Demographics and the Propensity to Consume

Marjorie Fox Utsey, Tulane University
Victor J. Cook, Jr., Tulane University
ABSTRACT - Demographics fell into disfavor as a tool of market segmentation when research showed weak links with product consumption. Today, consumer population surveys permit more rigorous testing of the dependencies between demographics and product consumption. In this exploratory study of 16 products, the hypothesis of independence is soundly rejected.
[ to cite ]:
Marjorie Fox Utsey and Victor J. Cook, Jr. (1984) ,"Demographics and the Propensity to Consume", in NA - Advances in Consumer Research Volume 11, eds. Thomas C. Kinnear, Provo, UT : Association for Consumer Research, Pages: 718-723.

Advances in Consumer Research Volume 11, 1984      Pages 718-723

DEMOGRAPHICS AND THE PROPENSITY TO CONSUME

Marjorie Fox Utsey, Tulane University

Victor J. Cook, Jr., Tulane University

[Doctoral student and Associate Professor of Marketing at the Tulane University School of Business, New Orleans, Louisiana, 70118]

[The authors thank the Tulane University School of Business for its generosity in funding this research. We also thank William Cook, Christopher Cook and Donna Mohr for their invaluable assistance with statistical and computer applications.]

ABSTRACT -

Demographics fell into disfavor as a tool of market segmentation when research showed weak links with product consumption. Today, consumer population surveys permit more rigorous testing of the dependencies between demographics and product consumption. In this exploratory study of 16 products, the hypothesis of independence is soundly rejected.

THEORETICAL FOUNDATIONS: ECONOMICS AND SOCIOLOGY

The idea demographics can explain the behavior of consumers has its roots in economics and sociology. The controversy in the literature during the past 30 years has failed to either disprove or confirm the value of demographics as a tool of market segmentation. The purpose of this article is to reopen the question.

Classical Heritage

When 20th century economists began to study consumption, their explanation was the by-product of theories of saying. The cornerstone of classical theory was Fisher's model of the saving decisions of an individual household (1930). Fisher's analysis focused on the impact of income, wealth and interest rates on saving (and therefore on consumption) and did not explicitly consider other socioeconomic variables or the family life cycle. However, the implication was clear that consumption need not be a function only of current income.

The Keynesian Revolution

Consumption occupied a central role in Keynes's general theory (1936). He defined consumption as a function of current net income and the propensity to consume. The marginal propensity to consume (mpc) was positive and less than one. A great deal of effort has been directed to tests and critiques of Keynes's hypothesis. There is general agreement that consumption is closely related in some way to current net income and that the mpc is positive and less than one. There is considerable disagreement concerning the exact nature of the relationships and the reaction of the mpc to changes in income.

The Relative Income Hypothesis

Duesenberry pointed out Fisher's analysis assumed each individual's preferences were independent of the purchase behavior of others (1949). Keynes also assumed preferences were independent in his general theory. In contrast, Duesenberry assumed that consumer preferences were interdependent. Pressure to increase consumption depended upon the ratio of a household's expenditures to the expenditures of those with whom it associated. Saving (and consumption) depended on the household's percentile position in the income distribution rather than on its absolute level of income. Duesenberry considered the household was influenced more by social contacts than by casual contacts. It therefore would seem logical that people of similar income, occupation and educational background should consume more homogeneously than people of different social status.

Neoclassical Thought

Friedman developed the permanent income hypothesis to incorporate the effects of wealth on consumption. Friedman divided wealth into a nonhuman component (assets) and a human component (education, occupation, ability and personality) (1957). Individuals held nonhuman wealth to smooth out consumption, earn interest and deal with the uncertainties of the income stream

Friedman defined both income and consumption as having permanent and transitory components. Permanent consumption of an individual household was some fraction of its permanent income. The particular fraction depended upon factors which in turn depended upon age, family composition, expectations and perhaps education. In addition, age, family size, education and occupation influenced the proportion of income considered permanent by the household. It rhus appeared age, family composition, education and occupation would contribute to a fuller understanding of consumption behavior.

Like Friedman, Ando and Modigliani were interested in the effects of wealth on consumption (1963). Their consideration of past, present and expected future incomes encompassed consumption behavior over the life cycle of the individual. Ando and Modigliani proposed "ln any given year, t, total consumption of a person of age T...will be proportional to the present value of total resources accruing to him over the rest of his life" (1963, p. 57). Proportionality depended upon interest, present age and utility. The theory suggested different age groups consumed differently; consumption behavior varied over the life cycle.

