GfK MEDIAMARK REARCH & INTELLIGENCE
SURVEY METHODOLOGY

 

 

I. SURVEY DESIGN: THE SAMPLE

 

II. THE SURVEY QUESTIONNAIRES

 

III. THE SURVEY EXECUTION

 

IV. THE STUDY REPORTS

 

 

 

 

 

I.  SURVEY DESIGN: THE SAMPLE

Back to top of page

A.        General Description:

The GfK MRI sample is a strict area probability sample of adults 18 years of age and older living in private households in the coterminous 48 states.   The sample, a multistage, known probability sample, is disproportionately over-allocated within the ten GfK MRI media markets   (New York, Los Angeles, Chicago, Philadelphia, San Francisco, Boston, Detroit, Washington, D.C., Atlanta[1], and Dallas-Ft. Worth[2]) and also within the upper 25% of the national income distribution.  The former enables GfK MRI to report stable estimates for each of the ten major markets. The sampling within the upper income population produces larger, more reliable samples, since many of the behaviors measured are more common among upper-income populations.

B.        The Sample Frame

The sample frame is a Survey Sampling International (SSI)-provided computer file of all Census Block Groups (BGs).  The entire land area of the US is subdivided into approximately 225,000 Block Groups. These are organized by state, county, tract and BG.    SSI, utilizing an estimating algorithm based on county household income data produced annually, produces a median HHI for each BG. BGs are arrayed by the updated median income, and the ranges for the upper 25%, the next 25% and the lowest 50% are determined.  Each listing in the upper range is assigned a weight of 4, each listing in the middle range a weight of 2, and the remaining lowest range a weight of 1.   These weights are used to produce income-weighted household counts used in the selection of primary sampling units and of clusters within the primary sampling units.

Back to top of page

C.         The Structure of the Sample

The sample consists of three major components: ten major media markets, each of which is self-representing; and, outside these ten markets, a sample of core based statistical areas; and a sample of non-core based statistical area counties.

           Within each of these, a sample of clusters (i.e., geographically compact areas) is selected.  All households located within the cluster are included in the sample.  Finally, one randomly selected adult in each of these households constitutes the final sample.

D.        Sample Selection (PSUs)

1.         Selection of Primary Sampling Units

Step One: List the income-weighted household counts for each core based statistical area (CBSA) and for each non-core based statistical area (non-CBSA) county (exclusive of the ten markets), ordered by 9 geographic regions, state, and weighted household count to achieve stratification by region, state, and county size.

Step Two: Determine the sampling interval—Divide the total weighted count by 8, since the original design calls for 8 clusters in each primary sampling unit.  All the CBSA and counties equal-to or greater-than the sampling interval are automatically included as self-representing primary units.

Step Three: Sample the remaining areas using a random starting point (a random number between 1 and the sampling interval).  Systematically apply the sampling interval to the accumulated, weighted count of the remaining core based statistical areas and counties such that the probability of any non-certainty unit being selected is proportionate to its weighted size. 

Back to top of page

2.         Cluster Selection Rate

A cluster selection rate is calculated for each of the ten major markets and for each primary sampling unit.  This rate is equal to the weighted count for the major market or primary sample unit divided by the number of clusters to be selected.  In the major markets, the number of originally ordered clusters is listed below. 

 

Waves 43-57

 

 

Waves 43-57

New York

999

 

Boston

400

Lost Angeles

999

 

Detroit

400

Chicago

999

 

Washington, D.C.

400

Philadelphia

600

 

Atlanta

400

San Francisco

400

 

Dallas-Ft.Worth

400

 

In the remaining primary sample units, the originally ordered number of clusters is a multiple of the number of clusters required for each selected PSU.  From this set of clusters, a random subset is selected for use in the actual study.

3.         Selection of Sample Clusters

Sample clusters are geographically defined compact areas within which the final selection of respondents will be made.  Within each primary sampling unit the ordered listing of addresses are subjected to a systematic, random selection process.  Beginning with a random starting point (between 1 and the cluster selection rate), every nth listing is selected by applying the selection rate to the weighted listing count.  In urban areas the addresses so obtained designate the starting point of the cluster.

Generally, fourteen listings immediately following each initial selection are extracted.   The last of these designates the terminal boundary of the cluster. The initial enumeration of the cluster comprises all the listed addresses. Pertinent information, i.e., name, address, telephone number, is extracted and printed for use by the field staff. 

4.         Designation of Sample Households—Urban Areas

All households located within the boundaries established by the first listing and the last listing are included in the sample.  The computer-prepared listing sheets are assigned to interviewers, who canvass the area and record any additional dwellings (new construction, non-telephone households, unlisted numbers, etc.), accounting for all dwelling units.  In some instances the beginning or the end of a cluster may be located within a multiple-dwelling-unit structure.  In these cases, the entire cluster is prelisted, and the limits of the cluster are established.  Generally, this is done alphabetically. All names in the structure that alphabetically follow the first listed name or precede the last listed name of the cluster are included within the sample.    The practice of including all additional dwellings between the first and last listed unit illustrates the principle of the closed interval.

Back to top of page

5.         Selection of Sample Individuals

The design calls for the selection of one person 18 years of age or older in each sample household.  As the initial cluster lists are prepared, each listed unit is randomly pre-designated with the sex of the prospective respondent.  The spaces provided for additional listings are also pre-designated.  Prior to beginning the selection process, the interviewer asks the household member answering the door whether anyone in the household is affiliated with the media.  A positive response eliminates any member of the household from study eligibility. In all other cases, when the interviewer first contacts a sample household, the names and ages of adults of the pre-designated sex are recorded on a grid that specifies an objective, random selection free of interviewer control.  If the household has no adult member of the pre-designated sex (a one-sex household), then all adult names and ages are listed and a sample respondent is selected.  Thus there are, in effect, two samples, one of men and one of women, in each of which the respondents are randomly selected from among the adult household members.

 

II. THE SURVEY QUESTIONNAIRES

Back to top of page

Two different questionnaires are used to collect data.  Data pertaining to media exposure—that is print (magazines and newspapers), radio and television, non-traditional media, and demographic data about the respondent and the household—are obtained in a personal, face-to-face interview.  Product and service usage, again both personal and household, are obtained from a respondent-completed questionnaire left with the respondent at the time of the personal interview and, in a substantial majority of cases, personally picked up by the interviewer. GfK MRI also makes numerous additional attempts, at the telephone validation stage and through other follow-up calls, to retrieve product booklets through the mail.  The interviewers personally retrieve some 60% -70% of all product booklets.

A.        The Personal Interview

The personal interview, conducted with the specifically selected sample respondent, is the technique used to collect data about the basic media exposure of the respondent and the demographic profile of the respondent and household.

1.         Newspaper Reading

The reading of both daily and Sunday/weekend newspapers is measured using an indirect questioning procedure. The questionnaire includes a listing of daily and Sunday newspapers that circulate in the particular area.  The respondent is asked which, if any, of the daily newspapers were read or looked into in the past seven days.  Then for each newspaper mentioned, a question about frequency of reading is asked.

This is followed by "When was the last time you read or looked into...?" for each daily newspaper read or looked into in the past seven days.

For interviews conducted on Sunday and Monday, "read yesterday" is defined as "last Friday" for a daily newspaper. A comparable procedure is followed for Sunday/weekend newspapers, using a four-week time span in the initial question.  The audience measure is based on the number of people who report reading the daily newspaper "yesterday" (or on the most recent weekday), or reading the Sunday paper within the past seven days.

