the societal context of perceived racial discrimination
TRANSCRIPT
The societal context of perceived racial discrimination
Rahsaan Maxwell Associate Professor
Department of Political Science University of North Carolina, Chapel Hill
Contact information: The Department of Political Science
322 Hamilton Hall • CB#3265 Chapel Hill, NC 27599-3265
[email protected] The research for this paper was funded by the University of Massachusetts, Amherst Faculty Research Grant/Healey Endowment Grant. Previous versions were presented at the 2012 Council for European Studies Conference of Europeanists, the 2012 Comparative Approaches to Immigration and Religious and Ethnic Diversity workshop (Princeton University), and “Identity Politics: The New World versus New (and Old) Europe” at the University of Texas, Austin (2013). The author would like to thank Rodolfo Espino, Marc Helbling, Daniel Hopkins, Tatishe Nteta, Brian Schaffner, and Alex Street for helpful comments on earlier drafts.
1
ABSTRACT Existing research has developed a wide range of evidence about predictors of
perceived discrimination among minorities, but often generates conflicting evidence across studies. I claim this may be due to the measurement strategy. Existing research often uses cross-sectional surveys that ask minorities about their experiences with discrimination, but these questions cannot determine whether subjects with higher levels of perceived discrimination are more likely to experience discrimination or are just more likely to interpret events through the lens of discrimination. In this article, I propose a new research strategy designed to measure when minorities use discrimination as a heuristic for understanding society. I use an original online survey of non-whites in the United Kingdom and provide vignettes that describe racial inequality in education, employment, politics, the arts, and business. I then ask respondents how they would explain each form of inequality. This ensures that all minority respondents are reacting to the same stimulus. In addition to the new measurement strategy, I make a theoretical contribution to the standard analysis of variation across minority individuals by emphasizing variation in perceived discrimination across societal contexts. This contextual approach has implications for debates about racial minorities in the UK and perceived discrimination in general.
2
Perceived discrimination is a key indicator of social cohesion. When racial and
ethnic minorities believe that society is biased against them they are likely to become
alienated (Safi 2010). Alienated minorities may then retreat from the political system,
weakening representation of their interests and the likelihood of improving their social
status (Schildkraut 2005). When they do act, dissatisfied minorities may choose tactics
of disruptive protest or violent revolt and in doing so contribute to societal disorder
(Lijphart 2012). Moreover, individuals who consider themselves fundamentally
disadvantaged due to their minority status may develop impermeable social and cultural
barriers with the majority population (Tajfel 1982: 23-4; Wimmer 2013: 174-203).
Despite widespread agreement that perceived discrimination is an important
measure of social cohesion, there is no consensus about which minorities are more likely
to have high levels of perceived discrimination. In fact, different studies often reach
contradictory conclusions. For example, some researchers find higher levels of perceived
discrimination among immigrant-origin minorities who recently arrived in their host
society while other studies find the opposite (Aguirre et al. 1989; Portes et al. 1980). One
explanation for these conflicting results is the way existing literature measures perceived
discrimination. Most research uses cross-sectional surveys that ask about experiences
with discrimination or opinions on the prevalence of discrimination in society. That
approach is vulnerable to measurement bias from differential item functioning.
Questions that explicitly ask about perceived discrimination cannot determine whether
subjects with higher levels of perceived discrimination are more likely to experience
actual discrimination or are just sensitive to the issue and more likely to use
discrimination as a heuristic for interpreting events. In addition, experiences with
3
discrimination and sensitivity to discrimination may influence survey responses in ways
that vary across subjects and are unobservable to the researcher. As a result, analysis of
the overall sample may yield unreliable relationships between the individual-level
characteristics of interest and perceived discrimination. In short, measurement error may
help explain the conflicting results across studies.
I propose that different aspects of discrimination be approached through distinct
research strategies. The most reliable way of studying when minorities are more likely to
experience discrimination is by using experiments that manipulate minority
characteristics (e.g. skin color, national origin, or religion) in a given situation (e.g. job
application or social interaction) and then observe majority individuals’ responses (Sen
and Wasow 2013). In contrast, studies of minorities are more appropriate for
determining which minorities are more likely to view life through the lens of racial
discrimination. In this article, I focus on the latter task with an online survey of non-
whites in the United Kingdom (UK).
I build on previous studies by proposing a new strategy to measure when
minorities use discrimination as a heuristic for understanding society. Instead of posing
questions that explicitly ask respondents to reflect on discrimination, I provide vignettes
that describe racial inequality in education, employment, politics, the arts, and business,
and then ask respondents to generate their own explanation for each form of inequality.
Previous research has asked minorities about their views on general racial inequality and
it has allowed minorities to express their views on discrimination with open-ended
questions. The key innovation in my design is the combination of open-ended questions
and no explicit prompt on racial discrimination. Research on survey methodology
4
suggests that open-ended questions are especially good at capturing how respondents
frame issues on their own terms (Achen 1975; Saris and Sniderman 2004). This may be
especially important for sensitive topics like discrimination, where respondents may be
reluctant to reveal their true attitudes to closed-ended questions that impose an external
framework on the issue (Tourangeau and Smith 1996). Moreover, the use of open-ended
questions that do not prompt respondents to think about discrimination provides a
standard stimulus and common benchmark for all respondents (Esses and Maio 2002;
Geer 1991). It is possible that previous experiences with discrimination may color the
ways in which minorities interpret future events (Kaiser, Vick and Major 2006), but tests
in the robustness checks section suggest that is not a concern for my analyses. In short,
the strategy of combining open-ended questions with no explicit prompt about
discrimination allows me to draw cleaner inferences about the conditions under which
minorities are more likely to view life through the lens of discrimination.
This article also makes a theoretical contribution to the study of perceived
discrimination. Existing literature primarily highlights variation in perceived
discrimination across minority individuals.1 In comparison, I emphasize variation across
societal contexts. I claim that the strength of specific individual-level predictors of
perceived discrimination will vary according to context. I also propose that there is more
variation in the likelihood of citing discrimination across context than across minority
individuals. Finally, I argue that contexts where minorities have less control over the
outcome will be more likely to elicit perceptions of discrimination.
1In part this focus reflects the ongoing debate mentioned earlier about the relationship between assimilation and perceived discrimination. There have also been a series of counterintuitive findings in the United States about the positive relationship between socio-economic attainment and perceived discrimination, which have sparked ongoing research into individual-level predictors of perceived discrimination (Chong and Kim 2006; Cose 1995).
5
My approach has implications for debates about racial minorities in the UK and
perceived discrimination in general. Most notably, my focus on context implies that all
minorities have the capacity to view life through racialized or non-racialized heuristics.
This complicates the search for which type of minority is more or less likely to view
society in racialized terms and has both optimistic and pessimistic implications for social
cohesion.
