The Elderly Consumer: One Segment Or Many?

Jeffrey G. Towle, The University of Michigan
Claude R. Martin, Jr., The University of Michigan
ABSTRACT - This paper reports on the segmentation of a sample of elderly consumers selected from the 1973 National Target Group Index Study. The six segments obtained were defined by self concepts of buying style using cluster analysis and were described in terms of their psychographic characteristics.
[ to cite ]:
Jeffrey G. Towle and Claude R. Martin, Jr. (1976) ,"The Elderly Consumer: One Segment Or Many?", in NA - Advances in Consumer Research Volume 03, eds. Beverlee B. Anderson, Cincinnati, OH : Association for Consumer Research, Pages: 463-468.

Advances in Consumer Research Volume 3, 1976      Pages 463-468


Jeffrey G. Towle, The University of Michigan

Claude R. Martin, Jr., The University of Michigan


This paper reports on the segmentation of a sample of elderly consumers selected from the 1973 National Target Group Index Study. The six segments obtained were defined by self concepts of buying style using cluster analysis and were described in terms of their psychographic characteristics.


The overall population growth of the past twenty-five years, coupled with rising incomes and the proliferation of consumption options in the market place has helped make it advantageous for marketers to recognize they are not dealing with a homogeneous mass of prospects, but with a diverse mixture of sub markets. Each of these submarkets or segments can be defined by a commonality of demographic, psychographic and/or behavioral characteristics. A review of the recent literature shows a sustained interest in segmentation and sub-segmentation which is still "based upon developments on the demand side of the market and representing a rational and more precise adjustment of product and marketing effort to consumer or user requirements" (Smith, 1956).

Given the general acceptance of segmentation as a useful and beneficial strategy to both producers and consumers (Haley, 1969), little work has been done sub-segmenting the elderly consumer group. The tendency of marketers is either to treat the elderly, over 65 consumers as a more or less homogeneous group, or to pay virtually no attention to them at all. This means that the more than 20-million elderly consumers are generally regarded as a single market segment, distinguishable from other segments on the basis of age, but homogeneous when treated alone, as a single sub-population. The limited attempts at sub-segmentation have been principally by suppliers of goods and services designed for the elderly and have concentrated on a few socio-economic factors, demographic characteristics and geographic location. For other marketers, grouping the elderly consumer together simplifies their conception of that market and simplifies their decision-making process. However, in line with segmentation theory it may also diminish their profit optimization with this overall class of consumers, (Frank, et al. 1972; Haley, 1969).

This paper reports on an attempt to sub-segment the over-65 market on demographic, psychographic and behavioral dimensions. The objective was to explore whether viable sub segments exist and to describe that diversity among the elderly so that marketers might recognize it and adapt to meeting the needs of parts of that market.


The elderly consumers studied were drawn from the 20,137 consumer data base collected by the Axiom Market Research Bureau (AMRB) as part of the national 1973 Target Group Index (TGI). The TGI is a comprehensive media and product survey of slightly over 20,000 adult consumers that also collected demographic, psychographic and buying style data on each respondent. The overall study consists of a probability sample of adults aged 18 and over in the continental United States. From that base we systematically selected every 10th respondent aged 65 years and over, arriving at a 10% sample consisting of 209 elderly respondents.

The variables used in our analysis included 13 demographic; 20 psychographic; and 10 buyer style measures (Figure 1). The demographic variables are standard and include most of those identified as major demographic segmentation variables (Kotler, 1972). The psychographics are a somewhat novel, but logical, approach by TGI.

Psychographic variables are those which describe the personality traits and attitudes of persons (Wells, 1974). The variables in the Target Group Index were developed by a pre-test survey conducted nationally by AMRB (TGI Report, 1972).

TGI uses twenty self-concept measures (adjective groups) which were designed to assess how the respondent views himself or herself as a person (Figure 1). These twenty measures are a distillation of 304 adjectives, which were selected from among all adjectives in the dictionary which could be used to describe people. A factor analysis of a pilot study using the original 304 adjectives produced a reduced set of adjectives which were then incorporated into the TGI survey.

