The definition of occupations is based on the 1990 occupational classification system of the Bureau of Labor Statistics. The IPUMS census and CPS databases both include a consistent variable across time based on the 1990 classification. For the GSS and NLS databases, I converted the 1960, 1970, 1980, and 2000 classifications to the 1990 coding.
Occupations were then classified into eight categories, shown below with examples of the occupations that fall within them.
The Belmont occupations consist of those in categories 1 and 2. The Fishtown occupations consist of those in categories 5 through 8.
The socioeconomic status (SES) of an unmarried adult living alone is determined by that person’s education, occupation, and income. For adults who are part of a married couple, the situation is more complicated.
In 1960, SES was almost always determined by the status of the husband both because of custom and because so few married women had a job in a higher-status occupation than the husband’s. Both factors changed over time, reflected in the designation of “head of household” in the Current Population Survey. The wife was designated as the head of household for only 1 percent of married couples in the 1960 census. By 2010, women were designated as the head of household for 42 percent of married couples in the CPS.
What then is the SES of a couple in which the husband works on an assembly line and the wife is manager of the company’s payroll department? No answer works for all cases, but I chose to assign people who are part of a married couple to Belmont or Fishtown based on the person who has the higher-ranking occupation, with “higher” based on the order of the eight occupational categories listed above. If only one spouse has an occupation, assignment is based on the person with an occupation. If both spouses have a Belmont occupation or both have a Fishtown occupation, I used the educational data for the spouse with the higher level of education.2 These criteria also define “head of household” as I use the phrase in the text.
Within the framework described above, persons were assigned to Belmont or Fishtown according to the following decision rules:
With regard to the last category, adults who are neither the head of household nor the spouse, I again did not have the option of choosing a perfect rule. A twenty-three-year-old who is living with his affluent parents probably still enjoys their socioeconomic status even if he is working as a bartender. But the rule becomes more consistently appropriate when dealing with persons ages 30–49, as almost all of the analyses in part 2 do. The older you get, the more your status depends on your own education and job, no matter with whom you live.
For a few tabulations, I needed to classify persons under the age of 21 who were neither the head of household nor the spouse. They were assigned to a neighborhood based on the occupation and education of the head of household.6
The three standard components of socioeconomic status are occupation, educational attainment, and income, and yet I created the neighborhoods without using income as a criterion. The reason is that including income in the definition of a neighborhood exaggerates tendencies that already exist. For example, if I require that everyone in Fishtown have a family income in the bottom quintile, I guarantee that Fishtown has a high percentage of single-parent homes (not all people with low incomes are single parents, but single parents disproportionately have low family incomes). If I require that everyone in Belmont have an income in the top 20 percent, I guarantee that almost every head of household is in the labor force (few households have high family income without the head of household being in the labor force).
By not using income, the people in Fishtown can include the blue-collar couple who both work and have a combined income of $90,000. The people in Belmont can include the divorced mother with a PhD on a college faculty who has a modest income because she is working only half time. Using an income criterion would have excluded both kinds of people. Some degree of artifact is unavoidable even using just education and occupation, because education and occupation are related to income. But omitting income reduces the artifact.
Trendlines running from 1960 to 2010 have to consider a major technical issue: The compositions of Belmont and Fishtown presumably changed.
The national numbers on the variables used to assign people to Belmont and Fishtown shifted radically from 1960 to 2010. Figure C.1 shows the situation with regard to education.
The proportion of prime-age whites without a high school diploma dropped from one out of two to one out of twenty-five. The proportion with a college degree grew from one out of ten to one out of three. It has to be assumed that high school dropouts in 1960 consisted of a pool that was very different from the pool of high school dropouts in 2010, and that the same is true of the pool of college graduates.
FIGURE C.1. CHANGES IN EDUCATIONAL ATTAINMENT
Source: IPUMS. Sample limited to whites ages 30–49.
FIGURE C.2. CHANGES IN THE PREVALENCE OF TWO BASIC JOB CATEGORIES
Source: IPUMS. Sample limited to whites ages 30–49 in the labor force.
The same thing happened with occupations. Figure C.2 illustrates this through two basic job categories: managerial jobs and skilled blue-collar jobs (categories 5 and 6 combined).
In 1960, 47 percent of prime-age white American workers were working at blue-collar jobs. By 2010, that proportion had been halved to 23 percent. Meanwhile, managerial jobs went from only 9 percent of the workforce to 18 percent.
