THE DECLINE OF SOCIAL CLASS?
The second half of the twentieth century brought enormous changes to the United States. Prosperity and labor peace characterized an “affluent society” (Galbraith, 1958) that most observers quickly interpreted as deadening the class consciousness of American workers.
It had not always seemed that workers were destined for such a fate. The first part of the century had been dominated by events that raised workers’ consciousness: the concentration of capital in giant corporations, the rise of the Socialist Party, and the ravages of the Depression with its sudden explosion of union organization.
But the prosperity of the post-World War II period ushered in a new era of more comfortable working-class existence and a softening of class conflicts. The American Socialist Party had long since faltered, split up into warring factions, and self-destructed. The bitter labor disputes that had marked the growth of unionization in the 1930s had subsided into well-orchestrated collective bargaining. Unprecedented prosperity from World War II through the Vietnam War endowed workers with a standard of living that would have been inconceivable to their parents and grandparents. Not even the stagnation of recent times has done much to resurrect a moribund labor movement.
The “End-of-Ideology” Thesis
The stereotypes of American workers that arose from these two eras are starkly contrasting. If we conjure up a labor scene from the Depression, we may observe Walter Reuther’s autoworkers battling company goons on the picket line in Detroit. But in the 1950s, the picture is one of workers moving to the suburbs, owning their own homes and cars, sending their kids to college, and generally winning their share of American abundance. The 1980s image shows workers quiescent or even defeatist, their politics inclining toward Ronald Reagan rather than militant protest.
Such stereotypes are mischievous at best. The supporting evidence is, and always has been, too thin to justify the pictures. In this chapter we describe some new evidence that portrays a more class-conscious transition during the twentieth century.
But it is also important to realize how the stereotype found its way into social science; indeed, academe shares much of the blame for foisting the distorted image on the American public. Grand theories about “the end of ideology” proclaimed a new society based on class accommodation. As Seymour Martin Lipset (1960:442) put it, “The fundamental political problems of the industrial revolution have been solved.”
The end-of-ideology thesis interpreted the midcentury affluence as the final refutation of Marx. Marx had written in the midst of nineteenth-century industrialization when an ever-increasing division between workers and capitalists seemed likely, when society was “splitting up into two great hostile camps . . . Bourgeoisie and Proletariat” (Marx and Engels , 1976:485). But the end-of-ideology school claimed Marx had been mistaken about the future. Working-class consciousness had waned, not waxed. The changes in modern society, which Marx could not have foreseen, had controlled and minimized the conflict between workers and capital. A widely held view of industrial relations (e.g., Kerr et al., 1960; Ross and Hartman, 1960) forecast that strikes would decline as the acceptance of collective bargaining and grievance procedures tamed the once-bitter labor strife. The successful containment of class conflict destined class consciousness to become a historic relic of the early industrial period.1
The end-of-ideology school did not have the facts straight, however, even on the observable levels of class conflict. The record of strikes during the twentieth century (Figure 6.1) hardly reveals a long-term secular decline. Identifiable peaks define the wavelike nature of class conflicts (an international, not merely American, phenomenon; see Korpi and Shalev, 1980). The gradual decline of strikes in the 1950s and 1960s—the trend that prompted many of the theories about the institutionalization of class conflict—was merely a trough in the cyclical pattern, not a final resolution of class divisions presaging the waning of class consciousness.
FIGURE 6.1. Strike involvement, 1880–1980
SOURCE: U.S. Census, 1973a, 1983.
NOTE: Strike rates are moving averages across ± 2 years.
But the main problem with the end-of-ideology thesis is that its proponents rarely studied workers’ class attitudes and perceptions. We know very little about how class consciousness changed or even when or in what direction; most of the theories only inferred changes from the observed outcomes of class conflict. The rise and decline of unions or radical political parties have been taken as the main indicators of class consciousness. This now familiar fallacy of psychological reductionism obscured the fact that there were rarely any appropriate data on workers’ consciousness itself. Some of this lack was inevitable, since “dead men cannot be interviewed, and humble workmen left no written testimony behind for the historian’s use” (Thernstrom, 1964:58–59; see also Laslett, 1970:6; Dubofsky, 1975:6). Even when workers did leave such testimony as the “fierce words and bitter feelings” uncovered by Stephan Thernstrom (see Chapter 2), historians often discounted its significance or twisted its meaning to reconfirm their preconceptions.
