Critics of Richard Centers’s class-perception question complain that it forces a particular class model on all interviewees. They have a point—that we need to explore other class images beyond those labeled working class and middle class—but they exaggerate its importance by claiming that the possibility of other class images necessarily invalidates the use of Centers’s fixed choices.
The study of alternative class images usually begins with a simple, open-ended question asking people how they interpret class divisions. The best-known of these studies is the 1962 British survey of “affluent workers” (Goldthorpe et al., 1969),1 which posed this question to a small sample of autoworkers: “People often talk about there being different classes—what do you think?”
The investigators eventually coded the varied responses into money models, 54 percent; prestige models, 8 percent; power models, 4 percent; a residual “other,” 26 percent; and no images, 7 percent (Goldthorpe et al., 1969:150). Workers with power models, although uncommon, were the most class conscious; they saw society divided into two opposing classes: “bosses and men” or “the employing class and the rest of us.” At the opposite extreme, workers with prestige models were the least class conscious; they saw social differences stretched out along multiple gradations based on lifestyle or social background.
Between the class conscious and the status conscious were the majority of workers, who interpreted class primarily as a matter of the amount of money one has. All these models pictured a large central class to which most wage and salary earners belonged—disregarding the blue-collar-white-collar division that has been so central in British sociology. There were several variations on the money theme. A number of workers saw two income classes, the “very rich” and the “working class”; others added a third class below those two, “the very poor”; some distinguished two distinct classes at the top, “millionaires” and “the well-to-do.”
Problems with open-ended questions
Goldthorpe et al.’s direct approach to studying class images, using open-ended questions, may seem at first to be the easiest and most natural approach: if we are interested in how Americans see class divisions, why not just ask them and record what they say. And, indeed, the method is well suited to in-depth interviews that permit relaxed and thorough probing (e.g., Bott, 1957; Lane, 1962; Sennett and Cobb, 1972; Garson, 1973; Rubin, 1976; Hochschild, 1981). But the luxury of extended interviews is not feasible for studying representative national samples. When open-ended questions are transferred to large-scale surveys, they raise four great problems.
First, open-ended questions are often more ambiguous; the more unstructured they are, the more baffling they tend to be for the people being interviewed (Schuman and Presser, 1979; Jackman and Jackman, 1983: 14–15). Usually the researcher has a very specific purpose for asking the question but leaves the respondent to guess at that purpose (“What does he mean, ‘What do I think about social class?’”). The open-ended format is sometimes championed as the more democratic method—let the people decide for themselves what they mean by “class”—but in fact, this is a counterfeit democracy: the researchers know their intention in asking the questions, but since “class” may take several different meanings, the respondent is asked to guess which of the possible meanings the interviewer has in mind.
Second, open-ended questions give the sociologist extraordinary leeway in classifying the responses into categories that the people may never have intended.2 From the bewildering assortment of replies that such questions provoke, what kinds of responses does the analyst encode as “class conscious?” The classifications are rarely clear-cut. American workers do not, standing on their doorsteps, expound a theory of class struggle based on the expropriation by capitalists of the surplus value workers produce. We cannot insist that workers be trained Marxist ideologues before we classify them as class conscious, nor is that necessary for radical protest (Moor-house, 1976; Tilly and Tilly, 1981:17). Since Marxist theorists themselves disagree about the correct definition of the working class, we can hardly expect workers to have definitively resolved all these issues for themselves.
As a result of the ambiguity in the responses, then, researchers develop idiosyncratic coding schemes that are rarely replicated in subsequent studies. The prestige, power, and money models of the affluent-worker studies did not match the categories used in earlier analyses (e.g., Bott, 1957; Willener, 1957; cf. Goldthorpe et al., 1969:147), nor have later studies adopted the affluent-worker categories. Thus little cumulative knowledge—such as has developed, for instance, from the steady stream of research with the Centers class-perception question—has resulted from these attempts.
