggregated distribution generator 218 may determine the incidence of co-viewing by persons in demographic groups i and j living in the same household.

For all pairs of demographic groups i and j, the example aggregate distribution generator 218 determines the probability P(D|L) _ij that a person in demographic group i and a person in actual viewer demographic group j living in the same household both access a mobile device of the selected device type (block 710).

For all pairs of demographic groups i and j, the example aggregated distribution generator 218 determines the probability P(Sx |D)ij that a person in the identified demographic group i (who is a database owner user) who lives in the same household and accesses a mobile device of the selected device type shares the same mobile device used to view media of the selected media category with a person in the actual viewer demographic group _j (block 712). For example _, the example aggregated distribution generator 218 may determine the incidence of people in demographic groups i and j both accessing media of the selected media category on a device type selected from the survey calibration data.

For all pairs of demographic groups i and j, the example aggregated distribution generator 218 determines the sharing probability γ _ij for the selected pair of demographic groups i, j for the false positive correction matrix associated with the selected device type and the selected media category (block 714). For example, the aggregated distribution generator 218 can calculate the sharing probability γ _ij using the example equation 1 described above.

For all pairs of demographic groups i and j, the example co-viewing matrix generator 224 determines the probability κ _{ij that} : (1) a person in the identified demographic group i is identified by the database owner as a viewer of the media, and (2) an impression also exists for a person in demographic group j who is an actual viewer (block 716). The set of probabilities κ _ij for pairs of demographic groups i and j can be aggregated into a co-viewing matrix. Example instructions for implementing block 716 are described below with reference to FIG8.

The example aggregated distribution generator 218 determines whether there are any additional device types for which the survey calibration data from the survey calibration data source should be processed (block 718). If there are additional device types (block 718), control returns to block 706. If there are no additional device types (block 718), the example aggregated distribution generator 218 determines whether there are additional media categories for which the survey calibration data from the survey calibration data source should be processed (block 720). If there are additional media categories, control returns to block 704. Otherwise, if there are no additional media categories (block 720), the example instructions 700 end, and control returns to, for example, block 404 of FIG. 4.

8 is a flow diagram of example machine-readable instructions 800 that may be executed to calculate a co-viewing matrix. For example, the instructions 800 may be executed by the example co-viewing matrix generator 224 of FIG. 2 to implement block 510 of FIG. 5 and/or block 716 of FIG. 7 .

For all pairs of demographic groups i and j, the example co-viewing matrix generator 224 determines the probability P(Cx|D)ij that a person in the identified demographic group i, who lives in the same household and has access to a mobile device of a selected device type, and a person in the actual viewer demographic group j, concurrently accesses (e.g., views ₎ media of a selected media category using the same mobile device ₍ block 802). In other words, assuming the persons live in the same household and have access to the same mobile device, the example co-viewing matrix generator 224 determines the probability or incidence of co-viewing for the selected device type, media category, and demographic groups i and j. The example co-viewing matrix generator 224 determines a probability _κij for the co-viewing matrix for the selected device type and the selected media category (block 804). For example, the example co-viewing matrix generator 224 may calculate the probability _κij as _κij = P(L) _ij × P(D|L) _ij × P( _Cx |D) _ij . The example probability _κij represents the probability of co-viewing for the selected pair of demographic groups i and j, the selected device type, and the selected media category.

The example instructions 800 end, and control returns to, for example, block 512 of FIG. 5 and/or block 718 of FIG. 7 .

9 is a flow diagram representing example machine-readable instructions 900 that may be executed to implement the example impression data compensator 200 of FIG 2 for calculating demographics (or non-coverage factors) associated with media audiences that are not covered by the database owner. The example instructions 900 may be executed by the example non-coverage calculator 210 of FIG 2 to implement blocks 404 and 405 of FIG 4 .

The example uncovered calculator 210 of FIG2 obtains survey calibration data from a survey calibration data source (block 902). For example, the uncovered calculator 210 may obtain survey calibration data (e.g., from the calibration data collector 202 of FIG2) that indicates the number of persons who are not registered with the database owner 104a and/or who are registered with the database owner 104a but are not logged into the database owner 104a on a particular type of device.

The example non-coverage calculator 210 selects a device type (block 904). The example non-coverage calculator 210 selects a demographic group (block 906). The example non-coverage calculator 210 selects a media category (block 908). In the example of FIG. 9 , the media category, demographic group, and/or device type are selected from the media categories, demographic groups, and/or device types represented in the survey calibration data. For example, the non-coverage calculator 210 can generate different alpha factors (e.g., the example alpha factor α=B/A of Table 17 ) for different combinations of demographic group, media category, and device type based on the survey calibration data obtained from the survey calibration data source.

The example non-coverage calculator 210 determines a probability B based on survey calibration data (e.g., data from a survey calibration data source), where B is a probability that a person in the selected demographic group views a media item in the selected media category on the selected device type (block 910). In the example of FIG9 , the non-coverage calculator 210 determines the probability B based on weights associated with the selected device type, the selected demographic group, and/or the selected media category determined from the survey calibration data.

The example uncovered calculator 210 also determines a probability A based on the survey calibration data, where A is the probability that a person in the selected demographic group views a media item in the selected media category on a device type other than the selected device type (e.g., a standard device type associated with the media item) (block 912). In the example of FIG9, the uncovered calculator 210 determines the probability A based on weights associated with the other device type determined from the survey calibration data, the selected demographic group, and/or the selected media category. In the example of FIG9, the probabilities B and A are determined for media that is accessible via both the selected device type and the other device type.

For example, for episodes of a television program in a particular media category, the other device type may be a television, and the selected device type may be a computing device (e.g., a mobile device and/or a more specific type of mobile device, such as a smartphone, tablet, and/or portable media player) on which the episodes of the television program can be accessed via streaming video. Such television programs are often made available via streaming video after an initial or premiere presentation on the television. Thus, a person can access episodes of the television program via streaming media on a television (e.g., the other device type) and/or on a computing device (e.g., the selected device type).

The example non-coverage calculator 210 determines an alpha factor (e.g., alpha = B/A of Table 17 above) for the selected demographic group, the selected media category, and/or the selected device type (block 914). Thus, in the example of FIG. 9 , each combination of demographic group, media category, and device type has a separate alpha factor. However, in other examples, the alpha factor may be the same for each demographic group, each media category, and/or each device type. Additionally or alternatively, the example non-coverage calculator 210 determines the alpha factor using a combination of factors including, for example, geographic region, site, time of day, and/or other factors.

The example no-cover calculator 210 determines whether additional media categories exist (block 916). If additional media categories exist (block 916), control returns to block 908 to select another media category. When no more media categories exist (block 916), the example no-cover calculator 210 determines whether additional demographic groups exist (block 918). If additional demographic groups exist (block 918), control returns to block 906 to select another demographic group. When no more demographic groups exist for the selected device type (block 918), the example no-cover calculator 210 determines whether additional device types exist (block 920). If additional device types exist (block 920), control returns to block 904 to select another device type. When no additional device types exist for the selected media category (block 920), the example instructions 900 of FIG. 9 end, and control returns to, for example, block 406 of FIG. 4.

10 is a flow diagram representing example machine-readable instructions 1000 that may be executed to implement the example impression data compensator 200 of FIG2 for adjusting impression and/or duration units based on a false positive correction matrix. The example instructions 1000 may be executed by the example false positive corrector 206 of FIG2 to implement block 414 of FIG4.

The exemplary misidentification corrector 206 of FIG. 2 determines misidentification-adjusted impressions by calculating the dot product of: a) the misidentification correction matrix corresponding to the selected media category and the selected device type, and b) the impression count identified by the database owner for each identified demographic group i (block 1002). The result of this dot product is the misidentification-adjusted impression for each demographic group. For example, the misidentification corrector 206 can calculate the dot product of the misidentification correction matrix of Table 14 and the impression information of Table 15 to obtain the misidentification-adjusted impressions.

The example false positive corrector 206 of FIG. 2 determines the duration units for the false positive adjustment by calculating the dot product of: a) the false positive correction matrix corresponding to the selected media category and the selected device type, and b) the duration units identified by the database owner for each identified demographic group i (block 1004). Example 1000 ends and control returns to, for example, block 416 of FIG. 4.

11 is a flow diagram representing example machine-readable instructions 1100 that may be executed to implement the example impression data compensator 200 of FIG 2 that adjusts impressions based on a non-coverage factor for a database owner. The example instructions 1100 may be executed by the example non-coverage corrector 212 of FIG 2 to implement block 416 of FIG 4 .