Socioeconomic Status and Social Class

Warner and others explored the relation of class membership to behavior (Warner, Meeker and Eells, 1949). Warner computed an index of status characteristics by scoring respondents on six (and later four) socioeconomic factors, multiplying by weights and summing the results. He then divided respondents into six classes according to their composite scores.

Warner found the index predicted class membership better than any single descriptor. He began with a composite of source of income, amount of income, occupation, education, dwelling type and dwelling area, then eliminated education and amount of income with little loss of explanatory power.

Warner conducted an empirical study of how the six social classes spent their money on 26 categories (Warner and Lunt, 1941). The different classes appeared to have different priorities, both in terms of the rank order of the proportion of budget spent for each item and the percentage of budget allocated to each item. Social class appeared to affect the propensity to consumer specific product categories.

Socioeconomic Status Reformulated

The Bureau of the Census developed a socioeconomic status index that could be computed from readily available data (1963). Census developed scoring scales for occupation, education and income. The chief income recipient in each household received a score of O to 100 on each of the three variables. An unweighted average of the scores was computed and the resulting scores were then used to construct a percentage distribution classifying families.

A Classification of Families by Life Cycle Stage

Loomis developed the family life cycle concept as a tool for studying the economic and social life of the family (1936). He divided the family life cycle into four stages: (1) childless couples of childbearing age, (2) families with the eldest child under 14, (3) families with the eldest child 14-36 and (4) old families.

Glick proposed a seven stage family life cycle and studied the age at which the average person entered each stage. He argued important changes occurred in many characteristics as family members passed through the stages. He studied variation in residence and selected economic characteristics by age of the husband. A distinct pattern emerged for each dependent variable (Glick, 1947).

Glick's analysis was not based strictly on life cycle stages, since only age was considered. Nevertheless, it provided an indication that the family life cycle could be useful in studying certain kinds of behaviors.

EMPIRICAL TESTS PRODUCE MIXED RESULTS

Early Findings in Favor of Demographic Segmentation

In the 1950's and 1960's researchers applied economic and sociological principles to explaining consumption of individual products. Many articles appeared on the relation of demographic variables to consumer behavior.

Lansing and Kish found the propensities to own a home, incur debt and buy a car or TV all varied with stage of the family life cycle (1957). Moreover, life cycle stage had more explanatory power for each of the dependent variables than an analysis based solely on age.

Pierre Martineau found Warner's concept of social class had substantial explanatory power with respect to spending behavior (1958). His results showed class differences in spending-saving patterns, choice of retail outlet and taste in products such as automobiles, clothing, furniture and housing. The study indicated "the lower-status person is profoundly different in his mode of thinking and his way of handling the world." (Martineau, 1958, p. 122)

Life Magazine conducted a study of "who spends how much on what" that was conceptually similar to Warner's study of spending patterns in Yankee City (Ostheimer, 1958). Life dealt with age, family life cycle, geographics, income, occupation and education. The study covered eight major categories of expenditures and 57 individual product categories. It concluded all the independent variables were useful in evaluating consumer behavior.

Demographics Fall From Favor

Marketers explored alternative methods of segmentation. Authors such as Wilson proposed the addition of new dimensions such as psychographics to enhance the understanding already provided by demographics (1966). Others offered new theories as alternatives.

Yankelovitch proposed markets be segmented by values, needs, attitudes and usage patterns relevant to the product (1964). He felt these characteristics were not usually reflected in demographic variables. Demographics were not a proxy for product specific characteristics and could not provide sufficient knowledge of how the segments differed.

Twedt suggested consumers be segmented according to their volume of product usage (1964a and 1964b). He divided purchasers of 18 mature packaged goods in half at the median of usage. He then observed the half above the median consumed 80 to 90 percent of the total volume for most products. This observation became his now famous "heavy half" theory. Twedt advanced five propositions concerning the heavy half. Among them was the allegation demographics is a poor predictor of heavy usage. Although Twedt provided no documentation or empirical support for these propositions, they were widely accepted. Marketers began to discard demographics as a viable classification system.

Haley recommended consumers be segmented according to benefits sought (1968). He created segments which differed in the relative importance their members attached to the various benefits available from the product. Product characteristics, media choice, advertising copy and other strategies were tailored to the values of the target segment .

Frank, Massy and Boyd tested Twedt's proposition demographics of heavy users did not differ systematically from those of the population (1967). They analyzed data for 57 grocery products with regression analysis. They judged the 14 demographic independent variables poor predictors of product consumption because the variables explained only small proportions of the variation.