The coding problem suggests a third difficulty. The direct strategy depends on how well Americans can articulate their class images, not whether they perceive class divisions. Perception and articulation are distinct processes. Workers may experience class conflict and perceive the divisions around which that conflict is organized without being able to articulate precisely those perceptions. Yet unverbalized perceptions often exercise an important influence over thought and behavior.3 One of the more important advances in psychological research has been the recognition that implicit cognitive models can determine behavior without ever becoming fully conscious. Our use of language provides a good example of this phenomenon: we learn to talk according to regular patterns long before we are taught explicit rules of grammar. These language patterns shape the way all people talk in everyday life, although we leave to professional linguists the task of articulating those rules.
In their research on patterns of language usage, linguists do not ask a sample of the population direct, open-ended questions about what rules they use in forming sentences. Those rules, however important and ingrained in patterns of speech, are not usually conscious. Indeed, if linguists did ask that question, the best answer they could hope for would be a recitation of the rules of “proper” grammar learned years before in school. Whether the school “rules” truly reflect actual patterns of speech or not would remain an open question. Nevertheless, sociologists have often studied the perception of class exactly this way—by asking respondents what rules they use to sort people into different classes and assuming that these answers accurately describe the pattern of usage.
The final flaw in open-ended questions is their assumption that people have only one image of the class structure. A person is classified as having either a money model or a power model; either a two-class or three-class or multiclass model. Nobody is classified as having both money and power models, or both two-class and multiclass models. For example, John Goldthorpe and his colleagues attached great significance to the predominance of the money models over the more conflict-oriented power models. The money models implied a less hostile image of society, appropriate for these more affluent workers. But H. F. Moorhouse (1976) later showed that money models and power models are not as mutually exclusive as were first supposed; in fact, he found that for much of the population, “statements about money are statements about power” (1976:474). And in John Leggett’s (1968) study of working-class consciousness in Detroit, it is clear that for many workers, “money is power.” When workers were asked who enjoyed special powers or privileges in Detroit, they replied:
Similarly, several studies have thought it important to determine who has a two-class, three-class, or multiclass image of society; supposedly, the two-class models are more conflict oriented, perhaps even proto-Marxian. But this is a futile enterprise. Someone with a two-class image can distinguish subgroups within each of the major classes (see Lenski, 1952). Conversely, someone with a multiclass image can be persuaded to collapse the categories into fewer and larger classifications. Whether such a person responds to an interview question with a two-class or multiclass model depends chiefly on the momentary demands of the situational context.
Our objection to the assumption that each person has just one class image is the same as our comment on the occupational prestige studies—people have quite varied and flexible perceptions of their society. The fact that under one set of circumstances someone describes a particular image of the stratification system—be it a two-class, continuous status-ranking, or money model—does not mean that at other times and in other circumstances the same person will not use quite different images.
Few studies have been as sensitive to the multiple nature of class images as the early interviews of Elizabeth Bott (1957). In her intensive probing into the class images of Londoners, she was struck by the many seeming inconsistencies in the interviews.4 The criterion of class position would shift from income to power and then to occupational prestige without arousing any sense of contradiction in the respondent. The level of focus would also adjust from one context to another, so that at one point the respondent would talk of what seemed like a quite homogeneous working class but would later break that class down into well-defined segments. From this she concluded that
usages vary according to the immediate social situation and the specific purpose of the comparisons and evaluations. It follows that there is no one valid way of finding out what people really think about class, for each method will reveal slightly different reference groups, although there is a strain of consistency and continuity running through each couple’s usages at different times. (Bott, 1957:171)
The direct approach of asking people their class images often fails, then, because people sometimes cannot articulate a class model they do perceive, and the class model they do verbalize may not be the only model they perceive.
We would like to put to rest, therefore, the still-prevalent idea that unstructured direct questions (“What do you think?”) are necessarily better than more structured questions that allow respondents to choose from a series of possible answers. Because answers to unstructured questions are more “spontaneous,” they have been thought to reflect better what respondents really think about class; structured questions, by comparison, are said to be guilty of gratuitously imposing on respondents a framework that may be completely foreign to them (Gross, 1953; Case, 1955; Kahl and Davis, 1955; Haer, 1957; Gordon, 1958; Lopreato and Hazelrigg, 1972; Schlozman and Verba, 1979).
