A Study of Media Preferences and Media Consumption

Citation:

George H. Haines Jr. and David C. Efron (1972) ,"A Study of Media Preferences and Media Consumption", in SV - Proceedings of the Third Annual Conference of the Association for Consumer Research, eds. M. Venkatesan, Chicago, IL : Association for Consumer Research, Pages: 759-782.

Proceedings of the Third Annual Conference of the Association for Consumer Research, 1972      Pages 759-782

A STUDY OF MEDIA PREFERENCES AND MEDIA CONSUMPTION

George H. Haines Jr., Graduate School of Management, University of Rochester

David C. Efron, Graduate School of Management, University of Rochester

[The research reported in this paper was supported in part by the Systems Analysis Program, The University of Rochester, under Bureau of Naval Personnel Contract Numbers N00022-70-C-0076 and N00022-71-(-006) and in part by the Brookings Institution Workshop on Consumer Affairs at the University of Rochester. However, the conclusions, opinions, and other statements in this paper are those of the authors and not necessarily those of the Systems Analysis Program or the Brookings Institution. The authors are indebted to V. Srinivasan for helpful comments.]

It is well known that there are different consumption rates of certain products (and brands) between demographic groups such as blacks ant whites. There are, of course, several possible explanations for such phenomenon. However, investigations on the topic of race, such as Bauer, et. al (1965), Stafford, Cox, ant Higginbottom, (1968), ant Oladipupo (1970), have usually aimed at determining whether differential consumption patterns exist, not on why they exist. One study (Oladipupo, 1970) on Clairol hair coloring reported evidence that differential information sources were one of the causes of differential product consumption rates. This could imply, if generally true, that the market for consumer goods may not be efficient in an economic sense and that this lack of efficiency may cause consumption rate differentials. [An efficient market is one in which market prices fully reflect all available information (Fama, 1970).]

Further, studies that have attempted to explain consumption rate differences tue to income ant education (ant some other demographic variables, but not race) on a basis of market segmentation have proved disappointing; for example, see Frank ant Massy (1967). Again, one explanation for such results is that consumer goods markets are not efficient.

If this is the case, systematic differences across demographic groups in media consumption ant program preferences within a media should be found. [This paper labels the time a consumer spends watching television, etc., as consumption; hence the phrase @'media consumption". However, watching television, etc., can be used to produce commodities, in the sense of Becker (1965), by combining inputs of goods and time; hence the view of time spent on media as being pure consumption is incorrect. Indeed, the Becker (1965) theory is a necessary link to explain why differential media use could be a reflection of market inefficiency.] Therefore, this paper studies the proposition that there are systematic differences in media consumption ant program preferences within media caused by race, income, ant education

LITERATURE REVIEW

Race, socio-economic class, and level of education have been fount to have distinct effects upon individual consumption of ant preference for media.

Samuelson, Carter, and Ruggels (1963) argued that education was a key variable in explaining media consumption. The argument presented was that as one's level of education increases, one's appetite for all media grows. At the same time participation in other activities grows. There being a fixed number of hours in a tay, the end result is a person must give up some of the media consumption he would desire if there were no time constraint. Television loses the greatest amount of educated audience time. These people, SCR claim, substitute job-connected working hours, hobbies, organizations, ant time with their children for some of the hours they would otherwise spent watching television. As education increases, media competing with television also gain. Better education brings better reading skills, ant so magazines ant books win readership at the expense of television's audience.

The authors present empirical evidence to support these assertions. Correlations between education ant 1) job-connected hours per week, ant 2) number of organizations active in both are significant at a Type I error of 0.05 ant positive, while correlations between television and newspaper consumption (in hours per week) and job-connected work hours per week are negative. The authors then adjust the data to remove the influences of jobs, volunteer organizations, hobbies, and dependent children. Partial correlation methods are used; the correlation between education and media now becomes positive for all media studied.

Media credibility has also been found to be related to educational achievement (Westley and Severin, 1964). The higher educated tend to trust newspapers more and the less educated tend to trust television more. "Social interaction," defined as participation in group activities, was also found to be related to media credibility by Westley and Severin (1964). Membership in three or more organizations is correlated (positively) with trust in newspapers, while the socially inactive put their faith in television. This finding corroborated the role involvement hypothesis of Samuelson, et. al. (1963).

