Professional Documents
Culture Documents
Sociological Forum [sofo] pp1028-sofo-475161 October 22, 2003 19:0 Style file version June 4th, 2002
There is an old Zen koan in which the master Shuzan Osho held up
his staff before his disciples and said, “You monks! If you call this a staff,
you oppose its reality. If you do not call it a staff, you ignore the fact. Tell
me, you monks, what will you call it?” The discomfort felt by the monks,
who had to choose between denying their insight into the fundamental one-
ness of the universe and making the absurd counterfactual denial of self-
evident fact, is also felt by many sociologists when it comes to the analysis
of race. Race, as Weber (1978:385–393) argued, is a matter of conscious-
ness, not biology; sociological analyses of the historical and social develop-
ment of racial categories make it difficult for sociologists to employ existing
racial categories as if they were unproblematic, biological terms. Yet even if
sociologists explicitly dissent from simplistic understandings of race, the re-
peated use of the conceptual category of race has the potential to restrict
1 Department of Sociology, Rutgers—The State University of New Jersey, 54 Joyce Kilmer
Avenue, Piscataway, New Jersey 08854-8045.
2 To whom correspondence should be addressed; e-mail: jlmartin@rci.rutgers.edu.
521
0884-8971/03/1200-0521/0 °
C 2003 Plenum Publishing Corporation
P1: GXB
Sociological Forum [sofo] pp1028-sofo-475161 October 22, 2003 19:0 Style file version June 4th, 2002
3 Littlefield et al. (1982) found a decrease in the proportion of anthropology textbooks affirming
the fundamental reality of race.
4 Niemonen (1997:16, 35) is somewhat unclear here as to the boundaries of his study—he first
says that he compiled a database of 677 papers “on racial and ethnic relations,” but later says
that of these, only 55 “could be placed within the context of a sociology of racial and ethnic
relations.” Since not all papers that employ race appear in his sample, our interpretation
is that he picked papers that he believed had substantially to do with race, but considered
few of these to be primarily oriented toward contributing to a sociology of racial and ethnic
relations.
P1: GXB
Sociological Forum [sofo] pp1028-sofo-475161 October 22, 2003 19:0 Style file version June 4th, 2002
5 Reynolds (1992) has demonstrated that among sociologists there are predictable social corre-
lates of whether one sees race as a social construct as opposed to an intrinsic category. Thus
women fall strongly into the former category, while men are more evenly divided. This pattern
also holds among anthropologists, where “overdogs” (those with dominant status) are more
likely to emphasize racial differences than “underdogs” (Lieberman and Reynolds, 1978; Stark
et al., 1979). Among anthropologists, however, these differences are strongly mediated by one’s
training (physical vs. cultural anthropology). There was also a study of papers in the subfield
of sociology of occupations (Smigel, 1954) that found a decreasing attention to race from the
late 1940s to the early 1950s.
P1: GXB
Sociological Forum [sofo] pp1028-sofo-475161 October 22, 2003 19:0 Style file version June 4th, 2002
6 It is important to distinguish this from the number of papers about race. Simpson (1961) found
that the number of papers in the ASR about race or ethnicity decreased from 1945 to 1959,
but during this same time, we find the number of papers mentioning race increased.
P1: GXB
Sociological Forum [sofo] pp1028-sofo-475161 October 22, 2003 19:0 Style file version June 4th, 2002
First of all, we have been looking at what we have generously (by cur-
rent standards) called “empirically minded” papers. Many papers in the first
period did not have a clearly identifiable source of data, but were forceful
statements of opinion peppered with references to facts collected by others
or to things generally believed to be true by educated persons of that time.
These papers may be unlikely to be coded as “analyzing” race, if only be-
cause they do not really analyze anything. This would inflate the apparent
increase in number of papers analyzing race, as there has been a steady in-
crease in the proportion of all papers that are based on data analysis, from
42% in the first period to 84% in the last.