But this critique of structured questions loses its validity once we realize that everybody has multiple images of society. A “spontaneous” response would be preferred only if we could assume that the respondent has only one class image. But if people have several complementary images, then we have no way of knowing whether the “spontaneous” image first elicited is a person’s only image or what additional kinds of class perceptions he or she may have developed.
By contrast, a series of more structured questions can be fairer to the respondent by providing a more explicit context, so that people seem better able to understand what the researcher is asking. Of course, answers to structured questions can be misleading if the context provided is meaningless to most respondents. The validity of structured questions depends on the skill of the question author; our point is just that structured questions cannot be judged a priori invalid. Nor are we suggesting that unstructured questions are always invalid. Indeed, there can be no substitute for intensive interviewing where the interviewer has the time and skill to explore the multiple perceptions of class structure. Studies such as Elizabeth Bott’s can elicit new and valuable insights. We doubt, however, that most survey research meets these conditions.5
The appropriate strategy for studying class images is to observe many concrete instances where such images must be used and to abstract the images from the specific behavior. The emphasis is on the “many,” since no single rule will determine all concrete instances. In some circumstances, an authority-subordinacy distinction might be the decisive criterion; in others, income or education may be more important. That is, we must allow for the possibility that respondents, like sociologists, can work with different models, depending on the circumstances and the appropriateness of the model. But differences among individuals can also be observed by tendencies to utilize different models in the same situation.
Skillful intensive interviewing uses precisely this strategy to elicit multiple instances of respondents’ use of class images. In the study reported in this chapter, we experiment with a structured format more suitable for survey research; we presented what is called a “triads” task to a small sample of Chicago respondents.6 We listed three occupations and asked respondents to tell us which two of the three were in the same social class. Then we presented another set of three occupations and again asked them to choose the two in the same class. We proceeded in this way until we had asked all combinations from a list of eight occupations. Different parts of the sample were given different lists, but the main set that we analyze here comprised eight occupations taken essentially from the industrial sector: big corporation executive, plant manager, industrial engineer, factory foreman, skilled machinist, truck dispatcher, payroll clerk, assembly-line worker.7 This list includes both high- and low-prestige positions, occupations at various levels of prestige with direct supervisory authority and others without authority, and some blue-collar and some white-collar occupations. The mix enables us to test what criteria the respondents would fix on to make their judgments.
Tendencies in class images
The objective of research on class images must be to look for the “strain of consistency” that Bott (1957) found in her London interviews. People do differ in the way they perceive the social order, but we need not classify every person as having one and only one image of the class system. Some of the problems of earlier research in this area might be resolved if we looked for general tendencies in constructing class images rather than trying to identify complete models that precisely fit each observer’s cognitive map of the social order. We believe that two such tendencies are basic to the broad range of class images that people most often report.
First, there is a tendency to see a strong division near the top of the social order that separates the bulk of the population into a large subordinate class. Most of the models that Goldthorpe et al. (1969:148) report for their English auto workers reflect this tendency. So does the image suggested by the class-placement analysis in the previous chapter, which puts above the dividing line mental labor, the self-employed, and those in authority—all of whom are minorities in the work force. (The size of the elite in Goldthorpe’s research would appear to be considerably smaller, however, since the English workers do not include all administrators and professionals in the top category; 1969:149). We call this tendency the working-mass tendency, since it groups most of the labor force in a subordinate, working-class position. The large mass at the bottom would include not only the less-skilled manual workers but also craftworkers and the white-collar proletariat.
A second common tendency draws the major division in society near the bottom: it sees the unskilled workers as a separate lower class but assimilates affluent blue-collar workers into the “middle mass” of society. This middle-mass tendency combines many manual occupations with truly middle-class positions so that this more successful majority greatly outnumbers the vestigial working class.8
These tendencies are not exclusive alternatives. Given an appropriate situation, many people recognize divisions at both ends of the social order. People who emphasize the separation of a small dominant middle class from the working masses are not necessarily ignorant of the internal divisions between foremen and assembly-line workers, for instance. Nor do observers who see a large middle mass necessarily deny a class division between corporate directors and factory foremen. Synthetic class images (Ossowski, 1963) that incorporate two or more tendencies are more understandable if we consider these two main cognitive tendencies to be independent of each other rather than psychologically opposed. In many situations, people will unite the tendencies in a composite image. The combination can be quite cognitively consistent and may even be the most widespread model within a given population.