We are now in a somewhat better position than earlier researchers because we have survey data on several generations of Americans whose work lives have spanned most of the century. These data have their limitations, but they are an improvement over the almost total lack of historical data on workers’ attitudes that confronted earlier theorists.
The evidence of the previous two chapters should caution us against too readily accepting any conclusions about the “end of ideology.” Contemporary American workers do recognize class divisions, and they are not distracted by the many status distinctions that have proliferated in this century. Nevertheless, we have not yet looked at direct evidence about changes in class perception, and it is entirely possible that the clarity of such perceptions has faded considerably over time. Perhaps the class perceptions reflected in our most recent data are only a pale shadow of earlier and sharper distinctions that American workers drew between themselves and their bosses. We turn now to these historical questions.
Changes in Class Perceptions
The growth of the middle class
Figure 6.2 reports the changes in middle-class placements in the ten election surveys from 1952 to 1978.2 The fitted trend line demonstrates the growth of the middle class during this period of postwar prosperity. The rate of change, about half a percentage point a year, is not dramatic. Even by the end of the period there is little support for the “massive majority middle class” claims (Wattenberg, 1974:51) that supposedly heralded the “end of ideology.” When we understand better the causes of this change, there is even less reason to believe that class divisions are fading from popular perceptions: the size of the classes may have changed, but the division between them remains as real as ever.
The first question we will ask is whether the increase in middle-class self-placements was the result of middle-class generations replacing working-class generations, or whether the prosperity of the period affected all generations equally to raise levels of middle-class placements within each generation. In the language of statistical analysis, is the change a cohort effect (generational replacement) or a period effect (a time trend that is uniform across all generations)?
Generational effects suggest a more enduring modification of American society. Two birth cohorts have had especially dramatic impact. The more recent, the “Baby Boom” generation born between 1946 and 1964, changed the face of America. This demographic bulge was responsible for such diverse results as the explosion of school construction in the 1950s and 1960s and the increased crime rate of the 1970s. The political impact of the Baby Boom generation has been less thoroughly analyzed. Only a few of its members get into the later surveys analyzed here; even so, we will see that they perceive a distinct class position for themselves.
Political analysts (e.g., Campbell et al., 1960; Oppenheim, 1970; Knoke and Hout, 1974; Converse, 1976) have more often singled out the “Depression cohort” of voters who entered the electorate during the 1930s and were captured by Roosevelt, the New Deal, and the Democratic Party.3 These voters created the major party realignment of 1932 (Burnham, 1970). Even studies that include data through 1972—a full 40 years later—show a clear persistence of Democratic party identification among the Depression cohort; their party loyalty is greater than that of cohorts either before or since (e.g., Knoke and Hout, 1974; Converse, 1976).
It seems reasonable that the class consciousness of the Depression cohort, which entered the labor market during the greatest economic catastrophe in American history (Elder, 1974:3), would be equally distinctive. The Depression was difficult for almost everybody but most difficult for the groups least integrated into secure positions in the economy: the working class, minorities, and youth (Campbell et al., 1960:153). Their subordination to the larger economic forces may have indelibly imprinted a working-class identity on their class perception.
We can easily rearrange the survey data to compute class perceptions according to the year of the respondent’s birth rather than the year of the interview. Figure 6.3 compares birth cohorts between 1887 and 1958.4 The results suggest that the Depression cohort is indeed distinctive in its class placements. Its earliest members, those born soon after the turn of the century who were in their late twenties as the Depression hit, do indeed constitute the most working-class cohort in U.S. history. Those born earlier are more middle class, and there is a consistent though slow increase in middle-class placements over most of the cohorts from 1902 to 1945. But those born immediately after World War II, the beginning of the Baby Boom generation, are again more working class.
Figure 6.3 may be a misleading description of cohort changes because the averages are contaminated by the increases in middle-class placements between the 1952 and 1978 interviews (as seen above in Figure 6.2). The more recent (middle-class) cohorts appear more often in the later (more middle-class) surveys, while the Depression cohort appears more often in the surveys of the (more working-class) 1950s.