The example uncover corrector 212 of FIG. 2 selects a demographic group (block 1102). The selected demographic group may be the identified demographic group i and/or the actual viewer demographic group j discussed above with reference to Tables 1-14. The uncover corrector 212 determines uncovered impressions by multiplying the misidentification-adjusted impressions by an alpha factor associated with the selected demographic group, the selected device type, and/or the selected media category (e.g., alpha = B/A of Table 17 above) (block 1104). For example, the uncover corrector 212 may determine an alpha factor (e.g., determined in block 404 of FIG. 4 and/or in the example instructions 900 of FIG. 6 ) that is applicable to the selected demographic group (e.g., selected in block 1102), the selected device type (e.g., selected in block 412 of FIG. 4 ), and/or the selected media category (e.g., selected in block 410 of FIG. 4 ). In block 1104, the example uncover corrector 212 multiplies the determined alpha factor by the misidentification-adjusted impressions to determine uncovered impressions.

The example uncovered corrector 212 determines whether there are additional demographic groups (block 1106). If there are additional demographic groups (block 1106), control returns to block 1102 to select another demographic group. When there are no more demographic groups to adjust (block 1106), the example uncovered corrector 212 scales the uncovered impressions for the demographic groups (e.g., all demographic groups in the same group calculation) so that the sum of the uncovered impressions for all demographic groups equals the number of observed uncovered impressions (block 1108). For example, as described above with reference to Table 17, the exemplary uncovered corrector 212 may 1) calculate an adjusted percentage (Adj% of Table 17) for each demographic group (Demos) as the product of an alpha factor (α=B/A) and a measured percentage (Meas% of Table 17); 2) normalize the adjusted percentage (Adj% of Table 17) to sum to 100%; and 3) multiply the normalized percentage (Norm% of Table 17) by the uncovered impressions (e.g., the number of impressions not associated to a demographic group (Demos of Table 17) by the database owner 104a) to obtain a scaled number of uncovered impressions (Uncovered Impressions of Table 17) attributed to each demographic group (Demos of Table 17).

The example uncover corrector 212 of FIG2 selects a demographic group (block 1110). The demographic group selected in block 1110 is the demographic group previously selected in the iteration of block 1102. In the illustrated example, blocks 1110, 1112, and 1114 are executed to process the scaled uncovered impressions for all demographic groups determined at blocks 1102, 1104, 1106, and 1108. The uncover corrector 212 determines the misidentified and uncovered-adjusted impressions for the selected demographic group by adding the scaled uncovered impressions (e.g., as determined in block 1108) to the misidentified-adjusted impressions for the demographic group (block 1112).

The example uncovered corrector 212 determines whether additional demographic groups exist (block 1114). If additional demographic groups exist (block 1114), control returns to block 1110 to select another demographic group. When there are no more demographic groups to adjust (block 1114), the example instructions 1100 end and control returns to, for example, block 418 of FIG. 4.

Although the above examples are described with reference to impressions, these examples can additionally or alternatively be applied to special audiences and/or duration units. For example, the redistribution audience matrix can be applied to the audience of the media of interest, rather than to the impressions described above. In some examples, only one impression needs to be counted as a person in the audience of the media, and a person is not counted in the audience of the media when he or she does not have an impression. Therefore, for example, in the example of Table 3, after the redistribution of impressions, the audience for males aged 12-17 will be reduced by 1, because males aged 12-17 are reported by database owner 104a as accessing media, but after redistribution, it is determined that they do not have impressions of media in the comedy category on the tablet, thereby reducing the audience for males aged 12-17 by 1. On the contrary, the audience for females aged 35-44 will increase by 1, because females aged 35-44 are not associated with any impressions by database owner 104a and do not form an audience, but after the redistribution of the audience, it can be determined that females aged 35-44 do in fact access the media of the category of interest on the tablet.

12 is a flow diagram representing example machine-readable instructions 1200 that may be executed to implement the example impression data compensator 200 of FIG 2 for calculating demographics (or non-coverage factors) associated with media audiences that are not covered by the database owner. The example instructions 1200 may be executed by the example non-coverage calculator 210 of FIG 2 to implement blocks 404 and 405 of FIG 4 .

The example non-coverage calculator 210 obtains data from a calibration survey (block 1202). For example, the non-coverage calculator 210 may obtain survey calibration data indicating the number of respondents who are not registered with the database owner 104a and/or who are registered with the database owner 104a but are not logged into the database owner 104a on a particular type of device.

The example non-coverage calculator 210 selects a device type (block 1204) and selects a demographic group (block 1206). The example non-coverage calculator 210 determines a non-coverage factor for the selected demographic group and the selected device type as a portion of respondents to the calibration survey in the selected demographic group that would not be recognized by the database owner 104a when viewing media on the selected device type (block 1208).

The example no-cover calculator 210 determines whether additional demographic groups exist (block 1210). If additional demographic groups exist (block 1210), control returns to block 1206 to select another demographic group. When no more demographic groups exist for the selected device type (block 1210), the example no-cover calculator 210 determines whether additional device types exist (block 1212). If additional device types exist (block 1212), control returns to block 1204 to select another device type. When no additional device types exist (block 1212), the example instructions 1200 of FIG. 12 end, and control returns to the calling function or process, such as the example instructions 400 of FIG. 4 , for example, to continue execution at block 406.

13 is a flow diagram representing example machine-readable instructions 1300 that may be executed to implement the example impression data compensator 200 of FIG 2 that adjusts impression counts and/or audience size based on a non-coverage factor for a database owner. The example instructions 1300 may be executed by the example non-coverage corrector 212 of FIG 2 to implement block 416 of FIG 4 .

The example non-coverage corrector 212 of FIG2 selects a demographic group (block 1302). The example non-coverage corrector 212 determines a special audience size adjusted for misidentifications and non-coverage errors by dividing the misidentification-adjusted special audience size (determined using the instructions of FIG10) by 1-(non-coverage factor for the selected demographic group and the selected device type) (block 1304).

The example non-cover corrector 212 determines a misidentified and non-cover adjusted impression count by multiplying the misidentified and non-cover adjusted special audience size by the frequency determined from the impression information for the selected demographic group and the selected device type (block 1306 ).

The example non-coverage corrector 212 determines the false positive and non-coverage error adjusted duration units by dividing the false positive adjusted duration units (determined using the instructions of FIG. 10 ) by 1−(non-coverage factor for the selected demographic group and the selected device type) (block 1308 ).

The example uncovered corrector 212 determines whether additional demographic groups exist (block 1310). If additional demographic groups exist (block 1310), control returns to block 1302 to select another demographic group. When there are no more demographic groups to adjust (block 1308), the example instructions 1300 end, and control returns to the calling function or process, such as the example instructions 400 of FIG. 4, to continue execution at block 418.

In some instances, data selection can be performed based on artistic genre. In some examples, artistic genre can be provided by media publisher 120 (e.g., a television network provider, a media provider, etc.). In other examples, artistic genre may not be available from the media provider. In such other examples, when artistic genre is not readily available, the artistic genre can be predicted using techniques such as those disclosed below in conjunction with FIG. 14 to FIG. 20 .

In cross-platform audience measurement, an AME (e.g., AME 108 in FIG. 1 ) monitors the audience for a television (and/or radio) program as it is broadcast by a television (and/or radio) station (e.g., a local television station, a local radio station) and provided via a delayed media service for a specified period of time after the broadcast. Exemplary delayed media services can be accessed on mobile devices or fixed computing devices via on-demand services, streaming service applications, apps, web pages, and the like. Exemplary delayed media services can also be provided on digital video recorders (DVRs), cable boxes, internet media delivery set-top boxes, smart TVs, and the like. In this manner, AME performs cross-platform audience measurement by collecting impressions for both the television (and/or radio) audience and the delayed audience. In some examples, the impressions are analyzed to generate media ratings. In some examples, art genres are used to determine adjustment factors for calibrating mobile census data. In such examples, the calibrated mobile census data is used to generate ratings. Art genres are media categories based on similarities in form, style, and/or subject matter, for example. Exemplary art genres include comedy, drama, sports, news, children's programming, etc.

In some examples, impressions are collected based on data encoded in the media. In some examples, the encoded data does not include genre information. For example, programs broadcast by local broadcasters do not have genre information, and the encoded data for local broadcasts does not include program names or identifiers. Furthermore, genre information or program names are not provided by the broadcaster or media owner in a timely manner to generate overnight ratings. Overnight ratings are often used to quickly assess the success of broadcast programs based on impression counts and audience size. In some instances, the success of a broadcast program reflected in overnight ratings can be used by a media network (e.g., a television network) to sell advertising space corresponding to delayed viewing access to the same media, for example, through an app or webpage of a media network (e.g., ABC Television) or a third party (e.g., Hulu.com) that owns the rights to provide the media through a delayed viewing service. The examples disclosed herein can be used to generate a model that predicts the genre of local television (and/or radio) impressions. Although the examples below are described with respect to television, the examples below can alternatively be implemented using radio or other media types.

Art Genre Predictive Modeling

The above examples of misidentification and/or non-coverage correction can include generating multiple misidentification correction tables (e.g., such as Table 14 above). In some examples, misidentification correction matrices are generated for different device types and/or art types for different media. The following examples can be used to predict the media art type of an impression for which no art type information is provided from the impression. By assigning art types to impressions, the examples disclosed herein select an appropriate misidentification correction matrix and/or non-coverage factor (the alpha factor of Table 19 and/or Table 21 above) to compensate for misidentification errors and/or non-coverage errors in impression counts and/or audience sizes as disclosed above.