Frank and others also studied the relation of demographic variables to private brand proneness, brand loyalty, package size proneness and average price paid (Summarized in Frank, 1967). All studies were regression analyses which assumed linearity and absence of interactions.

Frank found demographics did not discriminate well between purchasers of private brands and manufacturer's brands. He found little association of demographics with brand loyalty. Demographics were of some value in segmenting markets with respect to package size and price paid.

An Interesting Methodological Question

Bass, Tigert and Lonsdale argued the conclusion demographic segmentation was not viable did not necessarily follow from the low observed R2s (1968). "Market segmentation involves postulates about the characteristics and the behavior of groups, not persons. The absence of a satisfactory theory of individual behavior does not necessarily imply the absence of valid propositions about the groups' behavior" (Bass, Tigert and Lonsdale, 1968, p. 265.)

Bass et. al. tested the proposition groups of consumers with large differences in mean purchase rates could be identified with demographics. They examined purchases of 10 grocery products. The demographic variables were income, education, occupation, age and number of children. Bass et. al. recommended a cross-classification technique, but indicated analysis with a properly specified regression model should also produce valid results.

The regression model used dummy variables to avoid the questionable assumptions of linearity and continuity. Stepwise regression validated the significance of all independent variables. The analysis produced mean purchase rates that permitted clear identification of segments by volume of consumption.

Bass et. al. created contingency tables for each independent variable with each product. Chi square analysis indicated cell means differed significantly for 38 of the 50 tables. Contingency tables that used two independent variables simultaneously (and thus allowed for interactions) produced even greater differences in cell means. Bass et. al. concluded meaningful demographic segments existed and could be found through appropriate analysis.

Recent Applications

Ellis applied cross-classification to the residence telephone market (1975). He fitted income, education and occupation to numerical scales running from O to 100 and calculated socioeconomic status as an unweighted average of each household's numerical scores. He then sorted the composite scores into four categories.

Ellis grouped households into four family life cycle stages by ages of the household head and the youngest child. He then cross-classified households in matrix form by their socioeconomic and family life cycle status. He found "each cell...represents a specific and discrete class of people completely defined by the two factors." (Ellis, 1975, p. 489)

In a factor analysis designed to identify Bell system customer groups, socioeconomic status, family life cycle position, family housing and mobility accounted for 88 percent of the variation. Ellis also found socioeconomic status and family life cycle variables were highly correlated with lifestyle, values and attitudes in the residence telephone market.

In another cross-classification study, Blattberg, Bluesing, Peacock and Sen used demographic variables to identify a particular market segment (1978). They argued previous poor results in relating deal proneness to demographics stemmed from the methodological approach of previous studies and in some cases from improper specification of the independent variables.

Blattberg et. al. analyzed deal proneness for five grocery products. They found socioeconomic status variables predicted deal proneness for every product. The absence of small children produced a small increase in deal proneness for three of the products.

Is Demographics Really Dead?

We feel a number of problems may have prevented past studies from uncovering existing relationships. Investigation requires a sample of substantial size to provide sufficient observations in each cell of analysis. Many researchers in the past have been forced to work with small samples. The relationships among the variables appear to be quite complex. The presence of nonlinearities and interactions may have confounded past results. Bass et. al. find some variables are nonlinear. Ellis finds the variables interact in such a way that composite scores produce better results than analysis conducted with individual variables.

Demographics may interact with a general economic condition, such as consumer sentiment. Curtin finds income alone explains 30 percent of the variation in new car sales from 1966 through 1981 (1982). When the Index of Consumer Sentiment is added, the model explains 73 percent of the variation.

Cross-classification or an N-way discriminant analysis seem more appropriate than a regression model focused on individual differences. The three more recent studies that use cross-classification produce encouraging results.

Inappropriate specification of variable categories may mask relationships. Blattberg cites a study in which life cycle status predicts poorly when stages are defined by presence of children (1978). When Blattberg redefines the stages by age of children life cycle becomes a discriminating variable.

Rich and Jain point out the importance of recent changes in socioeconomic conditions (1968). Murphy and Staples raise the same issue for the family life cycle (1979). Both articles conclude socioeconomic status and family life cycle categories may require modification if the variables are to remain useful.

THE CURRENT STUDY

Research Design

This analysis is based on data from the annual Simmons Survey of Media and Markets for 1981, a nationwide probability sample of 15,029 households. To avoid selection bias and provide comparability, we restrict our analysis to 16 of the 18 grocery products studied by Twedt (1964). (The other two products are not covered by Simmons.) The demographic variables and their measurement categories are summarized in Figure 1.