Age differences may also contaminate the cohort averages reported in Figure 6.3; for example, the nineteenth-century cohorts, who appear quite middle class, are also quite old by the time they get into our election surveys. Perhaps they are more middle class because older people tend to see themselves as middle class, not because that generation is distinctive. We would like to compare the nineteenth-century cohorts with the later cohorts when they were all the same age. Only then can we decide whether their middle-class placements are a fundamental generational difference or just a normal consequence of advancing age.5
FIGURE 6.3. Class self-perceptions of birth cohorts, 1890–1954
SOURCE: American Election Surveys.
NOTE: Smooth curve: class placements = 817.45 – 19.265 cohort + .156 cohort2 – .00041 cohort3. R2 = .766
Investigating the true cohort effect, uncontaminated by age and period changes, requires a sophisticated statistical analysis that allows us to look at birth-cohort differences among people of about the same age, interviewed in the same year (or period).6
We have plotted these “true” generational changes in Figure 6.4 together with the best-fitting curve to the points. The smooth curve provides a somewhat clearer picture of the historic trends.7 Figure 6.4 confirms the extraordinary working-class orientation of the Depression generation. Middle-class placements are highest at 57 percent in the earliest birth cohort (1887–91). After that, middle-class placements decline sharply, reaching their nadir at 37–38 percent in the 1904–19 birth cohorts. These cohorts are the Depression generation: their early working experience occurred during the decade of the 1930s. Our more careful analysis, eliminating the possible contamination of age and survey timing, confirms the conclusion that the Depression cohort is the most working-class generation in American history.
FIGURE 6.4. Effects of birth cohort on class self-perceptions, 1890–1954
SOURCE: American Election Surveys.
NOTE: Effects are calculated with controls for age and survey year.
Smooth curve: middle-class placements = 977.46 – 22.354 cohort + .175 cohort2 – .00045 cohort3. R2 = .886
The cohorts immediately following the Depression adopt slightly more middle-class orientations, peaking at 48 percent in the 1945 cohort. An upswing after the Depression is not surprising, as these cohorts entered the labor force during the expanded economy produced by World War II. But the upswing is rather modest, not as dramatic as indicated in the previous figure suggesting that some of the apparent post-Depression increases are a general time-period effect, reflecting the good times during which these cohorts were interviewed. In fact, the post-Depression cohort appears more similar to the Depression generation (that is, more working class) than to the pre-Depression cohorts. Moreover, the second inflexion point at 1945 indicates that the post-Depression increases have not been maintained. After 1945, the curve dips again: the Baby Boom generation is more working class than the World War II generation. The most recent cohort in the sample, the post-1952 birth cohort, is 42 percent middle class, only 4 percentage points above the Depression cohorts.
The overall picture that emerges from this analysis is a Depression generation dramatically more working class than the cohorts that came before but only slightly more working class than the cohorts that came after. The Depression’s effect in changing class perceptions appears to have been permanent.8
Cohort succession or time-period trend?
The cohort curve indicates that the trend in recent interviews to more middle-class placements (Figure 6.2) did not result from cohort succession processes: that is, middle-class cohorts replacing working-class cohorts. The recent birth cohorts entering the labor force are no more middle class than those who are exiting through retirement or death. If we assume that the full cycle of cohort replacement takes about 45 years, then in the 1952 surveys the 1887 cohort (65–year-olds) were being replaced by the 1932 cohort (20–year-olds). Examining the curve at these two points reveals that the net effect of such replacements is toward greater working-class identification: the 1887 cohort is very middle class and the 1932 cohort is much more working class. This trend—working-class cohorts replacing middle-class cohorts—continues until about the year 1965. In 1965, the newer 1945 middle-class cohorts are replacing the equally middle-class 1900 cohorts. The net effect of this cohort replacement on the levels of middle-class placement is negligible.