Generating hypotheses about perceived discrimination in the UK from existing
literature
Racial discrimination is highly salient in the UK. The UK has a long history of
racial diversity (and racial inequality) dating back to the transatlantic slave trade. More
recently, non-whites from around the world immigrated to the UK in the decades after
World War II and often suffered integration difficulties due to their status as racial
minorities (Gilroy 1993; Phillips and Phillips 1998). There is also a long tradition in the
UK of using the concept of racial discrimination as a framework for understanding the
inequalities suffered by non-whites (as opposed to other European countries where
similar inequalities are framed as a function of culture, religion or national origin) (Bleich
2003; Givens and Case 2014). Yet, despite the salience of racial discrimination, non-
whites in the UK do not always view their lives through the lens of race. In fact, existing
literature generates three different predictions for why some minorities in the UK will be
more likely than others to use racial discrimination as a heuristic for interpreting events.
The first hypothesis focuses on social class. In some respects, this is an alternative
framework for understanding inequality because it is possible that some non-whites in the
6
UK will interpret inequalities as the result of differences in socio-economic resources
(Miles 1982). In addition, longstanding debates about the interaction between race and
class suggest that poorly-educated and working class minorities are the most likely to
suffer from racial discrimination (Miles 1993; Rex and Tomlinson 1979; Sivanandan
1976). This imbalance in disadvantage should shape how different minorities view the
world (Chong and Kim 2006).
H1: Non-whites with lower socio-economic status will be more likely to cite
racial discrimination as a cause of inequality.
A second argument focuses on generational differences. Most racial minorities in
the UK have immigrant origins and when immigrants arrive they tend to interpret their
place in society through the lens of their foreign origins. In contrast, second-generation
non-whites who are born in the UK will inevitably be socialized into British norms and
social categories. As a result, second-generation non-whites will be more likely to
interpret life through the racial categories commonly used in the UK (Goulbourne 2009;
Paul 1997). This is consistent with research from across Europe and North America,
which finds that first-generation immigrants are often more positive and optimistic about
life in the new country, compared to the second generation which is more likely to be
disappointed, alienated, and sensitive to their vulnerability to discrimination (Maxwell
2010a, 2010b; Waters 1999).
7
H2: Non-whites born in the UK will be more likely to cite racial discrimination as
a cause of inequality.
The third argument highlights differences among non-whites. In contemporary
Britain, there are many established non-white communities, the largest of which are
Bangladeshis, Caribbeans, Indians, Pakistanis and Sub-Saharan Africans.2 Some argue
that racial discrimination is of particular concern to black minorities (e.g. minorities with
origins in the Caribbean and Sub-Saharan Africa).3 Black minorities were the ones most
likely to be defined by ‘racial’ differences, because the concept of race dated back to the
transatlantic slave trade and the distinction between people of African and European
descent (Gilroy 1993). Moreover, the legislation and institutions that had been developed
to fight racial discrimination in Britain during the 1960s and 70s were focused on unfair
treatment of minorities due to their phenotype (Bleich 2003). Other non-whites faced
different integration challenges. Most notably, for South Asians the primary barrier has
been whether their religious needs would be respected by a British society constructed
around Christian norms (Modood 2010; Modood and Berthoud 1997).
H3: Black non-whites should be the most likely to cite racial discrimination as a
cause of inequality.
2These groups accounted for 66% of the non-white population in England and Wales according to the 2011 Census. There are also smaller non-white groups from across Asia and the Middle East (Office for National Statistics 2012). 3There was a brief period during the 1960s and 70s when British activists tried to unite all non-whites under the same ‘black’ label to maximize their political power, but by the 1990s that unified classification had failed (Modood 1988; Modood and Berthoud 1997).
8
The societal context of perceived discrimination
The main theoretical proposition in this article is that individual-level attributes
matter less than context for predicting perceptions of discrimination. The main way
context has been used in previous studies of discrimination is to analyze how majority
individuals’ likelihood of discriminating against minorities varies according to the
demographic composition of their geographic environment (Oliver and Mendelberg
2000; Taylor 1998). Recent work has attempted to extend that approach and analyze
spatial variation in minorities’ perceptions of discrimination, but found no differences
across geographic contexts (Hopkins et al. 2013; McDermott 2011). In this article, I take
an alternate approach and analyze ‘societal contexts’, which are social institutions and
structures with different behavioral norms and ideological principles. For example, one
would expect different types of interpersonal interactions in a secondary school, on the
labor market, and in a political campaign. Therefore, one may also have different ways of
evaluating racial discrimination across societal contexts.
My approach builds on a range of work that examines how social spaces are rich
with symbolic meaning that can shape our attitudes and behavior (Blumer 1969;
Bourdieu 1989; Ross and Nisbett 1991), and that the meaning of these social spaces
depends on subjective interpretations (Crocker and Quinn 2000). In addition, my
approach is informed by work on the microfoundations of racism, which analyzes how
minorities receive subtle cues in different social spaces which must constantly be
evaluated for evidence of potential racism (Essed 1991). These insights suggest that
minorities may be more likely to perceive discrimination in some contexts as opposed to
others.
9
H4: The relative importance of individual-level factors for predicting perceived
discrimination will vary according to the context.
H5: There is more variation in perceived discrimination across contexts than
across individual-level factors.
To predict the type of context in which minorities will be most likely to perceive
discrimination, I draw on the finding that when minorities are confronted with negative
outcomes and evidence of potential discrimination, they adopt numerous strategies to
preserve their self-esteem (Crocker and Major 1989; Ruggiero and Taylor 1997). When
minorities have less control over the outcome, they are more likely to use discrimination
to explain negative results and protect their self-esteem by not blaming themselves. But
when minorities have more control over the outcome they will avoid using discrimination
to explain negative results. This strategy protects minorities’ agency and allows them to
avoid being categorized in a lower social status (Crosby 1984; Kaiser and Miller 2001).
H6: Non-whites will be more likely to cite discrimination as a cause of racial
inequality in societal contexts where they have less control over the outcome.
Data
The data in this article are from a survey conducted February 24 – 27, 2012 by the
market research agency YouGovUK. At the time of the study, YouGov maintained a
panel of approximately 350,000 potential survey respondents in the UK. YouGov recruits
10
panel members through standard (online and offline) advertising campaigns and
partnerships with a broad set of websites. Upon completion of a survey, respondents are
compensated with points that are redeemable for cash or prizes. As with any public
opinion survey, YouGov respondents have consciously decided to participate and
therefore may be more socially and civically engaged than the broader population or may
have views about discrimination that are not representative. However, research suggests
that internet samples drawn from large pools of potential respondents are at least as (if
not more) likely to represent the general population as standard telephone or face-to-face
surveys (Stephenson and Crête 2011).