FIGURE 1

DEMOGRAPHICS AND THEIR CATEGORIES

Following Ellis, four categories for each variable are used and Warner's six social classes are combined by collapsing the top three. Ellis's socioeconomic status and family life cycle specifications are used here with income updated to allow for inflation. The age categories are compatible with Ellis's definition of family life cycle stages.

This exploratory study seeks to answer two questions:

(1) Is the propensity to use a product dependent on demographic characteristics?

(2) Do heavy users differ from light users with respect to demographic characteristics?

Simmons reports the number of individuals or households in the population base that use the product. Users are classified as heavy, medium and light. The per capita consumption rate that constitutes heavy usage is determined in consultation with industry sources for each product. This is a more exacting definition of heavy usage than the "heavy half" concept. For many products, it may be the top 10 percent or 20 percent of users that buys 80 to 90 percent of output. In our analysis, we use Simmons definition of heavy users and combine Simmons medium and light usage categories to represent light users.

If usage rates are independent of demographics, the proportion of users in a demographic category should be consistent with the proportion of the population in that category. To test for these dependencies we cross-classify each demographic descriptor with usage/nonusage and again with heavy/light usage of each product. The chi square test is then applied to determine whether the actual number of consumers in the cells differs from the expected number.

In most cases a statistical finding of dependence does not establish the direction of causality. However, it is unlikely consumption of most products would cause socioeconomic or family life cycle status. Thus significant results in our tests suggest a person's socioeconomic and life cycle status product differences in her consumption. If these differences exist, demographics is a valid basis for market segmentation.

Results

Table 1 summarizes the contingency tables that refer to user or nonuser status for the 16 products. Table 2 summarizes the contingency tables for heavy versus light user status. Each contingency table that produced a chi square statistic significant at the .05 level is marked with an S. The others are marked NS. The results of 135 of 155 tables are significant at the .05 level, indicating systematic dependence of consumption on the demographic characteristic in question. (The absence of per capita consumption data for lemon/lime soda eliminates five of the potential 160 tables.) At the .01 level, 129 of 155 tables are significant, showing the conclusions are insensitive to the significance levels selected.

TABLE 1

RESULTS OF CONTINGENCY TABLES OF DEMOGRAPHICS BY USER/NONUSER STATUS

Only four products have two or more nonsignificant tables in a grouping of five tables Usage of detergents does not depend on age, education or income. This finding is not surprising since 97 percent of the population uses these products. The same descriptors, however, are useful in evaluating the propensity to consume detergent and toilet paper.

TABLE 2

RESULTS OF CONTINGENCY TABLES OF DEMOGRAPHICS BY HEAVY/LIGHT USER STATUS

Demographics also produces mixed results regarding heavy usage of hair tonic and bourbon. Both products serve a restricted market; penetration levels are 26 percent and 21 percent respectively. It is possible variable categories tailored to the specific product would produce significant results. It is also possible other factors are more important for explaining heavy usage.

Tables 3 and 4 display sample contingency tables that refer to heavy usage of detergents. We have constructed a tentative profile of the heavy user of detergents based on the proportions of the population found in the respective cells of the tables. (Of course reliable profiles require analysis with mean consumption rates for each cell in place of population figures.) The highest concentration of heavy detergent users is found among middle aged people with children 12 to 17 years old, a high school diploma and an income of $10,000 to $24,999 per year. The concentration of these female homemakers in the "other" occupational category indicates most are full time homemakers.

The detergent example allows us to examine the two factors at work in creating demand for a product. One factor is the population distribution. Lower middle income consumers may buy the bulk of a product simply because they represent the largest category of the income distribution. The other factor is the differing propensities of groups to consume. For example, older families without children are the largest population group in the detergent base, but families with children 12 to 17 provide 2.5 times as many heavy users. We would argue these propensities indicate the ability of demographics to capture underlying differences in values, attitudes and lifestyles.

TABLE 3

THE PROPENSITY TO CONSUME DETERGENTS (SOCIOECONOMIC STATUS)

Finally, cross-classification on a single variable basis cannot include interactive effects. Multiple cross classification analysis of the type undertaken by Ellis will produce even more significant results.