Nor does the most recent period reveal any generational trends toward embourgeoisement. The Baby Boom generation is quite working class (unlike the World War II generation that immediately preceded it) and replaces an equally working-class generation from the early post-Depression era (the 1925–35 birth cohorts). The more working-class orientation of the Baby Boom generation prevented a sudden rise in middle-class placements. If the high middle-class orientations of the World War II cohorts had continued, middle-class cohorts would have replaced more working-class cohorts, thus accelerating the embourgeoisement of America. Instead, levels of middle-class placements fell among the Baby Boom cohorts. Even during the most recent surveys (1965–78) the entering cohorts were no more middle class than the departing ones.
Our conclusion, therefore, is that the recent increase in middle-class orientation remains independent of cohort succession trends. When we look at cohort replacements in the entire period 1952–78, at no point are recent middle-class cohorts replacing earlier, more working-class cohorts. The postwar trend to middle-class identification, therefore, must represent a true time-period change: a uniform move to more middle-class placements among all birth cohorts and all age groups. We will confirm this shortly, but first we must ask what causes the cohort differences.
Why is the Depression generation different?
Probably the most striking aspect of the cohort comparisons is the uniquely working-class sentiment of the Depression generation (the 1904–23 birth cohorts). The 1914 cohort is 10 percentage points more likely to identify as working class than their 1887 predecessors. The Depression cohorts are also more working class, by a few percentage points, than the cohorts that came after. Explaining the later trend toward more middle-class placements is relatively easy, so we take up that task first. Explaining the differences between Depression and pre-Depression cohorts is more difficult but, we feel, more interesting for understanding the long-term changes in American class consciousness.
The post-Depression cohorts. Americans born in the post-Depression cohorts see themselves as more middle class because they are more middle class: they more often hold managerial positions, and they have more college degrees than those of the Depression generation (see Table 6.1).
When we compare equivalent workers from different cohorts—that is, workers with the same class position, education, and income—the post-Depression generation appears very similar to the Depression generation. In Figure 6.5, the most working-class Depression cohort (in 1918) is 41 percent middle class; middle-class placements rise slowly to 46 percent in 1946—a small 5 percent increase, compared with the 11 percent increase for the same years in Figure 6.4. Thus the increased middle-class identifications of the recent cohorts are explained by the somewhat improved social structural conditions they entered.
SOURCE: American Election Surveys.
NOTE: Managerial positions include the census classifications of managers and professionals but exclude technicians.
FIGURE 6.5. Birth-cohort effects for equivalent workers, 1890–1954
SOURCE: American Election Surveys.
NOTE: Effects are calculated with controls for age, survey year, managerial position, occupational prestige, education, and family income.
Smooth curve: middle-class placements = 738.88 – 16.145 cohort + .123 cohort2 – .00031 cohort3. R2 = .903
The pre-Depression cohorts. The same explanation is not available for the pre-Depression cohorts. The early cohorts saw themselves as more middle class (Figure 6.5) despite being less often managerial or college educated. When we compare workers in pre-Depression cohorts with equivalent workers in later cohorts, the pre-Depression generation appears even more remarkably middle class. In fact, the most outstanding features of Figure 6.5 are the high rates of middle-class placements among the pre-Depression cohorts. There is a drop of 18 percentage points in middle-class placements from the 1890 cohort (59 percent) to the Depression cohort of 1918 (41 percent).
It is the pre-Depression Americans, not the Depression cohorts usually discussed, who are most unusual. No simple social structural differences can account for their high rates of middle-class placements; we think that a more psychological explanation is required. A further analysis suggests not only that the overall rates of middle-class placements are extraordinarily high, but that the class division between managers and workers is less clearly perceived before the Depression. Table 6.2 divides the birth cohorts into four main groups: the pre-Depression, Depression, World War II, and Baby Boom generations. When managers and equivalent workers are compared within each group, there are large differences in class perceptions for the last three cohorts but very small differences within the pre-Depression cohorts. Pre-Depression workers often place themselves in the middle class. It is the trauma of the Depression that solidifies working-class perceptions and changes the way Americans think about class. Pre-Depression workers appear more “confused” about class divisions in America. After the Depression, the division between managers and workers emerges as a main determinant of class perception and remains important through present generations.
SOURCE: American Election Surveys.