For this article, YouGov drew a sample of 2,008 non-white respondents.4 The
sample was then weighted to be representative of the national non-white population of
adults 18 years old and older (including those without internet access) according to sex,
social class, region, party identity and newspaper readership.5 Appendix table 1 provides
descriptive statistics for weighted and unweighted data, which are similar on many basic
demographic variables. This suggests that the results in this article are unlikely to be
biased by underrepresented demographic sub-groups.
Measures
To measure perceived discrimination I provide a series of vignettes about racial
inequality and then ask for subjects’ opinions on the most important explanation for the
inequality. I do not prompt them in any way, and respondents are free to provide any
4YouGovUK uses a ‘turbo sampling’ method in which members of their panel are invited to participate in a survey and then assigned to a specific survey once they have accepted the invitation. The overall panel response rate is approximately 20%. Of those who started taking the survey, 10.5% did not complete the survey and were not included in the final sample. 5For more details on the polling methodology: http://yougov.co.uk/publicopinion/methodology/.
11
explanation that they feel is relevant. The vignettes were presented one at a time, on
separate screens, and the full text is as follows:6
Recent studies suggest that many BME7 groups are less likely than whites in Britain to finish secondary school. In your opinion, what is the most important explanation for this finding?
Over the past few decades the unemployment rate for many BME groups has consistently been higher than for whites in Britain. In your opinion, what is the most important explanation for this dynamic? After the 2010 General Election, BMEs now represent 4% of the Members of Parliament but are roughly 11% of the United Kingdom population. In your opinion, what is the most important explanation for this underrepresentation?
In recent years, many BME artists have complained that they do not receive the same funding or publicity opportunities as white artists. In your opinion, what is the most important explanation for this dynamic?
In your opinion, what is the most important explanation for why there are few BME CEOs of large corporations in Britain?
These contexts were selected to test H6 by providing variation in the extent of
minority control over the outcomes. Minorities have the most control (and therefore
should have the least perceived discrimination) in secondary education where the
outcomes are largely driven by meritocratic testing and there is no limit on the number of
people who can succeed. Minorities have the least control (and therefore should have the
most perceived discrimination) in politics, arts and business because while all three
outcomes involve elements of merit they are also highly competitive and dependent on
networks and luck. Unemployment is in-between. It is not as tightly linked to merit and
performance as secondary school graduation, but it is also not randomly distributed
across the population and one can exert some control over the likelihood of finding a job.
6The vignettes were consistently presented in this order. 7‘BME’ stands for black minority ethnic, a common British term for non-whites.
12
I code responses to the open-ended questions into three general categories: racial
discrimination, blaming minorities, and socio-economic status. I also code whether
responses are a mixture of these categories, whether they pose an ‘other’ explanation, or
whether they should be classified as ‘don’t know’. I limit the coding to three main
substantive categories because they were the most common responses and the most
general way of organizing the responses.8 The coding scheme is summarized in table 1
and produces twelve response categories. Full details on coding for the independent
variables can be found in the appendix.
‘Table 1 about here’
Perceived discrimination across individuals and societal contexts
To evaluate H1-H5, table 2 presents multinomial logistic regression models that
estimate the likelihood of explaining racial inequality in each of the societal contexts by
either racial discrimination, socio-economic factors, blaming minorities, a mixture of the
explanations, or ‘other’/‘don’t know’.9 For each model the baseline category is
‘Discrimination’, so the coefficients indicate the likelihood that respondents choose the
respective classification as opposed to discrimination. In an ideal world I would estimate
8Most of the ‘other’ responses were idiosyncratic and could not be grouped into a category. All coding was done by the author. All responses were coded twice for reliability. Automated coding was inappropriate because computers can code for subject matter but not meaning and cannot distinguish between ‘it’s all about race’ and ‘it’s definitely not about race’. 9A potential concern with multinomial logistic regression is the independence of irrelevant alternatives (IIA) assumption, which requires that the odds of choosing between two response options do not depend on the presence or absence of the other response options. There are several statistical tests (e.g. the Hausman-McFadden test and the Small-Hsaio test) that attempt to test whether the IIA assumption has been violated. Unfortunately, research suggests that these tests are unreliable and should be avoided by applied researchers (Cheng and Long 2007). In this case, there are reasons to believe that the IIA assumption was not violated because my survey respondents did not have a formal choice among concrete response options. Although there is no way of knowing exactly what came into each subject’s head before answering, a reasonable assumption would be that each subject (consciously or subconsciously) surveyed the range of possible answers before responding.
13
models that account for each of the twelve classifications in my coding scheme.
However, that would require a significantly larger sample size to have enough cases in
each response option for all values of each covariate. Given that constraint, I select these
five categories as the most substantively meaningful.
‘Table 2 about here’
The results in table 2 support for H1 because having no educational qualifications
is consistently associated with being more likely to cite discrimination as opposed to a
mixture of explanations for inequality. In addition, having a higher education degree is
associated with being more likely to cite a mixture of explanations as opposed to
discrimination in four of the five societal contexts. There is limited support for H2 as
being born in the UK is consistently associated with being more likely to cite
discrimination as opposed to blaming minorities. However, for neither H1 nor H2 is the
predicted relationship consistent across each of the coding category comparisons. There
is very little support for H3, as the relationship between being black and citing
discrimination is uneven across contexts and categorical comparisons. Overall, the most
striking pattern in table 2 is that no covariate has a consistent relationship with perceived
discrimination across all contexts and each categorical comparison. This supports H4, as
it suggests that the individual-level factors associated with perceived discrimination vary
across contexts.
Across the models in table 2, few covariates are statistically significant at
conventional levels and the r-squared statistics are low. This suggests that the standard
explanations for perceived discrimination are not good predictors of whether non-whites
14
interpret racial inequality through the lens of discrimination. I now turn to H5 and ask
whether societal context is a better predictor of perceived discrimination.
The first evidence supporting H5 is the fact that very few individuals interpret
inequality the same way in each societal context. Less than one percent of non-whites
give the same explanation for each of the five examples of inequality. Even if I collapse
the coding scheme to six options (Only discrimination, Only blame, Only economic
difficulties, a Mixture, Other, Do not know), only two percent of respondents answer the
same way across each question.
For a formal comparison of individual and contextual factors, I estimate a series
of multilevel logistic regression models without any covariates and in which the variation
in citing discrimination is decomposed across the five questions and across individuals. I
expand the dataset to create five observations (one for each open-ended question) per
respondent. The likelihood of citing discrimination is the outcome variable, the question-
level grouping variables are categorical variables with five values corresponding to the
values for each of the five inequality questions, and the individual-level grouping
variable is the unique respondent identification code.