The examples disclosed below of generating an art type prediction model use a home media meter installed locally at a client computer or television to collect information about media accessed or tuned to to facilitate generating an art type prediction model. However, the examples disclosed above generate misidentification correction factors without relying on panelist metering software installed locally at the client computer to collect database owner registration status data and/or collect data about media accessed by family members. The use of a home media meter installed locally at a client computer or television in art type prediction modeling and art type assignment is limited to collecting impressions and/or assigning art types to impressions. Such locally installed home meters are not used to generate misidentification correction matrices or non-coverage factors as disclosed herein. The art type prediction modeling and art type assignment disclosed below can be used to correct impressions for misidentification and/or non-coverage as disclosed above.

FIG14 illustrates an exemplary system 1400 that generates an art genre prediction model and predicts an art genre and/or assigns an art genre to non-art genre impression data 1402 (e.g., impression data that does not include art genre information). In the illustrated example, a household media meter 1404 (e.g., a set-top meter, a personal meter, etc.) sends local television impression data 1402 to an audience measurement entity (AME) 108. The exemplary AME 108 of FIG14 may be the AME 108 of FIG1 .

In the illustrated example, impression data 1402 generated from local broadcast television is not encoded with artist type information because the local television broadcaster has not yet encoded the media at the time of its broadcast. In the illustrated example, in order to monitor cross-platform audiences, in addition to collecting local television impression data 1402 for television audiences, an exemplary collector 1406 is provided in the network-enabled media device 1410 to collect impressions of media accessed using streaming services via the network-enabled media device 1410. In the illustrated example, the collector 1406 is implemented using instructions incorporated into a webpage, application, or application (e.g., an on-demand application, a streaming application, a DVR access application, etc.) executed by the network-enabled media access device 1410 (e.g., a computer, television, tablet, smartphone, e-book reader, etc.).

Examples of using the exemplary collector 1406 to collect impressions are disclosed in U.S. Patent No. 8,370,489, issued on February 5, 2013; U.S. patent application serial number 14/127,414, filed on December 18, 2013; U.S. patent application serial number 14/329,779, filed on July 11, 2014; U.S. provisional application serial number 61/952,726, filed on March 13, 2014; U.S. provisional application serial number 62/014,659, filed on June 19, 2014; and U.S. provisional application serial number 62/023,675, filed on July 11, 2014, the entire contents of which are incorporated herein by reference. The example media meter 1404 and/or the example collector 1406 communicate with the example AME 108 via an example network 1412 (e.g., the Internet, a local area network, a wide area network, a cellular data network, etc.) via a wired connection and/or a wireless connection (e.g., a cable/DSL/satellite modem, a base station, etc.).

14 , AME 108 processes non-art type impression data 1402 to determine an art type for impressions recorded at AME 108 based on the non-art type impression data 1402. In the illustrated example, AME 108 includes a collection database 1414, an art type prediction modeler 1416, an art type predictor 1418, a historical art type database 1420, and a prediction database 1422. The example collection database 1414 is provided to store impressions recorded based on the example non-art type impression data 1402 received and/or retrieved from the example media meter 1404 and/or the example collector 1406.

The example art genre prediction modeler 1416 generates one or more art genre prediction models 1428 based on historical art genre data stored in the example historical art genre database 1420. In the illustrated example, the historical art genre data includes historical program records 1430 that, for example, identify the program's art genre, the program's time slot, the program's day category, and the program's duration. The example historical program records 1430 are based on information provided by a local broadcaster 1432 after the program is broadcast (e.g., a program identifier, one or more timestamps, a station identifier, a member identifier, etc.). For example, the local broadcaster 1432 may provide records 1430 for every quarter-hour of the locally broadcast program for a week following the local broadcast.

The example art type predictor 1418 uses one or more art type prediction models 1428 generated by the example art type prediction modeler 1416 to predict art type information and assign predicted art types to impressions recorded in the collection database 1414 based on the non-art type impression data 1402. In this manner, the example art type predictor 1418 generates predicted art type impression records 1424 (e.g., non-art type impressions recorded in the collection database 1414 combined with predicted art type information). In the illustrated example of FIG14 , the predicted art type impression records 1424 are stored in the prediction database 1422. In some examples, the predicted art type impression records 1424 are used as uncorrected impressions that can compensate for misidentifications and under-coverage to determine the demographic group to which the predicted art type impression belongs, as discussed above.

FIG15 illustrates an exemplary implementation of the art type predictor 1418 of FIG14 , which determines art type information for a non-art type impression record 1501 recorded in the collection database 1414 based on the non-art type impression data 1402 of FIG14 . In the illustrated example, the art type predictor 1418 receives and/or retrieves the non-art type impression record 1501 from the collection database 1414 and generates a predicted art type impression record 1424, which is stored in the prediction database 1422 and/or reported in the prediction report 1426. In the illustrated example of FIG15 , the art type predictor 1418 includes a data classifier 1500 and an art type analyzer 1502. The exemplary data classifier 1500 is configured to transform the exemplary non-art type impression record 1501 into classified impression data 1504. The classified impression data 1504 will be described in more detail below with reference to FIG16 .

The example art type analyzer 1502 is configured to apply one or more art type prediction models 1428 generated by the example art type prediction modeler 1416 to the example classified impression data 1504. The example art type analyzer 1502 applies the one or more art type prediction models 1428 to the classified impression data 1504. The example art type analyzer 1502 then assigns a predicted art type to the classified impression data 1504 to generate an example predicted art type impression record 1424. The example art type analyzer 1502 stores the example predicted art type impression data 1424 in the example prediction database 1422 and/or includes the example predicted art type impression data 1424 on an example prediction report 1426.

FIG16 illustrates an exemplary method by which the data classifier 1500 of FIG15 classifies a non-artistic impression data record 1501, which is then used by the art genre analyzer 1502 of FIG15 to predict an art genre. In the illustrated example, the non-artistic impression data record 1501 includes an exemplary market identifier (ID) 1600, an exemplary station ID 1602, an exemplary affiliate ID 1604, an exemplary start timestamp 1606, an exemplary end timestamp 1608, an exemplary program ID 1610, and an exemplary source ID 1612. Any one or more of these fields may be omitted, and one or more additional fields may be present. For example, a non-artistic impression data record 1501 corresponding to a national program may include a program ID 1610 and a source ID 1612. Additionally, a non-artistic impression data record 1501 corresponding to a locally broadcast program may omit the program ID 1610 and the source ID 1612. Exemplary market ID 1600 identifies the market (e.g., geographic location, municipality, etc.) in which the corresponding television program is broadcast. In some examples, market ID 1600 is assigned by AME 108 ( FIG. 14 ) or any other suitable entity. Station ID 1602 identifies the local station that broadcasts the television program. In some examples, station ID 1602 is an alphanumeric number assigned by AME 108 ( FIG. 14 ) or any other suitable entity. In some examples, station ID 1602 is a call letter used by a local television station to identify itself (e.g., WXTV, KVEA, WSNS, etc.).

Example affiliate ID 1604 identifies the network affiliate of a local television station (e.g., ABC, Telemundo, Fox, etc.). In some examples, affiliate ID 1604 is a number assigned by AME 108 that represents the network affiliate (e.g., 100 for ABC, 101 for NBC, etc.). In some examples, affiliate ID 1604 may be an alphabetical representation of the name of the network affiliate (e.g., "ABC" for ABC, "TMD" for Telemundo, etc.). Example start timestamp 1606 includes the start date and start time of the program associated with non-artistic impression record 1501. Example end timestamp 1608 includes the end date and end time of the program associated with non-artistic impression record 1501. Example program ID 1610, if included in non-artistic impression record 1501, identifies the television program associated with non-artistic impression record 1501 (e.g., "Arrested Development," "News Radio," "Sabado Gigante," etc.). In some examples, program ID 1610 is a numeric or alphanumeric identifier uniquely associated with the title and/or other information of the corresponding television program in a reference database.Exemplary source ID 1612 identifies the source of the television program (eg, production company, etc.).

16 , the classified impression data 1504 includes an example market ID 1600, an example station ID 1602, an example member ID 1604, an example day category 1614, an example time period 1616, and an example duration 1618. The example data classifier 1500 uses the start timestamp 1606 and the end timestamp 1608 of the non-art type impression record 1501 to determine the example day category 1614, the example time period 1616, and the example duration 1618 of the classified impression data 1504. In the illustrated example, the duration 1618 of the classified impression data 1504 is calculated by the data classifier 1500 as the time difference between the start time marked in the start timestamp 1606 and the end time marked in the end timestamp 1608.

