A Modern Sociological Approach to the Stratification of Material Life Styles

Marcus Felson, University of Illinois at Urbana
[ to cite ]:
Marcus Felson (1975) ,"A Modern Sociological Approach to the Stratification of Material Life Styles", in NA - Advances in Consumer Research Volume 02, eds. Mary Jane Schlinger, Ann Abor, MI : Association for Consumer Research, Pages: 33-38.

Advances in Consumer Research Volume 2, 1975      Pages 33-38


Marcus Felson, University of Illinois at Urbana

After discussing the fallacies of the Warnerian model of social class, the author criticizes modern marketing research for its lack of adequate theoretical models. Special criticism is reserved for "life style" research, a current fad which fails to specify solid models of consumer decisions. An alternative approach is an opportunity model, taking time and knowledge as resources in addition to income. The causal impact of one consumer decision on another must be formulated and precise models of how social opportunities affect consumption must be specified.

One of the least successful areas of social research is the sociological study of material life styles. Many market researchers are inclined to neglect modern sociology, while modern sociologists generally neglect the study of consumer taste. The present paper is designed to bring market researchers up to date on developments in stratification research over the last decade and to suggest how they might take advantage of these developments in planning their own research. Recent literature reviews are available elsewhere (cf. Lasswell and Benbrook, 1974; Dietrick, 1974). My purpose here is to summarize the state of the art and to suggest applying its research tools to studying the stratification of consumers.

What most market researchers call "social class" is a 20-year-out-of-date concept popularized by W. Lloyd Warner. The Warnerian Model of social class is characterized by (1) reasonably clear-cut class boundaries, (2) a single dimension summarizing social inequality, and (3) a single type of relationship between social class and other variables.

Warner's model simply cannot stand up to the evidence collected in modern America for several reasons. First, the correlations among stratification variables taking individuals or households as units of analysis are not so high that one can reasonably treat social class as a single dimension. Often these correlations are .4 or .3, seldom over .6. Second, stratification variables have causal relationships to one another which Warner ignored. Education has an effect on occupational attainment which has an effect on income. Yet Warner mixed all these distinct variables together, ignoring their causal relationships. Third, education, occupation, income, race, and other stratification variables often have different sorts of consequences for social life, consequences which have been decomposed by more careful researchers. Fourth, Warner failed to consider how various rewards come at various stages of the life cycle or even change in the period of a few years. Poverty, prosperity and occupation are conditions rather than properties of people, so that they are not demographic variables at all. Being an electrician at Company X is not an innate or immutable characteristic, so that treating it as "demographic" completely misses the point. regrettably, this bad habit has been adopted by modern market research. Fifth, Warner did not realize that clear boundaries between social strata simply do not exist with respect to most stratification variables. The terms "blue collar," "white collar " and "middle class," however convenient, obscure the actual distributions of occupational prestige and income, which are actually smooth curves with few clear fissures. Education does have clearer gaps because diplomas are offered mainly after the 12th and 16th year, but even this so -called gap is mitigated by the fact that each year of education has a payoff regardless of whether diplomas are received. The use of class boundaries or class names is largely arbitrary and their reification is a good example of careless inference from data. The major exception has to do with race, which tends to be a dichotomous variable with virtually no one holding a position intermediate between black and white.