Second, it may be irrelevant to include cases where the data to be an-
alyzed come from a country other than the United States, where either the
emphasis on, or presence of, racial divisions is absent. Third, it may be that
numerical methods make it easier to measure and analyze race. If there were
changes in the proportion of studies looking at data from the United States,
or using numerical methods, this would also affect our temporal compar-
isons. Further, since there has been a change in the type of data sets used
(in particular, fewer U.S., White-only, individual-level data sets), we might
expect changes if only because some types of data make it easier to analyze
race.
If we then restrict our attention to those papers that analyzed U.S. sam-
ples, including individual-level data from both Whites and Blacks—those
that could actually analyze race—we find that the increase in the propor-
tion actually carrying out such analyses is quite extraordinary (see Table II).
While three fourths of the papers in the first period using such data would
fail to bring up race (quite possibly because the data came in tabular form,
and hence authors were unable to analyze it further), only one quarter do
not mention race in the last period.
Thus we see an increase both in the proportion of papers that are at
risk for examining race because they contain individual-level U.S. data from
more than one race, and in the proportion of those papers at risk that actually
Table II. Mention of Race by Period, Numerical Analyses of U.S. Data, Including
Blacks and Whites Only
Period Analyzed Race-specific Other mention No mention Total
1937–40 15.9% 4.5% 6.8% 72.7% 44
1952–55 19.6% 1.0% 11.3% 68.0% 97
1967–70 19.4% 13.9% 66.7% 72
1982–85 51.6% 1.1% 6.5% 40.9% 93
1996–99 74.6% 1.4% 23.9% 71
Total 37.4% 1.1% 8.2% 53.3% 377
Note. (χ)2 = 89.9 at 12 df ( p < 0.001).
P1: GXB
Sociological Forum [sofo] pp1028-sofo-475161 October 22, 2003 19:0 Style file version June 4th, 2002
do analyze race. However, it could also be that there has simply been a
shift in attention such that certain problems more likely to be associated
with race occupy a position of greater importance, explaining the seeming
increase in attention to race. We coded every article in terms of the main
disciplinary area into which it fell; each article was coded on two areas if (as
was frequently the case) an article straddled areas. (See the Appendix for
information on this coding and others.) Of course, the meanings of certain
topics have shifted over the course of the century—for example, comparisons
of international development that were couched in evolutionary theory in
1937 were more likely to be couched in world-systems theory in 1999—but
our analysis allows for such change.
To examine whether any topic was associated with the treatment of
race, we entered each in a multinomial regression with treatment of race
as the independent variable. (Since each article was coded 1 as opposed
to 0 for at most two topics, we could examine one topic at a time without
danger of bias; we were unable to identify parameters when examining all
topics together.) We then controlled for the changing composition of the
ASR in terms of topics of interest by entering these terms (and interac-
tions with period when necessary) into a multinomial logistic regression,
with our four-category division used in Tables I and II as the dependent
variable. Thus Table II can be restated as the parameters resulting from
such a regression on “period” (considered a categorical variable). Figure 1
presents the results (here we look at all analyses of U.S. data, not only those
with Whites and Blacks in the sample, so as to facilitate an examination of
changes in the proportion of race-specific studies; this only decreases the
size of the change in the proportion of studies analyzing race.)7 Each line
gives the logarithm of the relative odds that an article analyzes race, includes
a race-specific question, or mentions race in some other fashion, all relative
to not mentioning race (the excluded category). The results can be inter-
preted as follows: the odds of analyzing race have increased from period 3
to period 4, and then again from period 4 to period 5 (relative to not men-
tioning race), while the odds of collecting data from a race-specific sample
have basically declined, although there was a small increase from period 3 to
period 4.8
Figure 2 presents the same parameters when we take the substantive
composition of papers in each period into account by subtracting the effect
of each of the topics found to be related to race (and allowing this effect
7 The −2LL for this model was 66.6 in comparison to a model with only an intercept having
a −2LL of 143.1, leading to a chi-square test of fit of 76.5 at 12 df, p < .001, indicating the
significant effect of period on treatment of race.