NOTE: Adjusted percentages are calculated after controls for occupational prestige, respondent’s education, and family income.
In short, our results contradict the implications of the end-of-ideology thesis. Recent generations see class divisions as clearly as their parents did. A few more see themselves as middle class, but that is what we would expect from the increases in college education and managerial positions. Baby Boom generation workers are just as likely as those from the Depression generation to consider themselves working class; neither group has any special middle-class illusions. Instead, what emerges most strikingly from our analysis is the indication that the pre-Depression cohorts are the workers most confused about class divisions in America. The Depression transformed class perceptions in the United States, and, as far as we can see, the transformation was lasting.
Time-Period Changes in Class Perceptions
Causes behind the growth in middle-class placements
We began the analysis of birth cohorts by asking whether generational replacements could explain the middle-class increases between the first survey in 1952 and the last in 1978. By now it should be clear that it does not, although we think we have learned along the way much about how Americans have changed the way they think about class. The time trend pictured in Figure 6.2 is a true period effect: all ages and all generations increased their rates of middle-class placements during those decades. When comparisons are made within birth cohorts and age groups, the time-period effect remains virtually unchanged: the average yearly increase in middle-class placements changes from 0.515 percent per year to 0.502 percent per year.
An increase in middle-class placements should not be surprising. The period in which the surveys were conducted spans one of the most prosperous eras in American history. Managerial positions opened up, schooling lengthened, and incomes grew (see Table 6.3). It seems reasonable to expect that increased prosperity would increase middle-class placements, and the structural changes do in fact account for all the increases in middle-class placements. A comparison of workers in the same occupational position, with the same educational background, and earning the same salaries shows no measurable change in the rate of middle-class placements between 1952 and 1978. The overall increase, about 0.51 percentage point per year in Figure 6.2, drops to a decrease of 0.03 percentage point when comparing equivalent workers, as in Figure 6.6. This is no more than we might expect from chance fluctuations. Middle-class placements increased because there were more college-educated managers earning high incomes in 1978 than there were in 1952. But any given worker who stayed in the same position throughout this period was not likely to change his or her self-perception.
FIGURE 6.6. The effects of survey year for equivalent workers, 1952–78
SOURCE: American Election Surveys.
NOTE: Effects are calculated with controls for age, birth cohort, managerial position, occupational prestige, education, and family income.
Middle-class placements = 43.366 – .034 (year-1900). R2 = .023.
This finding would merit little attention except that at least two schools of thought would lead us to expect other results. The first, the “embourgeoisement” school, argued that growing prosperity infused all workers with a sense of middle-class attainment so that they all changed the way they thought about themselves. Mark Evers (1976:13), for instance, interprets growth in middle-class placements among a Detroit sample as reflecting “part of a general trend toward a universal ‘middle class’ existence and lifestyle, as so often heralded by commentators in the popular media.”
In fact, however, there was nothing “universal” in the changes brought about by the postwar prosperity. Workers did not change, even though their relative proportions in the labor force did; that is, the decline in working-class self-placement was not a decline in working-class consciousness. Instead, it was a decline in the relative number of working-class positions. The point is similar to the one we made in Chapter 2: there are structural explanations for changes that are often attributed to psychological causes.
Absolute or relative standards of class perceptions
Studies of perceived well-being. A second school of thought is more subtle, but its failings are more revealing for understanding the psychology of class perceptions. In the early 1970s, as appropriate data became available, several social scientists began to ask whether the unparalleled prosperity of the postwar decades had translated into greater personal satisfaction. To their surprise, they found that most measures of well-being were unrelated to changes in national prosperity. Our class-placement results are less surprising in that regard, since they do reflect the increased prosperity. The different results for class perceptions and personal satisfaction suggest that different psychological processes link the national prosperity and individual perceptions. The personal satisfaction studies therefore will bear some closer scrutiny.