            Beginning in Wave 23, GfK MRI introduced a separate set of additional questions for Sunday and Monday interviewing.  In addition to the standard readership question asking “when last time read,” GfK MRI also asks the respondent whether he/she read the weekday issues “this Saturday or Sunday.”  In the case of USA Today and The Wall Street Journal, GfK MRI credits readership if the respondent answers he/she read “this Saturday or Sunday” or “Friday.”  This procedure accounts for any additional readership of Friday issues of the papers over the weekend.

            In addition, questions regarding location of reading and how the newspaper was obtained are asked for the nationally circulated newspapers.

            Beginning in Wave 57, GfK MRI added select qualitative questions for national newspapers measurement.  These are: 1) time spent reading, 2) percent of pages read, 3) overall rating and 4) interest in advertising.  In order to maintain clarity in the survey, these qualitative questions along with the magazine qualitative questions are asked after the newspaper and magazine readership questions have been administered.

Back to top of page

2.         Magazine Reading

 GfK MRI’s procedure for measuring magazine audiences is a recent reading technique specifically developed for the magazine environment in the United States, taking full account of experiences gained in other countries using similar techniques.  The principle of the recent reading technique is that the number of people reading any issue of a magazine during its publication period (recent reading) is equal to the total number of people reading any particular issue over its total life (average issue audience).  Important to note, beginning with Wave 56, GfK MRI removes web-exclusive readership claims from the reported average-issue audience for magazines and national newspapers. The average-issue audiences for these publications are any reading of the hard or printed copy, whether or not the reader has also visited the magazine’s website.  In effect, we are removing the exclusive Web/Internet reads from the average-issue calculation.  In order not to change already published numbers, GfK MRI will not make any adjustment to Wave 55 or earlier waves. 

It is essentially a two-step procedure.   The first step, a screening procedure, serves to eliminate magazines the respondent has not read or looked into in the last six months.  The second step, applied only to magazines seen by the respondent in the last six months, ascertains reading within each magazine’s publication period. 

            The interviewer first produces a binder containing sort boards and a deck of cards on which are printed black-and-white logos of some +/- 250 magazines.  Black-and-white reproductions are used following the practice of past through-the-book studies.  Some magazine logos change color with successive issues while others retain the same color.  The logo deck is therefore neutral in this respect.   The deck is shuffled in front of the respondent to ensure that it is in random order.

The respondent is then asked to sort the cards on the sort board into three groups, indicating whether they were read or looked into within the last six months.  The questioning begins as follows:

 

"The titles of magazines and other publications are printed on these cards.  Some of these publications are weekly newspapers."

 

The interviewer then opens the "in the last six months" sort board and continues:

 

"This is a sort board.  I'd like you to sort these cards into piles on the board to show whether or not you've read or looked into them in the last six months.  If you are sure that you have read or looked into the publications, put the cards in this position."  The interviewer points to the "yes—sure have" block on the board.

 

Then, "If you are not sure if you have read or looked the publications in the last six months, put the cards in this position." The interviewer points to the "not sure" block on the board.

 

            Finally, "If you are sure that you have not read or looked into the publications, place the cards in this position."   The interviewer points to the "no—sure have not" block on the board.

 

Before handing over the deck of cards, the interviewer reads the following explanation to the respondent:

"We want to know whether you've read or looked into any copy, whether it belonged to you or not."

 

"It could have been in your home, someone else's home, or any other place at all, such as the beauty (barber) shop, doctor’s office, etc."

 

"It doesn't matter whether you read it, or just looked into it."

 

The interviewer then hands the deck of cards to the respondent, saying:

 

"Now, would you sort these cards to show whether you've read or looked into the magazines and other publications in the last six months?   Please take your time and consider each one carefully."

  Back to top of page

Actual card sorting takes some six to eight minutes on average.  In-flight publications are screened in a similar way by using separate decks (up until Wave 50, cable publications were also screened in using a separate deck).  Additionally, in Wave 52 GfK MRI added a Spanish language title deck and procedure.

            After the initial sorting, the respondent is asked to read to the interviewer the names and code numbers on the logo cards he/she has sorted into the "yes" and the "not sure" positions.   In addition to retaining the involvement of the respondent, who would otherwise have nothing to do while the interviewer records the results, this has the advantage of removing stray cases of confusion due to initial misreading of the cards, such as New York for The New Yorker, or Travel Holiday for Travel & Leisure, and so forth. On average, about 14 publications are screened in, with wide variation: some respondents screen in none or very few and others 30 to 40 or more.

The interviewer then asks the frequency-of-reading questions about each screened-in publication: that is, "On the average, out of 4 issues that are published, how many issues of (Name of Magazine) do you read or look into? Is it 0,1,2,3 or 4?” This frequency question serves several purposes.  First, it gives the respondent an opportunity to say what is most natural to him/her, and what he/she generally supposes the interviewer wants to know—how often he/she reads the particular magazine. Second, it contributes to the process of familiarization with the magazines that have been screened in, a process that begins with the initial sorting. Previous research suggests title confusion is minimized when respondents have multiple opportunities to consider titles that at first sight they think they may have read.   Third, the frequency data are used directly to estimate cumulative audiences.

            Then the recency question follows.   First, the interviewer separates the cards into weeklies, monthlies, and so forth (each publication is identified on its logo card by publication interval), and the corresponding sort boards are opened.  The respondent is asked to consider very carefully when he/she last read or looked into each publication, excluding today.   A date is provided to facilitate the accurate identification of the reading period—for example, for weeklies, "the seven days since last Wednesday" for interviews conducted on a Wednesday; for monthlies, “the 30 days since September 10th" for interviews on October 10th, and so on.  The card for each screened-in magazine is placed by the respondent in one of three positions on the sort board: "Yes, sure have," "Not sure," or "No," and the response recorded by the interviewer.

Only those respondents who place a logo card in the "Yes—sure have" position—that is, those who have read or looked into a magazine during the period equal to its most recent publication interval—are classified as members of the total audience of the publication.  The remaining two categories, "No" and "Not sure," are not classified as such.

Upon completion of this second card sort, a series of questions is asked about each publication for which the respondent is classified as a reader, having read the publication within the most recent publishing interval.  These questions, sometimes termed the “qualitative” aspects of magazine reading, include place of reading, reading days, reading time, reader actions, source of copy, percent of pages looked at, rating, and interest in advertising. The responses to these questions are used to define in-home and out-of-home audience, primary and pass-along readers, reading days, and page exposures.  As appropriate, the questions are asked using show cards displaying the range of possible responses.

Back to top of page

Four versions of the questionnaire are employed.  In two, weeklies are listed first, followed by bi-weekly, tri-weekly, monthly and bi-monthly magazines.  In the remaining two versions, the order is reversed.  Within the publication interval-ordered sets, titles are listed in one version in alphabetical order and in the other in reverse alphabetical order.

3.         Radio Listening

The interviewer displays cards on which are listed five time periods.   While showing this card, the following questions are asked:

"These are time periods during which people can listen to or hear a radio.  Thinking about YESTERDAY, to the nearest half hour, how much time, if any, did you spend listening to or hearing a radio during the time period of (TIME PERIOD)—either in your home, car or any other place?"  and "During the period (TIME PERIOD), what station or stations did you listen to?  Please give me the Call Letters of each station and whether it was AM, FM, Sirius, XM, or the Internet."  These two questions are asked for "yesterday," for "last Saturday" and for "last Sunday."

4.         Television Viewing

The interviewing procedure employed for television is similar to that for radio.    A show card indicating a list of time periods is shown and the following question asked:   "These are time periods during which people can watch television.  To the nearest half hour, how much time, if any, did you spend watching television in each of these time periods yesterday?  How about (TIME PERIOD)?"

            As with radio, this is also asked for "last Saturday" and for "last Sunday,” providing the basis for audience estimates of time slot and average half-hour viewership for weekdays and weekends.