8 This increase appears in Table II not as an increase in the number of race-specific papers, but
as a decrease in the number of papers in which race was not mentioned.
P1: GXB
Sociological Forum [sofo] pp1028-sofo-475161 October 22, 2003 19:0 Style file version June 4th, 2002
9 Thismodel fits with −2LL = 570.1 in comparison to a model with only the intercept −2LL
= 1020.7; this chi-square difference of 450.6 at 138 df was significant: p < .001. If we nest the
former model in this one by ignoring all factors other than period, we get −2LL = 944.1; since
126 additional parameters were identified in the former case, we have a test of a chi-square
of 374.0 at 126 df, p < .001. (Thus the contribution of the coefficients for the topic areas are
jointly important.) Because most of the parameters for these models could not be identified
because of the many a priori zeros in the data matrix, we do not present them, but only
present the crucial (and identifiable) period parameters. While the use of parameters from
incompletely identified models is suspect, this is the single most rigorous way of demonstrating
that our results do not depend on changing composition; analyses of particular fields support
our interpretation.
P1: GXB
Sociological Forum [sofo] pp1028-sofo-475161 October 22, 2003 19:0 Style file version June 4th, 2002
data-analytic papers into three categories. The first included what we call
“informal” analyses of qualitative data, such as might be based on historical,
ethnographic, or in-depth interview data. If some formal analysis was applied
to such qualitative data (e.g., Ragin’s procedure Ragin, 1987), we placed the
article in the second category, “formal” approaches, which also included sta-
tistical models. (Early formal approaches tended to involve mathematically
derived models for diffusion or other specific social processes.) Analyses
that were mainly numerical but simply involved the presentation of “raw”
numbers in tables or unprocessed correlation coefficients were placed in our
third category, “numerical” but not formal.
While qualitative work specifically focused on race might present the
most complex treatment of race, qualitative analyses of other topics tended
not to “take race into account” other than by selecting a race-specific sample;
we accordingly turn our attention to the quantitative analyses. If we look at
only papers using U.S. data, we find that the proportion of numerical or
formal papers analyzing race increased markedly over the last three periods
(see Fig. 3). There were only four formal analyses of U.S. data in the first
period, one of which analyzed race; hence, the 25% for the first period should
not be given too much weight. After this, however, we see a pattern of
monotonic increase: 11% of the formal papers analyzed race in the second
period, 13% in the third, 35% in the fourth, and 54% in the fifth.
Yet the “formal” category includes a number of approaches. Some ap-
proaches (for example, loglinear modeling) can make it difficult to incorpo-
rate additional variables into the model; adding a dichotomy for race doubles
P1: GXB
Sociological Forum [sofo] pp1028-sofo-475161 October 22, 2003 19:0 Style file version June 4th, 2002
Fig. 3. Proportion of papers using U.S. data analyzing race by method (N = 136).
Note. (χ)2 = 64.4 at 2 df ( p = 0.002) for effect of methods on proportion analyz-
ing; χ 2 = 91.3 at 4 df ( p < 0.001) for effect of period on proportion analyzing; both
from multinomial regression with analysis as dependent variable and independent
variables treated categorically; chi-square compares models with and without inde-
pendent variable in question.
the number of cells and makes it less likely that the statistics used to deter-
mine fit will be applicable. Other formal approaches, most notably those
based on linear models with continuous covariates, allow one to take addi-
tional factors into account at a minor loss of 1 degree of freedom per covari-
ate. We call all such models that make a single dependent variable a function
of a set of other variables including continuous covariates “regression-type
models,” or “regression analyses” for short. Such regression analyses, to be
blunt, make it easy to take race into account, and they are strongly corre-
lated with the analysis of race. Indeed, they are so strongly associated with
the analysis of race that the period effect becomes insignificant when the use
of regression models is taken into account.