Table 3 summarizes results from the multilevel models and indicates that when
predicting the likelihood of citing only discrimination as opposed to all other responses,
80 percent of the variation is across individuals and 20 percent is across contexts. The
next column indicates the exact same results for a model predicting the likelihood of
citing any discrimination. However, when I exclude those who answered ‘Do not know’
and estimate models only among respondents who offered an interpretation of the
inequality, roughly two-thirds of the variation is across individuals and one-third is across
15
contexts. The final column limits the analysis to those who answered either
Discrimination only or Blame only and here the variation is split evenly across
individuals and across contexts. These results suggest that among non-whites who are
able to provide a heuristic for interpreting inequality, much may depend on the context.
The results in table 3 do not support H5 as there is more variation in perceived
discrimination across individuals than contexts for most specifications. Yet, the models
summarized in table 3 decompose variation across all individual level characteristics,
including those which are unobserved, poorly understood by the literature, and random
noise. A more focused test of H5 would compare the importance of contexts with the key
independent variables cited in H1, H2, and H3. For this test, I again estimate logistic
regression models on the expanded dataset. I include individual-level covariates as well
as dummy variables for four of the five open-ended questions about inequality. Full
results are in appendix table 2, but the relationship between each covariate and the
likelihood of citing discrimination is summarized with a graph of average marginal
effects in figure 1.
‘Figure 1 about here’
Figure 1 provides strong evidence in favor of H5, as the average marginal effects
are much larger for the question variables than they are for the individual-level
characteristics. In the top graph, the average marginal effect for the inequality questions
ranges from 0.13 to 0.26, while the only effect for the individual covariates that is
statistically significant at the 5 percent level is 0.09 for having no qualifications.
Similarly, in the graph for responses that have any mention of discrimination, the average
marginal effect for the inequality questions spans from 0.11 to 0.26 while the largest
16
effect for an individual-level covariate is 0.09 for having no qualifications. In summary,
results thus far offer limited support for H1, H2, or H3 but provide stronger support for
H4 and H5. Societal contexts may be at least as important as all individual-level factors,
but they are much more important than the individual-level variables cited in existing
literature as explanations for why minorities are more or less likely to view life through
the lens of racial discrimination.
Variation in perceived discrimination across societal contexts
Figure 2 explores my hypothesis (H6) about the contexts in which non-whites are
more likely to perceive discrimination with graphs of the percentage of respondents who
mentioned discrimination as either the only explanation or as one of multiple
explanations for inequality.10 In each graph the pattern is the same. Discrimination is
more likely to be cited as an explanation for inequality among CEOs, artists and
politicians than as an explanation for educational inequality. This supports H6 as it
suggests non-whites are more likely to perceive discrimination in contexts where they
have less control over the outcomes. Moreover, appendix figure 1 indicates that ‘Blaming
minorities’ is a more common response to the question about educational inequality than
the other four contexts. This is also consistent with the logic behind H6, as minorities
should be more likely to emphasize their own responsibility (as opposed to
discrimination) in contexts like education where they have more control over the
outcome.
‘Figure 2 about here’
10Full results with the percentage of respondents in each of the twelve coding categories for each of the five contexts are in appendix figure 1.
17
A result from figure 2 that is inconsistent with H6 is the fact that discrimination is
more likely to be cited as an explanation for employment inequality than for political
inequality and it is mentioned by roughly the same percentage of respondents to explain
employment and arts inequality. I expected fewer mentions for employment inequality
because I claimed that minorities have more control over employment outcomes than
political and art outcomes. However, it is possible that non-whites feel less control over
employment outcomes than I anticipated. Alternatively, another form of variation across
societal contexts may be more important. For example, public debates about anti-
discrimination strategies in the UK have focused on the labor market and politics but not
as much on education and this may influence how non-whites interpret inequalities in the
different contexts (Givens and Case 2014).11 Future research should explore these
possibilities more closely.
Robustness checks
To ensure that my results are not dependent on coding decisions, I explore
alternate specifications of the independent variables. In the main analysis I coded mixed-
race respondents as ‘black’ because mixed-race individuals in Britain are likely to be
treated by society as black and are vulnerable to racial discrimination (Song 2003). Yet,
mixed-race individuals may be less sensitive to racial issues than other blacks, so I
conduct additional analyses with an operationalization of ‘black’ that excludes all mixed
individuals. A second issue is that socialization into the use of British racial categories
may occur within generations as immigrants spend more time in the UK. Therefore I
conduct additional analyses among non-whites born abroad and include a control variable
11I would like to thank one of the anonymous reviewers for suggesting this possibility.
18
for how long respondents have been in the UK. Finally, in the previous models I did not
account for interactions between non-white subgroups and socio-economic outcomes or
generational status. This may be an oversight because being black and having no
qualifications or between being black and being born in the UK may be more important
for perceived discrimination than merely being black, having no qualifications, or being
born in the UK (Modood and Berthoud 1997).
I use the expanded datasets to estimate a series of logistic regression models with
the alternate specifications. I estimate models predicting the likelihood of citing only
discrimination and models predicting the likelihood of citing any discrimination.12 Full
results are in appendix table 3, but the main finding is that none of these alternate
specifications challenge my earlier findings. Neither the narrower definition of black nor
the amount of time spent in the UK among first-generation immigrants is a statistically
significant predictor of perceived discrimination. For the interaction terms, there is no
statistically significant relationship between being black, having no educational
qualifications, and perceiving discrimination. There is a relationship between being
black, being born in the UK, and perceiving discrimination, although it is not the
expected relationship. For both the ‘discrimination only’ and ‘any discrimination’
specifications, blacks born in the UK are less likely to cite discrimination than blacks
born abroad, non-blacks born in the UK, or non-blacks born abroad. However, the
differences among these groups in the predicted probabilities of citing discrimination are
minor (less than five percentage points), suggesting there is not much substantive
12These models are a useful overview of relationships with perceived discrimination but results are similar when I estimate models predicting the likelihood of citing discrimination in each of the five contexts.