Maxwell E. McCombs (1968), studying sources of political information, found race to be related to media consumption while education was important in explaining preferences for various mass communications material. Race, in part, explained the quantity of political information sought, but education was related to choice of media. Those of high school education or more, regardless of race, chose newspapers most often for political news; those who had not attained a high school degree were consistent viewers of television reports.

Gerson (1966) identified race as the dominant variable in media consumption patterns. This finding concurs with Carey's (1966) results. Carey argued the Negro experience was a set of factors which explained the significance of the race variable. Gerson argued that teenage blacks first join the dominant white sub-culture when they enter high school. Lacking preparation for the dating relationships common in this environment, they seek the mass media for advice. The media provide "...a mechanism...through which individuals learn to be motivationally and technically adequate in the performance of certain roles" (Gerson, 1966, p. 49). Therefore, one should find blacks using mass media, particularly television, in greater amounts than their white counterparts.

As can be seen, up to this point the research tradition had essentially studied demographic variables one at a time. Greenberg and Dervin (1970, 1970) attemPted to rectify this weakness. The variable they added to race was income.

Greenberg and Dervin (1970, 1970) found that, although their sample interviewees differed when broken into low income (OEO designated areas; personal interviews used to collect data) and "general" populations (telephone survey used to collect data), the differences disappeared when the low income group was split black and white. For example, the low income sample spent 5.2 hours of the typical 16-hour working day watching television whereas the "general" population viewed for only two hours. Another difference lay in program preferences. Of twelve top-rated shows, a rank-order correlation between the low income sample and the general population sample is 0.03. However, when the low income sample was separated into racial groups no consistent differences appeared in program preferences. Similar findings are reported for media preferences for world news. The low income sample definitely prefer television, the general population is equally satisfied by newspapers and TV, and both black and white sub-samples of low income respondents prefer television.

Preference for local news produces a sharp contrast. The low income and general population samples named radio and newspapers as first choice, respectively. The low income blacks preferred people as sources of local news, but whites of similar economic background were indifferent between TV, radio and newspapers. One possible explanation for this, advanced also by Honig, et. al. (1972), is a lack of news coverage in the inner city which creates dissatisfaction among blacks with the media's presentation of local news.

That the media, especially the broadcast media, do not program news with black interest in mind is further substantiated by Carey (1966). This paper, however, is in agreement with Greenberg and Dervin only on this point of news coverage. In fact, Carey (1966) generally contradicts all they say about the racial subsamples being not significantly different. He finds hardly any coincidence in preference among the top 40 television shows in 1963

Carey (1966) concludes that the Negro experience forms a basis for black preferences which not only are different from white program and media preferences, but are also usually counter to the "cliche content" of most programs. Bauer and Cunningham reach the same conclusion: "Negroes are more concerned with matters close to their own life situation, and correspondingly they use the media proportionately more than do equal-income whites for recreation, diversion, and escape, and less as a way of maintaining contact with the realities of the world around them -- except for using advertising to learn what is going on in the market" (Bauer and Cunningham, 1970, p. 124). Bauer and Cunningham's figures show blacks to be heavier consumers of television and radio, lighter consumers or magazines and newspapers, than whites.

So the research findings end on a note of almost total disagreement between Greenberg ant Dervin and the previous literature. There are three possible explanations for this:

1)  Everyone else is wrong;

2)  Greenberg and Dervin's results are idiosyncratic ant cannot be replicated;

3)  Greenberg ant Dervin's results are replicable, but the conclusions Greenberg ant Dervin draw from them are incorrect. This could be the case if (a) their analysis technique was a poor one for the problem at hand, ant/or (b) their exclusion of the effects of education tended to mask racial effects.

REPLICATION OF GREENBERG AND DERVIN

The Greenberg and Dervin Data

Two samples were taken by Greenberg and Dervin in their 1967 survey: (1) a "low income" sample consisting of random personal interviews in an area of Lansing, Michigan designated by the Office of Economic Opportunity (OEO) as having a high concentration of low income residents, ant (2) a general population sample taken randomly from the Lansing telephone book. This latter sample was interviewed by telePhone.