But of course the use of such models is itself a period effect, and so
examining the relation between these models and race analysis restricts us to
the last three of our five periods. Given this curtailed sample, and restricting
attention to analyses of U.S. data that includes Blacks and Whites, we find
that regression analyses were no more likely to include race as a variable
than were other quantitative analyses in the third period (21% as opposed to
20%, respectively). But in the fourth period, 62% of the regression analyses
took race into account, as opposed to 25% of the other quantitative analyses,
and in the last period, an overwhelming 78% of regression analyses used
P1: GXB
Sociological Forum [sofo] pp1028-sofo-475161 October 22, 2003 19:0 Style file version June 4th, 2002
race, as opposed to 25% of the other quantitative methods. This does not
mean that formal methods in general have been increasingly likely to lead
to the analysis of race. If we look at papers that use formal analyses other
than regression-type models, we find the proportion of papers that take race
into account shrinks precipitously, from 20% in the third period to 14% in
the fourth period, and in the last period we find none. But this decreasing
tendency of nonregression formal analyses to take race into account makes
little difference for the overall trend, as such models were rapidly becoming
extinct. While there were 26 such models used in the third period, their
number decreased to 14 in the fourth period, and there was only 1 formal
nonregression model of U.S. data on Blacks and Whites in the last period.
In sum, the hegemony of regression-type models that make it easy to add
covariates corresponds closely to the increasing attention to race in sociology,
and this methodological change better explains the increase in attention than
do shifts in substantive concern. Yet it is not that the increase in regression
models led to an increase in the analysis of race simply because the new
technique made it easier to do so. That is, there is no reason to assume that
there was a constant propensity to include race. While regression analysis is
indeed associated with analyzing race, the degree of this association increases
with time. The decreased costs of taking race into account may have helped
legitimate a standard procedure in which race tended to be examined, thus
increasing the general propensity to take race into account even further.
ANALYTIC APPROACHES
Given that race is analyzed, precisely how it is taken into account may
have serious, if subtle, implications for the working theory of race held by
sociologists. To simply add race as a control variable in a regression model
(or to do the equivalent with simpler numerical methods) implies that, while
race makes a difference, it is not a very profound one, in that race does
not affect the relationships between other variables. For example, if one is
looking at the relationship between education and socioeconomic status, one
may “take race into account” by adding it as a control to better estimate what
we may call a “deracialized” coefficient for education on status. This is in
effect to say that, while there are compositional differences between Blacks
and Whites, on average, in terms of their education and socioeconomic status,
the basic process of status attainment is the same.
But it is also possible for the researcher to add an interaction, allowing
the coefficient of education to vary by race, which is in effect to say that
the fundamental processes of status attainment among Whites and Blacks
are somewhat different. This seems to assert a greater degree of difference
P1: GXB
Sociological Forum [sofo] pp1028-sofo-475161 October 22, 2003 19:0 Style file version June 4th, 2002
between Whites and Blacks (and other races, should these be taken into
account), and here race has a qualitatively privileged position over the other
independent variables. If interactions with race are added for all coefficients,
this is equivalent to splitting the sample by race and conducting parallel
analyses. Despite the mathematical equivalence of these approaches, there
is still an important difference between the use of interaction coefficients
and separate analyses. The former facilitates the statistical test of significant
difference between the Black and White coefficients, while the latter facili-
tates the examination of the effects in each subgroup considered separately.
Thus the choice of mode of presentation seems to suggest a preexisting sense
of how great the racial differences will be: The former suggests that one is
not already convinced that racial differences are important, while the latter
suggests that the basic issue of racial difference is settled. Finally, an even
more radical assertion of difference would follow if different variables were
used to predict the same outcomes among Whites and Blacks.
We coded every numerical article analyzing race according to how race
was treated—as a control, as a crucial independent variable (when race itself
was of interest), whether interactions between race and other independent
variables were included, or whether separate analyses were presented for
different races. (No analyses ever introduced wholly different models for
Whites and Blacks.) As Table III shows (looking only at papers using U.S.
data from both Blacks and Whites), the increasing use of regression-type
methods allowed analysts to increasingly enter race as a control variable, as
opposed to dividing the sample by race. Indeed, the proportion of papers
dividing the sample by race decreased dramatically. While there has been an
increasing use of interaction terms in the last two periods, this is still a rela-
tively rare strategy, and does not make up for the decrease in the percentage
of analyses consisting of separate treatment of Whites and Blacks.