The twenty psychographic dimensions are, therefore, basic ways in which people describe themselves. The list was systematically arrived at, although it does not depend on any particular psychological theory of personality except to the extent that a measure of "real-self" or "ideal self" is hoped for. Landon found a relatively high correlation between real and ideal self concept and noted that "the debate over which self concept is more important may be inconsequential" (Landon, 1974).

The data base also measures "buying style," and this is also based on a factor analytic reduction of several originally proposed measures.

For purposes of this research, the data consist of straightforward demographics, self evaluation of some psychographic dimensions (what they like as people); and self-evaluation of "buying style" (what they like as consumers). The data base was a representative sample of 209 elderly (over 65 years of age) consumers.


The use of buyer behavior variables for segmentation has been more successful than utilization of personality traits or other possible dimensions of a psycho-segmentation (Rewoldt, et al, 1973). The authors agree with this evaluation and suggest that the viability of buyer behavior as a segmentation base was predicated on its operationalizability. Based on this we decided that the 10 buyer behavior variables might form a reasonable starting point for defining sub-segments in the market, with the 33 demographic and psychographic dimensions as potential describers of those behavioral segments. The basis for our analysis was to form natural segments, based on self-concepts of buying style, that used individual respondents as the basic analytical unit in a cross-sectional study. The purpose was to define purchasing behavior groups or segments and describe those groups in terms of the psychographic and demographic variables.




While a reasonably precise process was followed by AMRB in constructing the Target Group Index, including definition of the psychographic and behavioral variables, the decision was made to undertake another verification of these dimensions.

Canonical analysis was used to check the redundancy (Cooley and Lohnes, 1971, pp. 12 and 170-72) of the information in each set of measures using the psycho-graphic variables as one variable set and the buyer behavior variables as the other. "The statistical decision rule is to determine a linear combination of variables in each set so that correlation between the sets is maximized" (Cox and Enis, 1972). Such an analysis of sets of personality and behavioral variables was reported by Kernan in 1968. Further support for using canonical analysis in this way is discussed by Alpert and Peterson (1972). They also observe that canonical analysis can be applied effectively in conjunction with other multivariate techniques. The suggestion has been, as is done by the TGI construction of the 30 psychographic and behavioral variables, that one might factor analyze variables to remove multicollinearity within variable sets prior to canonical analysis. Our analysis showed a fairly high canonical correlation between the first pair of factors (.5427), the second pair (.4991), and the third pair (.4800).

There has been some discussion (Ginter, 1974; Bass and Wilkie, 1973) of the problem of univariate responders when scaling techniques are used. To check for this phenomenon in the TGI data, a histogram of the variability of response across the 30 behavioral and psychographic variables was constructed. That histogram showed 12 per cent of the respondents had zero variation and another 2.3 per cent who had a very low degree of response variation (their standard deviation around the mean response was less than .600). It was decided that such responders, who we named "univariate responders" should not be included in subsequent analyses since they provided no information about themselves from the information theoretic standpoint (zero variance = zero information). [The information content of a message symbol is the negative of the logarithm of the probability that this symbol will be emitted from the source, e.g., p=1.0 for univariate responders, log of 1.0 = 0.0 (Engels, 1971).]

The strictly univariate responders were extracted from our data base and a distribution of their responses showed that all had answered "not sure" (the mid-point on the five-point scale) for all of the psychographic and behavioral dimensions.

A verification of the data base, again using canonical analysis, was undertaken with the univariate responders excluded. This procedure lowers the erroneously higher canonical statistics which tended to "cloud" the analysis using all respondents. Clearly, the redundancy between the variable sets is absolute for the univariate responders, so their exclusion should be expected to reduce the canonical correlations in the variable sets. The revised canonical correlation shows similarly high relation between the first pair of factors (.5528); the second pair (.4896); and the third pair (.4060). Overall these correlations are lower than previously measured, as expected.