Nevertheless, we are interested also in the differences among people in their readiness to adopt one or the other tendency. Americans may have multiple images of their class system, but they may lean toward one or another model as a preferred perspective. The two tendencies therefore represent the “strains of consistency” that Bott found to hold across the many variations that occur in situational contexts.
Class Image Patterns
Two statistical strategies have been developed to abstract the general tendencies from concrete similarity judgments such as those in our triads task. We rely first on a technique known as “individual differences multidimensional scaling” (INDSCAL); we then supplement this technique by grouping the occupations through cluster analysis methods.
INDSCAL attempts to summarize the differences among the occupations as distances in a one-, two-, or multidimensional space (Carroll and Chang, 1970). If occupations are frequently chosen as being in the same social class, they will be close together in this space; if they are rarely chosen together, then they will be quite far apart. INDSCAL has a great additional benefit in that the scaling dimensions also describe differences among the people making the judgments. Thus we can uncover both the major tendencies of ways of grouping occupations in a social class, and the different preferences of people for each of these tendencies.
We begin by looking at the one-dimensional solution that INDSCAL generates from the triads data. Not surprisingly, this dimension closely resembles a status ranking. Figure 5.1 reports the values assigned to the occupations by the INDSCAL analyses as well as their scores on the NORC prestige scales (Siegel, 1971). The two are highly correlated (0.96). There is a tendency in the INDSCAL results to cluster the occupations into three separate groups, but the rankings are remarkably similar. (Both the high correlations and the clustering tendency are observed also in the two other variants we used in these interviews: see the chapter appendix, Table 5.A.)
While the one-dimensional results reproduce the conventional prestige scale quite well, less than half of all responses to the triads task conform to the implications of the prestige scales. The distances along the NORC scale can be used to predict which two occupations should be chosen in the same social class, but only 47 percent of the actual responses agree with the NORC prestige predictions. Some 14 percent of the triad responses reject even the ordinal properties of the NORC prestige scale: That is, respondents chose the highest and lowest prestige positions as being in the same class. For instance, for the triad factory foreman (NORC prestige = 45.1), payroll clerk (41.3), truck dispatcher (33.5), the modal response (40.5 percent) placed the highest status position (foreman) and the lowest (dispatcher) in the same class. These two can be chosen only if prestige ranking is rejected as a model of social class judgments. The modal response to this particular triad suggests a predominance of an authority model over a prestige model: both the foreman and the dispatcher exercise authority over other workers; the payroll clerk does not. While the predominance of authority over prestige does not hold for all the triads, the extent of the poor fit to the prestige models suggests the existence of multiple criteria in the population.
FIGURE 5.1. One-dimensional representation of class image ratings
SOURCE: Images Survey, Siegel (1971).
Our real interest with INDSCAL lies in the multidimensional solutions that identify the alternative criteria people use in constructing class images, the INDSCAL results represent the “meaningful psychological dimensions” in the data (Carroll and Chang, 1970); that is, the “strains of consistency” that run through people’s many concrete judgments.
Figure 5.2 displays the two-dimensional results and identifies the “working-mass” and “middle-mass” tendencies described above. The vertical dimension separates the three managerial positions from the remaining five working-class positions. The big corporation executive is especially distinct in this perspective. On the other hand, there is little difference among the five working-class occupations; they represent an undifferentiated mass of workers subordinate to their managerial “bosses.” Direct supervisory authority is not the issue here. The factory foreman scores only slightly higher than the machinist, while the industrial engineer is in the dominant group together with the more clearly supervisory plant manager. The division is similar to a “mental labor” differentiation, as in Poulantzas’s emphasis on the middle-class positions that help to reproduce the existing capitalist system. The criterion appears to be a worker’s function in the larger system rather than the specific work situation.