19
significance to the relationship. In short, my earlier findings are robust to several
alternate specifications.13
Another concern is that although I provide all subjects with the same stimuli in an
effort to obtain a common benchmark, individuals may have different experiences with
discrimination that influence how they respond to the questions about inequality. For
example, the relationship between having no qualifications and being more likely to
interpret inequality as discrimination may reflect the fact that individuals with no
qualifications are more likely to have experienced discrimination. To test for this
possibility, I use a survey item that asks how frequently subjects experience
discrimination. I then re-estimate the models from table 2 and appendix table 2 to see if
the relationships between individual-level covariates and perceived discrimination
become more consistent after controlling for self-reported exposure to discrimination.14
Full results from the models with an additional covariate for self-reported
discrimination are in appendix tables 4 and 5, and they suggest that the results are similar
to models without a covariate for personal discrimination. In all models, all of the
previously documented relationships are the same. In the multinomial logistic regression
models, having more personal experience with discrimination is mostly associated with
being less likely to cite discrimination as an explanation for inequality, although the
relationships are generally not statistically significant at the 10 percent level. In the
logistic regression models using the expanded dataset, having more experience with
13I also explored the possibility of gender differences in perceived discrimination. Non-white women in the sample are generally more likely than men to cite discrimination as an explanation for inequality, although the differences are fairly small and only statistically significant at the 10 percent level for the question about education. Moreover, the inclusion of a control variable for gender does not change any of the results presented in this article. 14In an ideal world I would also have controlled for variation in personal experiences with discrimination across the five societal contexts, but those questions were not included in the survey.
20
discrimination is also associated with being less likely to mention discrimination as an
explanation for inequality, and in this case the relationships are statistically significant at
the 1 percent level.
These findings are counterintuitive as one might have expected individuals who
report more experience with discrimination to be more likely to interpret inequality
through the lens of discrimination. One possibility is that non-whites’ views on
discrimination in their personal lives and in public are two distinct dynamics. However,
one should not over-interpret results based on the closed-ended question about
discrimination. As discussed earlier, closed-ended questions that explicitly ask about
discrimination measure multiple dynamics that are difficult to disentangle. Nonetheless,
at a minimum these robustness checks suggest that the results in this article are not biased
by variation in personal experiences with discrimination.
The final area of concern is survey effects. Open-ended questions are more
challenging than closed-ended questions because they require subjects to formulate their
own response. As a result, subjects are more likely to not answer open-ended questions
(Geer 1988). As seen in appendix figure 1, ‘Do not know’ ranges from 27 to 45 percent
across the five questions about inequality and it is the model response for each question.
In comparison, ‘Do not know’ was less than 8 percent of responses to the closed-ended
question about discrimination. In some respects, these greater cognitive demands are
appropriate for my study because they mirror the real-world challenge of appraising a
situation and determining whether it involves discrimination (Essed 1991). However,
there are two ways in which the cognitive demands of open-ended questions may bias my
results.
21
One possibility is that variation in perceived discrimination across individuals is
primarily about who is sufficiently cognitively engaged to answer the questions. This
could depress the relationships between other individual-level covariates and perceived
discrimination in the overall sample. To explore this possibility, I re-estimated the models
in table 2 only among non-whites who did not answer ‘Do not know’ for each question
(i.e. the most cognitively engaged sub-set of the sample) and the results were the same as
in table 2. In addition, if cognitive demands were the main factor driving ‘Do not know’
responses (as opposed to ambivalence or indecision) we would expect a large and
consistent amount of ‘Do not know’ responses to all five questions because people who
are vulnerable to the cognitive demands should immediately opt out of all open-ended
questions. Yet, while seventy percent of the sample answered ‘Do not know’ at least
once, less than five percent of the sample answered ‘Do not know’ to all five questions.
Finally, if I use educational attainment as a proxy for cognitive sophistication it is not a
statistically significant predictor at the 10 percent level of responding ‘Do not know’ for
any of the open-ended questions.15
The other potential survey effect is that variation in perceived discrimination
across contexts may not be due to the contexts themselves but rather the effect of
answering a series of challenging questions. This could lead to increasing numbers of ‘Do
not know’ responses as subjects progressed through the five questions and became more
vulnerable to the cognitive demands. Or, it could lead to decreasing numbers of ‘Do not
know’ responses as subjects became more habituated to the challenge and better able to
15This is based on a series of logistic regressions predicting the likelihood of answers being coded ‘Do not know’ as opposed to all other categories. I model education in three different ways. One is as the dummy variable: Higher education degree/No higher education degree. A second is as the dummy variable: Some higher education/No higher education. The third is: No qualifications/Some qualifications.
22
answer. Future work should randomize the order of open-ended questions to fully
mitigate this concern, but my results do not indicate that response patterns changed
consistently across questions. As seen in appendix figure 1, the number of ‘Do not
know’ responses is fairly constant across the first three questions, peaks at the fourth and
then drops for the final question.
Conclusion
This article offers a new way of studying perceived discrimination. I posit that
the standard practice of explicitly asking minorities about discrimination is likely to
produce inconclusive and misleading results because it cannot distinguish between the
extent to which minorities are responding based on their actual experiences with
discrimination or their levels of sensitivity to the issue of discrimination. To address this
concern, I focus on the latter dynamic and provide a series of open-ended questions that
measure the use of racial discrimination as a heuristic for understanding society.
My results break new theoretical ground by suggesting that the individual-level
variables cited in the existing perceived discrimination literature are less important than
variation across contexts. Furthermore, I argue that non-whites are more likely to cite
discrimination as a cause of racial inequality in societal contexts where they have less
control over the outcome. This builds on literature that connects minorities’ perceptions
of discrimination to strategies for preserving self-esteem.
A major implication of this article is that societal contexts are underappreciated
predictors of how minorities interpret events. For better or worse, this does not provide
easy answers for concerns about social cohesion in the UK and elsewhere. This article
23
suggests that all minorities – irrespective of socio-economic status, generational status, or
the particular non-white group – have the capacity to view life through racialized or non-
racialized heuristics. This raises new challenges, but the extent that it is an accurate
representation of how perceived discrimination operates, it is a useful contribution.
24
References Achen, C. (1975). Mass Political Attitudes and the Survey Response, American Political
Science Review, 69, 1218-1231.
Aguirre, B., Saenz, R. and Hwang, S.S. (1989). Discrimination and the Assimilation and
Ethnic Competition Perspectives, Social Science Quarterly, 70, 594-606.
Bleich, E. (2003). Race Politics in Britain and France: Ideas and Policymaking since the
1960s. New York: Cambridge University Press.
Blumer, H. (1969). Symbolic Interactionism: Perspective and Method. Englewood, NJ:
Prentice-Hall, Inc.
Bourdieu, P. (1989). Social Space and Symbolic Power, Sociological Theory, 7, 14-25.
Cheng, S. and Long, J.S. (2007). Testing for IIA in the Multinomial Logit Model,
Sociological Methods and Research, 35, 583-600.
Chong, D. and Kim, D. (2006). The Experiences and Effects of Economic Status among
Racial and Ethnic Minorities, American Political Science Review, 100, 335-351.
Cose, E. (1995). The Rage of a Privileged Class. New York: Harper Perennial.
Crocker, J. and Major, B. (1989). Social stigma and self-esteem: The self-protective
properties of stigma, Psychological Review, 96, 608-630.