The result is that the distribution of prime-age whites into the two neighborhoods also changed drastically from 1960 through 2010, as shown in figure C.3:
FIGURE C.3. THE CHANGING BALANCE OF THE TWO NEIGHBORHOODS
Source: IPUMS. Sample limited to whites ages 30–49.
In 1960, 64 percent of prime-age white Americans qualified for Fishtown, a number that had fallen to 30 percent by 2009. In 1960, only 6 percent of prime-age white Americans qualified for Belmont, a number that had risen to 21 percent by 2010.
This raises a question: Isn’t it possible that Fishtown didn’t really deteriorate at all? The hypothesis is that behavior in Fishtown changed from 1960 through 2010 because the composition of the neighborhood changed. In effect, more than half of the people in Fishtown in 1960 had moved out by 2010. Which people left Fishtown? Presumably those with the most ability to move up in the world. The cream was skimmed from Fishtown. A similar artifact might be working in Belmont, which more than tripled as a proportion of the population between 1960 and 2010. Perhaps changes in Belmont merely reflect a dilution of the quality of its population, as people who formerly wouldn’t have completed college or entered the professions moved in.
These hypotheses are likely to explain something, so we need a constant yardstick based on the proportions of people in Belmont and Fishtown as of 2010—in round numbers, 20 percent and 30 percent, respectively. For example, suppose we are looking at divorce, a trendline that begins in 1960 and ends in 2010. The questions to be asked are “What was the divorce rate for the 30 percent of the population who had the least education and were employed in the lowest-level jobs in both years?” and “What was the divorce rate for the 20 percent of the population who had the most education and were employed in the highest-level jobs in both years?”
Choosing the educational attainment measure was straightforward. I used the highest grade completed. Choosing an occupational measure that ranks occupations from “lowest-level” to “highest-level” was more complicated. The eight occupational categories listed at the beginning of this appendix are too broad. A continuous scale is required.
One option was to use one of the indexes of job prestige that have been created over the years, based on the answers that social scientists get when they ask large samples of people to say which of two occupations has more prestige in their eyes. With enough people making enough comparisons, it is possible to combine the results into a continuous scale. I used one of the best of those scales, created by Keiko Nakao and Judith Treas, for analyses conducted early in the research for this book.7 But it was ultimately unsatisfying. The orderings even in the best indexes often don’t pass the face-validity test—we don’t look at them and say, “Yes, that makes sense.” On the Nakao-Treas scale, for example, a sociology teacher has higher prestige than a judge, a math instructor has higher prestige than a chief executive, an air traffic controller has higher prestige than an electrical engineer, and a registered nurse has higher prestige than a space scientist. Actors and professional athletes—people who are idolized in our celebrity culture—have lower occupational prestige than all of the above. Any scale of occupational prestige is riddled with such examples. Robert Hauser and John Warren, creators of another major occupational prestige index, reviewed the evidence for measures of occupational prestige and concluded that the educational level required for an occupation was a more useful indicator than composite measures of occupational prestige and that “the global concept of occupational status is scientifically obsolete.”8
But sticking with the educational level required for occupations is not much help for discriminating among the people who held a large variety of blue-collar occupations in 1960. The number of years of formal K–12 education required to be a carpenter and a menial laborer are probably about the same—many stevedores and highly skilled carpenters in 1960 had identical levels of education, having dropped out of school as soon as the law allowed—but the cognitive demands of the two jobs are quite different. Of those who were carpenters and stevedores in 1960, we would expect that the proportion of carpenters who had the capacity to move into technical or white-collar occupations was higher than the proportion of stevedores who had the capacity to do so.
To use information about a person’s occupation for assigning him to the top 20 percent or bottom 30 percent, I adapted the work of psychometricians Earl Hunt and Tara Madhyastha, who used the Department of Labor’s O*NET ratings to assign cognitive requirements to the entire range of jobs.9 The O*NET database in the years used for the analysis contained ratings by the incumbents of jobs of the skills required for 801 jobs, using anchored questions. For example, for the characteristic “arm-hand steadiness,” the incumbent was asked to rate the requirements for that job on a 1–7 scale in which 2 was “light a candle” and 6 was “cut facets in a diamond.” Hunt and Madhyastha focused on twenty cognitive demands covering verbal abilities, idea generation, reasoning abilities, quantitative abilities, memory, perceptual abilities, spatial abilities, and attentiveness. A factor analysis of these twenty cognitive demands produced the generic result that has characterized factor analyses of batteries of mental measures for a century: The first factor, representing the general mental factor known as g, dominated the results. I used the factor loading for each occupation—its “g-loading,” expressed in the IQ metric, with a mean of 100 and a standard deviation of 15—as the measure of the cognitive demands of that occupation.10
Why is a measure of the cognitive demands of a job useful for discriminating among blue-collar jobs? Because it has been determined that cognitive ability affects job productivity throughout the range of jobs, from nuclear physicist to janitor.11 This doesn’t negate the importance of small-motor skills in being, say, a carpenter, but being an outstanding carpenter also requires good visual-spatial skills, which are part of what IQ tests measure, and also the problem-solving abilities that are part of what IQ tests measure.