TABLE 4

THE PROPENSITY TO CONSUME DETERGENTS (FAMILY LIFE CYCLE)

CONCLUSIONS

Demographics is the most readily available and actionable vehicle for market segmentation. Although media decisions are important, the product manager must also plan strategy for product, distribution and pricing. Purchase data linking product consumption to media usage do not serve these other purposes. Moreover, the availability of large, syndicated data bases, more refined statistical techniques and better computer support makes analysis of demographic data more feasible than ever before.

This study finds dependencies between demographic characteristics and product usage as well as between demographics and volume of usage. We believe more sophisticated statistical techniques can define the nature of these dependencies and produce reliable profiles of market segments.

REFERENCES

Ando, Albert and Modigliani, Franco (1963), "The Life Cycle Hypothesis of Saving," American Economic Review, 53, 55-84.

Bass, Frank M., Tigert, Douglas J. and Lonsdale, Ronald T. (1968), "Market Segmentation: Group Versus Individual Behavior." Journal of Marketing Research, 5, 264-270.

Blattberg, Robert, Bluesing, Thomas, Peacock, Peter and Sen, Subrata (1978), "Identifying the Deal Prone Segment," Journal of Marketing Research, 15, 369-377.

Bureau of the Census (1963), Methodology and Scores of socioeconomic Status, Washington: Government Printing Office.

Curtin, Richard T. (1982), "Indicators of Consumer Behavior: The University of Michigan Survey of Consumers," Public Opinion Quarterly, 46, 340-352.

Duesenberry, James (1949), Income, Saving and the Theory of Consumer Behavior, Cambridge: Harvard University Press.

Ellis, Richard B. (1975), "Composite Population Descriptors: The Socio-Economic/Life Cycle Grid," in Mary Jane Schlinger, ed. Advances in Consumer Research, vol. 2, Ann Arbor. Mich.: Association for Consumer Research, 480-491.

Fisher, Irving (1930), The Theory of Interest, New York: Mcmillan.

Frank, Ronald E (1967), "Correlates of Buying Behavior for Grocery Products," Journal of Marketing, 31, 48-53.

Frank, Ronald E., Massy, William F. and Boyd, Harper W. (1967), "Correlates of Grocery Product Consumption Rates,' Journal of Marketing Research, 4, 184-190.

Friedman, Milton (1957), A Theory of the Consumption Function, Princeton, N.J.. Princeton University Press.

Glick, Paul C. (1947), "The Family Cycle," American Sociological Review, 12, 164-174.

Haley, Russell I. (1968), "Benefit Segmentation: A Decision Oriented Research Tool," Journal of Marketing, 32, 30-35.

Keynes, John M., (1936), The General Theory of Employment Interest and Money, New York: Harcourt, Brace and Company

Lansing, John B. and Kish, Leslie (1957), "Family Life Cycle as an Independent Variable," American Sociological Review, 22, 512-519.

Loomis, Charles (1936), "The Study of the Life Cycle of Families," Rural Sociology, 1, 180-199.

Martineau, Pierre (1958), "Social Classes and Spending Behavior," Journal of Marketing, 23, 121-130.

Murphy, Patrick E. and Staples, William A. (1979), "A Modernized Family Life Cycle," Journal of Consumer Research, 6, 12-22.

Ostheimer, Richard H. (1958), "Who Buys What? LIFE's Study of Consumer Expenditures," Journal of Marketing, 22, 260-272.

Rich, Stuart U. and Jain, Subhash C. (1968), "Social Class and Life Cycle of Predictors of Shopping Behavior," Journal of Marketing Research, 5, 41-49.

Simmons Market Research Bureau (1981), Study of Media and Markets, New York: Simmons Market Research Associates.

Twedt, Dik Warren (1964), "How Important to Marketing Strategy is the Heavy User?" Journal of Marketing, 28, 71-72.

Twedt, Dik Warren (1964), "Some Practical Applications of Heavy Half Theory," Advertising Research Foundation 10th Annual Conference, 13-16.

Warner, W. Lloyd, Meeker, Marchia and Eells, Kenneth (1949), Social Class in America, Chicago: Science Research Associates.

Warner, W. Lloyd and Lunt, Paul S. (1941), The Social Life of a Modern Community, New Haven: Yale University Press.

Wilson, Clark L. (1966), "Homemaker Living Patterns and Marketplace Behavior - a Psychometric Approach," in Wright, John S. and Goldstucker, Jac L., ed., New Ideas for Successful Marketing, Chicago: American Marketing Association Proceedings, 305-331.

Yankelovitch, Daniel (1964), "New Criteria for Market Segmentation," Harvard Business Review, March/April, 83-90.

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