It was not possible to segment the elderly market in this study on the usual a priori basis. We did not have knowledge about usage or non-usage of a specific product or product category or the criteria usually employed in designating heavy, medium or light users. Rather we were seeking a natural and more catholic segmentation, based on self concepts of buyer behavior. Ours is mainly a descriptive, rather than predictive, study. With no predefined segments, a clustering algorithm, Ward's method using an euclidean metric (Everett, 1974, p. 15) was employed. This clustering routine developed six segments based on the self-concept buyer behavior set. The criterion for using the six cluster solution was to strike a balance between minimizing the increase in variance resulting from combining respondents into clusters or clustering clusters to reduce their number (Anderberg, 1973) and the small cell sizes resulting from increasing the number of clusters produced.

Our analysis assumed an interval scale in measuring the buyer behavior characteristics and our initial description of each of the six clusters used the highest and lowest mean response for each characteristic. However, we assumed a null hypothesis that the mean response for each characteristic would be equal across clusters. Thus, a high level of significance of the F-test statistic would be grounds for rejecting the null hypothesis and accepting the alternative hypothesis that a significant difference does exist between the mean responses across groups. Therefore, we adopted a decision rule that any characteristic having en F-statistic significance greater than .10 be stricken from our analysis. This procedure eliminated the self-concept of impulsive buying (when in store, I often buy an item on the spur of a moment) with a significance of .1020. It was also decided to eliminate the final characteristic (all products that pollute the environment should be banned) since this was not a self-concept of buyer behavior, but rather an attitude toward a social problem.

The analysis produced behavioral descriptions of each segment that are relative, not absolute. The overall profile of these elderly consumers (Figure 2) should be used as a guide or norm. The cluster descriptions (Figs. 3-4) are relative to that overall elderly profile.

For the sake of identification we have collapsed those descriptions into the following more operational identifications:


1. Saver/planner

2. Brand Loyalist

3. Information Seeker

4. Economy Shopper

5. Laggard

6. Conspicuous Consumer


The next step in our analysis was to broaden the definition of each of these segments by trying to describe each in terms of their corresponding demographic and psychographic characteristics:

Demographic Definitions

The TGI demographic variables are mostly categorical and the analysis applied .to them was a chi-square test. We used a simple cross-tabulation of each demographic characteristic for each of the six behavioral clusters. The chi-square null hypothesis in this case is that the demographic variable is independent of cluster membership. Thus, the number of respondents in each cell is a function only of the marginal distributions of the variables, i.e., the demographic variable and the cluster membership variable. A significant chi-square statistic would reject the null hypothesis and lead to acceptance of the alternative hypothesis that a relationship does exist between the demographic variable and cluster membership. Once again our decision-rule was a significance figure of .10 or below to reject the null hypothesis. The data produced significance levels for chi-square considerably above .10. Thus, we cannot further describe the behavioral segments using demography at this time. However, because of the small cell sizes in the cross-tabulation we do not reject the possibility of demographic differences among clusters.

For those demographic variables which approximate an interval scale, or at least are of a higher than nominal scale of measurement, we ran Univariate one-way ANOVA's to detect significant mean response differences between clusters. Although none of these demographics showed significant differences between means, one of the variables showed some tendency toward discrimination. Education level(F-statistic significance of .175) was highest for group 3 (6.83) and lowest for group 2 (5.6) indicating that information seekers tend to have completed more formal education than brand loyalist consumers. This variable was coded as follows:

3 - some grammar school

4 - completed grammar school

5 - some high school

6 - completed high school

7 - some college

8 - completed college

9 - some graduate school

10 - completed graduate school

Psychographic Descriptions

In the case of the psychographic dimensions we again assumed an interval scale and used the highest and lowest mean response in our descriptive process. Consistently we used the decision rule of .10 or less for the F statistic significance to employ the psychographic characteristic in the analysis. This procedure eliminated eleven of the twenty self-concepts of psycho-graphics from further descriptive use. The resultant descriptions for each buying style segment are detailed in Figure 4.