The Rochester Data

The data in the present study are drawn from two surveys performed in the Rochester, New York, area. The first, conducted during 1969, collected data from a random sample of residents of Monroe County, New York. The sampling plan was designed to collect a higher proportion of respondents living in areas where black people lived than from white ghetto areas. All sampling was tone on a personal interview basis by a specially trained force of black interviewers. [The collection of these data was supported by a grant from the Consumer Research Institute, Inc., and would not have been possible without the aid of Marcus Alexis and Leonard S. Simon. We wish to express our gratitude to all the above, for without the data this paper would not have been possible.]

The 1971 survey was designed to be a random survey from Monroe and surrounding outlying counties. The sampling plan was designed to draw randomly and proportionately. Professional interviewers, all white, were used. All interviews were personally collected. However, there was an under-representation of blacks in the sample because the professional white interviewers tended to be unwilling to conduct in home interviews in certain inner-city areas. The effects of this selection bias will be discussed later. The present analysis uses only the data from Monroe County (the city of Rochester is entirely within Monroe County). [The collection of these data would not have been possible without the aid of James Peck and Daniel Braunstein. Once again we wish to express our gratitude for this help.]

Greenberg and Dervin, using the standard x2 measure of association test, show that no significant differences exist between their law income and general populations on the number of working television sets, but do find significant x2 values on color television ownership, hours television viewed yesterday, and the medium preferred for world news. However, when they split their low income sample into blacks and whites they found no significant differences at a 5% Type I error level. The replication analysis is concerned with these four questions.

The first question is whether the results are an artifact of the statistical test used. Therefore, Greenberg and Dervins' data was first reanalyzed treating the percentage of respondents in any category as an estimate of the parameter of a multinomial distribution. Differences across samples or race are then tested for by using a x2 test of homogeneity of multinomial distributions (Potthoff and Whittinghill, 1966). [Copies of a BASIC program for performing the computations of this test are available from the authors on request.] Table I presents the results of this reanalysis of the Greenberg and Dervin data. The results with the homogeneity test statistical procedure are exactly the same, in a qualitative sense, as the results reported by Greenberg and Dervin (1970, 1970) using the traditional x2 association test.

Tables II and III present replications of these tests with the 1969 and 1971 Rochester survey data. As with Greenberg and Dervin, respondents are divided into groups according to whether they live in the central city (Rochester) or in the surrounding suburbs. The samples of city residents are split according to race.

Something has gone wrong! The results certainly replicate Greenberg and Dervin's in the sense that low income blacks look like low income whites (or vice versa), but now there seem to be no significant differences between city and suburban residents, aside from a couple which could have occurred by chance. What could have happened?

One possibility might be the kind of breakdown used by Greenberg and Dervin. They split their sample geographically according to whether a person lived in a low income area or not. However, the Rochester surveys collected actual income data from respondents. One discovery which arose from this was that many people who live in low income areas do not have low incomes, and, obviously, there aye some low income people living in the so-called medium or high income areas. [This is really not a very startling discovery. The reader should also see Harrison (1972, p. 26), where frequence distributions of male weekly earnings in March 1966 by race by residential area type are presented.]

Therefore the analysis was redone, breaking on the income data directly rather than attempting to use place or residence as a proxy for income. The results, for the 1969 and 1971 data, are presented in Tables IV and V respectively. Everyone living in the city of Rochester and earning less than (or just) $5,000 per year was compared to the suburban sample. The low income city sample was also split on race; all respondents who were neither black nor white are excluded on this racial comparison. These results show no significant differences except when color television ownership is compared across low income and Monroe County residents in the 1969 survey results. Everyone is alike! Greenberg and Dervin's claim that suburban people are different is not replicated in these results. Their claim that low income blacks and whites are alike is replicated, but this runs counter to all other findings in the research literature.