The horizontal dimension describes the middle-mass perspective, grouping the intermediate positions of machinist and foreman with the managers. It is only the unskilled assembly-line worker and the low-prestige white-collar workers who are separated into a distinct (lower?) class—a vestigial working class that has not yet made it into the comfortable majority.
Two-dimensional results based on a somewhat different set of occupations are quite similar (see Figure 5.3). The horizontal or middle-mass dimension collapses the intermediate occupations (represented in this set by the carpenter and payroll clerk) with the three managerial and professional positions. The vertical or working-mass dimension collapses the intermediate positions with the assembly-line worker, janitor, and unemployed laborer. These are precisely the same tendencies we identified with the industrial occupations analyzed in Figure 5.2.
None of the respondents could be classified as “pure” types in weighting one dimension to the complete exclusion of the other. These dimensions should be thought of, therefore, as cognitive tendencies. The actual images are some weighted combination of the two tendencies. The image of the many respondents who weight the two dimensions almost equally approximates the one-dimensional results reported in Figure 5.1. When the results of respondents with both types of models are aggregated, we again obtain the familiar status gradations. Thus, the typical strategy of aggregating responses across individuals obscures the systematic “distortions” that some respondents are making at either end of the scale.
As interesting as what does appear in these results is what does not. Neither a supervisory authority distinction nor a manual-nonmanual division appears to be meaningful in sorting these occupations into social classes. A manual-nonmanual division can be obtained only if a four-dimensional solution is computed. (These results are reported in the chapter appendix, Table 5.A.) The skilled machinist and assembly-line worker score low on the fourth dimension (–0.44 and –0.72), while the remaining positions score from +0.06 to +0.40. The major division here is a manual labor distinction. Status is of less concern: similar status positions are separated while quite disparate positions are grouped together.
The four-dimensional results with the alternative occupational titles also yield one dimension that describes a manual-nonmanual division. In this model the eight occupations are sorted into two groups with the payroll clerk (+ 0.41) joining the other three nonmanual positions (+ 0.30, + 0.31, and +0.33), while the carpenter (-0.14) is seen as closer to the manual positions (-0.45, –0.25, and –0.51).
FIGURE 5.3. Two-dimensional representation of class image ratings: alternative titles
SOURCE: Images Survey.
The manual-nonmanual division is apparently a weak criterion, since there is no suggestion of it in the two-dimensional results. Nevertheless, earlier work that dismissed it as irrelevant in stratification judgments (Duncan, 1966; Glenn, 1975) is not supported by these techniques, which are more sensitive to individual differences. On the other hand, the weakness of this dimension suggests that for this American sample, at least, the manual-nonmanual gap is not so crucial a division as some researchers have maintained (Lockwood, 1958).
The authority-subordinacy distinction does not appear in any of these solutions. From results with our third subsample, however, we know that people are cognizant of an authority distinction among these occupations; they just don’t apply it to social class judgments. In the third version of the questionnaire, we kept the same industrial occupations as shown in Figures 5.1 and 5.2. But instead of asking people to choose the two in the same social class, we asked them only to choose the two “most similar to each other.” This reveals criteria of similarity that are quite removed from social class. In fact, the one-dimensional results do replicate a prestige hierarchy quite well. But the second dimension represents a work-setting criterion, distinguishing the strictly factory jobs (machinist, assembly-line worker, and factory foreman) from the office jobs (corporation executive and payroll clerk), with the plant manager and industrial engineer appropriately in between.
The third dimension is of special interest in this version of the interview: here, all authority positions score positively and all nonauthority positions score negatively. The plant manager is separated from the almost equally prestigious industrial engineer; the factory foreman from the skilled machinist; and the truck dispatcher from the payroll clerk. These distinctions show that the respondents recognize authority as an important criterion for differentiating between jobs. But the emergence of an authority dimension here rather than in questions of class implies that while authority may be a salient aspect of work, it is not especially relevant in determining class position.