In the illustrated example, data classifier 1500 determines example day category 1614 using the start date in example start timestamp 1606 and the end date in example end timestamp 1608. In some examples, day category 1614 may have a value for a day of the week (e.g., Monday, Tuesday, Wednesday, etc.). In some examples, day category 1614 may be a reduced set of days of the week. For example, the reduced set may include values for "weekend," "Saturday," or "Sunday." In some examples, when the start date indicated in example start timestamp 1606 and the end date indicated in example end timestamp 1608 are different, data classifier 1500 may assign a day to day category 1614 based on the amount of time the program was broadcast on that day. For example, if start timestamp 1606 is "7/25/2014 23:30" (e.g., 30 minutes on a Friday) and end timestamp 1608 is "7/26/2014 1:00" (e.g., 60 minutes on a Saturday), then "Saturday" would be assigned to day category 1614. In such examples, the data classifier 1500 assigns days with a larger portion of broadcast programming to the day category 1614 .

The example data classifier 1500 uses the start time indicated in the example start timestamp 1606 and the end time indicated in the example end timestamp 1608 to determine an example time period 1616 of the example classified impression data 1504. As discussed further below in conjunction with FIG17 , a time period is a time segment or partition of a day (e.g., nighttime, weekend mornings, weekday daytime, evening hours, prime time, weekend daytime, etc.) during which different types of television programs are broadcast, advertising is purchased, and/or audience ratings are measured. In some examples, the time period 1616 of FIG16 may be a time period that includes the start time indicated in the start timestamp 1606 and the end time indicated in the end timestamp 1608.

Although an exemplary manner of implementing the art type predictor 1418 of FIG. 14 is shown in FIG. 15 , one or more of the elements, processes, and/or devices shown in FIG. 15 may be combined, divided, rearranged, omitted, eliminated, and/or implemented in any other manner. Additionally, the exemplary data classifier 1500, the exemplary art type analyzer 1502, and/or, more generally, the exemplary art type predictor 1418 of FIG. 14 may be implemented using hardware, software, firmware, and/or any combination of hardware, software, and/or firmware. Thus, for example, any of the exemplary data classifier 1500, the exemplary art type analyzer 1502, and/or, more generally, the exemplary art type predictor 1418 may be implemented using one or more analog or digital circuits, logic circuits, one or more programmable processors, one or more application specific integrated circuits (ASICs), one or more programmable logic devices (PLDs), and/or one or more field programmable logic devices (FPLDs). When any of the device or system claims of this patent are read to cover a purely software and/or firmware implementation, at least one of the exemplary data classifier 1500, the exemplary art type analyzer 1502, and/or the exemplary art type predictor 1418 is hereby expressly defined to include a tangible computer-readable storage device or storage disk, such as a memory, a digital versatile disk (DVD), a compact disk (CD), a Blu-ray disk, etc., storing the software and/or firmware. Additionally, the exemplary art type predictor 1418 may include one or more elements, processes, and/or devices in addition to or in place of those shown in FIG. 15 , and/or may include more than one of any or all of the illustrated elements, processes, and devices.

FIG17 shows an exemplary chart depicting time periods (such as those indicated in the exemplary time periods 1616 of FIG16 ) and day categories (such as those indicated in the exemplary day categories 1614 of FIG16 ) used by the art genre predictor 1418 of FIG14 and FIG15 . In the illustrated example, each day of the week 1700 is divided into a plurality of time periods 1702 . In the illustrated example, Monday through Friday are divided into identical time period segments or partitions (e.g., time periods having the same duration and delimited by the same time period start and end times). Alternatively, each day of the week 1700 , or any combination thereof, may be divided into different time periods. In the illustrated example, the day categories 1704 are divided into three categories: “weekdays,” “Saturdays,” and “Sundays.” Alternatively, the day categories 1704 may include separate categories for each day of the week (e.g., a “Monday” category, a “Tuesday” category, etc.). In some examples, the day categories may have different day groupings based on similar characteristics between the different days being grouped together (e.g., temporal proximity, audience viewing habits, etc.). For example, day categories 1704 may include an “early week” category (e.g., including Monday and Tuesday), a “mid-week” category (e.g., including Wednesday and Thursday), a “late week” category (e.g., including Friday), and a “weekend” category (e.g., including Saturday and Sunday).

FIG18 illustrates an example of the art type prediction modeler 1416 of FIG14 , which constructs one or more art type prediction models 1428 to be used by the art type predictor 1418 of FIG14 . In the illustrated example, the art type prediction modeler 1416 includes an exemplary art type prediction model builder 1802, an exemplary art type prediction model evaluator 1804, and an exemplary art type prediction model comparator 1806. The exemplary art type prediction model builder 1802 uses a dataset from the historical art type database 1420 to generate one or more candidate models using various modeling techniques. Exemplary modeling techniques that can be used to implement the art type prediction modeler 1416 include logistic regression, linear discriminant analysis, quadratic discriminant analysis, k-nearest neighbor algorithms, and the like. The exemplary model builder 1802 retrieves an exemplary training dataset 1808 from the historical art type database 1420. In the illustrated example, a training dataset 1808 is used to establish a predictive relationship between the inputs of one or more art-type prediction models 1428 (e.g., variables included in the classified impression data 1504 of FIG. 15 ) and the outputs of one or more art-type prediction models 1428 (e.g., predicted art-type). The exemplary training dataset 1808 includes classified impression data (e.g., exemplary market identifier 1600, exemplary station ID 1602, exemplary member ID 1604, exemplary day category 1614, exemplary time period 1616, and exemplary duration 1618) and assigned art-types. Using the training dataset 1808, the exemplary art-type prediction model builder 1802 generates one or more candidate models 1800. The candidate models 1800 are art-type prediction models that are evaluated for their suitability (e.g., accuracy) for use by the art-type analyzer 1502 ( FIG. 15 ) to assign art-types to non-art-type impression records 1501 ( FIG. 14 ). In some examples, after generating the exemplary candidate model 1800, the art genre prediction model builder 1802 calculates a correct discrimination rate (CCR) using the training dataset 1808. In some examples, the CCR is the percentage of impressions in the training dataset 1808 that are correctly predicted by the candidate model.

In the example shown in FIG18 , an art type prediction model evaluator 1804 receives and/or retrieves a candidate model 1800 generated by an art type prediction model builder 1802. The art type prediction model evaluator 1804 also retrieves a test dataset 1810 from the historical art type database 1420. The example art type prediction model evaluator 1804 validates the candidate model 1800 using the test dataset 1810. In some examples, the art type prediction model evaluator 1804 calculates a CCR for the candidate model using the test dataset 1810. In some such examples, the art type prediction model evaluator 1804 calculates a CCR for the art type of interest included in the test dataset 1810.

18 , the art type prediction model comparator 1806 uses the selection criteria defined by the AME 108 ( FIG. 14 ) to select one or more of the candidate models 1800 as the one or more art type prediction models 1428. In some examples, the selection criteria include the highest CCR, the shortest processing time, the smallest resource requirements, etc. In some examples, the art type prediction model comparator 1806 selects the candidate model 1800 with the highest CCR.

Although an exemplary manner of implementing the art type prediction modeler 1416 of FIG. 14 is shown in FIG. 18 , one or more of the elements, processes, and/or devices shown in FIG. 18 may be combined, divided, rearranged, omitted, eliminated, and/or implemented in any other manner. Furthermore, the exemplary art type prediction model builder 1802, the exemplary art type prediction model evaluator 1804, the exemplary art type prediction model comparator 1806, and/or more generally, the exemplary art type prediction modeler 1416 of FIG. 14 may be implemented using hardware, software, firmware, and/or any combination of hardware, software, and/or firmware. Thus, for example, any of the exemplary art type prediction model builder 1802, the exemplary art type prediction model evaluator 1804, the exemplary art type prediction model comparator 1806, and/or more generally, the exemplary art type prediction modeler 1416 of FIG. 14 may be implemented using one or more analog or digital circuits, logic circuits, one or more programmable processors, one or more application specific integrated circuits (ASICs), one or more programmable logic devices (PLDs), and/or one or more field programmable logic devices (FPLDs). When any of the device or system claims of this patent are read to cover a purely software and/or firmware implementation, at least one of the exemplary art type prediction model constructor 1802, the exemplary art type prediction model evaluator 1804, the exemplary art type prediction model comparator 1806, and/or the exemplary art type prediction modeler 1416 is hereby expressly defined as comprising a tangible computer-readable storage medium or storage disk, such as a memory, a digital versatile disc (DVD), a compact disc (CD), a Blu-ray disc, etc., storing software and/or firmware. Additionally, the exemplary art type prediction modeler 1416 may include one or more elements, processes, and/or devices in addition to or in place of those shown in FIG. 18 , and/or may include more than one of any or all of the illustrated elements, processes, and devices.