Richard Easterlin (1973) compared measures of personal happiness in 19 countries. Despite the fact that the countries ranged in overall prosperity (based on their gross national product) from the United States with $2,790 GNP per capita to Nigeria with $134 GNP per capita, there was no noticeable pattern of national averages in the happiness ratings. Easterlin found, to nobody’s surprise, that within each country people with higher income reported greater happiness. Money, it seemed, could buy happiness for people but not for nations. Easterlin interpreted the paradoxical findings by concluding that people judge their own well-being relative to some social norm of what goods they ought to have. This norm varies among societies: the wealthier the society, the higher the norm against which one measures one’s own well-being. Thus there are no national differences in perceived well-being because the people living in the more prosperous countries have a stiffer standard for what constitutes happiness. Every increment in prosperity for the whole country is matched by an increment in the standard to which personal well-being is compared.
Otis Dudley Duncan (1975) found the same paradoxical result when comparing changes in satisfaction in the United States between 1955 and 1971. Despite the increase in average income over the 16 years, Duncan found no increase in satisfaction with the standard of living. Within each year, however, Duncan also found that the wealthier Americans were, in general, happier Americans. Again, more money bought more satisfaction for individual Americans but not for the country as a whole.
Duncan also interpreted his results in relative terms: “The relevant source of satisfaction with one’s standard of living is having more income than someone else, not just having more income. And satisfaction measures as such cannot tell us whether a population with a higher average income is really ‘better off’ than a population with a lower one” (1975:23). Even adjusting for inflation, people did not find the $10,000 they earned in 1971 to be as satisfying as the $10,000 they earned in 1955. A family earning $10,000 in 1971 was better off than 48 percent of 1971 families, but a family earning $10,000 (in 1971 dollars) in 1955 had been better off than 85 percent of 1955 families (U.S. Bureau of the Census, 1972) and therefore felt more satisfied than the 1971 family.
Lee Rainwater (1974) found evidence supporting this change in income standards. Results from 18 Gallup Polls between 1946 and 1969 showed that Americans had raised their standards for “the smallest amount of money a family of four needs to get along in this community.” In 1946, Americans on the average thought an income of $4,614 (in constant 1971 dollars) was necessary; after 23 years of steady prosperity, the amount had increased by about half, to $6,878.
Class perceptions as absolute judgments. Class perceptions do not obey this logic of relative comparisons. Unlike satisfaction with living standards, middle-class placements increased between 1952 and 1978—there were more managers, more college degrees, and higher incomes in 1978—but they had the same meaning in 1978 as in 1952. Comparing equivalent people in the two years yields a statistically nonsignificant change over the time period.9
We interpret these findings to mean that the class perception process obeys a more absolute logic. The meaning of “working class” and “middle class,” unlike that of happiness and satisfaction, remains the same over time. The labels have specific referents (a managerial position or a college degree, for example) rather than a relative ranking in a status hierarchy (such as more education than average, or a higher prestige occupation than most people). This is exactly the point we made in Chapter 3 in comparing class perception and prestige studies, and in Chapter 4 in examining the determinants of class perceptions. Americans do interpret class in categorical terms with some absolute standard of reference; that is, they perceive class, not just status. When appropriate (in judgments of personal well-being or occupational prestige, for example), Americans can use a relative-comparisons model that ranks people along a continuum of social standing. But class divisions are also meaningful, and the evidence is that the terms “working class” and “middle class” are used in this categorical way.
1. See Blumberg, 1980, for a good review and refutation of these theories. Many scholars have noted that class conflict is most intense in early industrialization (see note 5, Chapter 1). In the concluding chapter we suggest an explanation for this phenomenon that does not rely on changes in working-class consciousness.
2. The sample in this analysis and throughout the rest of this chapter comprises respondents between the ages of 21 and 64 who have recorded values on occupation, education, and income. See Schrieber and Nygreen (1970) for an earlier analysis of similar data.
3. Unfortunately, research has not been consistent in its definition of the Depression cohorts. Some research has defined this generation very broadly as all birth cohorts from 1900 to 1930 (e.g., Knoke and Hout, 1974), or quite narrowly as only those from 1905 to 1912 (e.g., Converse, 1976). For our purposes, entering the labor force—not entering the voting population—during the Depression should be the important influence on class perceptions; therefore, we look especially at the birth cohorts from 1904 to 1923. These people were either in their mid-twenties in 1930 or had reached at least age 16 by the end of the decade—the ages when most Americans first enter the labor force.