            If the interview is conducted on a Sunday or Monday, then "last Friday” instead of "yesterday" is asked to determine weekday viewership.

5.         Cable Television/Satellite TV

A series of questions is asked to establish:

                         a. The availability of cable in the respondent's neighborhood;

                         b. Subscription to cable TV; Have Satellite TV; Subscription to digital cable service;

                         c. The company through which household accesses programming on satellite dish;

                         d. Subscribe to Fiber Optic or Internet Protocol TV

                         e. Availability of Pay-Per-View services;

            f. Whether Pay-Per-View has been watched in the last year and the number of times    types of programs have been viewed;

g. Whether Video-On-Demand service is offered through their cable company and if it was watched in the last year.

                         h. The number of hours viewed for specified cable and premium cable channels;

i. Whether the household has a DVR, its brand and how it was obtained.

  Back to top of page

6. Outdoor Travel/Yellow Pages

            A series of questions is asked to establish:

a.       Miles traveled in past week, past month;

b.      Last time rode in car, how many people, including self, were in it and how many of these passengers were 18 and older;

c.       Last time referred to yellow pages;

d.      Number of times referred to yellow pages in last week;

e.       Number of times referred to yellow pages in phone book and on Internet in last week;

f.       Number of times referred to yellow pages at work.

 

7. On-Line Services/Internet Usage

           A series of questions is asked to establish:

a.       The availability and use of the Internet;

b.      How connected to the internet at home;

c.       Which Internet Service Provider household subscribes to;

d.      If no internet in household, any alternative access locations;

e.       Connect to internet via Wi-Fi, wireless connection or Cell phone/PDA/ Smartphone/ handheld organizer/other similar mobile device.

f.       Activities on the Internet;

g.      Search engines used;

h.      Types of Instant Messenger used;

i.    How often Internet looked at in a typical month;

j.    Time using the internet in the past 30 days (not including IM);

k.      Specific websites visited (websites asked in past 30 days and yesterday);

8. Demographic Information

A complete set of demographic characteristics of the respondent, the household head and the household itself is obtained.  This includes age, sex, marital status, occupation, industry, household and individual employment income, education, household composition, race, and home ownership.

This information is obtained by the use of straightforward questions and show cards that contain the range of possible responses.  The recording of the replies requires minimal effort on the part of the interviewer, since almost all responses are pre-coded on the questionnaire in the same manner as on the show cards.

  Back to top of page

* Public Activities

The respondent is given a list of activities that they may have engaged in relating to politics, public or civil affairs in the past 12 months

 

B.         Marketing Questionnaire

            Data on usage of an extensive range of goods and services are obtained using a questionnaire completed by the respondent and, if the respondent is not the Principal shopper, the Principal shopper.    Upon completion of the media and demographic personal interview, a marketing questionnaire is left with the respondent.   A ten-dollar incentive was offered for its completion through Wave 34, since which the incentive has been twenty dollars.  In Wave 38, GfK MRI conducted additional retrieval efforts (in-person, over the phone or by mail) among non-respondents to the initial product booklet attempts.  In these cases, GfK MRI offered a $50 incentive for completion.  These additional efforts at product booklet retrieval are now part of GfK MRI’s standard protocol for collecting product booklets.

        In most cases, an appointment is made for the collection of this questionnaire.  If necessary, additional efforts, such as those discussed above, are made to retrieve the self-administered questionnaire via mail. In general, this questionnaire is designed to measure:

1.         Ownership and/or use of products or services;

2.         The brand (kind, type, variety, etc.) used;

3.         Quantities used within specified time periods;

4.         Participation in the decision to buy or use.

Product data are of two types: personal product questions answered by the respondent and household product questions answered by the Principal shopper (who may or may not be the respondent).

Although questions are necessarily tailored to particular subjects, every effort is made to use standardized wording and standard time frames, as well as to ask simple, unambiguous questions. The questionnaire is also designed to minimize the amount of recording entry by respondents.  Whenever possible the questionnaire is constructed so that a check mark or a number completely records the response.

In addition, viewing of network TV programs, sports, and specials is also obtained in this questionnaire. And, a series of psychographic type questions are also included in the product booklet.

 

III. THE SURVEY EXECUTION

Back to top of page

GfK MRI works with LHK Partners to develop the protocols for executing the study, including training and evaluating the field staff.

 

 

A. Staffing the Fieldwork

The study is conducted by a staff of some 100-125 interviewers recruited, trained, and supervised by eight LHK field supervisors and a staff of 8 recruiters and trainers who are, in turn, directed and supervised by a full-time Field Director and the three LHK senior partners. Since the study is continuous, a great deal of effort is expended to recruit, train and maintain an experienced field staff.  The performances of supervisors and of interviewers are reviewed continually.

Prior to each wave, training materials, including manuals and a taped model interview are prepared, in addition to the questionnaires, show cards, sort boards, and other materials needed for the execution of the interview.

All interviewers are trained or retrained, with new interviewers receiving more intensive instruction.  Included in the training are instructions on locating and listing the geographic cluster, making the initial contact, selecting the sample respondent, and executing the survey.  Interviewers are instructed in the handling of difficult or unusual interviewing circumstances, including gaining access to security buildings. Interviewers assigned to large apartment buildings or rural areas are instructed accordingly.  Interviewers are briefed on the organization and planning of callbacks and the importance of gaining the cooperation of respondents.

Continuous quality checks are undertaken during the course of data collection and appropriate action is taken when necessary.  No new interviewer may begin interviewing until he/she has been judged acceptable by the LHK trainers.  The work of each interviewer is validated by telephone, or by mail or, on occasion, by personal contact. GfK MRI attempts to validate 100% of all personal interviews.  In practice, GfK MRI achieves a 55%-60% validation rate.

           LHK Partners maintains frequent contact with the local field supervisors, who in turn maintain similar contact with the interviewers.  In this manner, tight control is maintained over the flow and the quality of the work.   The computerized control system employed by GfK MRI/LHK Partners has a complete record, organized by cluster, of the entire sample which provides information about the current status of every cluster in the study.

B. Data Collection

The listed addresses for each cluster, as described in the “Selection of Sample Clusters” section above, form the foundation of interviewing.  The interviewer lists and interviews all households located between the first and last original, prelisted addresses.  If the listing contains a multiple dwelling unit, the interviewer proceeds to the unit and describes its layout and then provides LHK Partners with the names and apartment numbers, if possible, based on the alphabetic interval chosen in the sample.  The lists expanded by this method are used to make a mailing to all known, prospective respondents explaining the nature of the study and emphasizing the confidential nature of responses.  In rural areas, where the cluster is defined as a starting point with a series of listing instructions, the listing is done at the time of interviewing.

The sample then comprises all dwelling units starting with the initial dwelling and continuing to and including the last dwelling.

At the start of W54, GfK MRI increased personal interview incentives for households in PSU 001 (New York DMA) from $5, $20, and $40 to $10, $30, and $50, respectively (with the exception of Manhattan which remains at $25, $50, $75). These are permanent incentive changes and thus, will continue into W55 and thereafter.

In addition, beginning March 3 for clusters not yet started, GfK MRI made the following adjustments:

Top 10 Clusters: All clusters in the Philadelphia, Boston, Detroit, and Washington D.C. DMAs increased incentives from $5, $20, and $40 to $20, $40, and $50, respectively.

Non-Top 10 clusters:

g.      $5 clusters that are Stratum 1 and 2 increased to $10.

h.      $5 clusters that are Stratum 4 increased to $20.

i.        $20 clusters increased to $40

j.        $40 clusters increased to $50.