A flowchart representing exemplary machine-readable instructions for implementing the exemplary art genre prediction modeler 1416 of FIG. 14 and FIG. 18 is shown in FIG. 20 . A flowchart representing exemplary machine-readable instructions for implementing the exemplary art genre predictor 1418 of FIG. 14 and FIG. 15 is shown. In these examples, the machine-readable instructions include instructions for executing one or more programs by a processor, such as the processor 2112 shown in the exemplary processor platform 2100 discussed below with reference to FIG. 21 . The one or more programs may be embodied in software stored on a tangible computer-readable storage medium, such as a CD-ROM, floppy disk, hard drive, digital versatile disk (DVD), Blu-ray disc, or memory associated with the processor 2112, although all of the one or more programs and/or portions thereof may alternatively be executed by a device other than the processor 2112 and/or embodied in firmware or dedicated hardware. 19 and 20 , many other methods of implementing the example art type predictor 1418 and/or the example art type prediction modeler 1416 may alternatively be used. For example, the order of execution of the blocks may be changed, and/or some of the blocks described may be changed, eliminated, or combined.

As described above, the exemplary processes of Figures 19 and/or 20 may be implemented using coded instructions (e.g., computer and/or machine readable instructions) stored on a tangible computer-readable storage medium, such as a hard drive, flash memory, read-only memory (ROM), compact disc (CD), digital versatile disc (DVD), cache, random access memory (RAM), and/or any other storage device or storage disk in which information is stored for any duration (e.g., for an extended period, permanently, temporarily, temporarily buffered, and/or a cache of information). Additionally or alternatively, the exemplary processes of Figures 19 and/or 20 may be implemented using coded instructions (e.g., computer and/or machine readable instructions) stored on a non-volatile computer and/or machine readable medium, such as a hard drive, flash memory, read-only memory, compact disc, digital versatile disc, cache, random access memory, and/or any other storage device or storage disk in which information is stored for any duration (e.g., for an extended period, permanently, temporarily, temporarily buffered, and/or a cache of information).

FIG19 is a flow diagram representing exemplary machine-readable instructions 1900 that can be executed to implement the exemplary art type prediction modeler 1416 of FIG14 and FIG18 for constructing one or more art type prediction models (e.g., one or more art type prediction models 1428 of FIG18 ) for predicting art type for non-art type impressions recorded based on non-art type impression data 1402 ( FIG14 ). Initially, at block 1902, the art type prediction model builder 1802 ( FIG18 ) selects variables or parameters (e.g., market identifier, station ID, member ID, day category, time period, and duration) from among the impressions included in a training dataset (e.g., training dataset 1808 of FIG18 ). In some examples, the art type prediction model builder 1802 selects the variables by performing statistical significance tests on the variables (e.g., calculating p-values, performing chi-square tests, etc.). At block 1903, the example art-type prediction model builder 1802 excludes impressions from the training dataset 1808 based on selection criteria (e.g., excluding impressions associated with art-types of no interest, excluding impressions whose duration of contact or visit is below a threshold, etc.). For example, some records 1430 ( FIG. 14 ) in the historical art-type database 1420 ( FIG. 14 ) may be classified in the “unclassified” art-type. In some instances, it may be undesirable for the art-type prediction model 1428 to predict the “unclassified” art-type for impressions. In such examples, to prevent the art-type prediction model 1428 from assigning non-art-type impressions 1402 ( FIG. 14 ) to the “unclassified” art-type, records 1430 in the historical art-type database 1420 that are classified in the “unclassified” art-type are excluded from the training dataset 1808.

At block 1904, the example art-type prediction model builder 1802 constructs a candidate model 1800 ( FIG. 18 ) using the variables or parameters selected at block 1902 based on the training dataset 1808. Example modeling techniques that can be used to generate the candidate model 1800 include logistic regression, linear discriminant analysis, quadratic discriminant analysis, the k-nearest neighbor algorithm, and the like. For example, the art-type prediction model builder 1802 can construct the model using the k-nearest neighbor algorithm, where k equals 9. In some examples, the art-type prediction model builder 1802 calculates the CCR of the candidate model 1800 using the training dataset 1808. At block 1906, the art-type prediction model evaluator 1804 ( FIG. 18 ) evaluates the candidate model 1800 generated at block 1904. In some examples, the art-type prediction model evaluator 1804 retrieves one or more test datasets 1810 ( FIG. 18 ) from the historical art-type database 1420 ( FIG. 14 ) and uses the one or more test datasets 1810 to calculate one or more CCRs for the candidate model 1800. In some such examples, the art genre prediction model evaluator 1804 calculates separate one or more CCRs for each art genre represented in the one or more test datasets 1810 .

At block 1908, the art type prediction model evaluator 1804 determines whether the one or more CCRs generated at block 1906 are acceptable. If the one or more CCRs are acceptable, program control proceeds to block 1912. Otherwise, if the one or more CCRs are unacceptable, program control proceeds to block 1910. At block 1910, the art type prediction model builder 1802 refines or adjusts the candidate model 1800 generated at block 1904. In some examples, the art type prediction model builder 1802 adjusts model parameters (e.g., selecting a different value for k, etc.), adjusts input variables (e.g., using different day categories, adding and/or removing variables, etc.), and/or adjusts the training data set 1808 (e.g., increasing the size of the training data set 1808, changing the composition of the training data set 1808 to include impressions that were previously excluded based on, for example, the type of art type, etc.).

At block 1912, the art type prediction model evaluator 1804 determines whether to generate another candidate model 1800. In some examples, the art type prediction model evaluator 1804 uses a CCR threshold to determine whether to construct another candidate model 1800. In some examples, the art type prediction model evaluator 1804 evaluates the candidate models 1800 based on processing speed and/or computational resource requirements. If another candidate model 1800 is to be constructed, program control returns to block 1902. Otherwise, if another candidate model 1800 is not to be constructed, program control proceeds to block 1914. At block 1914, the art type prediction model comparator 1806 ( FIG. 18 ) compares the candidate models 1800 generated by the art type prediction model builder 1802. In some examples, the art type prediction model comparator 1806 compares the CCRs of the candidate models 1800. In some such examples, for the art type of interest, the art type prediction model comparator 1806 may compare the individual CCRs corresponding to the candidate models 1800. For example, a first candidate model may have an overall CCR of 70% (e.g., the average of the CCRs for the art genre of interest, etc.), but a CCR of 30% for the "family" art genre. A second candidate model may have an overall CCR of 65%, but a CCR of 57% for the "family" art genre.

At block 1916, the art-type prediction model comparator 1806 selects a candidate art-type prediction model 1800 as the art-type prediction model 1428 to be used by the art-type prediction model predictor 1418. In some examples, the art-type prediction model comparator 1806 selects the candidate model 1800 with the highest overall CCR. For example, from the example above, the art-type prediction model comparator 1806 would select the first candidate model. In some examples, the art-type prediction model comparator 1806 selects the candidate model 1800 with the highest CCR for a particular art-type of interest. For example, from the example above, if the "family" art-type is of particular interest, the art-type prediction model comparator 1806 would select the second candidate model. This is because the second candidate model's CCR for the "family" art-type is higher than the first candidate model's CCR for the "family" art-type. In some examples, the art-type prediction model comparator 1806 selects the candidate model 1800 based on performance factors (e.g., the processing power required to predict the art-type, the processing speed required to predict the art-type, etc.). The example process 1900 of FIG. 19 then ends.

FIG20 is a flow diagram representing exemplary machine-readable instructions 2000 that may be executed to implement the exemplary art genre predictor 1418 of FIG14 and FIG15 for predicting the art genre of collected impressions of local television programs that are not encoded with art genre information (e.g., non-art genre impression data 1402 of FIG14 ). At block 2002, the exemplary data classifier 1500 ( FIG15 and FIG16 ) retrieves non-art genre impression data (e.g., non-art genre impression record 1501 based on non-art genre impression data 1402 of FIG14 and FIG15 ) from the collection database 1414 ( FIG14 and FIG15 ). At block 2004, the data classifier 1500 transforms the non-art genre impression record 1501 into classified impression data 1504 ( FIG15 ). In some examples, the data classifier 1500 uses the timestamp values included in the non-art type impression record 1501 (e.g., the start timestamp 1606 and the end timestamp 1608 of FIG. 16 ) to generate variables that are used by an art type prediction model (e.g., the art type prediction model 1428 of FIG. 14 and FIG. 18 ) to classify the non-art type impression record 1501 to be input into the art type prediction model 1428. For example, the data classifier 1500 may generate a day category value 1614 ( FIG. 16 ), a time period value 1616 ( FIG. 16 ), and/or a duration value 1618 ( FIG. 16 ).

At block 2006, the art type analyzer 1502 ( FIG. 15 ) applies the art type prediction model 1428 generated by the art type prediction modeler 1416 ( FIG. 14 and FIG. 18 ) to the classified impression data 1504 to predict art type information. In some examples, the art type prediction may include a rating of accuracy. In some such examples, the rating of accuracy may be based on one or more CCRs of the art type prediction model 1428 calculated by the art type prediction modeler 1416. At block 2008, the art type analyzer 1502 associates the art type prediction with the non-art type impression data 1402 to create a predicted art type impression 1424 ( FIG. 14 and FIG. 15 ). In some examples, the art type analyzer 1502 stores the predicted art type impression 1424 in a prediction database 1422 ( FIG. 14 and FIG. 15 ). Alternatively or additionally, the art type analyzer 1502 includes the predicted art type impression 1424 in a prediction report 1426 ( FIG. 14 and FIG. 15 ). The example process 2000 of FIG. 20 then ends.

