(Focus Volume 17:2, Fall/Winter 1995)

The Bell Curve: A perspective from sociology

Sociologists Robert Hauser and Wendy Y. Carter also direct
attention to failures of empirical analysis by the authors of The
Bell Curve, in a paper prepared for the August 1995 meetings of
the American Sociological Association.[1] Below is a summation of
their analysis of some analytic flaws in the treatment of the
relationships between cognitive ability and occupation,
socioeconomic status, and education, respectively.

Ability and occupation

Herrnstein and Murray argue that, in the twentieth century,
ability has come to play an increasingly large role in
occupational and economic success. That, in turn, has tended to
segregate people in the higher reaches of the occupational
distribution. The consequent social isolation has led to elitism.
Socializing and intermarrying within a restricted circle, members
of the cognitive elite pass on both ability and status to their
children. This argument, according to Hauser and Carter, combines
real data about the growth of knowledge-based occupations with
speculation about the constancy of measured ability among the
incumbents of those elite occupations. It also assumes that the
range of ability varies inversely with the social standing of
occupations. If one is to become a brain surgeon, one must fall
above a very high ability threshold, and hence within a narrow
range, whereas persons of any ability can collect garbage or
clear tables. 

This is a standard finding in differential psychology, usually
supported by test score distributions for the many thousands of
white enlisted men who took the Army General Classification Test
(AGCT) during World War II. In these data, the median test scores
decline regularly with social standing, whereas the range of test
scores varies inversely with social standing (see Figure 1).
Thus, high-status occupations appear to draw on a much narrower
range of the general ability distribution than do low-status
occupations. But the finding is undoubtedly wrong. The Army data
show this pattern because men were assigned to the enlisted ranks
partly on the basis of measured ability--the higher the average
ability in the civilian occupation, the less likely were
high-scoring members of that occupation to enter the enlisted
ranks. 

Better data exist. Hauser and Carter examined test score
dispersions within occupations relative to their socioeconomic
status, taking their data from four sources: results from the
General Aptitude Test Battery of the U.S. Employment Service; the
Armed Forces Qualification Test results for men from the early
years of the National Longitudinal Survey of Youth; Henmon-Nelson
IQ scores among high school graduates in the Wisconsin
Longitudinal Study; and a short verbal test given to a sampling
of the general U.S. population in the General Social Survey
(GSS), conducted by the National Opinion Research Center. Results
from these multiple populations and tests provide scant evidence
of a tendency toward a narrower range of ability within
occupational groups where measured ability is high. 

To test the proposition that the relationship between high
ability and incumbency in cognitively demanding occupations has
increased over time, Hauser and Carter looked at data for 13,000
persons in the GSS from 1974 to 1993, years when the 10-item
verbal ability test was administered. About 6 percent of employed
persons aged 25-64 answered all ten questions correctly. The
authors classified occupations into those that represented the
cognitive elite, as defined by Herrnstein and Murray, versus all
other occupations, and calculated the chances that persons in
elite occupations answered ten questions correctly relative to
the chances that persons in nonelite occupations answered ten
questions correctly.[2] Over the twenty years covered by the GSS,
cognitive performance has been stable in nonelite occupations. In
elite occupations, it is higher than in nonelite occupations, but
appears to have declined slightly. The finding is unchanged if
minorities are excluded from the sample, and still unchanged by a
less stringent criterion of cognitive performance--getting nine
out of ten words right. Thus the GSS data not only fail to
support the Hernnstein-Murray thesis, but tend to contradict it:
if anything, there has been a slight decline in the difference in
verbal ability between persons in elite and those in nonelite
occupations.


Ability and socioeconomic status

Much empirical analysis in The Bell Curve is based upon two data
sets, the National Longitudinal Survey of Youth (a large sample
of American youth, aged 14-22 in 1979, who have been followed
annually since then) and the Children of the NLSY, which matches
women in the NLSY with their children. Both data sets contain
good measures of cognitive ability, but, say Hauser and Carter,
are used poorly by Herrnstein and Murray. Most of the original
analysis in the book consists of graphical displays of
reduced-form logistic or linear regression equations in which
some measure of educational or socioeconomic attainment, contact
with the criminal justice system, or child-rearing success has
been regressed on two variables, AFQT score in the IQ metric,
adjusted for age at administration, and a composite measure of
the socioeconomic status (SES) of the family of orientation. This
measure is limited in content to father's and mother's
educational attainments, father's occupational status, and family
income in 1979, the first year of the NLSY. This is a minimally
adequate specification, but it tends to understate the effects of
social background by omitting such variables as number of
siblings, intact family, rural or metropolitan origin, and
regional origin. Thus, in Herrnstein and Murray's analysis, the
social background variable becomes a straw man, largely used to
highlight the effects of ability. From the study of
stratification,  it is known that the explanatory power of
measured social background is modest, but it is also known that
the effects are important and worth understanding. No measures of
the explanatory power of the equations are reported in The Bell
Curve, so that the inexpert reader never learns that most of the
variation remains unexplained. 