Back to top of page

Beginning April 6, 2006 LHK made the following additional changes to clusters assigned for rework, where the re-work had not yet started (and where incentives were not already increased):

Personal interview incentives were increased from $5, $20, and $40 to $20, $40, and $50, respectively. This includes PSU 001 (the New York DMA), with the exception of Manhattan, as described above.

In Wave 56, GfK MRI also conducted a test of incentive for the Personal Interview in those clusters previously assigned to receive $5. In prior waves, approximately 40% of the sample was pre-designated to receive the $5 incentive. In the current wave, half of the $5 clusters will randomly be selected to receive $10 instead. The remaining clusters’ incentives remain unchanged at $20, $40 or $75 for personal interview completion. GfK MRI is conducting this test in order to evaluate the effect of increased response rates in like clusters.

   GfK MRI attempts up to five additional calls at different times and on different days in order to contact “difficult-to-reach” respondents, but sometimes interviewers are unable to complete all five additional attempts for each household.  In some instances, “traveling interviewers” must leave the cluster or primary sampling unit before all desired attempts could be made.

 LHK also tries to assign interviewing services or interviewers with Spanish-speaking capabilities to areas known to have substantial Spanish-speaking populations.  GfK MRI does not, however, specifically assign a bilingual interviewer in every instance requiring bilingual capabilities.  When necessary, we rely on another household member to translate the questions into another language (e.g., Spanish) for the selected respondent. Beginning with Wave 48, the GfK MRI questionnaire and product booklet were made available in both English and Spanish.

The sample respondent is selected by the established procedure.  The interviewer lists, from oldest to youngest, all adult respondents of the pre-designated sex currently living in the household and then follows computer-generated instructions to select the respondent. On average, slightly longer than sixty minutes is required to complete this interview.

Upon completion of the personal interview, the product questionnaire is introduced and the respondent is asked to complete it; the respondent is briefed on how to complete the booklet, and arrangements are made, in a majority of cases, for the interviewer to retrieve the completed questionnaire at a specified time and date.

In Wave 56, for the first time since 1996, GfK MRI raised the initial, general product booklet incentive from $20 to $40. In prior waves, approximately 70% of GfK MRI’s respondents who completed the product booklet received the $20 incentive. Respondents who haven’t completed the booklet by a certain date may receive secondary or tertiary offers, to a maximum of $75. Also, in a number of pre-designated and/or hard-to-reach clusters the initial incentive has been and will remain $75. 
Back to top of page

C. Data Processing

All of the data collected using the two basic survey questionnaires are processed as described below, and all data then reside as data files.   These files are the sources for the various report volumes and discs.  In addition, access to these files is afforded to subscribers for the further tabulation of data.

1.         Initial Editing and Coding

All completed questionnaires are reviewed by LHK Partners to ensure the interviewers are executing the study properly.  Questionnaires that fail to meet completeness and internal consistency checks are referred to the field for correction.   Most data are self-coded, excepting items such as names of newspapers and occupations. In addition, internal editing checks are applied to ensure interviewers are following instructions.  The results of these editing checks are fed back to the field.  The product book is also checked, since it must meet completeness standards to be included in the study.

2.         Data Capture

Two separate operations are utilized for data capture: one for the personal interview and another for the product booklet. In both instances, key-to-disc techniques are used, incorporating logical and self-checking steps.  The personal interview key entry is 100 percent validated.  The product booklet is subjected to a minimum of 25% validation, with additional validation as may be required.  All of these data are eventually combined into a single set of data files.

3.         Data Ascription

The sample comprises all respondents who are personally interviewed.  On average, about 60% - 62% of these respondents also complete the product questionnaire.  In order to avoid problems created by shifting bases, an ascription process for product questionnaire non-respondents is utilized.  This process is embodied in a computer program that finds the best match between a non-booklet respondent and a booklet respondent.  "Best match" is defined as a pair of respondents who most closely resemble each other on a prioritized list of critical variables including sex, geography, age, education, family status, and other demographic and behavioral items.  Once the best available match is identified, the product questionnaire data of the responding member of the pair are assigned to the respondent who did not complete the product questionnaire.

 

a. Special Personal Computer/Cell Phone/ In-Home Internet Access/Pet Ownership Ascription

 

GfK MRI collects data for personal computers, cell phone ownership, in-home Internet access and pet ownership in both the media/demographic booklet (the personal interview) and the product booklet. Special ascriptions are used for respondents who provide conflicting information.

 

The basic premise for these ascription procedures is that the information provided by the respondent in the media/demographic booklet overrules the information provided in the product booklet.

 

For example, if a respondent indicates no to household computer ownership in the media/demographic booklet but indicates yes in the product booklet, the information provided in the product booklet is removed. This holds true for cell phone ownership, Internet access and pet ownership as well.

 

If a respondent indicates yes to household computer ownership, cell phone ownership, in-home Internet access, pet (dog and/or cat) ownership in the media/demographic booklet but indicates no or no answer in the product booklet, then the product booklet data for those variables are ascribed from a donor who responds yes to any of these questions, respectively, in the personal interview.

 

For the personal computer ascription, the donor is selected by placing each potential donor (a respondent who indicated yes in both questionnaires) into one of eight cells based on sex and geography (2 sex by 4 geography). The geographic variables are the North East, North Central, South and West census regions.

  Back to top of page

Selection of a specific donor within these cells is performed identical to the process for selecting donors in product booklet ascription described above. Accordingly, special personal computer ascription is essentially performed twice, once for household computer ownership and once for personally using a computer at work. Consistent with product booklet ascription, the maximum number of times a donor can be used is three.

 

The cell phone, in-home Internet access and pet ownership ascription works on a similar principle. However, because these are household use/ownership questions, a limited number of variables (e.g. age, sex of Principal shopper, household income, presence of children) are used. Once again, the maximum number of times a donor can be used is three.

 

 

b. Special Ascription Pertaining to Psychographic Batteries

 

GfK MRI has historically released psychographic data for only those respondents who have completed all or almost all of the battery of questions in that topic area (e.g., Buying Styles). This restriction necessarily led to a unique sample balancing solution for each of the batteries and, in turn, unique weights for each psychographics sub-sample. Accessing these bases and unique weights had the potential to cause confusion and tabulation errors among our users. Beginning in Fall 04, GfK MRI employed a new ascription procedure that allowed users to access almost all of GfK MRI’s psychographic batteries using a single population weight.

The new ascription procedure uses the following three-step approach to ascribing items for a given psychographic battery:

(1) For those who filled out at least one item within the battery, the missing items are ascribed collectively based on respondents’ responses to other psychographic items, as well as their responses to both demographic and behavioral questions

(2) For those who returned the product booklet and did not answer any items within the battery, the missing items are ascribed collectively based on respondents’ responses to only demographic and behavioral questions

(3) For those who did not return the booklet, all psychographic batteries are ascribed collectively based on GfK MRI’s traditional booklet ascription procedure.

This ascription procedure is currently used for the following psychographic batteries:

Intent to Purchase, Buying Styles, Category INFLUENTIALS Segments (first released in Wave 58), Category-Specific Attitudes (Automotive, Food, Finance, Vacation Travel, Technology, Media), Cellular/Mobile Opinions (first released in Wave 58), Consumer Confidence, Fashion & Style Attitudes (first released in Wave 58), Health Attitudes, Intent to Purchase, Interest in Advertising, Interest in Sports (first released in Wave 53), and Alternative Advertising Places (first released in Wave 55).

 

c. Special Ascription for Hispanic Television Programs

The addition of measured Spanish television programs in the product booklet created a special ascription procedure.  All analyses of these data indicated that Spanish-language capability was the critical predictor for viewing these programs.  Accordingly, GfK MRI modified the ascription process for these variables by adding language spoken in home as a required variable in the ascription process.