Crocker, J. and Quinn, D. (2000). Social Stigma and the Self: Meanings, Situations and
Self-Esteem, in Heatherton, T., Kleck, R., Hebel, M. and Hull, J. eds. The Social
Psychology of Stigma. New York: The Guilford Press, pp. 153-183.
Crosby, F. (1984). The Denial of Personal Discrimination, American Behavioral
Scientist, 27, 371-386.
25
Essed, P. (1991). Understanding Everyday Racism: An Interdisciplinary Theory.
London: Sage Publications.
Esses, V. and Maio, G. (2002). Expanding the Assessment of Attitude Components and
Structure: The Benefits of Open-ended Measures, European Review of Social
Psychology, 12, 71-101.
Geer, J. (1991) Do Open-Ended Questions Measure “Salient” Issues?, Public Opinion
Quarterly, 55, 360-70.
Gilroy, P. (1993). The Black Atlantic: Modernity and Double Consciousness.
Cambridge, MA: Harvard University Press.
Givens, T. and Case, R.E. (2014). Legislating Equality: The Politics of
Antidiscrimination Policy in Europe. New York: Oxford University Press.
Goulbourne, H. (2009). Ethnicity and Nationalism in Post-Imperial Britain. Cambridge:
Cambridge University Press.
Hopkins, D., Mummolo, J., Esses, V., Kaiser, C., Marrow, H., and McDermott, M.
(2013). “Out of Context: The Unexpected Absence of Spatial Variation in U.S.
Immigrants’ Perceptions of Discrimination.” Working Paper.
Kaiser, C., Vick, S.B., and Major, B. (2006). Prejudice Expectations Moderate
Preconscious Attention to Cues that are Threatening to Social Identity, Psychological
Science, 17, 332-338.
Kaiser, C. and Miller, C. (2001). Stop Complaining! The Social Costs of Making
Attributions to Discrimination, Personality and Social Psychology Bulletin, 27, 254-263.
Lijphart, A. (2012). Patterns of Democracy: Government Forms and Performance in
Thirty-Six Countries. New Haven, CT: Yale University Press.
26
Maxwell, R. (2010a). Trust in Government among British Muslims: The Importance of
Migration Status, Political Behavior, 32, 89-109.
-. (2010b). Evaluating Migrant Integration: Political Attitudes Across Generations in
Europe, International Migration Review, 44, 25-52.
McDermott, M. (2011). Racial Attitudes in City, Neighborhood and Situational Contexts,
The Annals of the American Academy of Political and Social Science, 634, 153-173.
Miles, R. (1982). Racism and Migrant Labour. London: Routledge and Kegan Paul.
-. (1993). Racism after ‘race Relations’. London: Routledge.
Modood, T. (1988). ’Black’, racial equality and Asian identity, Journal of Ethnic and
Migration Studies, 14, 397-404.
-. (2010). Still Not Easy Being British: struggles for a multicultural citizenship. Stoke-
on-Trent: Trentham Books.
Modood, T. and Berthoud, R. (1997). Ethnic Minorities in Britain: Diversity and
Disadvantage. London: Policy Studies Institute.
Office for National Statistics. (2012). “Ethnicity and National Identity in England and
Wales 2011.”
Oliver, J.E. and Mendelberg, T. (2000). Reconsidering the Environmental Determinants
of White Racial Attitudes, American Journal of Political Science, 44, 574-589.
Paul, K. (1997). Whitewashing Britain: Race and Citizenship in the Postwar Era. Ithaca,
NY: Cornell University Press.
Phillips, M. and Phillips, T. (1998). Windrush: The Irresistible Rise of Multi-Racial
Britain. London: Harper Collins.
27
Portes, A., Parker, R.N. and Cobas, J. (1980). Assimilation or Consciousness: Perceptions
of U.S. Society among Recent Latin American Immigrants to the United States, Social
Forces, 59, 200-224.
Rex, J. and Tomlinson, S. (1979). Colonial Immigrants in a British City: a class
analysis. London: Routledge and Kegan Paul.
Ross, L. and Nisbett, R. (1991). The person and the situation: Perspectives of social
psychology. New York: McGraw-Hill.
Ruggiero, K. and Taylor, D. (1997). Why minority group members perceive or do not
perceive the discrimination that confronts them: The role of self-esteem and perceived
control, Journal of Personality and Social Psychology, 72, 373-389.
Safi, M. (2010). Immigrants’ Life Satisfaction in Europe: Between Assimilation and
Discrimination, European Sociological Review, 26, 159-176.
Saris, W. and Sniderman, P. (eds.) (2004). Studies in Public Opinion: Attitudes,
Nonattitudes, Measurement Error, and Change. Princeton: Princeton University Press.
Schildkraut, D. (2005). The Rise and Fall of Political Engagement among Latinos: The
Role of Identity and Perceptions of Discrimination, Political Behavior, 27, 285-312.
Sen, M. and Wasow, O. (2013). “Reconciling Race and Causation: How and When to
Extract Meaningful Causal Inferences About Race” Working Paper
Sivanandan, A. (1976). Race, Class and the State: The Black Experience in Britain, Race
and Class, 17, 347-368.
Song, M. (2003). Choosing Ethnic Identity. Cambridge: Polity Press.
28
Stephenson, L. and Crête, J. (2011). Studying Political Behavior: A Comparison of
Telephone and Internet Surveys, International Journal of Public Opinion Research, 23,
24-55.
Tajfel, H. (1982). Social psychology of intergroup relations, Annual Review of
Psychology, 33, 1-39.
Taylor, M. (1998). How White Attitudes Vary with the Racial Composition of Local
Populations: Numbers Count, American Sociological Review, 63, 512-535.
Tourangeau, R. and Smith, R. (1996). Asking Sensitive Questions: The Importance of
Data Collection Mode, Question Format, and Question Context, Public Opinion
Quarterly, 60, 275-304.
Waters, M. (1999). Black Identities: West Indian Dreams and American Realities. New
York: Russell Sage Foundation.
Wimmer, A. (2013). Ethnic Boundary Making: Institutions, Power, Networks. New
York: Oxford University Press.
29
Figures and Tables
Figure 1: Average marginal effects for the likelihood of citing discrimination as an explanation of inequality (expanded dataset with five observations per respondent)
Q.EmployQ.MPsQ.Arts
Q.CEOsHigher edNo quals
Prof.Manual
Born UKBlack
-.1 0 .1 .2 .3
Discrimination only
Q.EmployQ.MPsQ.Arts
Q.CEOsHigher edNo quals
Prof.Manual
Born UKBlack
-.1 0 .1 .2 .3
Any discrimination
Note: Results are calculated from models in appendix table 2.
The omitted question is about educational inequality. Bars are surrounded by 95% confidence intervals.