This would suggest that the weak effect of supervisory authority on class placements (Chapter 4) may not be just a consequence of the poor measurement of authority. Supervisory authority may be less important to American class perceptions than was first supposed. On the other hand, both the class-images data in this chapter and the class-placement data in the last chapter demonstrate the primary importance of the division between managerial professional positions and the remaining workers not engaged in mental labor. For Americans, the mental-labor division appears more important than an authority distinction or a blue-collar-white-collar division.
If the working-mass and middle-mass tendencies are the principal influences on class images, do the working and middle class differ in their preferences for these tendencies? Our small sample sizes (only 103 respondents for the main interview version) prevent a detailed multivariate analysis of the INDSCAL tendencies, but the simple zero-order relationships suggest that the middle-mass perspective may be class-linked. The INDSCAL analysis generates two scores for each respondent describing how closely his triad judgments match the two scales reported in Figure 5.2. Managers and professionals score significantly higher than workers on the middle-mass dimension.9 Thus, workers are less likely to put much weight on a perspective that lifts foremen and skilled workers out of the working class and locates them closer to the more dominant middle-class positions of plant managers and engineers.
The tendency to see such a division within the working class is, in fact, more characteristic of the middle class; workers themselves place less importance on the division. For example, on the triad industrial engineer, factory foreman, assembly-line worker, managers preferred (63 percent) to group the foreman with the engineer, whereas only 41 percent of the workers did so. Their preferred response (47 percent) put the foreman in the same class with the assembly-line worker.
There is no parallel preference of workers for the working-mass dimension. Working-class respondents do score somewhat higher than managers and professionals on this dimension, but the difference is not greater than what we would expect by chance.10
The triad data also permit a direct calculation of the best-perceived division among the eight occupations. Any division of the eight into two groups implies a unique pattern of responses to the triads. For example, a working-mass division that sorts out the executive, engineer, and plant manager at the top implies that in a triad of plant manager, foreman, assembly-line worker, the respondent will class the foreman and the assembly-line worker together; a middle-mass division implies that the respondent will class the plant manager and foreman together. The perceived fit of these two divisions can be calculated from the proportion of triad responses that match what is implied.
There are 127 logically possible divisions of eight occupations into two groups, but only three are widely endorsed. There are significant differences between managers and workers on two of these three divisions.
As would be expected from the foregoing discussion, the division best perceived by the managers and professionals is the middle-mass division that separates the clerk, dispatcher, and assembly-line worker at the bottom and merges the foreman and machinist with the three managerial positions. This division matches 51.8 percent of the managers’ responses to the relevant triads. For workers, the match is only 44.1 percent, a statistically significant difference.11
The working-mass division also matches the triad responses quite well, but there is little difference between managers (50.7 percent matching) and workers (50.2 percent). This confirms the nonsignificant differences in INDSCAL scores reported above.
The workers’ preferred division is an even more elitist model than the working-mass division. A division that isolates the occupation of big corporation executive into a class by itself matches 56.7 percent of the workers’ triad responses. Managers, however, are less likely to isolate the executive; 48.2 percent of their responses match this division.12 Thus the best match for workers is the opposite of the best match for managers: workers prefer to isolate a small elite at the top; managers prefer to separate a small lower/working class at the bottom.
The managers’ tendency to a middle-mass division has a clear conservative connotation. It lumps most Americans together in a large middle mass, recalling Ben Wattenberg’s (1974) complacent image of a “massive majority middle class.” Some problems of inequality may be recognized at the bottom, but the larger part of American society is subsumed in a comfortable category that ranges from corporate executive to machinist.
In contrast, the workers’ tendency toward an elitist model is potentially more radical. The capitalist is isolated at the top. Differences are recognized below that level, of course, but the fundamental division in the class structure is at the very top.
This is, in fact, not unlike the images that Goldthorpe et al. (1969) discovered with their English autoworkers. Both their power models and their various money models isolated small elites at the top, merging the middle class and working class in a large subordinate group. That study emphasized distinctions among these various elitist models, but our data suggest that such variants may be less important (cf. Moorhouse, 1976) than the difference between the elitist model and the middle-mass perspective. Goldthorpe and his colleagues did not find evidence for such middle-mass models, but their samples excluded managers and professionals—precisely the people who favor such models. Indeed, the popularity of middle-mass models in establishment social science may be mainly a reflection of a distinctively but understandably middle-class perspective within the U.S. social science establishment. The mistake has been to generalize this perspective to all Americans rather than recognizing its limitations as a middle-class tendency.