 

4.         Database Merging

In addition to the questionnaire items, a considerable amount of additional information is developed for each respondent by incorporating other databases.  There are three major types:

a. Geographic Classification: For each interviewing wave, a master file for each cluster in the sample is available details the following:

                         1) Geographic division and region;

                         2) County size;

                         3) Metropolitan area;

                         4) DMA and metropolitan area classification;

                         5) Zip code;

                         6) Local area median income.

These data are incorporated in the record of each respondent.

b. Media Classification Data: Three industry-prepared databases are used to provide media classification data.  These are:

1)         A file of carrier newspapers for newspaper-distributed magazines (Parade, Sunday

                        and USA Weekend) and comics (Metro - Puck);

            2)         A file of radio stations detailing formats and network affiliation for each station;

3)         A magazine file containing subject matter classification for each surveyed

                        magazine.

The data on these files are merged into the respondent data file for each wave so that each wave is as current as the industry source.

c. Geo-demographic Life-Style Classification: Proprietary systems of classifying populations by geo-demographic and life­style parameters have been developed.  Among these are PRIZM and ACORN.  Each wave of GfK MRI data is processed through these systems and the appropriate classifications are incorporated in the database.  Subscribers to these sources may have access to these classification systems on the GfK MRI database and utilize their conceptual structures on GfK MRI data.

5.         Projection

GfK MRI reports have been designed to quantify media and marketing behavior of the adult household population.  This is accomplished in two stages: weighting, which is the fulfillment of the sample design; and sample balancing, the precise tuning of major study demographics to the most recent independent estimates.

a.         Weighting:  If a sample were to be selected by choosing, say, every tenth member of a population, then the sample result could be projected to the population simply by multiplying by ten.  In general, if N is the sampling interval—that is, every Nth member of a population is selected—then N times the sample result is a straightforward, unbiased estimate of the population. This is how the GfK MRI sample is weighted. However, since the sample selection is a multistage process, the weighting, which is essentially the reciprocal operation, must also be multistage. The original sample is selected separately and independently for the separate strata.  In addition, the male and female portions constitute separate samples. Therefore, weighting (and subsequent balancing) must be undertaken for each of these separate populations.   Within these strata the following factors are evaluated as part of the weighting:

1)         Income Strata:

Because of differential sampling rates, respondents in the three income strata are assigned weights equal to the reciprocal of the sampling rate, adjusted for differential sample recovery.

2)         Number of Persons of Designated Sex:

Since each respondent is selected at random from all adults of the designated sex in the household, each respondent is weighted by this number.  For example, a male respondent in a household with two male adults has a 50% probability of selection and therefore has a weight of two.

3)         Additional Households:

The design calls for the inclusion in the sample of all additional households located after the first and before the last listings.    The differential probability of these non-listed households is N-1/N, where N is the number of original listings and N-1 is the number of intervals.   Therefore the weighting factor applied to these households is N/(N-l).  Prior to Wave 38, GfK MRI used “non-telephone listing status” to define “non-listedness” in the SSI frame.  Beginning in Wave 38, any added households between two listed households qualify as non-listed households.

4)         Two Residences:

Persons dividing their time between two residences during the four weeks preceding the interview have two chances of being included in the sample.   They are therefore assigned a weight of .5.

5)         One- and Two-sex Households:

By design, two-sex households have a 60% chance of being included in the male sample and a 40% chance of being included in the female sample.  One-sex households are included with certainty.     Respondents in these households are weighted to reflect this differential.

Back to top of page

6)       Non-response Factor:

            Non-response adjustment factors are applied on the basis of income stratum and the ten Mediamarkets vs. the balance of the sample.  These factors are equal to the ratio of eligible respondents/completed respondents, calculated separately within the cross classifications of the three income strata and the two major geographic strata.  

            The product of these six factors yields the intrinsic sample weights which, multiplied by the projection factor for each stratum, produces the sample weight.  The projection factor for any stratum is the independent estimate of its population divided by the sum of the corresponding intrinsically weighted respondents.

b.   Sample Balancing

Sample balancing is a widely accepted and used technique in sample surveys.   It was first discussed thoroughly by W. Edwards Deming in his book Statistical Adjustment of Data. Sample surveys produce a large number of estimates.  In some instances, more reliable and more precise estimates are available from other sources; either from larger, more comprehensive samples or from total counts and censuses. For example, a sample survey can produce an estimate of the population by age.   However, the Bureau of the Census reports data on the age distribution more accurately and precisely than most other sources.  Sample balancing is a technique for incorporating into a sample survey’s results the estimated counts from an external or independent source. The rationale is that this type of incorporation improves the accuracy and precision of the sample survey. As with sample weighting, the basic idea of sample balancing is quite simple.  Consider a basic illustration:

A sample survey estimates 4,500 men and 5,500 women in a particular population.  A valid, reliable, independent source reports 4,700 men and 5,300 women for the same population.  If the weight assigned every male respondent is multiplied by 47/45 and that of every female respondent by 53/55, the resultant estimates will conform to the desired distribution between men and women.  This is termed a ratio adjustment; i.e., multiplying each weight by the ratio of the desired number to the obtained number.   As such, it has a very important advantage: namely, it is a least squares adjustment. This means the sum of the squared difference between the original and the final weights is smaller than that of any other type of adjustment producing the same results.  The change necessary to obtain the desired result has been held to a minimum, and the maximum amount of the original weight structure has been maintained.

Sample balancing is simply a series of successive and reiterative ratio adjustments—successive in that only one set of factors such as age or sex can be balanced at one time, and therefore there is a succession of them.  It is reiterative because each successive adjustment partially obfuscates the previous ones.   Therefore, the process of balancing all the variables is essentially one of successive adjustments and is repeated until the desired parameters are obtained.

The GfK MRI sample is balanced within sex on the following sets of population parameters:

a.       Ten Mediamarkets;

b.      Remainder of the country by metropolitan versus non-metropolitan areas within census region;

c.       DMA Size;

d.      Age;

e.       Household income;

f.       Education;

g.      Employment status and occupation;

h.      Race within region;

i.        Marital status;

j.        County size;

k.      Marketing region;

l.        Household size;

m.    Hispanic Origin within region (Added in Wave 35);

n.    Language personally spoken in the home - Hispanics only (Added in Wave 64).

Each wave of fieldwork is weighted and balanced separately to population estimates corresponding to the midpoint of the fieldwork for that particular wave.  The independent sources of data used for sample balancing are the U.S. Bureau of the Census (beginning with Doublebase 2008, GfK MRI began using the Public-Use Microdata Samples, PUMS, data for establishing targets for the local markets), Claritas, Employment and Earnings (a monthly publication produced by the Bureau of Labor Statistics), and Nielsen's universe estimates of language use among Hispanics.

Back to top of page

6.         Final Weight Trimming

The sampling tolerances associated with a given sample are affected by the distribution of weights. In particular, extremely high weights disproportionately increase sampling error estimates.  Therefore, after sample balancing, the distribution of weights is inspected and respondents with weights greater than 5.75 average weight are each assigned the average weight for the respective group.  Weight trimming effectively reduces the highest weights, in turn reducing the sampling error.  GfK MRI also trims the weights of all respondents whose weight is under 1,000.  The trimming is done within sex by race, thus preserving the sample-balanced totals for these groups.

7.         Household Weight

Each household’s weight is obtained by dividing the population weight by the number of adults in the household.

8.         Rebalancing the Doublebase

Each year, to prepare two years’ data for release, the four most recent waves are subjected to additional sample balancing, incorporating demographic and geographic estimates for each of the ten major markets along with the national demographic and geographic estimates employed in the initial balancing.