Ability and education

Hauser and Carter note the absence of explicit or sustained
discussion of the main effects of schooling and their
relationship to ability. Indeed, Herrnstein and Murray
ingenuously claim that the relationship between ability and years
of schooling is too close, statistically, to untangle and that
the relationship is, in any event, beyond the scope of their
analysis. Only twice does educational attainment appear
explicitly in an equation, both times in the appendices.[3]
Herrnstein and Murray make much of the fact that African
Americans obtain more schooling than whites and--in some 
cases--have better jobs, once ability is controlled (this consequence
they attribute to affirmative action policies). Their analyses,
however, ignore the persistently high unemployment of African
Americans. Neither do they report black-white differentials in
schooling when socioeconomic background alone is controlled. Yet
such controls have for many years accounted for black-white
differentials in schooling, before the era of affirmative action
and in situations where affirmative action is irrelevant. For
example, Hauser and Phang, using October CPS data, have shown
that white high school dropout rates substantially exceeded those
of African Americans in the 1970s.[4] These differentials
probably have their roots both in labor market discrimination,
which decreases the opportunity cost, for African Americans, of
remaining in school relative to leaving it, as well as in the
continuing belief of African Americans in the value of schooling. 

The meager discussion of ability in relation to attainment (years
of schooling) in The Bell Curve is mirrored in a distorted
analysis of change in achievement (test scores). Yet the evidence
is clear: differences in academic achievement between African
Americans and majority white students have narrowed in the last
twenty years. The trends for proficiency scores reported by the
National Assessment of Educational Progress, depicted in Figure
2, show a consistent convergence from 1970 through the late
1980s, whether in science, math, or reading. That convergence is
acknowledged by Herrnstein and Murray, but its size is
misestimated and its importance discounted in their
conclusions.[5]

Schools are the major instrument of public policy affecting the
functional competence of adults in the United States. The
exclusion or misrepresentation of their effects in Herrnstein and
Murray's analysis is yet another instance of the flawed and
selective nature of their evidence. #

---------------------------------------------------------------

1. Robert M. Hauser and Wendy Y. Carter, "The Bell Curve as a
Study of Social Stratification," paper presented at the annual
meeting of the American Sociological Association, August 1995.

2. The elite occupations identified by Herrnstein and Murray are:
accountants, architects, chemists, college teachers, computer
scientists, dentists, engineers, lawyers, mathematicians, natural
scientists, physicians, and social scientists (The Bell Curve, p.
56). "Nonelite" in the text refers to all other occupations.

3. The Bell Curve, pp. 590, 652. For further light on black-white
differentials in educational attainment, see Kurt Bauman's
dissertation, "Family Background and Racial Differences in
Educational Attainment: Explaining Black Net Educational
Advantage," UW-Madison Department of Sociology, 1995. 

4. Robert M. Hauser and Hanam Samuel Phang, "Trends in High
School Dropout among White, Black, and Hispanic Youth,
1973-1989," IRP Discussion Paper no. 1007-93.

5. See, e.g., p. 293 of The Bell Curve. For a full discussion of
the treatment of the NAEP trends in The Bell Curve, see Hauser
and Carter's paper, pp. 17-21. 

Figure 1 is available on the gopher as "hausf1.eps", and can be
used with a PostScript previewer or a PostScript printer.  The
full caption for that figure is:

Figure 1.  Army General Classification Test (AGCT) standard score
distribution, for selected occupations with 50 cases or more.  
Percentiles 10, 25, 50, 75, and 90 are marked.

Source: Based on white enlisted men in War Department machine
records survey taken June 30, 1944 (Naomi Stewart, "A.G.C.T. 
Scores of Army Personnel Grouped by Occupation." 
Occupations 26:18-19)

Figure 2 is available on the gopher as "hausf2.eps", and can be
used with a PostScript previewer or a PostScript printer.  The
full caption for that figure is:

Figure 2. National Assessment of Educational Progress trend 
assessments for reading, math, and science, among African-American
and white students at age 13.  Black line = African-American 
students, gray line = white students.

Source: Ina V.S. Mullis, John A. Dossey, Jay R. Campbell, and others,
NAEP 1992 Trends in Academic Progress, National Center for 
Education Statistics Report No. 23-TR01 (Washington, D.C.: U.S.
Government Printing Office, July 1994)