30
Figure 2: Percentage of non-white respondents citing discrimination as an explanation for racial inequality
Note: The graphs plot responses to open-ended questions with the percentage of
respondents who either only cited discrimination or who cited discrimination as well as another explanation for the five different kinds of inequality.
31
Table 1: Coding scheme for responses to the open-ended questions about explanations for racial inequality
Single-factor Mixed-factors Other Racial discrimination Discrimination & Economic Other
Economic factors Discrimination & Blame Do not know Blame minorities Economic & Blame
Disc/Econ/Blame Discrimination & Other Economic & Other Blame & Other
32
Table 2: Multinomial logistic regression results for responses to open-ended questions about inequality
Education Unemployment MPs Arts CEOs Socio-
economic Higher education -.646 (.261) ** .297 (.404) .208 (.289) .251 (1.14) .363 (.460) No qualifications -1.46 (1.17) -.362 (1.09) .209 (.712) -9.96 (1.31) *** .747 (1.13)
Professional .589 (.263) ** -.264 (.400) -.664 (.324) ** 1.27 (1.43) -.647 (.477) Manual .991 (.471) ** -.040 (.575) .240 (.395) 1.55 (1.37) -.721 (.778)
Born in UK -.039 (.233) .229 (.346) -.170 (.259) -.124 (1.15) .214 (.413) Black -.080 (.265) -.213 (.374) .430 (.261) .002 (1.20) .131 (.409)
Constant .045 (.259) -2.55 (.455) ** -1.47 (.305) -5.89 (.164) *** -3.18 (.521) ***
Blame minorities
Higher education .052 (.215) -.015 (.186) .067 (.175) -.066 (.200) -.116 (.220) No qualifications .164 (.704) -.533 (.555) -.629 (.562) -.466 (.599) -1.52 (1.05)
Professional .088 (.215) .007 (.187) -.258 (.175) .028 (.202) -.106 (.233) Manual .990 (.393) ** -.115 (.300) .289 (.268) .059 (.280) .260 (.314)
Born in UK -.049 (.182) -.137 (.166) -.254 (.152) * -.038 (.176) -.001 (.202) Black -.044 (.203) .149 (.182) -.092 (.170) .148 (.193) .136 (.218)
Constant 1.03 (.210) *** -.576 (.194) *** .328 (.174) * -.963 (.208) -1.42 (.231) ***
Mixed Higher education .078 (.222) .172 (.152) .095 (.200) .018 (.212) -.015 (.152) No qualifications -.235 (.745) -.979 (.491) ** -.503 (.580) -2.20 (1.05) ** -.563 (.489)
Professional .276 (.222) .065 (.153) -.154 (.196) .079 (.211) -.106 (.153) Manual 1.38 (.394) *** .304 (.216) .434 (.291) .020 (.292) -.136 (.220)
Born in UK .169 (.189) .122 (.131) -.242 (.171) .074 (.183) .205 (.131) Black .338 (.205)* -.009 (.146) -.074 (.190) -.029 (.200) -.023 (.145)
Constant .472 (.224) ** .004 (.156) -.167 (.203) -1.07 (.214) *** -.313 (.150) **
Other/ Don’t Know
Higher education -.244 (.209) .050 (.148) -.094 (.156) -.168 (.133) -.052 (.134) No qualifications -.225 (.689) -.606 (.413) -.704 (.452) -.657 (.374) * -.155 (.400)
Professional .213 (.216) -.038 (.150) -.089 (.154) .031 (.135) -.073 (.134) Manual .958 (.387) ** .142 (.213) .354 (.238) -.039 (.189) .069 (.189)
Born in UK .125 (.179) .049 (.128) -.103 (.134) -.074 (.113) .110 (.115) Black .081 (.196) -.002 (.142) -.025 (.146) .059 (.126) -.040 (.126)
Constant 1.31 (.210) *** .296 (.147) 1.02 (.160) *** .899 (.129) .284 (.135) N 2008 2008 2008 2008 2008 Pseudo R2 0.01 0.00 0.00 0.00 0.00
Weighted data Cells provide coefficients with robust standard errors in parentheses
*=p<.10, **=p<.05, ***=p<.01 Baseline category is ‘Discrimination only’
33
Table 3: Individual-level and question-level variation in responses Disc. only Any disc. Disc. only
(w/o dk) Any disc. (w/o dk)
Disc. vs. Blame
Individual 80% 80% 62% 64% 51% Question 20% 20% 38% 36% 49% Note: Results are calculated from multilevel logistic regression models with variation
across individuals and across questions specified as random effects. The first two columns include all respondents. The third and fourth columns exclude respondents who answered with a form of ‘I do not know’. The fifth column includes only respondents who answered Racial discrimination only or Blame minorities only.
34
Online Appendix Dependent variable: “Recent studies suggest that many BME groups are less likely than whites in Britain to finish secondary school. In your opinion, what is the most important explanation for this finding?” “Over the past few decades the unemployment rate for many BME groups has consistently been higher than for whites in Britain. In your opinion, what is the most important explanation for this dynamic?” “After the 2010 General Election, BMEs now represent 4% of the Members of Parliament but are roughly 11% of the United Kingdom population. In your opinion, what is the most important explanation for this underrepresentation?” “In recent years, many BME artists have complained that they do not receive the same funding or publicity opportunities as white artists. In your opinion, what is the most important explanation for this dynamic?” “In your opinion, what is the most important explanation for why there are few BME CEOs of large corporation in Britain?” Independent variables Education: What is the highest educational or work-related qualification you have?
0- No formal qualifications 1- Youth training certificate/skillseekers 2- Recognized trade apprenticeship completed 3- Clerical and commercial 4- City and Guild certificate 5- City and Guild certificate - advanced 6- ONC 7- CSE grades 2-5 8- CSE grade 1, GCE O level, GCSE, School Certificate 9- Scottish Ordinary/ Lower Certificate 10- GCE A level or Higher Certificate 11- Scottish Higher Certificate 12- Nursing qualification (eg SEN, SRN, SCM, RGN) 13- Teaching qualification (not degree) 14- University diploma 15- University or CNAA first degree (eg BA, B.Sc, B.Ed) 16- University or CNAA higher degree (eg M.Sc, Ph.D)
35
17- Other technical, professional or higher qualification 18- Don't know 19- Prefer not to say
Higher education (14, 15, 16, 17) No qualifications (0)
Current Occupation:
0- Professional or higher technical work 1- Manager or Senior Administrator 2- Clerical 3- Sales or Services 4- Foreman or Supervisor of Other Workers 5- Skilled Manual Work 6- Semi-Skilled or Unskilled Manual Work 7- Other 8- Have never worked
Professional (0) Manual (5, 6) Place of birth: 0 – Not in the UK, 1 – UK Race/ethnicity:
0- White and Black Caribbean 1- White and Black African 2- White and Asian 3- Any other mixed background 4- Indian 5- Pakistani 6- Bangladeshi 7- Any other Asian background 8- Black Caribbean 9- Black African 10- Any other black background 11- Chinese 12- Other ethnic group [open]
Black (0, 1, 8, 9, 10) Personal experiences with discrimination: “How often, if at all, would you say you experience discrimination in Britain?”