D.        Audience Estimating Procedures

1.         Magazines

a. Total Audience (average issue audience): The total audience of a magazine includes all respondents who read a copy of the magazine during the past N days, where N is the publication interval of the magazine (7 for weeklies, 30 for monthlies, etc.). These responses come from the card-sorting technique described in Section II of this guide.

            b. Primary Audience: The primary audience of a magazine is defined as readers who live in a household in which the magazine was obtained by either subscription or newsstand purchase.  During the personal interview, questions are asked about how the magazine was obtained and who obtained it.  Generally, purchase and subscription tend to be over claimed.  When over claims exist, the accuracy of these estimates is improved by randomly reducing the number of purchasers and/or subscribers to the known circulation and the number of other primary readers to the same level (see Page 35, Primary Reader Adjustment).

            c. In-Home Audience: Respondents are asked where the reading of the most recent publication interval took place and are shown a list of possible places. Those responding "at home" are classified as “in-home readers.”

            d. Magazine Groups: In some instances, individual magazines are reported as parts of magazine groups. For the most part these are gross audiences—the sum of the audiences of the constituent magazines.

            e. Cumulative Audience: During the personal interview a frequency of reading question (0, 1, 2, 3, or 4 of the average 4 issues) is asked. Responses to this question, along with the responses to the publication-interval reading question, are used to estimate, first, two-issue reach and, second, reach and corresponding frequency for any number of issues greater than two.  This can best be shown by an illustration:

TABLE 1

 

(1)

(2)

(3)

(4)

(5)

(6)

(7)

Frequency of Reading Answers

Total
In tab

Read
In tab

Pct. Read Within

Pct. Not Read Within

Pct. Non-Read 2 Issues

Pct. Read 1 or 2 of 2 issues

No. Read 1 or 2 of 2 issues

0

200

2

1.0

99.0

98.01

1.99

4

1

100

10

10.0

90.0

81.00

19.00

19

2

100

38

38.0

62.0

38.44

61.56

62

3

200

100

50.0

50.0

25.00

75.00

150

4

400

300

75.0

25.0

6.25

93.75

375

Screens

1000

450

 

 

 

 

610

Non-Screen

9000

 

 

 

 

 

 

 

This table reads as follows:

Columns 1 and 2 are the basic survey data.               

Column 3 = Column 2 / Column 1                                  The percent of each group reading

Column 4 = 100.0 - Column 3                                         The percent not reading

Column 5 = (Column 4)2                                                 The probability of not reading either of two

Column 6 = 100.0 - Column 5                                         The percent reading at least one of two issues

Column 7 = Column 6 X Column 1                                 The number reading at least one out of two issues

 

The foregoing is straightforward probability mathematics used to estimate higher orders of reach.  However, there is a limitation to this method: the calculated cumulative audience, no matter how many issues are considered, could never exceed total screenings, in this instance 1000.  This is an artificial limit. Another approach, the widely used “beta binomial," does not have this limitation.  Briefly stated, the beta binomial method assumes a continuous distribution of probabilities of reading from 0 to 1 (compared to the 5-point distribution), and the solution is in fact the integral or sum of all of these probabilities, extended to the appropriate number of issues. The data required for this solution can be obtained directly from a two-issue measurement.  Moreover, the solution is in fact simpler than the straight binomial expansion, particularly for more than two issues:

C1 = proportion reached by one issue

C2 = proportion reached by two issues

A = (C2 - C1) / (2 x C1 - C2 - (C1)2)

 
B = A x (1 - C1)

The proportion reached by t issues, C, is:

       Ct = Ct-1 + (B + t - 2) / (A + t - 1) x (Ct-1 – Ct-2)

 Using the formula and the above illustration,

C1 = .045, C2 = .061, A = .593, B = .566

This produces the following results:

                 Cumulative Audience

                                                Number of Issues                    Proportion Reached

                                                            3                                              .0707

                                                            4                                              .0776

                                                            5                                              .0829

 

The frequency distribution for any reach can be obtained by using the same set of input in a slightly different format: Let D = A - B.  Then the formula for obtaining the frequency s out of a total of t issues is:

formula1                                     

 

where initially

 

Formula 2

 

 

It should be borne in mind that all extensions beyond the empirical data are hypothetical and although useful, based on assumptions that may or may not be warranted.  These assumptions are:

            1)         Each issue has the same audience.

2)         The turnover (or its corollary, the duplication) is the same between every pair of issues.

  Back to top of page

The method is useful, therefore, when the audience of a magazine is reasonably stable.   The method can also be applied to demographic and marketing segments of the audience, although as the bases become smaller, reliability tends to decrease.  Moreover, an additional assumption; i.e., fixed composition, is now implied.

2. Newspaper Audiences

            a. Daily Newspaper Audience: All respondents who read a daily newspaper yesterday      (or on the most recent weekday).

            b. Sunday/Weekend Audiences: All respondents who read a Sunday (weekend) newspaper within the past seven days.

             c. Newspaper Cumulative Audience: Cumulative audiences of newspapers are obtained using a frequency question in the same manner as magazines.

            d. Newspaper-Distributed Magazines: The audiences reported for newspaper-distributed magazines are the measured audiences of their carrier newspapers, which is standard practice in newspaper research.

3.  Broadcast Data

Data are collected for both radio and television for an average weekday (based on yesterday or last Friday) and for each of the most recent two weekend days.  In each instance, the number of half-hours watched (listened) within major time slots is obtained.  This is used to produce two types of data:

a. Cumulative audience: The total number of people viewing (listening) within a day or day-part.  In addition, radio estimates are obtained by format and network.

b. Average half-hour audience: The average half-hour audience within each time period is obtained from a weighted average; i.e., the number of half hours viewed divided by the total number of half hours in the time period.

c. Television program audiences: Viewing of current television programs is obtained using a series of respondent-completed questions in the product questionnaire. These questions are:

1)          How many times a (month) (week) do you usually watch... (Followed by a list of weekly or daily programs).

2)          Did you watch the program in the past seven days (yesterday)?

3)          If you watched the program in the past seven days (yesterday), how much attention were you paying?

4)          If you watched the program in the past seven days (yesterday), where did you watch it? The responses to these questions are used to develop audience estimates for programs (“Yes" to watched in the last seven days, or yesterday for daily programs). The frequency question is used to develop cumulative audiences, and the other two questions are used to produce estimates of in-home audience and degree of attentiveness.

4.        Cable Networks:

           Data are collected from respondents living in households subscribing to cable or households with a satellite dish or disc.  The following question is asked for a list of approximately 64 cable networks and 5 premium cable channels:

a)      Have you watched in the past 30 days?

b)      About how many hours have you watched (network) in the past 7 days?

           Responses are used to develop both weekly cumulative audience estimates and average number of hours-per week estimates for individual cable and premium channel networks.

5.                   Yellow Pages:

Data are collected about frequency of using the yellow pages by means of the following questions:

a.        When was the last time you, yourself, had occasion to refer to the yellow pages either at home, at work, or elsewhere, using the phone book or the Internet?

b.         (If in the past seven days)  Now, thinking of the just the past 7 days, how many times did you refer to the yellow pages either at home, at work, or elsewhere using the phone book or the Internet?

c.         How many of these times in the past 7 days did you use the yellow pages in the phone book itself?

d.        How many of these times in the past 7 days did you use the yellow pages on the Internet?

e.        Now thinking back to the time(s) you used the yellow pages in the past 7 days, how many of these times did you refer to the yellow pages at work either by the phone book itself or on the Internet?

          

             Responses are used to develop yellow page usage estimates, including the creation of terciles of yellow page-usage and a media code for use in cost ranking analysis.