21 is a block diagram of an exemplary processor platform 1200 capable of executing the instructions of FIG4 , FIG5 , FIG6 , FIG7 , FIG8 , FIG9 , FIG10 , FIG11 , FIG12 , and/or FIG13 to implement the exemplary calibration data collector 202 , the exemplary sharing matrix generator 204 , the exemplary false positive corrector 206 , the exemplary impression information collector 208 , the exemplary non-coverage calculator 210 , the exemplary non-coverage corrector 212 , the exemplary impression information adjuster 214 , the exemplary household distribution generator 216 , the exemplary aggregate distribution generator 218 , the exemplary matrix corrector 220 , the exemplary matrix normalizer 222 , the exemplary co-viewing matrix generator 224 , and/or more generally, the exemplary impression data compensator 200 of FIG2 . The processor platform 2100 may be, for example, a server, a personal computer, a mobile device (e.g., a mobile phone, a smartphone, a tablet computer such as an iPad ^™ ), an Internet appliance, or any other type of computing device.

The illustrated example processor platform 2100 includes a processor 2112. The illustrated example processor 2112 is hardware. For example, the processor 2112 can be implemented by one or more integrated circuits, logic circuits, microprocessors, or controllers from any desired family or manufacturer. The exemplary processor 2112 of FIG. 21 can implement the exemplary calibration data collector 202, the exemplary shared matrix generator 204, the exemplary false positive corrector 206, the exemplary impression information collector 208, the exemplary non-coverage calculator 210, the exemplary non-coverage corrector 212, the exemplary impression information adjuster 214, the exemplary family distribution generator 216, the exemplary aggregate distribution generator 218, the exemplary matrix corrector 220, the exemplary matrix normalizer 222, and/or the exemplary co-viewing matrix generator 224.

The processor 2112 of the illustrated example includes a local memory 2113 (e.g., a cache). The processor 2112 of the illustrated example communicates with a main memory via a bus 2118. The main memory includes a volatile memory 2114 and a non-volatile memory 2116. The volatile memory 2114 can be implemented by synchronous dynamic random access memory (SDRAM), dynamic random access memory (DRAM), RAMBUS dynamic random access memory (RDRAM), and/or any other type of random access memory device. The non-volatile memory 2116 can be implemented by flash memory and/or any other desired type of memory device. Access to the main memories 2114 and 2116 is controlled by a memory controller.

The processor platform 2100 of the illustrated example also includes an interface circuit 2120. The interface circuit 2120 may be implemented by any type of interface standard, such as an Ethernet interface, a universal serial bus (USB), and/or a PCI express interface.

In the illustrated example, one or more input devices 2122 are connected to the interface circuit 2120. The one or more input devices 2122 allow a user to enter data and commands into the processor 2112. The one or more input devices can be implemented by, for example, an audio sensor, a microphone, a camera (still or video), a keyboard, buttons, a mouse, a touch screen, a trackpad, a trackball, a touch screen, and/or a voice recognition system.

One or more output devices 2124 are also connected to the interface circuit 2120 of the illustrated example. Output device 2124 can be implemented by, for example, a display device (e.g., a light emitting diode (LED), an organic light emitting diode (OLED), a liquid crystal display, a cathode ray tube display (CRT), a touch screen, a touch-type output device, a light emitting diode (LED), a printer, and/or a speaker). The interface circuit 2120 of the illustrated example therefore typically includes a graphics driver card, a graphics driver chip, or a graphics driver processor.

The interface circuitry 2120 of the illustrated example also includes communication devices such as transmitters, receivers, transceivers, modems, and/or network interface cards to facilitate data exchange with an external machine (e.g., any type of computing device) via a network 2126 (e.g., an Ethernet connection, a digital subscriber line (DSL), a telephone line, a coaxial cable, a mobile telephone system, etc.).

The processor platform 2100 of the illustrated example also includes one or more mass storage devices 2128 for storing software and/or data. The exemplary mass storage device 2128 of FIG. 21 may implement the AME media image storage 134 of FIG. 1. Examples of such mass storage devices 2128 include floppy disk drives, hard drives, optical disk drives, Blu-ray disk drives, RAID systems, and digital versatile disk (DVD) drives.

The encoded instructions 2132 of Figures 4, 5, 6, 7, 8, 9, 10, 11, 12 and/or 13 may be stored in a mass storage device 2128, a volatile memory 2114, a non-volatile memory 2116, and/or a removable tangible computer-readable storage medium such as a CD or DVD.

22 is a block diagram of an exemplary processor platform 1200 capable of executing the instructions of FIG19 and/or FIG20 to implement the exemplary collection database 1414, the exemplary art type prediction modeler 1416, the exemplary art type predictor 1418, the exemplary historical art type database 1420, the exemplary prediction database 1422, the exemplary data classifier 1500, the exemplary art type analyzer 1502, the exemplary art type prediction model builder 1802, the exemplary art type prediction model evaluator 1804, the exemplary art type prediction model comparator 1806, and/or more generally, the exemplary audience measurement entity 108 of FIG1. The processor platform 2200 may be, for example, a server, a personal computer, a mobile device (e.g., a mobile phone, a smartphone, a tablet computer such as an iPad ^™ ), an Internet appliance, or any other type of computing device.

The example processor platform 2200 of the illustrated example includes a processor 2212. The example processor 2212 of the illustrated example is hardware. For example, the processor 2212 can be implemented by one or more integrated circuits, logic circuits, microprocessors, or controllers from any desired manufacturer. The example processor 2212 of FIG. 22 can implement the example art type prediction modeler 1416, the example art type predictor 1418, the example data classifier 1500, the example art type analyzer 1502, the example art type prediction model builder 1802, the example art type prediction model evaluator 1804, and/or the example art type prediction model comparator 1806.

The processor 2212 of the illustrated example includes a local memory 2213 (e.g., a cache). The processor 2212 of the illustrated example communicates with a main memory via a bus 2218. The main memory includes a volatile memory 2214 and a non-volatile memory 2216. The volatile memory 2214 can be implemented by synchronous dynamic random access memory (SDRAM), dynamic random access memory (DRAM), RAMBUS dynamic random access memory (RDRAM), and/or any other type of random access memory device. The non-volatile memory 2216 can be implemented by flash memory and/or any other desired type of memory device. Access to the main memories 2214 and 2216 is controlled by a memory controller.

The processor platform 2200 of the illustrated example also includes an interface circuit 2220. The interface circuit 2220 may be implemented by any type of interface standard, such as an Ethernet interface, a universal serial bus (USB), and/or a PCI express interface.

In the illustrated example, one or more input devices 2222 are connected to the interface circuit 2220. The one or more input devices 2222 allow a user to enter data and commands into the processor 2212. The one or more input devices can be implemented by, for example, an audio sensor, a microphone, a camera (still or video), a keyboard, buttons, a mouse, a touch screen, a trackpad, a trackball, a touchscreen, and/or a voice recognition system.

One or more output devices 2224 are also connected to the interface circuit 2220 of the illustrated example. Output device 2224 can be implemented by, for example, a display device (e.g., a light emitting diode (LED), an organic light emitting diode (OLED), a liquid crystal display, a cathode ray tube display (CRT), a touch screen, a touch-type output device, a light emitting diode (LED), a printer, and/or a speaker). The interface circuit 2220 of the illustrated example therefore typically includes a graphics driver card, a graphics driver chip, or a graphics driver processor.

The interface circuitry 2220 of the illustrated example also includes communication devices such as transmitters, receivers, transceivers, modems, and/or network interface cards to facilitate data exchange with an external machine (e.g., any type of computing device) via a network 2226 (e.g., an Ethernet connection, a digital subscriber line (DSL), a telephone line, a coaxial cable, a mobile telephone system, etc.).

The processor platform 2200 of the illustrated example also includes one or more mass storage devices 2228 for storing software and/or data. The example mass storage device 2228 of FIG22 may implement the AME media impression memory 134, the example collection database 1414, the example historical art type database 1420, and/or the example prediction database 1422. Examples of such mass storage devices 2228 include floppy disk drives, hard drives, optical disk drives, Blu-ray disk drives, RAID systems, and digital versatile disk (DVD) drives.

The coded instructions of FIG. 19 and/or FIG. 20 may be stored in the mass storage device 2228 , the volatile memory 2214 , the non-volatile memory 2216 , and/or a removable tangible computer-readable storage medium such as a CD or DVD.