TABLE I

GREENBERG AND DERVIN [12] DATA REANALYZED USING MULTINOMIAL X2 HOMOGENEITY TEST

TABLE II

REPLICATION OF GREENBERG AND DERVIN 1969 ROCHESTER SURVEY DATA

TABLE III

REPLICATION OF GREENBERG AND DERVIN 1971 ROCHESTER SURVEY DATA

TABLE IV

REPLICATION-OF GREENBERG AND DERVIN USING REPORTED INCOME DATA -- 1969 SURVEY RESULTS

TABLE V

REPLICATION OF GREENBERG AND DERVIN USING REPORTED INCOME DATA -- 1971 SURVEY RESULTS

Before assessing the implications of these results one final question must be asked: are the samples the same? The low income samples on television ownership are, again taking a 0.05 level of Type I error (for whites, x2 - 8.41, 3.49 d.f.; for blacks, x2 - 5.56, 3.16 d f.). The general populations differ on responses to this question (x2 - 12.19*, 3.49 d.f.). All samples differ on color television ownership, reflecting the increase in ownership of color television. Hours viewed yesterday are also all quite different; preferences for news are the same in the two low income samples but are different in the general populations. [The results are as follows:

Color Television Ownership:

General Population X2 = 35.86*, 1.95d.f

Low Income White X2 = 90.37*, 1.99d.f.

Low Income Black X2 = 15.74*, 2.02d.f.

Hours Viewed Yesterday:

General Population X2 = 103.47*, 5.76d.f.

Low Income White X2 = 25.09*, 5.91d.f.

Low Income Black X2 = 34.28*, 6.02d.f.

Preferences for News:

General Population X2 = 13.95* 2.68d.f.

Low Income White X2 = 0.31, 3.27d.f.

Low Income Black X2 = 0.11, 2.01d.f.]

Now, what can be concluded? First, it appears that Greenberg and Dervin have really shown one gets different responses to questions on media use, consumption, and preferences if one uses telephone rather than personal surveys. Second, it appears that using homogeneity tests may not be a very good technique for studying the general problem. These tests are at once too general and too restrictive. Previous research suggests race, income, and education all effect media consumption and preferences. But homogeneity tests tend to push a researcher into simple comparisons; an example is the exclusion of education in the Greenberg and Dervin analysis. In this sense they are too restrictive. They are also too general: they ask "are there any differences," when past research has provided clear evidence for the direction differences should take if they exist. Simple homogeneity or association tests make no use of such important information.

The purpose of the next section is to investigate this issue further, using an appropriate analysis technique, one which will allow for simultaneous study of effects of race, income, and education, and for use of knowledge of the direction of the effect of these variables on media consumption and program preferences.

ANALYSIS METHOD

The analysis model devised by James Coleman (1964) is used to see whether systematic differences in media consumption and media preference "caused" by demographic factors exist. Three independent demographic variables are examined: race, income, and education.

The basic structure of the Coleman analysis may be briefly outlined for the case of one independent variable (see Figure 1). [A more extended exposition of the Coleman analysis applied to a marketing problem is given in Whittaker (1972); see also Alexis, Haines, and Simon (1972).] There are two possible states a respondent may be in: that of being a light user of television, and that of being a heavy user of television. Whether a respondent is in one of these states depends upon the initial state of the respondent and, in Figure 1, the respondents race and random factors e1 and e2. e1 is a random shock tending to lead a heavy user to being a light user of television; e2 is a random shock in the opposite direction. Bx is the effect being white has on creating a light user of television, ax is the effect being black has on leading a respondent to be a heavy user of television. It is assumed ax = Bx; that is, the effects of race are equal, but operate in different directions. This is a useful and Usual assumption in cross-sectional analysis; relaxing it in any meaningful way requires time series data.

FIGURE 1

BASIC STRUCTURE OF COLEMAN ANALYSIS MODEL

EXHIBIT I

MEDIA CONSUMPTION QUESTIONS

MEDIA CONSUMPTION

The questions asked and analyzed are presented in Exhibit I. A light user of television is defined as a respondent who answers less than ten hours to question 1; the same convention is used for question la. Since the analysis method yields m estimate of the direction of effects, alternative hypotheses were set down prior to running the analyses (in all cases the null hypothesis is that the demographic variables have no effect):

(1) Higher educated respondents tend to be light users of television,

(2) Higher income respondents tend to be light users of television, and

(3) White respondents tend to be light users of television.

The proportion of respondents answering yes is taken as the dependent variable in the other analyses (instead of the proportion of light users). The null hypothesis remains the same; the alternative hypotheses are equivalent to the above (e.g., white respondents tend to answer yes to question 2).