5.         Internet/On-Line Usage:

A series of questions are asked about Internet availability and usage in the last 30 days, place of access, activity on the Internet and use of the World Wide Web.  Similar questions are asked about using or looking at an on-line service in the last 30 days.

These responses are used to develop estimates of:

a.       Where is the Internet available;

                        b.   Where Internet used in the last 30 days;

            c.    Internet activities done in the last 30 days;

            d.    How often look at or use Internet in a typical month;

e.       Commercial on-line computer services looked at or used anywhere in the last 30 days (whether household subscribes or not);

f.       How often look at or use on-line service in a typical month;

g.      Access of the World Wide Web in the last 30 days;

h.      Websites visited in the last 30 days; Websites visited yesterday.

7.         Quintiles

Quintiles of exposure to the six media are generated from the recorded data, separately for men and women.  In each instance quintiles are generated so that, if required, a single frequency may be assigned to either adjacent quintile.  The specific definition for the quintiles is based on the most recent wave of data. These are contained in the appendix of this guide.  The measures used to define these are as follows:

a.          Magazines:  The total number of magazines read in a 30-day period, obtained by weighting reading a weekly by 4, reading a bi-­weekly by 2, reading a tri-weekly by 3, and reading a monthly by 1, etc., and then summing the total of these weights.

b.       Newspapers:  The number of newspapers read in a 28-day period, obtained by multiplying the number of daily newspapers “read in the past week” (using issue frequency claims times “read yesterday” newspapers) by 4 (the number of weeks in a 28-day period) and multiplying the number of weekend/Sunday newspapers ”read in the past 4 weeks” (using issue frequency claims times “read in past 7 days” weekend/Sunday newspapers) by 1, and summing the total of these two products.

c.          Outdoor:  Based on the number of miles traveled by motor vehicle in the last week.

d.         Radio:  The number of half hours of radio listening per week, developed by adding the sum of the weekend half-hours to five times the sum of the daily half-hours.

e.          Television:

Prepared in the same manner as radio using the counts of half-hours viewed daily and on the two weekend days.  Two quintiles are developed, one for total TV and one for primetime TV, the latter based on the reported half hours viewed in primetime.  (Terciles are created in a similar manner for daytime television viewing.)

f.          Internet: Based on how often the Internet is used or looked at in a typical month.

8.         Media Comparatives:

In addition to the quintiles, the same measures are used to develop comparatives – moieties or half codes - for each medium.  The total population is divided into two equal parts based on exposure to each of the five media, then identified as heavy and light exposure groups.   These can be combined across media into any desired combination of heavy and/or light exposure populations.

9.         Qualitative Magazine Measures

In the personal interview, a series of questions is asked of all readers of each magazine.  The questions are administered using show cards that display all responses and their corresponding codes.  These are:

a.          Where the magazine was read (at home, at work, etc.);

b.         On how many different days the magazine was read;

c.                    How much time was spent reading on the last reading day and how many issues were read that day;

d.         What actions the reading precipitated (clipping, couponing, etc.);

e.          What percentage of the pages were read or looked at;

f.          How the magazine was obtained (subscription, newsstand, borrowed, etc.);

g.          The overall rating the reader assigns to the magazine;

h.         How much interest the magazine’s advertising holds for the reader.

This range and variety of data provides media analysts with a multidimensional array of attributes for evaluation and media planning.  It affords the opportunity for scaling and other types of augmenting and discounting.  By detailing attributes of the exposure experience, these data can be used to measure in a more detailed way the advertising value of various types of readers of the measured magazines.

10.       Primary Reader Adjustment

A primary reader is defined as a reader residing in a household in which some household member either subscribes to or purchases the magazine at a newsstand.  Any reader who claims the magazine was so obtained is initially classified as a primary reader.  However, in this study (and in most readership studies that attempt to measure source of copy) the purchase and subscription claims, compared with Audit Bureau of Circulation statements, appear to be fairly consistently overstated.  Unadjusted, this would lead to an overstatement of primary readers.  It is a long­standing and widely accepted practice in survey research to utilize reliable and accurate external data to adjust, scale, or weight survey data.  In readership surveys it has become standard practice to adjust primary claims to circulation data.  In the GfK MRI study this is accomplished by the following procedure:

a.                    For each wave of fieldwork, the circulation of each magazine is obtained.  An upper limit of two primary readers per copy is set.  The primary readers of all magazines having two or fewer primary readers per copy are not adjusted.

b.                   For each magazine having more than two primary readers per copy, the number is reduced to two by randomly designating the requisite number of primary readers and recoding them as secondary readers.  The reduction selection is designed to maintain the observed distribution of male and female readers.

c.                    When primary readers per copy within sex exceeds 1.35, another random procedure is performed to reduce the level to no greater than 1.35.

d.                   Similarly, if the projected number of single-copy purchasers or sub­scribers exceeds a magazine’s total circulation, the requisite number of these is randomly selected and reclassified to “other primary” prior to the overall evaluation of primary readers.  In this selection, the reduction is designed to maintain the observed distribution of male and female single copy purchasers/subscribers.

            11.       Page Exposures

            Page exposures are a measure of the average number of times the average page of a magazine is seen by an average reader.  It is derived as follows, respondent by respondent, for each magazine read:

a. The number of days multiplied by the number of issues read on the most recent day produces an estimate of issue-reading days.  If this statistic is in excess of 50 for any magazine for any respondent, as it is on very rare occasions, it is reduced to 50.

b. The number of issue-reading days multiplied by the percentage of pages read on the most recent reading day produces total page exposure.  If this statistic is greater than 0 and less than .1, it is made equal to .1.  All values greater than 9.9 are made to equal the mean of all such values (approximately 16.0).

These two types of alterations (1 and 2) reduce the variance of the estimates that is otherwise drastically affected by extreme values.

E.         Marketing Data Estimates

            Mainly, two types of data are collected in the leave- behind marketing questionnaire; i.e., users and usage.   “Users” refers to the number of people who report the purchase or use of a product or service within a specified period of time.   This segment can be described in terms of demography, media exposure, and other of consumption behavior. The second type of data, “usage,” refers to a quantitative measurement of product or service use, such as “amount used” (number of rolls of aluminum foil), “number of times or occasions” (three or more trips to a department store) or “dollars spent” (amount spent for men's suits in the past year).  In many instances, the usage time frame is shorter than that for users.  These two types of data are used to generate further descriptions of users and usage as follows:

  Back to top of page

1.         Volume Usage

Users are classified as light, medium, or heavy users depending on their relative consumption or use of a particular product.  In general, the goal is to divide product users into three user groups each including about one-third of all users.

2.         Brand Users

Users of branded products are classified into one of three types for each brand used, based on evaluation of the brand used and corresponding volumes, as:

a.                      Sole users: Use only one brand

b.                     Primary users: Use more than one brand, but one more than of all the others

c.                                            Secondary users: Use more than one brand but do not qualify as primary users.

 

IV.         THE STUDY REPORTS

Back to top of page

Reports are based on the two most recent waves of fieldwork. The semi-annual reports are, in fact, one year moving averages, with each wave of data being utilized in two successive reports.

A.        Tabulation Plan - Fall

The tabulation plan calls for the production of three volumes each Fall:

1.         M-1 Magazine Total Audience Volume:  This volume contains audience estimates for magazines, cross-tabulated by the major standard demographic categories such as age, income, etc., and a selected set of nested demographics. Estimates are shown separately for adults, men, women, female Principal shoppers, male Principal shoppers, managers/professionals and working women.

 

2.         M-2 Magazine Qualitative Audiences: Primary Audiences, place of reading, reading days, reading time, reader activities, reader attitudes and cumulative audience.