From the foregoing, it will be appreciated that methods, apparatus, and articles of manufacture have been disclosed for enhancing the operation of a computer to improve the accuracy of impression-based data (such as special audiences, impression counts, and duration units), such that the computers and processing systems herein can be relied upon to generate audience analytics information with a higher degree of accuracy. In some examples, based on the equations and techniques described above for determining special audience sizes for misidentification corrections and/or non-coverage corrections, impression counts for misidentification corrections and/or non-coverage corrections, and/or duration units for misidentification corrections and/or non-coverage corrections, computer operation can be made more efficient. That is, by using these processes, a computer can operate more efficiently by relatively quickly determining parameters and applying those parameters to data for determining misidentification corrections and/or non-coverage corrections through the techniques disclosed above. For example, using the example processes disclosed herein, a computer can more efficiently and effectively correct misidentification errors (e.g., applying the misidentification correction matrix of Table 14 above to impressions, special audience sizes, and/or duration units) and/or correct non-coverage errors (e.g., applying non-coverage factors or alpha factors to impressions, special audience sizes, and/or duration units) in development or test data recorded by AME 108 and database owners 104a-b without using a large amount of network communication bandwidth (e.g., conserving network communication bandwidth) and without using a large amount of computer processing resources (e.g., conserving computer processing resources) to continuously communicate with individual online users to request survey responses about their online media access behavior, and without relying on such continuous survey responses from such online users. Survey responses from online users may be inaccurate due to the users' inability or unwillingness to recollect online media access. Survey responses may also be incomplete, which may require additional processor resources to identify and supplement incomplete survey responses. Thus, the examples disclosed herein more efficiently and effectively determine misidentification-corrected data. Such misidentification-corrected data can be used in subsequent processing to identify the engagement performance of different media so that media providers, advertisers, product manufacturers, and/or service providers can make more informed decisions about how to spend advertising dollars and/or media production and distribution costs.

Although certain example methods, apparatus, and articles of manufacture have been disclosed herein, the scope of coverage of this patent is not limited thereto. On the contrary, this patent covers all methods, apparatus, and articles of manufacture fairly falling within the scope of the claims of this patent.

Claims (52)