A coefficient with a negative sign that is statistically unlikely indicates the demographic variable has an effect opposite to the alternative hypothesis specified above. An example of this is the effect of income on television consumption; that is, the results given in Tables VI and VII indicate that higher income respondents tend to be heavy users of television when effects of race and education are removed.

The results from the 1969 survey are presented in Tables VI m d VII. In Table VI, A1, A2, and A3 are the coefficients of Education, Income and Race respectively; U1, U2, and U3 are standard normal deviates used to test the null hypothesis. R is a random shock toward the dependent state (for example, being a light user of television); S a random shock away from the dependent state. TAi is the standard deviation of the coefficients in the analysis. Table VII is a "small sample" analysis. B1, B2, B3 are the adjusted coefficients of Education, Income, and Race; U1', U2', U3' are standard normal deviates used to test the null hypothesis. Tables VIII and IX are analogous, except these tables report results for the 1971 survey data.

The results given in Tables VI and VII indicate important racial effects. Race is significant in every category except "book rather than television." Income is also important. Education is never significant, but the sign is always in accord with the direction predicted by the theoretical argument of Samuelson, Carter, and Ruggels (1963).

There are some additional conclusions which these results would support. First, the coefficient values and significance levels in the two alternative forms of the television consumption question are quite close, indicating that either question is an equally satisfactory measure of television consumption. Second, it is instructive to compare the television consumption results in Tables VI and VII with those in Table IV. The results in Table IV show no significant differences between the hours viewed yesterday distributions. The difference demonstrates the ability of the Coleman analysis procedure to yield more information from a set of data than even the parametric frequency distribution homogeneity test employed in Table IV. Finally, the results in Table VII are very close to those in Table VI and are the same qualitatively. This indicates that the large sample analysis is adequate in this case.

TABLE VI

1969 SURVEY RESULTS - REGULAR ANALYSES

TABLE VII

1969 SURVEY RESULTS - SMALL SAMPLE ANALYSES

TABLE VIII

1971 SURVEY RESULTS - REGULAR ANALYSIS

TABLE IX

1971 SURVEY RESULTS - SMALL SAMPLE ANALYSES

The results from the 1971 survey are quite different. There are no significant coefficients in Table VIII; Table IX, the small sample analysis, indicates, an important education effect, with income being significant once. Race is never significant, but Table IX, at least has the expected sign (Sample size problems did not allow a small sample adjustment to be made in the case of television consumption.). The most striking feature of Table VIII is the size of the standard deviation of the coefficients compared to the size of the standard deviation of the coefficients in Table VI: roughly five times greater. The variances are significantly different. For example, comparing the variances for "hours a day TV set is on" yields an F value of 32.2 (134, 150 d.f.).

Earlier it had been noted that in the 1971 sample difficulties were encountered in getting the random sampling plan fulfilled because the professional interviewers were loath to work in inner-city areas. Now the effects of this can be seen. Many people doing commercial market research work seem to feel that low income urban areas can be excluded because, after all, "who needs data on some poor blacks anyway?" What these results show is the fallacy of such an argument, for what has happened is that the information content of the entire sample has been greatly reduced by the interviewing forces' actions. [A second possible explanation is that the world has changed so as to greatly increase random error. While there is no way to exclude this interpretation on statistical grounds, there is also no reason to suppose it is correct.]

Television Program Preferences

The question asked and analyzed are presented in Exhibit II. The alternatives presented represent the actual programs scheduled on the three UHF television stations in Rochester in the spring of 1969; the nature of the alternatives leads to the supposition that there might be a strong racial effect on expressed preferences.

The analyses were run under the following alternative hypotheses (in all cases the null hypothesis is that education, income, and race have no effect):

1) as education level (A1) increases one tends to move toward the being interested in the program (health and birth control, for example),

2) as income (A2) increases one tends to move toward being interested in the alternative program, and

3) blacks (A3) are more interested than whites in the alternative program.