In sum, we seem to see a widespread normalization of race—if one is
analyzing the United States, one is probably going to collect data on all races,
and if one does so, one is very likely to use a regression model and enter
Table III. Use of Race by Period, U.S. Analyses With Black and White Data Only
Simple Crucial Control Crucial Separate
Period control term interaction interaction interaction by race Total
1937–40 42.9% 14.3% 42.9% 7
1952–55 5.3% 21.1% 73.7% 19
1967–70 28.6% 71.4% 14
1982–85 26.5% 36.7% 2.0% 4.1% 30.6% 49
1996–99 46.9% 24.5% 4.1% 8.2% 16.3% 49
Total 26.8% 29.7% 2.2% 5.1% 36.2% 138
Note. (χ)2 = 42.1 at 16 df ( p < 0.001).
P1: GXB
Sociological Forum [sofo] pp1028-sofo-475161 October 22, 2003 19:0 Style file version June 4th, 2002
race as a control variable. Thus the pattern here differs dramatically from
epidemiology, where an increasing focus on race coincided with a decreasing
propensity to collect data on nonwhites (Jones et al., 1991:1081f.). While
race is generally seen as irrelevant in some fields (e.g., science) and is widely
considered relevant in others (e.g., demography), there has been an across-
the-board increase in the propensity to “control” for race. As far as we can
impute a prototheory to most sociologists on the basis of the results, we might
say that there is a widespread presumption that race makes a difference,
and the difference it makes is “broad but shallow.” Racial differences are
expected in many substantive areas, and these differences are not assumed to
be orthogonal to other causes that are of interest (and hence simply ignored).
But these racial differences are only differences in degree, not in kind, and
they can more or less be compensated for by controls (as opposed to parallel
analyses or even different sets of variables). Entering race as a control with
an effect, as Zuberi (2001) stresses, may then imply that it is a “cause” of the
phenomenon in question, and not itself a phenomenon to be explained.
CONCLUSION
10 Contra the proposal to replace “race” with “ethnicity,” Bonilla-Silva argues that the historical
constructions of the two concepts are different, the former in the colonial encounters and the
latter in the process of the formation of nation states, and that “the conceptual elimination of
race and the utilization of ethnicity as the mantra for interpreting ethnic, racial, and national
phenomena are usually associated with the unwillingness of members of the dominant race
‘to accept responsibility for the problem of racism.”
P1: GXB
Sociological Forum [sofo] pp1028-sofo-475161 October 22, 2003 19:0 Style file version June 4th, 2002
11 And this increased use of race in a causal model, despite the point stressed by Zuberi (2001:95,
129; cf. 2000) that since race cannot be experimentally altered, it cannot be considered a
“cause” that has an “effect” as opposed to a difference to be explained.
12 Further, we do not see a decreasing attention to race in the fourth period, the heart of the
Reagan years; we doubt any simple mapping between general social trends and sociological
practice would be at all revealing. Orr (1997:33–38) argues on principled theoretical grounds
that dominant as opposed to subordinate classes will always de-emphasize the starkness or
conflictual nature of divisions between rulers and ruled; he then applies this idea to changes in
the treatment of race in sociology. We find this general sociology of knowledge quite doubtful:
One can think of any number of cases where dominant groups deliberately foment conflict
with subaltern ones, and adopt a general philosophy accordingly.