It would be another mistake to overextend our interpretation by calling these two images the working-class and middle-class images of American society. Both classes can and do recognize the relevance of both images. But the “strain of consistency” that Bott urges us to identify is different for workers than it is for their bosses. American workers tend toward what is, for them, a more class-conscious image: a society dominated by a small elite.
Our general argument on American exceptionalism is thus supported by these data. When we gather evidence on what American workers actually think, even a small exploratory study such as this one finds them to be more class conscious than they are generally believed to be. The weight of our evidence is properly cumulative, but the pieces are beginning to fit together. The American exceptionalism thesis suggests a lack of difference between American workers and the American middle class, but the results in Chapter 4 and the widespread acceptance of the proletarianization perspective demonstrate that this class division is in fact well perceived in American society.
Respondents were 317 male residents of the Chicago metropolitan area, 25 years or older, interviewed by telephone during October and November 1974. Unfortunately, women were excluded, but this was done in order to simplify analysis of the effects of occupation on class images. Since other research (see Chapter 8) shows that women use both their own and their husbands’ occupations in determining subjective class placement, the small sample of this study would have made it impossible to separate these effects reliably. Men younger than 25 were also eliminated in order to facilitate the analysis of the relationship of occupation to class image.
The response rate for this survey was disappointingly low: 53 percent. Some of this might be attributed to the post-Watergate, post-Vietnam alienation of the times; much is probably due to the difficulty of the triads task. In any event, the resulting sample was heavily skewed toward better-educated, more middle-class respondents. The sample included roughly twice the proportion of college-educated respondents as was reported in the 1970 census of the Chicago metropolitan area. To counteract the worst effects of this sample bias, we weighted the respondents to match the census distribution across six educational categories.
The question used for two-thirds of the sample was as follows: “Now I’m going to read the names of some occupations, three at a time. For each set of occupations, please tell me which two of the three you think are in the same social class.” The remaining one-third of the sample was asked to choose “which two of the three you think are most similar to each other.”
The disadvantage of the triads method is the large number of judgments required even with a small number of occupations: eight occupations form 56 triads. The method has the advantage of asking respondents for concrete judgments. In retrospect, it seems to us perhaps to carry this virtue to an extreme. After many repetitions, the task becomes quite trying. Moreover, many of the judgments are not easy, especially in a telephone interview. These problems may account for the low response rate. Reducing the occupations to five (by dropping the supervisory authority dimension, for instance) would present a more reasonable series of 10 triads.
A somewhat similar and very promising technique is the “vignette” method developed by Peter Rossi and his coworkers (see, for example, Alves and Rossi, 1978; Rossi and Nock, 1982). The vignettes offer the advantage of presenting respondents with a social composite including many characteristics beyond occupations. Laumann and Senter (1976) have resurrected a social distance methodology to study perceived differences between occupations. The analysis closest to our own is Coxon and Jones’s (1978) multidimensional scaling of perceived occupational similarities. All three of these techniques could provide valuable new data if suitably adapted to incorporate class factors.
The 56 triads were first analyzed with the INDSCAL program, an extension of multidimensional scaling methods to incorporate individual differences (Carroll and Chang, 1970). The triads data were converted into perceived distances for each respondent by counting how often a pair of occupations was selected as most similar. With eight occupational stimuli, the maximum possible score was six, indicating that a pair was perceived as most similar each time it was included in a triad with any of the other occupations. The minimum score was zero, indicating that the pair was never chosen as most similar. “Don’t know” responses were scored as one-third for each of the three pairs. In this way, a similarity score was computed for every pair of occupational stimuli (eight stimuli produce 28 possible pairs) for each respondent. These scores were the basis of the INDSCAL analysis.
The second analysis is a variant on a divisive clustering algorithm (Bailey, 1974). It computes directly the match between the actual pattern of triad responses and the patterns implied by each possible division of the eight occupations. Each possible division leaves some of the triad responses indeterminate: these triads are excluded from the calculation of the proportion of matching responses.