3.         M-3 Multimedia Audiences: This volume includes cable TV, newspapers, radio, on-line services, and demographics, cross-tabulated by demographics (Available on Reporter only beginning in Fall 1997)

4.         Product Summary P-S:  This report contains top-line estimates for all of the products and services contained in the product booklet.  Included are total users and usage, kinds, types, brands and volume data, where applicable. Products and services studied for the first time are reported on for the most recent six-month wave, while all other products are reported for the two most recent waves. The definitions of heavy, medium, and light users, as well as the corresponding volumes of consumption are shown. (Available on Reporter only).

B.         Tabulation Plan - Spring

The Spring program calls for a much broader and more extensive system of reporting.

            1. M-l Magazine Total Audience Volume: This volume’s contents are the same as in the Fall.

2. M-2 Magazine Qualitative Audiences: This volume includes primary audiences, place of reading, reading days, reading time, reader activities, reader attitudes and cumulative audiences.

            3. M-3 Multimedia Audiences: This volume includes cable TV, newspapers, radio, on-line services, and demographics, cross-tabulated by demographics (Available on Reporter only beginning in Fall 1997)

            4. P-l through P-22: These reports contain a product summary and the details of each product/service, including brand, types and number, cross-tabulated by demographic segments and media audiences, thus providing a very detailed segmentation of the product data. (Available on Reporter only).

C.         Doublebase Reports

The Doublebase consists of four consecutive waves (two years) of data and is updated annually.  The Doublebase reports are:

1.         Mediamarket Reports: These reports are available in MEMRI and the electronic codebook; codebook pages are also sent to clients

A report for each of the 10 major markets and a national summary (see appended list and definition of these markets).

          2.         Upper Deck Report: These reports are available in MEMRI and the electronic codebook; codebook pages are also sent to clients

A report on the demography, media exposure and product/service consumption of the affluent population (upper ten percent of households ranked by income).

         3.         Business to Business Report: These reports are available in MEMRI and the electronic codebook; codebook pages are also sent to clients

  Back to top of page

This is a report on the demographic and business characteristics and business-related product/service usage of business decision-makers.

D.         Format of Report Data

For the most part, the tables in the various reports are cross-tabulations of one set of data by another, for specified population groups.  A standard format is employed, showing four different numbers, as follows:

1).         Projected Number:  The projected number in thousands;

2).         Vertical Percentage:  The proportion of the column total;

3).         Horizontal Percentage:  The proportion of the row total;

4).         Index of Selectivity.

            The index shows the ratio of the horizontal percentage of the detail row to the total row.  In other words, this index shows the extent to which the reported data have a higher or lower concentration in the population segment represented by the detail line compared to the total population.  An index over 100 means greater concentration, and one under 100 less concentration.

            In study reports, projected numbers based on fewer than 50 respondents are indicated by an asterisk (*), indicating that these estimates should be used with caution.  This standard is also used for estimates reported in MEMRI.  The two sigma tolerances on these types of estimates generally are at least 40% of the estimate itself.  Percentages and indices are not shown where a row (or column) total is based on fewer than 50 respondents.

E.         Sampling Tolerances

All sample surveys are characterized by sampling tolerances. Sampling tolerance is the difference that can be expected between the results of a sample survey and the results of a full survey or census, using the same procedures and techniques.  This is the difference due to the chance selection of one group of respondents or another.  In sample surveys, the actual sampling tolerance is not known.  What can be determined is what the samples of the specified size and design can be expected to have. Sampling tolerances are dependent on the size of the sample, the incidence of the particular characteristic and its homogeneity in the population.  Other things being equal, larger samples and higher incidences tend to have lower relative sampling tolerances, and characteristics that are evenly distributed tend to have smaller relative sampling tolerances than those that have uneven occurrences.  The sampling tolerance is a very specific statement.  It states, "In 95% of the samples of this size and type, the difference between the sample estimate and true value will not exceed plus or minus N, where N is the sampling tolerance."

Sampling tolerances for the magazine and other media audiences are tabulated for each report series, and are contained in the respective Reports.  These are computed using a set of eight replicated sub-samples of the total sample.  The differences among the eight sub-samples, which are chance differences, are used to estimate the sampling tolerances of the total sample.

The sample tolerances should be used to evaluate the precision of an estimate and the degree of confidence that can be placed in it.

The tolerance tables specify two-sigma tolerance limits for particular estimates.   Frequently users of data may want to evaluate whether the difference between two estimates is significant or due to chance.  This can be done as follows:

formula3

 

 

Back to top of page

where A is the sampling tolerance of the first estimate and B is the sampling tolerance of the second estimate.  K then equals the chance variation or sampling tolerance of the difference between A and B.  If the actual difference divided by K is higher than 2, it lies outside the two-sigma range and can be accepted as a real difference; if it is equal to or lower than 2, it may be due to chance factors in the sample process, since it lies within the two-sigma range.

F.          Access to the GfK MRI Database

Each Spring and Fall, as the data become available, computer tapes and other forms of data storage, containing all of the available data in coded form for each respondent and the weights necessary for the projections, are released to the subscribers and on-line services.  Codebooks specifying the code and location of each data item are also provided.   Subscribers are thereby afforded the capability of accessing this database and extracting their own specific analyses.  Since all of the data come from a single source, all types of cross-tabulations are possible.

The Doublebase files are updated annually, as are the special files containing volumetric product data.

 

G.         Limitations

1.  Non-responding and non-reporting persons may have media habits which differ from those of respondents.  Therefore, non-responding persons and other limitations in the original sample prevent the in-tab from being a perfect probability sample.  In addition, effort is made to exclude households with media affiliation.  The inclusion or exclusion of such households from the sample is dependent upon information revealed by the sample household in response to GfK MRI’s media affiliation question at the time of the personal interview.

2.  The personal interviewer may not always follow GfK MRI’s instructions.  Also, the interviewer may not be under the direct control of GfK MRI Research & Intelligence, LLC, as GfK MRI uses independent marketing research suppliers.

3.  The sample design and/or response patterns may preclude proportional representation of certain groups within the population such as ethnic groups, racial groups, persons in certain income or education groups, or any persons whose primary language is other than English (or Spanish).  Such persons may have media habits that differ from other persons.

4.  Estimates from the U.S. Bureau of the Census and from Market Statistics are used by GfK MRI. to make population estimates.  These estimates are based upon the most recent available decennial U.S. census and are subject to all limitations inherent therein.  In addition, population estimates are subject to limitations such as sampling errors, errors in locating undocumented populations and processing and recording errors.  Furthermore, the sources used by Market Statistics to update populations between decennial census dates may not include adjustments for known or unknown over– or undercounts of various segments of the population, including undocumented population groups.  In addition, annual population updates may be based on the results of sample surveys and are subject to their respective limitations.

5. Self-administered product booklets may be completed improperly if the respondent does not follow the booklet instructions.

6.  Human and computer processing errors may occur before or after GfK MRI receives the personal interview and the product booklet.  Consequently, the degree of variance in the data may be greater than that expected from sampling variance alone.

7. The data upon which GfK MRI has based its in-tab sample weighting, including racial or ethnic identification may not be precise.

8. Defects and limitations found in data supplied by others (e.g., SSI, Audit Bureau of Circulation, Survey Sampling Inc.) are inherent in GfK MRI estimates based thereon.

 



   [1] Atlanta replaced Cleveland in the sample beginning in the Fall of 2005 (Wave 53).

[2] Dallas-Ft Worth replaced St. Louis in the sample beginning in the Spring of 1986 (Wave 15).

 

Back to top of page

 

Updated February 2009

For more information regarding GfK Mediamark Research & Intelligence, contact GfK MRI at info@gfkmri.com

Copyright 2010, GfK Mediamark Research & Intelligence, LLC