P1: GXB
Sociological Forum [sofo] pp1028-sofo-475161 October 22, 2003 19:0 Style file version June 4th, 2002
For the case at hand, sociologists had a rather good tool for partialing
out the variance associated with different independent variables and hence
avoiding spurious causal claims, which increased their tendency to “take
race into account.” The shared belief that taking race into account is the
standard thing to do then may have reinforced this practice, since scientific
publications are largely preemptive responses to imagined critiques (see
Latour, 1987). Indeed, there is reason to suspect that the threshold at which
researchers decide to take race into account has dropped. We coded each
article, as noted earlier, regarding the degree to which the topic in question
implied attention to race; we have collapsed this into three categories: race
the focus or central subject; race peripheral but potentially related; and race
not clearly related or irrelevant.13 While there is room for disagreement in
such a coding, if there is any change in the direction of nonrandom error
over time, it would seem likely that we would understate the relevance of
race in earlier periods, not having the same sense as to what themes naturally
“hung together,” which would lead to an apparent increase over time in the
association between the relevance of race and the likelihood of analyzing
it. But quite the contrary, we find that the recent periods show a dramatic
increase in the proportion of cases in which race was not really central to
the question at hand but was taken into account anyway, perhaps simply
because it was there (see Fig. 4, which presents data for analyses of U.S. data
on Whites and Blacks).
To make this more concrete, consider the subfield of the family. While
only 0–33% of papers in this field analyzed race in the first three periods,
80% did so in the last two periods. This increase has to do largely with the
integration of data sets—as more data sets included Whites and Blacks (from
52% in the first period to 88% in the final), there was a greater tendency
for analysts to take race into account. Table IV demonstrates that race has
been increasingly treated as a control in family studies, especially in the
most recent period, which includes an unusually large number of analyses
that separated Whites and Blacks.
Studies of the family might well encapsulate the changes we have seen
in sociology as a whole. The questions with which this research field is
concerned—marriage formation and dissolution, for example—are gener-
ally not intrinsically related to race. At least they are not obviously more
closely related to race than they are to other characteristics of individuals
such as religion, size of the city they live in, or region, though these are less
widely employed as generic controls. Yet there may indeed be differences of
13 While this is a somewhat subjective matter, the correlation between both coders’ initial ratings
was .763 before collapsing, which is not bad for a 5-point scale. This would (after collapsing)
lead to agreement on 75.4% of the cases.
P1: GXB
Sociological Forum [sofo] pp1028-sofo-475161 October 22, 2003 19:0 Style file version June 4th, 2002
Fig. 4. Percentage of papers analyzing race by the centrality of race to the ques-
tion, U.S. data from blacks and whites only (N = 337). Note. (χ )2 = 52.0 at 2 df
( p < 0.001) for effect of degree of relation of race to question on proportion ana-
lyzing; (χ)2 = 77.9 at 4 df ( p < 0.001) for effect of period on proportion analyzing;
both from multinomial regression with analysis as dependent variable and indepen-
dent variables treated categorically; chi-square compares models with and without
independent variable in question.
degree between Whites and Blacks, and in the clear majority of cases, when
race is taken into account, it is statistically significant (this was true in 11
of 14 cases of family research employing race as a control, a higher rate of
significance than in nonfamily research, where the figure is 37 out of 55).
Taking race into account in the sociology of the family is certainly de-
fensible according to statistical criteria, yet the dramatic increase in the
proportion of papers that do so provides food for thought. Are we simply
realizing that previous findings were incorrect because of a failure to ad-
just for differences between Blacks and Whites (and perhaps Hispanics and
Table IV. Use of Race by Period, U.S. Analyses With Black and White Data, Family and
Marriage Analyses Only
Simple Crucial Control Crucial Separate
Period control term interaction interaction interaction by race Total
1937–40 50.0% 50.0% 2
1952–55 10.0% 10.0% 80.0% 10
1967–70 0
1982–85 37.5% 37.5% 12.5% 12.5% 8
1996–99 58.3% 16.7% 25.0% 12
Total 34.4% 21.9% 3.1% 0% 40.6% 32
Note. (χ)2 = 16.0 at 9 df ( p < 0.067).
P1: GXB
Sociological Forum [sofo] pp1028-sofo-475161 October 22, 2003 19:0 Style file version June 4th, 2002
Asians)? Are we taking race into account even though not doing so would
leave our conclusions unaltered? Or are we emphasizing those aspects of
family life that are most intrinsically racialized, and ignoring those that do
not distinguish between Whites and Blacks?