The results using the cognitive demands of a job have a few anomalies of their own—are the cognitive demands of being a veterinarian really higher (if only slightly) than those of being a physician, as the g-loadings of the two jobs say? There are also problems produced by the way that the 1990 Census Bureau job categories are defined—directors and producers are in the same occupational category as actors, even though the skill sets required by those jobs are different. But the orderings work reasonably well, especially for the blue-collar jobs that are most important for understanding whether Fishtown was subject to a creaming effect.12
The index for ranking people from high to low on this combination of education and occupation was created as follows:
Educational attainment. Educational attainment was expressed as the standardized score for highest grade completed, based on the mean and standard deviation of whites ages 30–49 in the year in question.
Cognitive demands of an occupation. A standardized score based on the g-loadings was computed for the distribution of occupations among whites ages 30–49 who had occupations in the year in question.
The standardized scores for educational attainment and cognitive demands of the occupation were combined. For persons without an occupation, the standardized score for educational attainment was doubled. The combined scores were rank-ordered, from lowest to highest. These were the assignment rules:
Because of its large and nationally representative samples, the CPS was used as the template for determining the cutoffs for the top 20 percent and bottom 30 percent when other databases were used. That is, the means and standard deviations for educational attainment and cognitive demands of the CPS sample were applied to the data from other databases that were smaller or with less representative samples. For the GSS, with its comparatively small sample sizes, I did not use a single year in determining cutoffs. I combined the four surveys from 1972 through 1975 to use as the opening baseline and all four surveys from 2004 through 2010 as the closing baseline.
The exercise created a shadow population for Fishtown in the opening year of a data series that plausibly represents the people who would have remained in Fishtown even in 2010. Take 1960, the opening year I use whenever I can, as an example. The bottom 30 percent was an amazingly badly educated group of people by today’s standards. Fifty-nine percent of people in the bottom 30 percent in 1960 had no more than eight years of education. Only 12 percent of the bottom 30 percent had high school diplomas.
The high proportion of the bottom 30 percent with just an eighth-grade education is especially indicative of a shadow Fishtown with low levels of ability. As noted in chapter 9, everyone in their thirties and forties in 1960 had grown up at a time when children were already legally required to remain in school until age sixteen. In the ordinary course of events, children finish eighth grade when they are fourteen. A high proportion (an exact estimate is not possible) of those with no more than eight years of education had repeated a grade in elementary school or junior high, which is a strong indicator of serious learning difficulties.
The exercise produced parallel results for Belmont, drastically lowering the educational distribution. Everybody assigned to Belmont in 1960 had a college degree, compared to just 53 percent of those in the top 20 percent. Forty-one percent in the top 20 percent had no more than twelve years of education.
In almost all of the graphs in part 2, you can see from the markers for the top 20 percent and bottom 30 percent that changes in the composition of the neighborhoods make remarkably little difference. How can this be?
It is easy to see why the increase in the percentage of people qualifying for Belmont between 1960 and 2010 didn’t make much difference. In 1960, the people ages 30–49 had been of college age from the late 1930s to the early 1950s, when many people who had the ability to get college degrees were not even trying to go to college. The college sorting machine discussed in chapter 2 had not yet kicked in, and a large pool of college-qualified students was not being tapped. Thus the increased number of people who qualified for Belmont by 2010, decades after the college sorting machine had been doing its work, didn’t necessarily mean that the upper-middle class as a whole consisted of much smarter people (the effects of the sorting machine are strongest for people at the very top of the ability distribution). The case of managerial jobs makes the point. In 1960, 80 percent of people holding managerial jobs and who therefore qualified for Belmont occupationally did not have college degrees and didn’t qualify for Belmont educationally. By 2010, that proportion had dropped to 47 percent. Many of the people in Belmont in 2010 were holding the same jobs that their counterparts held in 1960; the only difference was that in 2010 they had a piece of paper saying they had been awarded a college degree. In terms of ability, the pool of people in Belmont was not necessarily diluted.