Modern stratification models generally decompose "socioeconomic status" into components which are then related in a causal model. Assumptions about cause and effect should be justified before setting out and intervening variables should be precisely specified. Indirect effects should be allowed, and variables having no direct effect upon one another are welcome for consideration. Blind factor analysis or any computerized searching of correlation matrices before specifying a causal model is strongly discouraged in modern stratification research. Market research will need to follow the same path, not only in relating social stratification to consumption but also in relating consumer variables to one another. I will come back to the latter point, but first I wish to comment upon a tendency to try to maximize the proportion of variance explained. Under the leadership of Otis Dudley Duncan sociologists have largely escaped from that quicksand (see Blau and Duncan, 1967, as an example). It does science little good to have a high correlation if you have a spurious causal model explaining that correlation. Of course, raw and spurious correlations can be used by businessmen to decide in which magazine to advertise their product. But any creative endeavor to develop new products cannot proceed without understanding. Furthermore, marketing as the science of consumer taste needs to go beyond selling products and to build a model of consumption in a larger sense. The Williamsberg Conference on Social Structure, Family Life Styles, and Economic Behavior (see Sheldon, 1973) took a major step in this direction, but marketing as a discipline needs to carry out the task. I do not deny that plenty of very general schematic diagrams are found in marketing journals and books, but these do little good if they are ignored when data is collected and analyzed. Causal modeling, as done by the Duncan School and its followers in sociology, forces researchers to operationalize their theoretical statements. If it is later discovered that lots of variance cannot be explained, this may reveal that non-systematic or random factors influence behavior. The degree of freedom in a family budget and the thousands of goods and services available tend to reduce the correlations between any one consumer trait and its predictors for good substantive reasons. The proper measure relating many types of consumption to their predictors is often an odds ratio or conditional probability, rather than a standardized regression or correlation coefficient. In a sense, any correlation above .1 relating very specific consumption to serious predicters is a miracle. In many cases, the small regression coefficients linking independent to dependent variables may arise from an incomplete model. Measurement error may attenuate the relationships. Or the failure to specify intervening variables may give the illusion of no effect. Consider the case where four variables form a causal chain: A affects B which affects C which affects D. Let us say that each of these causal coefficients is a respectable .5, standardized. Now suppose that the researcher omits intervening variables B and C. It will appear that the effect of A on D is (.5) (.5) (.5) = .0625. Though this is small, it obscures a significant process which helps one to understand how consumer behavior works, Perhaps the main reason why the multiple R2 derived from regressing any particular consumption variable against any set of background variables fails to be very large is that these models are misspecified in a very serious way. Consumer variables tend to have negative impacts on one another, insofar as consumer goods and services are either substitutable for one another or compete with one another for scarce resources. It is interesting that consumer research models tend to ignore this basic principle of economics. This produces small R2 for the simple arithmetic reason that one cannot have high correlations between background variables and consumer traits themselves strongly negative in relationship. Such a situation would lead to correlations greater than 1, as can be discovered by applying the fundamental theorem of path analysis. Similarly, elementary arithmetic will show that competing consumer goods must have rather small correlations with any sort of background variable. Multiple regression analysis based upon such misspecified models will inevitably lead to small coefficients and small R2. The conclusion that socioeconomic status and other background variables have little consequence for consumption is an artifact of this misspecified model and totally unjustified. Regression analysis should only be used if the researcher keeps in the back of his or her mind the fact that regression will suppress the true effects of background variables on consumption. Use of more elaborate causal models is preferable, once problems of identifiability can be worked out. In the meantime, conclusions that socioeconomic status is unimportant for consumption are best treated as false.

I would also argue that the relationship between background factors and consumer traits is often obscured by the following suppressor variables:

(1) Quantity of time available to shop or to consume

(2) Scheduling of work, school and other obligations which structure use of time

(3) Knowledge about consumer goods and their attributes

(4) Life cycle and family variables, carefully partialled out

(5) Female labor force participation

My general thesis is that consumption is organized by the structure of opportunities available to consumers. Material resources are only one limitation upon opportunities, and the weakness of the relationship of income to many consumer traits is that so many other variables also affect opportunity to consume a given product. Indeed, the rising median level of per capita prosperity over the past half century has decreased the family's dependence upon income, thereby reducing income stratification of life styles (see Felson, 1974). New resources are likely to take on importance. Time is first on the list because a person needs time to buy and to use consumer goods. If time is limited, certain goods facilitate its conservation. Not only is the quantity of available time important but also the nature of one's schedule. A man and wife each of whom works 40 hours a week will consume differently if one of them must work a night shift. How many cars they need, which groceries they purchase, who does the shopping, what activities are shared--all of these are affected by the scheduling of worklife.

Knowledge is a very important and neglected resource. How can someone deliberately purchase a Product which he or she has never heard of? How can a product's high class image sell it to people who think it has a low class image? Much of the so-called unexplained variance in preference for products or brands probably has to do with variance in public knowledge about them. Thus any model of consumption needs to include knowledge or lack of knowledge about consumption. Age, life cycle, family size and other such attributes condition consumer choice by putting people into positions which require or encourage certain behavior patterns. These variables have different effects, depending upon which consumer trait is under study. The failure of many earlier attempts to use these variables as predictors is probably attributable to careless model specifications. In some cases, whether you have any children is more important than how many you have. Sometimes teenage children affect consumption in ways that younger children do not. Age is not a direct measure of life cycle or role obligations, so that its impact on consumption requires consideration of other traits (e.g., marital status). The point is that social roles and statuses affect opportunities to consume various goods, but do so in precise ways which must be disentangled.

Female labor force participation is important because it lessens a woman's free time and alters her exposure to other people. It also may increase her power in her own family. In any case, it affects her opportunities to consume.

None of these suggestions will do a bit of good unless market researchers analyze more fully the opportunity to consume and build a precise model which breaks down family consumption step by step. Poor theory yields poor correlations and even if by chance one finds large correlations, poor theory will render them meaningless.

Present marketing theory, interestingly enough, is weakest in its own backyard. While market researchers scramble in search of independent variables, they neglect their own dependent variables and usually fail to study the order and organization of consumer tastes themselves. The reason for this may be that market researchers work for a finite set of clients who dictate the dependent variable and have little or no funding, or time for basic research. Most of the recent life style research effort seeks to relate social or psychological traits to one another but fails to relate consumer traits as behaviors to one another in a meaningful causal model.