1. A method for compensating for impression data, comprising: At a first server in a first Internet domain, a first network communication is received from a first type of computing device, the first network communication indicating access to media at the first type of computing device; By utilizing the instructions executed by the first server, a second network communication is sent from the first server to request demographic information, the demographic information corresponding to the first network communication and the third network communication received from the first type of computing device in the first Internet domain; By utilizing instructions executed by at least one processor, a count of media impressions occurring on the first type of computing device is accessed, wherein a first portion of the media impressions corresponds to individuals whose first demographic information can be identified by a second server of the database owner, and an uncovered error generated by the second server for second demographic information of individuals whose second demographic information cannot be accessed by the database owner corresponds to a second portion of the media impressions; and By executing instructions using the at least one processor, the undercoverage error generated by the second server of the database owner is corrected by determining the second demographic information for the second portion of the media impression based on: 1) the first demographic information for the first portion provided by the second server of the database owner in response to the second network communication; 2) a first probability of accessing the media corresponding to the media impression on a first type of computing device; and 3) a second probability of accessing the media on a second type of device, thereby improving the accuracy of the computer in relation to generating audience analysis information by correcting the undercoverage error generated by the second server. 2. The method of claim 1, wherein determining the second demographic information comprises multiplying the following: 1) the ratio of the first probability to the second probability, and 2) the number of media impressions belonging to the first demographic group. 3. The method of claim 2, wherein the first probability is the probability that a person in the first demographic group accesses the media on a computing device of the first type, and the second probability is the probability that the person in the first demographic group accesses the media on a device of the second type. 4. The method of claim 2, further comprising: adjusting the media impressions to compensate for incorrect attribution of a subset of the media impressions to a second person in a second demographic group, wherein the number of media impressions attribution to the first demographic group is determined from the subset of media impressions adjusted to compensate for the incorrect attribution. 5. The method of claim 1, wherein the first type of computing device includes a mobile device, and the second type of device includes a television. 6. The method of claim 1, wherein the computing device of the first type includes at least one of a smartphone, a tablet computer, and a portable media player. 7. The method of claim 1, wherein the first probability and the second probability correspond to the media category of the media. 8. The method of claim 7, wherein the media category is at least one of comedy, drama, political, realistic, and combined media categories. 9. The method of claim 1, further comprising: prior to determining the second demographic information, adjusting the media impression for cases where the media impression is incorrectly attributed to persons who did not generate the media impression. 10. The method of claim 1, further comprising: calculating the first probability and the second probability based on a survey response to a person survey, wherein the calculation of the first probability includes determining a weight from the survey response associated with at least one of a demographic group, a media category, a type of computing device, and a geographic region, the weight indicating the probability of accessing the media associated with the media impression on a device type of interest. 11. The method of claim 10, wherein the survey targets at least one of a randomized group and a group of audience members maintained by an audience measurement entity. 12. The method of claim 1, wherein determining the second demographic information for the second portion of the media impression comprises: Determine the proportion of media impressions belonging to different demographic groups corresponding to the individuals; and The scale of the media impression is scaled to the second portion of the media impression. 13. The method of claim 1, further comprising: providing instructions to an issuer, the instructions being provided by the issuer to the computing device, and, when executed by the computing device, causing the computing device to send the first network communication. 14. The method of claim 1, further comprising: conserving computer processing resources by not communicating with individual online users about their online media access activities and by not requesting survey responses from the online users to determine the first probability of accessing the media corresponding to the media impression on the first type of computing device or to determine the second probability of accessing the media on the second type of device. 15. The method of claim 1, further comprising: saving network communication bandwidth by not communicating with individual online users about their online media access activities and by not requesting survey responses from the online users to determine a first probability of accessing the media corresponding to the media impression on a first type of computing device or a second probability of accessing the media on a second type of device. 16. An apparatus for compensating impression data, comprising: An impression collector, the impression collector being used for: In a first Internet domain, access is made to first network communication received from a first type of computing device at a first server, the first network communication indicating access to media at the first type of computing device; and A second network communication is sent from the first server to request demographic information, the demographic information corresponding to the first network communication and the third network communication received from the first type of computing device in the first Internet domain; An impression information collector is configured to access counts of media impressions occurring on a computing device of the first type, wherein a first portion of the media impressions corresponds to individuals whose first demographic information can be identified by a second server of a database owner, and an uncovered error generated by the second server for individuals whose second demographic information cannot be accessed by the database owner corresponds to a second portion of the media impressions; and An uncovering corrector is configured to correct the uncovering error generated by the second server of the database owner by determining second demographic information for the second portion of the media impression based on a first probability of accessing the media corresponding to the media impression on a first type of computing device and a second probability of accessing the media on a second type of device, thereby improving the accuracy of the computer associated with generating audience analysis information. 17. The apparatus of claim 16, further comprising an uncovered calculator for multiplying: 1) the ratio of the first probability to the second probability, and 2) the number of media impressions belonging to a first demographic group. 18. The apparatus of claim 17, wherein the first probability is the likelihood that a person in the first demographic group accesses the media on a computing device of the first type, and the second probability is the likelihood that the person in the first demographic group accesses the media on a device of the second type. 19. The apparatus of claim 17, further comprising an error identification corrector for adjusting the media impressions to compensate for incorrect attribution of a subset of the media impressions to a second person in a second demographic group, the number of media impressions attributing to the first demographic group being determined from the subset of media impressions adjusted to compensate for the incorrect attribution. 20. The apparatus of claim 16, wherein the first type of computing device includes a mobile device, and the second type of device includes a television. 21. The apparatus of claim 16, wherein the computing device of the first type includes at least one of a smartphone, a tablet computer, and a portable media player. 22. The apparatus of claim 16, further comprising an uncovered calculator for calculating the first probability and the second probability based on a survey response to a person survey, the uncovered calculator being configured to calculate the first probability by determining from the survey response a weight associated with at least one of a demographic group, a media category, a type of computing device, and a geographic region, the weight indicating the probability of accessing the corresponding media associated with the media impression on the device type of interest. 23. The apparatus of claim 22, wherein the survey targets at least one of a randomized group and a group of audience members maintained by an audience measurement entity. 24. The apparatus of claim 16, wherein the first probability and the second probability correspond to the media category of the media. 25. The apparatus of claim 24, wherein the media category is at least one of comedy, drama, political, realistic, and combined media categories. 26. The apparatus of claim 16, wherein the uncovered corrector is configured to determine the second demographic information for the second portion of the media impression in such a manner as follows: Determine the proportion of media impressions belonging to different demographic groups corresponding to the individuals; and The scale of the media impression is scaled to the second portion of the media impression. 27. A tangible computer-readable storage medium comprising computer-readable instructions, which, when executed, cause a processor to at least: At a first server in a first Internet domain, access is made to a first network communication from a first type of computing device, the first network communication indicating access to media at the first type of computing device; A second network communication is sent requesting demographic information, which corresponds to the first and third network communications received from the first type of computing device in the first Internet domain; Access to a count of media impressions occurring on the first type of computing device, the first portion of which corresponds to a person whose first demographic information can be identified by a second server of the database owner, and the uncovered error generated by the second server for the second demographic information of persons whose second demographic information cannot be accessed by the database owner corresponds to a second portion of the media impression; and By determining the second demographic information for the second portion of the media impression based on a first probability of accessing the media corresponding to the media impression on a first type of computing device and a second probability of accessing the media on a second type of device, the uncovering error generated by the second server of the database owner is corrected, thereby improving the accuracy of the computer in relation to the generation of audience analysis information. 28. The tangible computer-readable storage medium of claim 27, wherein the instructions cause the processor to determine the second demographic information by multiplying the ratio of the first probability to the second probability by the number of media impressions belonging to the first demographic group. 29. The tangible computer-readable storage medium of claim 28, wherein the first probability is the likelihood that a person in the first demographic group accesses the medium on a computing device of the first type, and the second probability is the likelihood that the person in the first demographic group accesses the medium on a device of the second type. 30. The tangible computer-readable storage medium of claim 28, wherein the instructions further cause the processor to adjust the media impressions to compensate for incorrect attribution of a subset of the media impressions to a second person in a second demographic group, the number of media impressions attributing to the first demographic group being determined from the subset of media impressions adjusted to compensate for the incorrect attribution. 31. The tangible computer-readable storage medium of claim 27, wherein the computing device of the first type includes a mobile device, and the device of the second type includes a television. 32. The tangible computer-readable storage medium of claim 27, wherein the computing device of the first type includes at least one of a smartphone, a tablet computer, and a portable media player. 33. The tangible computer-readable storage medium of claim 27, wherein the first probability and the second probability correspond to the media category of the medium. 34. The tangible computer-readable storage medium of claim 33, wherein the media category is at least one of comedy, drama, political, realistic, and combined media categories. 35. The tangible computer-readable storage medium of claim 27, wherein the instructions further cause the processor to adjust the media impression for the following situation prior to the determination of the second demographic information: the media impression in the media impression is incorrectly attributed to a person who did not give rise to the media impression. 36. The tangible computer-readable storage medium of claim 35, wherein the instructions further cause the processor to calculate the first probability and the second probability based on a survey response to a person survey, the instructions causing the processor to calculate the first probability by determining from the survey response a weight associated with at least one of a demographic group, a media category, a type of computing device, and a geographic region, the weight indicating the probability of accessing the corresponding media associated with the media impression on a device type of interest. 37. The tangible computer-readable storage medium of claim 36, wherein the survey targets at least one of a randomized group and a group of audience members maintained by an audience measurement entity. 38. The tangible computer-readable storage medium of claim 27, wherein the instructions cause the processor to determine the second demographic information for the second portion of the media impression by: Determine the proportion of media impressions belonging to different demographic groups corresponding to the individuals; and The scale of the media impression is scaled to the second portion of the media impression. 39. A method for compensating for impression data, comprising: Use a processor to collect media impressions from a first-type computing device; The processor requests demographic information for the media impression from the first server of the database owner, the first part of the media impression corresponding to the person whose first demographic information is stored on the first server of the database owner, and the overwrite error generated by the first server for the second demographic information of the person whose second demographic information is inaccessible to the database owner corresponds to the second part of the media impression; Receive the first demographic information corresponding to the first portion of the media impression from the first server of the database owner; The processor is used to determine the number of media impressions in the second portion of the media impressions; and By using the processor to determine the second demographic information for the second portion of the media impression based on a first probability of accessing the media corresponding to the media impression on a first type of computing device and a second probability of accessing the media on a second type of device, the processor corrects the uncovering error generated by the first server of the database owner, thereby improving the accuracy of the computer in relation to the generation of audience analysis information. 40. A method for compensating for impression data, comprising: At a first server in a first Internet domain, a first network communication is received from a first type of computing device, the first network communication indicating access to media at the first type of computing device; At the first server in the first Internet domain, a second network communication is received from a computing device of the first type, the second network communication indicating access to a duration unit of the media at the computing device; By utilizing the instructions executed by the first server, a third network communication is sent from the first server to request demographic information, the demographic information corresponding to the first network communication and the second network communication received from the first type of computing device in the first Internet domain; By executing instructions using at least one processor, a count of duration units occurring on the first type of computing device is accessed, the first portion of which corresponds to a person whose first demographic information can be identified by a second server of the database owner, and the uncovered error generated by the second server of the database owner's inability to access the second demographic information of persons corresponds to a second portion of the duration unit; and By executing instructions using the at least one processor, and by determining the second demographic information for the second portion of the duration unit based on a first probability of accessing the media corresponding to the duration unit on a first type of computing device and a second probability of accessing the media on a second type of device, the computer corrects the uncovering error generated by the second server of the database owner, thereby improving the accuracy of the computer in relation to the generation of audience analysis information. 41. An apparatus for compensating impression data, comprising: An impression collector, the impression collector being used for: At a first Internet domain, a first network communication received at a first server from a first type of computing device, the first network communication indicating access to media at the first type of computing device; Receiving second network communication from a first-type computing device in the first Internet domain, the second network communication indicating access to a duration unit of the media at the first-type computing device; and A third network communication is sent from the first server to request demographic information, the demographic information corresponding to the first network communication and the second network communication received from the first type of computing device in the first Internet domain; An impression information collector is configured to access counts of duration units occurring on a computing device of the first type, wherein a first portion of the duration unit corresponds to a person whose first demographic information can be identified by a second server of the database owner, and an uncovered error generated by the second server of the database owner for persons whose second demographic information cannot be accessed corresponds to a second portion of the duration unit; and An uncovering corrector is configured to correct the uncovering error generated by the second server of the database owner by determining second demographic information for the second portion of the duration unit based on a first probability of accessing the media corresponding to the duration unit on a first type of computing device and a second probability of accessing the media on a second type of device. Correcting the uncovering error generated by the second server improves the accuracy of the computer associated with generating audience analysis information. At least one of the impression information collector and the uncovering corrector is implemented by logic circuitry. 42. An apparatus for compensating impression data, comprising: A component for collecting impressions, the component for collecting impressions being used to access first network communication received at a first server from a first type of computing device in a first Internet domain, the first network communication indicating access to media at the first type of computing device; and A component for collecting demographic information, the component for collecting demographic information being used to send a second network communication from the first server to request demographic information, the demographic information corresponding to the first network communication and the second network communication received from the first type of computing device in the first Internet domain; A component for collecting impression information, the component for collecting impression information being used to access counts of media impressions occurring on the first type of computing device, a first portion of the media impressions corresponding to persons whose first demographic information can be identified by a second server of the database owner, and an uncovering error generated by the second server of persons whose second demographic information cannot be accessed by the database owner corresponding to a second portion of the media impressions; and The component for correcting uncovered parts is used to correct the uncovered error generated by the second server of the database owner by determining the second demographic information for the second part of the media impression based on a first probability of accessing the media corresponding to the media impression on a first type of computing device and a second probability of accessing the media on a second type of device. Correcting the uncovered error generated by the second server improves the accuracy of the computer associated with generating audience analysis information. 43. The apparatus of claim 42, further comprising a component for calculating uncovered areas, the component for calculating uncovered areas being configured to multiply by: 1) the ratio of the first probability to the second probability, and 2) the number of media impressions belonging to a first demographic group. 44. The apparatus of claim 43, wherein the first probability is the likelihood that a person in the first demographic group accesses the media on a computing device of the first type, and the second probability is the likelihood that the person in the first demographic group accesses the media on a device of the second type. 45. The apparatus of claim 43, further comprising a component for correcting misidentification, the component for correcting misidentification being configured to adjust the media impressions to compensate for incorrectly attributing a subset of the media impressions to a second person in a second demographic group, the number of media impressions attributing to the first demographic group being determined from the subset of media impressions adjusted to compensate for the incorrect attribution. 46. The apparatus of claim 42, wherein the first type of computing device includes a mobile device, and the second type of device includes a television. 47. The apparatus of claim 42, wherein the computing device of the first type includes at least one of a smartphone, a tablet computer, and a portable media player. 48. The apparatus of claim 42, further comprising a component for calculating uncovered areas, the component for calculating uncovered areas being configured to calculate the first probability and the second probability based on a survey response to a person survey, the component for calculating uncovered areas being configured to calculate the first probability by determining from the survey response a weight associated with at least one of a demographic group, a media category, a type of computing device, and a geographic region, the weight indicating the probability of accessing the corresponding media associated with the media impression on a device type of interest. 49. The apparatus of claim 48, wherein the survey targets at least one of a randomized group and a group of audience members maintained by an audience measurement entity. 50. The apparatus of claim 42, wherein the first probability and the second probability correspond to the media category of the media. 51. The apparatus of claim 50, wherein the media category is at least one of comedy, drama, political, realistic, and combined media categories. 52. The apparatus of claim 42, wherein the component for correcting uncovered areas is used to determine the second demographic information for the second portion of the media impression in such a way as follows: Determine the proportion of media impressions belonging to different demographic groups corresponding to the individuals; and The scale of the media impression is scaled to the second portion of the media impression.