Similarly, Fishtown surely suffered some loss of talent as it went from 64 percent of the prime-age population to 30 percent, but that loss wasn’t necessarily huge. The NLSY-79, with its large nationally representative sample of whites with a good measure of IQ, helps make that point. The occupational data for the following numbers refer to the early 2000s, when all the members of the sample were in their late thirties through mid-forties.
If we look at mean IQ by occupational category, the relationship is as we would expect, as shown in table C.1.
TABLE C.1. g-LOADINGS OF JOBS AND THE MEAN IQ OF THE PEOPLE WHO HOLD THEM
Occupational category | Mean g-loading of the occupations | Mean IQ of the job holders |
High-status professions | 120 | 117 |
Managerial positions | 116 | 107 |
Mid-level white-collar occupations | 111 | 107 |
High-skill technical occupations | 107 | 109 |
The blue-collar professions | 109 | 100 |
Low-level white-collar occupations | 92 | 103 |
Other skilled blue-collar occupations | 89 | 98 |
Low-skill service and blue-collar occupations | 83 | 94 |
Source: NLSY-79. Sample limited to whites.
The ordering of the g-loadings is about what one would expect, with the high-status professions on top, the low-skill service and blue-collar jobs at the bottom, and others spaced with modest differences. The similarity of the requirements for the mid-level white-collar jobs, high-skill technical jobs, and the blue-collar professions also makes sense. Intuitively, there is no reason to think that you need to be smarter to be a paramedic than to be an electrician, nor that there should be a difference between them and people holding down mid-level jobs in an office.
The ordering of the mean IQ of whites in the NLSY-79 who actually held those jobs generally follows the same order, but with much more bunching. The people who held managerial positions, mid-level white-collar jobs, and high-skill technical jobs were all about the same. In part, this probably reflects measurement error—people who actually hold mid-level white-collar jobs can easily give their job a description that leads the interviewer to code it as a managerial job. In part, it reflects the aggregation of different kinds of jobs. Except for “chief executives and public administrators,” the 1990 occupational categories for managers do not discriminate between senior managers and junior ones, and no one in the NLSY-79 sample had become a chief executive. The fifty-one “accountants and auditors” had a mean of 113, suggesting that, not surprisingly, jobs have an IQ gradient within the managerial category.
Despite these problems, the important pair of points from table C.1 are that (1) yes, occupational sorting by IQ exists, but (2) it is very far from perfect. While a higher percentage of carpenters than stevedores have the capacity to become paramedics, as I wrote a few pages ago, table C.1 indicates that the difference in those percentages are modest.
Thus part of the explanation for the generally small differences in the results using the Belmont-Fishtown method and the Top 20 Percent–Bottom 30 Percent method is that there was a great deal of slack in the sorting of people by SES in 1960, and that many of the people who moved out of Fishtown between then and 2010 moved into jobs that were no more demanding than the jobs they had left. But that is unlikely to be the whole story. It remains remarkable that even when we limit the sample in 1960 to people who not only qualified for Fishtown but were in the lower half of Fishtown with regard to both their education and the cognitive demands of the jobs they held, their records on marriage, employment, crime, and religiosity were about the same as those in the rest of Fishtown. The result suggests that powerful norms of social and economic behavior in 1960 swept virtually everyone into their embrace.
The Current Population Survey sample is so large that restricting the analyses to whites ages 30–49 poses no problems with sample sizes, but the same cannot be said of the General Social Survey. The Belmont samples for individual survey years had a median of only 81, and dipped as low as 48. For Fishtown, the comparable figures were 216 and 143. I therefore originally conducted the GSS analyses with a broader age group that included everyone from ages 25 to 64. This had the effect of expanding the Belmont and Fishtown samples for individual survey years to medians of 122 and 373, respectively, but I discovered that the results were virtually identical to analyses restricted to ages 30–49. Table C.2 illustrates, showing the beginning values for Belmont (combining results from the 1972 to 1976 surveys), the ending values (combining results from the 2006 to 2010 surveys), and the difference between the two.
TABLE C.2. GSS RESULTS FOR BELMONT USING AGES 30–49 COMPARED TO AGES 25-64
The results for Fishtown, with its larger sample sizes, were even closer. Because the results for ages 30–49 were so similar, I decided to maintain consistency in the presentation, using the 30–49 age range for the GSS as I did for the CPS.