The most successful modeling of consumer behavior was carried out in a simpler period of history by rural sociologists, who developed scales of socioeconomic status during the 1930's based upon consumer traits of various sorts (Chapin, 1935; Sewell, 1940). The weight of evidence indicates that the American stratification system during the Great Depression was characterized by a fairly clear hierarchy of material life styles strongly reflective of occupational prestige and income. In my own work, I have shown that major social change has probably occurred in American society since that time (Felson, 1974). It appears that the present consumer goods market is probably not as clearly stratified and that stratification probably is only partially manifested in consumer goods. Thus the discontent with so-called social class as an independent variable in market research is partly justified, yet the proper response to that discontent is to build more sophisticated models relating stratification to material life styles, including knowledge, time, family, life cycle and the like. Most important, modern models must consider how one consumer decision affects another. For example, once you buy an 8-room house, you need more furniture, telephones, and consumption of utilities. Once your daughter sets up a separate household in another city at your expense, you have a whole set of costs tied to that first decision. Once you buy a large car and move 20 miles from work your gasoline consumption is largely determined. Thus, consumer decisions have causal impacts on one another. These impacts can be thought out and then measured. In any case, no one consumer decision should be taken in isolation from others. Clients are perhaps too short-sighted to see this. A large auto manufacturer might be surprised to learn that one of its major competitors is Gerber Baby Food, since families sometimes choose between a new car or a new baby. Yet an auto company is probably too worried about other auto companies as competitors and pay too little attention to the less obvious competition. The latter can be understood only on the basis of understanding the whole consumption process.

I will turn now to a discussion of a current market research fad-life style research. Interestingly, what some of the "life style" adherents see as its greatest strength is actually its Achilles' heel, namely the generation of so-called precise profiles of consumers as individuals. If you really wanted to do the latter, just take interviewer identification code as a set of dummy variables and you can explain 100 percent of the variance in any dependent variable or you can establish that people who read Playboy probably agree with Playboy's ideas. Such information is too specific to generate much understanding, whatever the size of the correlations. Taking 300 independent variables in a sample of 1000, it is little wonder that some researchers have increased their R . Much of this increment is sampling error. The rest it theoretically meaningless. Life style cannot help one to understand consumer behavior if life style variables are a disorderly, nongeneral, nonhierarchical, or atheoretical set of vaguely related traits whose causal relationships to each other and to anything else are unspecified. Much life-style research could better be termed "idiosyncracy research," since it uses the computer to group people with similar idiosyncracies. The central point is that one purchases generality only at the price of specificity and vice versa. Life style research today contributes no understanding with 300 independent variables, unless these can be reduced to a few meaningful dimensions.

Nor does life style research have a right to claim that it is studying "the whole man." No one studies whole men or women. To do so one would have to include psychoanalyses, eye examinations, family tree investigations, and autopsies. The basic research process is one of abstracting certain dimensions from the whole and looking only at those dimensions (i.e., variables). The life style researchers are no different from any other researchers, except that they sometimes abstract 300 variables from the "whole man."

My central point in this paper is to look for a small number of organizing principles which structure consumer behavior, and especially to work out how one type of consumption helps organize another. Fundamental conflict between the short-run goals of business and the long-run goals of science will need to be faced. A good model of family consumption will need to combine and scale products and to study consumption as a system. The prospect of selling more widgets is far from guaranteed. Only basic research can do the job, and it is more likely to come from the ivory tower people than from the Ivory Soap people--though the latter might benefit from it if they wish.

One of the first steps is to abandon discriminant analysis with its multiplicity of dependent variables. Consumer traits will need to be combined into indexes which economize on information. Fancy statistical models cannot compensate for underdeveloped theoretical models.

To summarize what I have said, modern stratification research has abandoned W. Lloyd Warner for O.D. Duncan; concentrating on multivariate causal modeling, using continuous variables whose effects on one another are decomposed, often with path analysis or other equation systems. Market research needs to take a similar tack and to develop models of material life styles which go beyond the immediate needs of single companies, taking into account at least some of the interrelationships among consumer traits.


Blau, Peter and Duncan, O.D. The American occupational structure. New York: Wiley, 1967.

Chapin, F.S. Contemporary American institutions. New York: Harper, 1935.

Dietrick, B.A. Social mobility: 1969-1973. Annals of the American Academy of Political and Social Science, 1974, 414, 138-147.

Felson, M. Social Stratification in household attributes: changes since the Great Depression. University of Illinois at Urbana-Champaign, Department of Sociology, Program in Applied Social Statistics. Working Paper #7416, 1974.

Lasswell, T.E. and Benbrook, S.L. Social stratification: 1969-1973. Annals of the American Academy of Political and Social Science, 1974, 414, 105-137.

Sewell, W. The construction and standardization of a scale for the measurement of the socio-economic status of Oklahoma farm families. Agricultural and Mechanical College, Stillwater, Oklahoma, Agriculture Experiment Station. Technical Bulletin Number 9, 1940.

Sheldon, E.B. Family economic behavior: Problems and prospects. New York: Philadelphia; Lippincott, 1973.

Warner, W.L. Yankee city. New Haven; Yale, 1963.