As before, a coefficient with a negative sign that is statistically unlikely indicates the demographic variable has an effect opposite to the alternative hypotheses specified above. R is a random shock toward preferring proposed program; S is a random shock away from preferring the proposed program. TAi is the standard deviation of the coefficients. These results are given in Table X. These results indicate a clear racial effect: the race variable is significant in every case except health and birth control. Education is never significant; in 3 of the 8 programs studied Income is significant. [A "small sample" analysis could be performed in four of the eight cases. The results were not greatly different. If anything, they tended to indicate the importance of the race variable may be slightly overstated in Table X. Copies of these results are available from the authors on request.]

CONCLUSIONS

Overall, what can be said? First that the time has clearly come to stop examining demographic variables' influence on consumer media use one at a time. The results plainly indicate that income, race, and education can all significantly effect media consumption. Similarly, income and race can both significantly effect television program preference. These are expected results given the initial "theory of market inefficiency" explanation of observed consumption rate differences.

EXHIBIT II

TELEVISION PROGRAM PREFERENCE QUESTIONS

TABLE X

1969 SURVEY RESULTS - REGULAR ANALYSIS

The results also tend to reinforce the notion that Greenberg and Dervins' interpretations of their results are suspect. Finally, from a methodological viewpoint, the results show very clearly the importance of ProPer execution of random sampling plans.

REFERENCES

Alexis, Marcus, Haines, George H. & Simon, Leonard S. Neighborhood effects, family decision-making patterns, and consumption expenditures. In Fred C. Allvine (Ed.), Combined Proceedings 1971 Spring and Fall Conferences. Chicago: American Marketing Association. 1972.

Bauer, Raymond, et.al. The marketing dilemma of Negroes. Journal of Marketing, 1965, 29, 1-6.

Bauer, Raymond A. & Cunningham, Scott M. Studies in the Negro Market. Cambridge, Mass.: Marketing Science Institute, May 1970.

Becker, Gary S. A theory of the allocation of time. The Economic Journal. 1965, 75, 493-517.

Carey, James W. Variations in Negro/White television preferences. Journal of Broadcasting, 1966, 10, 199-211.

Coleman, James. Introduction to Mathematical Sociology. New York: The Free Press. 1964.

Fama, Eugene F. Efficient capital markets: a review of theory and empirical work. Working paper, University of Chicago, January, 1970.

Frank, Ronald E. & Massy, William F. Effects of short-term promotional strategy in selected market segments. In Patrick J. Robinson (Ed.) Promotional Decisions Using Mathematical Models. Boston, Mass.: Allyn and Bacon, Inc. 1967.

Gerson, Walter M. Mass media socialization behavior: Negro-white differences. Social Forces, 1966, 45, 40-50.

Greenberg, Bradley S. & Dervin, Brenda. Mass communication among the urban poor. Public Opinion Quarterly, 1970, 224-235.

Greenberg, Bradley S. & Dervin, Brenda. Use of the Mass Media bs the Urban Poor. New York: Praeger, 1970 (esp. Ch. 1).

Harrison, Bennett. The intrametropolitan distribution of minority economic welfare. Journal of Regional Science, 1972, 12, 23-44.

Honig, David, Bomens, James, Brooks, James & Hickman, Darwin. Some sources of consumer revolt: the market research project, July-September 1971. Paper presented at the Conference on Consumer Affairs, Rochester, New York, June 9, 1972.

McCombs, Maxwell E. Negro use of television and newspapers for political information, 1952-164. Journal of Broadcasting, 1968, 12, 261-266.

Oladipupo, Raymond O. How Distinct is the Negro Market. New York: Ogilvy and Mather. 1970.

Potthoff, Richard F. & Whittinghill, Maurice. Testing for homogeneity: I. The binomial and multinomial distributions. Biometrika, 1966, 53, 167-182.

Samuelson, Merrill, Carter, Richard F. & Ruggels, Lee. Education, available time, and use of mass media. Journalism Quarterly, 1963, 491-496.

Stafford, James E., Cox, Keith K. & Higginbottom, James B. Some consumption pattern differences between urban Whites and Negroes. Social Science Quarterly, 1968, 49, 614-630.

Westley, Bruce & Severin, Werner J. Some correlates of media credibility. Journalism Quarterly, 1964, 325-335.

Whittaker, William S. The Relationship Between Individual Difference Variables, Media Usage, and Product Choice. Ph.D. Dissertations: University of Rochester, Rochester, N.Y., 1972.

----------------------------------------