4. We have coded birth cohorts into a series of dummy variables for this and following analyses. The sample was divided into 16 four-year intervals representing the birth cohorts from 1887–91 to 1947–51 and a seventeenth cohort including all respondents born between 1952 and 1958. In the analyses where age and time period were included, they too were categorized: age was divided into 11 dummy variables, one for each four-year interval between ages 21 and 64; time period into 12 dummy variables for the 12 survey years from 1952 to 1978.
It may seem that a cohort analysis requiring 40 dummy variables offers an unnecessarily complex alternative to the more obvious and parsimonious model including only the three interval variables: actual age, year of interview, and year of birth. However, this case dictates the more cumbersome approach: first, it seems unreasonable to assume that all age, period, and cohort effects are linear; second, to include all three dimensions in a single analysis results in a linear dependency among the measures—Cohort = Period – Age—which precludes a unique solution to the normal equations of the regression analysis. Dividing each dimension into a series of dummy variables and constraining some to equal effects helps eliminate that dependency; see also note 6, below.
5. Because earlier studies were based on a single survey, they could not separate the effects of age and birth cohort. Schlozman and Verba (1979), with their more ambiguous open-ended question, found that older workers were more likely to place themselves in the working class, a result consistent with a more working-class Depression generation. They also found higher levels of working-class placements in a comparable 1939 survey. Centers (1949:167) found slightly more working-class placements among manual workers 30 to 49 years old, roughly the Depression cohorts in our analyses.
6. The technique follows each cohort throughout its life cycle and measures class placements at all ages. If birth cohort is the critical factor, we would expect the Depression generation to identify as more working class at all ages; if age is the critical factor, we would expect the older ages to identify as more working class in all surveys, regardless of even wide differences in birth cohort. But to trace such patterns—and thus separate age, period, and cohort effects—requires several surveys at widely separated points in time.
Constraining some of the time intervals to have the same level of class placements enables parameter estimation (Mason et al., 1973), but the results can be affected by which age groups, birth cohorts, or time periods are pooled. We have tested 15 different models constraining different combinations of cohorts, ages, and survey periods to be equal. All analyses produced similar results. We have reported the one with the fewest constraints. Largely for reasons of data management, ordinary least-squares regression is used in all cohort analyses rather than probit analysis. The number of cases and variables in these several analyses made the use of probit analysis computer routines impractical.
7. To plot the smooth curve, we regressed the middle-class percentage of each cohort interval on four polynomials (first on a linear cohort variable, then quadratic, cubic, and fourth-power cohort variables). Even the plot makes it clear that a straight line (the first-order polynomial) will not adequately describe the trend. Neither does the addition of a quadratic term (representing a U-shaped curve) fit the data well: R2 = 0.62. A better fit is the cubic equation with an R2 of 0.89. A fourth-order polynomial is unnecessary, since the additional term is not statistically significant and the increment to R-squared is negligible. The two inflexion points (bends) for the cubic curve estimate the times when the curve reaches a local minimum and maximum; a local minimum is observed at 1914 and a maximum at 1945.
8. To understand why the cohort-class relationship took on this shape after controls, one needs to examine the parallel age and period curves. Although not presented here, the age-class relationship is an inverted-U pattern that counterbalances the U-shaped pattern of the birth cohort relationship. The age-class relationship therefore suppresses the early-cohort-middle-class relationship. Specifically, the early birth cohorts are observed mainly in their older ages when the tendency is to working-class identification. When the competing effects of age are eliminated through statistical controls, their relatively high levels of middle-class identification become apparent. Likewise, the recent cohort trends are confounded with period changes so that recent cohorts appear more middle class than they actually are. Recent cohorts entered the sample in more prosperous, middle-class periods. When the period trend toward middle-class identification is controlled, the recent cohorts are more similar to the Depression cohorts. In sum, separating the life-cycle and period trends in class perceptions from those across generations does not alter the strength of the cohort-class relationship but changes its form.
9. A relative-comparisons model would predict a significant negative time coefficient, once individual social structural variables were held constant, as Duncan (1975) found. It is, of course, not possible to prove the null hypothesis, but the analysis reveals no support for the relative-comparisons model.