1. For other examples, see Willener, 1957 and 1975; Popitz et al. , 1969; Lopreato and Hazelrigg, 1972; Bulmer, 1975; Coleman and Rainwater, 1978:18.
2. This problem also confronts unstructured and in-depth interviews—resulting in such divergent interpretations as those of Lane (1962), Sennett and Cobb (1972), and Garson (1973). Sennett and Cobb interpret their working-class interviewees as suffering from a lack of respect—such as being addressed by their first names while their white-collar neighbors are always called “Mister.” But their interviews also reveal how the workers resent the power that the middle class exercises over their lives. For example, an electrician’s apprentice gripes, “It’s a question of, like am I working for someone? I . . . I feel like I’m taking shit even when, actually, even when there’s nothing wrong” (1972:34). The authors neglect the more class-conscious themes of dominance and subordination in such material and prefer a status-oriented interpretation. Both interpretations may be correct given the multiple consciousness of each American worker, but the neglect of the class themes creates the erroneous impression of weak class consciousness.
3. The converse of this is often true, too. As Michael Burawoy (1979:139) observes in his observation of machine shop workers, “The idiom in which workers couch and rationalize their behavior is no necessary guide to the patterns of their actual behavior.” (See Chapter 11)
4. Intensive interviewing often corroborates the complexity of workers’ understanding of society. David Garson (1973) found that Massachusetts autoworkers would both endorse the standard American ideology of individual responsibility for success and advocate class solidarity to confront bosses. He interprets their ideas as typical of the multiple consciousness of American workers.
5. How are we to interpret the answers people give to unstructured questions about class perception? It seems likely that the class images described often simply comprise categories and phrases picked up from popular culture—like the recitation of school grammar rules. In the United States, to be sure, the popular culture is largely bereft of class categories (Marwick, 1980). Neither the dominant American cultural institutions (the media and the schools) nor working-class-oriented organizations (unions and the Democratic Party) supply workers with the vocabulary they need to articulate the class divisions that surround them (Mann, 1973). It is expecting too much to presume that workers should develop this vocabulary de novo.
6. The details of the survey and its statistical analysis are presented in the methodological appendix to this chapter.
7. Some earlier research with occupational differences (e.g., Burton, 1972; Coxon and Jones, 1974) found that respondents often distinguished occupations by organizational setting: differentiating bureaucratic occupations from occupations where work was individualized (e.g., gardeners, police officers, real estate agents). Because we wanted to avoid this “situs” distinction in the main analysis, we chose the list mainly from the industrial setting. See also Blau and Duncan (1967) and Laumann (1973) for similar organizational setting dimensions in patterns of occupational mobility and friendship choices.
8. The working-mass image reflects the results of a proletarianization process that lowers many white-collar workers into the working class. The middle-mass image reflects the results of an embourgeoisement process that raises affluent blue-collar workers into the middle class (cf. Blumberg, 1980).
9. The difference between the managers’ average (0.48) and the workers’ (0.36) is 3.1 times the standard error of the difference.
10. The workers’ average (0.48) is only slightly greater than the managers’ (0.45). The difference amounts to only 0.62 times the standard error of the difference. Similarly, the manager-worker differences on form B of the questionnaire (listing the alternative occupation titles shown in Figure 5.3) were not statistically significant. The problem with this form was that dropping the two intermediate occupations, foreman and machinist, made the gap between the top and bottom groups so large that it forced most respondents to use the working-mass perspective for most of the triads. Thus, this form became less sensitive to individual differences within the sample.
11. Chi-square = 25.2 with 1 degree of freedom. The rather low proportion of matching responses for even this best division reflects the range of perspectives among managers but also the variety of models that any one respondent may apply to the series of triads. We interpret the indeterminacy of this analysis—the fact that no one model is made to fit any respondent perfectly—as part of the merit of the triads method. The appropriate research question reveals the relative strengths of the different perspectives that each person combines in his or her own way.
12. Chi-square = 14.3 with 1 degree of freedom.