Figure 5 shows the matching of the buying style definitions of each cluster or segment to those psychographic descriptions and the relative size of each of these elderly sub-segments


The analysis identified six buying style segments of the elderly market and relates psychographic characteristics to each segment. The largest of these are identified as conspicuous consumers who self evaluate themselves as relatively more stubborn, egotistical and dominating than their consuming peers. The second largest segment are the saver/planners who tend to buy unknown brands and are self-described as more candid and confident. The fact that these two segments constitute almost 60 per cent of the elderly consumers and that only 8.4 per cent are brand loyalists is significant to the marketing strategist.

However there are methodological cautions the authors give to that strategist in evaluating this report.

First, the clustering used in determining the elderly sub-segments used an algorithm which limits the shape of the resultant clusters imposing the same shape on all (Everitt, 1974, pp. 46-48).

Second, we recognize that the canonical analysis may capitalize on factors which are not necessarily the same factors which would be extracted in a principal components analysis (Cooley and Lohnes, 1971, p. 171).

Third, there is no consideration of a possible halo effect in the response (Wilkie, et al, 1974). We also recognize there may not be independence in responses, although the factor analytic routines used by AMRB in constructing the TGI data base were designed to offset this possibility.

In summary, this study did find the existence of natural segments of the elderly market defined by buying style characteristics and it did fit psycho-graphic descriptions to those natural segments that are sensible and operational.


Alpert, Mark I., and Robert A. Peterson, "On Interpretation of Canonical Analysis," Journal of Marketing Research, May, 1972, pp. 187-92.

Anderberg, Michael R., Cluster Analysis for Applications, (New York: Academic Press, 1973).

Bass, Frank M., and W. L. Wilkie, "A Comparative Analysis of Attitudinal Predictions of Brand Preference," Journal of Marketing Research, October, 1973, pp. 262-69.

Cooley, William W., and Paul R. Lohnes, Multivariate Data Analysis, (New York: John Wiley and Sons, Inc., 1971).

Cox, Keith K., and Ben M. Enis, The Marketing Research Process, (Pacific Palisades: Goodyear Publishing Co. Inc., 1972).

Engels, Franklin M., Information and Coding, (Scranton, Pennsylvania: International Textbook Company, 1971).

Everitt, Brian, Cluster Analysis, (New York: John Wiley and Sons, Inc., 1974).

Prank, Ronald E., and William F. Massey and Yoram Wind, Market Segmentation, (Englewood Cliffs, New Jersey: Prentice-Hall, Inc., 1972).

Ginter, James L., "The Effects of Normalization on the Multivariate Model," Advances in Consumer Research, Volume 1, Scott Ward and Peter Wright, eds., (Urbana, Illinois: Association for Consumer Research, 1974), pp. 302-309.

Haley, Russel I., "The Implications of Market Segmentation,'' Conference Board Record, March, 1969.

Kotler, Phillip, Marketing Management: Analysis, Planning and Control, 2nd ed. (Englewood Cliffs, New Jersey: Prentice-Hall, Inc., 1972).

Landon, E. Laird, "Self Concept, Ideal Self Concept, and Consumer Purchase Intentions," Journal of Consumer Research, September, 1974, pp. 44-51.

Smith, Wendell R., "Product Differentiation and Market Segmentation as Alternative Marketing Strategies," Journal of Marketing, July, 1956, pp. 3-8.

Wells, William D., ed., Life Style and Psychographics, (Chicago: American Marketing Association, 1974).

Wilkie, William L., John McCann, and David J. Reibstein, "Halo Effects in Brand Belief Measurement: Implications for Attitude Model Development," Advances in Consumer Research, Vol. 1, Scott Ward and Peter Wright, eds. (Urbana, Illinois: Association for Consumer Research, 1974), pp. 280-290.