Whatever the reason, in some sense, sociology has become more “racial-
ized,” as analysts are more likely to take race into account whatever the ques-
tion they ask. And yet the implicit theory of race that guides this approach
emphasizes that racial differences are not so great that analyses must be car-
ried out separately, or interaction terms added. We called this the “broad but
shallow” understanding of race (cf. Hartmann et al., in press). Race is just
one of many “variables” that affects whatever is of interest, and data can be
“deracialized” without substantially changing the character of whatever is of
interest. Further, we see a greater tendency in the last period to see race as
something beyond the White/Black dichotomy, as well as other ways of cod-
ing race which tend not to emphasize the normality of whites.14 If we were
simply to extrapolate recent trends indefinitely, we might expect a sociology
consisting largely of regression-type analyses with a set of dummy variables
for an ever-increasing number of racial–ethnic categories that would join the
usual suspects for sociological investigation.
Final Remarks
In sum, we hypothesize that the “broad but shallow” conception has be-
come dominant because of the combination of three elements: first, the im-
portance of race in American lived experience; second, the ease with which
controls may be added in regression models; third, the nature of the selective
mechanisms (gatekeeping such as peer review) in the social sciences, which
leads researchers to neutralize as many possible alternative explanations as
possible. This led to an implicit understanding that the goal of sociological
research in a racialized society is to “deracialize” its findings. That is, main-
stream sociologists seem to assume that most sociological phenomena come
to us in a racialized form, and that it is our task to take this into account
and present a final version that reflects how various causal processes would
appear were there no racial differences.
14 Technically, we may say that most schemes treated Blacks as “marked,” a term from struc-
turalist linguistics that denotes which side of an opposition (originally, between phonemes) is
actively highlighted; this term has been applied to social identities such as race (e.g., Brekhus,
1996:528; Williams, 1997:13). In this case, the “marked” identity in an opposition such as
Black/White is one that brings the fact of the opposition to the fore, while the “unmarked”
identity—the one assumed to be present in the absence of other information, the one that is
somehow seen as “normal”—is not, in the dominant culture, felt to be intrinsically related to
this opposition.
P1: GXB
Sociological Forum [sofo] pp1028-sofo-475161 October 22, 2003 19:0 Style file version June 4th, 2002
general, race should be treated. Indeed, there may not be any general answer
independent of specific research questions and social contexts. Our research
cannot imply the superiority of any particular approach, but it does strongly
suggest that our choice should now be a deliberate and reflexive one, not an
unintended consequence of methodological innovations.
ACKNOWLEDGMENTS
APPENDIX: ON CODING
We coded every article and research report, with two exceptions. First,
we did not code the ASA president’s yearly speech when this was printed.
Second, we did not code the equivalent of research reports from the Volume
17–20 period, although they existed, because they tended to be less formal
than the later reports; they fell into the category “Notes on Research and
Teaching,” and did not necessarily purport to be reports on primary research.
Each paper was read through; the JSTOR system was also searched
for the words Black, white, race, Negro, ethnic, colored, African, and racial
to make sure that we did not overlook any discussions of race. The second
author coded every paper, and the first author coded every paper for 1 of the
20 volumes so that consistency could be checked. It turned out that consis-
tency was high across coders with one exception, the “area” of the research.
The inconsistency arose because many papers span two areas; paradigmati-
cally, research would be of interest to researchers in two fields, one pertaining
to the dependent variable and another to the independent variable. (One of
us tended to code on the dependent variable and the other on the indepen-
dent variable.) As a result, the first author then recoded every paper, adding
the second area, and checking the coding regarding the treatment of race,
the methods used, and the presence of a regression-type model.
When coding areas, we attempted to code what subarea of sociology
would actually be concerned with the analyses in question, as opposed to
P1: GXB
Sociological Forum [sofo] pp1028-sofo-475161 October 22, 2003 19:0 Style file version June 4th, 2002
REFERENCES