0 – Rarely/not very much, 1 – Some of the time, 2 – Often
36
Appendix Figure 1: Responses to open-ended questions about the most important explanation for various forms of racial inequality in Britain
Note: Weighted data.
The percentage of non-whites coded in each of the twelve response categories for each of the five questions about racial inequality.
‘E’ – Education, ‘U’ – Unemployment, ‘P’ – Politicians, ‘A’ – Arts funding, ‘C’ - CEOs
37
Appendix Table 1: Descriptive Statistics – Percentage of non-whites in various demographic categories
Weighted sample Unweighted sample Mean SD Min Max Mean SD Min Max
Age 38.4 14.2 18 83 38.6 13.6 18 83 Sex 0.52 0.50 0 1 0.51 0.50 0 1
Born in UK 0.55 0.50 0 1 0.56 0.50 0 1 Higher ed. 0.64 0.48 0 1 0.66 0.47 0 1 No quals. 0.02 0.14 0 1 0.02 0.14 0 1
Professional 0.29 0.46 0 1 0.31 0.46 0 1 Man. laborer 0.11 0.31 0 1 0.11 0.31 0 1
Mixed 0.12 0.32 0 1 0.13 0.34 0 1 Black 0.29 0.45 0 1 0.28 0.45 0 1
Black Caribbean 0.12 0.32 0 1 0.11 0.31 0 1 Black African 0.12 0.32 0 1 0.12 0.32 0 1 South Asian 0.48 0.50 0 1 0.47 0.50 0 1
Indian 0.27 0.45 0 1 0.27 0.44 0 1 Pakistani 0.12 0.32 0 1 0.11 0.32 0 1
Bangladeshi 0.05 0.22 0 1 0.05 0.22 0 1 Note: Sex: 0-Male, 1-Female.
The composite categories ‘Black’ and ‘South Asian’ include respondents with ‘Mixed’ ethnicity.
38
Appendix Table 2: Logistic regression results for the likelihood of citing discrimination as an explanation of inequality (expanded dataset with five observations per respondent)
Discrimination only Any discrimination Unemployment question 1.16 (.098) *** 1.03 (.078) ***
MPs question .822 (.101) *** .525 (.082) *** Arts question 1.29 (.097) *** .881 (.079) ***
CEOs question 1.48 (.096) *** 1.27 (.078) *** Higher education .033 (.062) .074 (.054) No qualifications .545 (.177) *** .425 (.161) ***
Professional .025 (.062) .083 (.054) Manual -.175 (.089) ** -.068 (.077)
Born in UK -.010 (.053) .010 (.047) Black -.019 (.058) -.015 (.051)
Constant -2.36 (.099) *** -1.66 (.078) *** N 10,040 10,040
Pseudo R2 0.04 0.03 Weighted data
Cells provide coefficients with robust standard errors in parentheses *=p<.10, **=p<.05, ***=p<.01
The omitted question is about educational inequality
39
Appendix Table 3: Robustness checks: Logistic regression results for the likelihood of citing discrimination as an explanation of inequality (expanded dataset with five
observations per respondent) with covariates for non-mixed black respondents, time spent in the United Kingdom, and interactions between black and no qualifications and black
and being born in the UK Discrimination only Any discrimination I II III IV V VI VII VIII
Black (excluding mixed) -.037 (.060)
-.035 (.053)
Time since arrival in UK .002 (.003)
.005* (.003)
Black (including mixed) -.026 (.057)
.157* (.085)
-.020 (.050)
.121 (.075)
No qualifications .420** (.183)
.255 (.167)
Born in the UK .079 (.061)
.067 (.053)
Black (including mixed) * no qualifications
.346 (.505)
.563 (.465)
Black (including mixed) * Born in the UK
-.336*** (.114)
-.248** (.101)
Constant -1.31*** (.029)
-1.34*** (.078)
-1.32*** (.031)
-1.35*** (.045)
-.795*** (.026)
-.902*** (.070)
-.805*** (.027)
-.835*** (.040)
N 10,040 4,420 10,040 10,040 10,040 4,420 10,040 10,040 Pseudo R2 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
Weighted data Cells provide coefficients with robust standard errors in parentheses
*=p<.10, **=p<.05, ***=p<.01
40
Appendix Table 4: Robustness checks: Multinomial logistic regression results for responses to open-ended questions about inequality with additional covariate for prior
exposure to discrimination Education Unemployment MPs Arts CEOs
Socio-economic
Higher education -.672 (.271) ** .274 (.402) .199 (.289) .285 (1.09) .338 (.464) No qualifications -1.51 (1.17) -.413 (1.08) .194 (.712) -9.48 (1.24) *** .696 (1.12)
Professional .572 (.263) ** -.281 (.403) -.670 (.323) ** 1.28 (1.43) -.669 (.479) Manual .995 (.472) ** -.042 (.576) .238 (.395) 1.56 (1.36) -.730 (.786)
Born in UK -.055 (.233) .215 (.347) -.175 (.261) -.107 (1.14) .199 (.412) Black -.112 (.264) -.243 (.371) .420 (.261) .044 (1.20) .099 (.420)
Personal disc. .274 (.157) * .251 (.212) .086 (.150) -.211 (.283) .289 (.245) Constant -.160 (.282) -2.75 (.493) *** -1.53 (.315) *** -5.77 (1.69) *** -3.40 (.527) ***
Blame
minorities Higher education .042 (.214) -.020 (.186) .058 (.176) -.065 (.200) -.127 (.221) No qualifications .147 (.702) -.543 (.556) -.646 (.558) -.464 (.559) -1.55 (.105)
Professional .080 (.216) .004 (.187) -.265 (.175) .029 (.202) -.113 (.234) Manual .990 (.393) ** -.115 (.300) .288 (.268) .060 (.279) .259 (.314)
Born in UK -.056 (.182) -.140 (.167) -.260 (.151) * -.037 (.178) -.008 (.201) Black -.057 (.203) .142 (.182) -.103 (.170) .150 (.194) .122 (.201)
Personal disc. .117 (.125) .051 (.109) .090 (.100) -.013 (.119) .109 (.133) Constant .951 (.232) *** -.613 (.210) ** .260 (.191) -.953 (.226) *** -1.50 (.252) ***
Mixed Higher education .062 (.222) .165 (.153) .090 (.201) .020 (.215) -.031 (.152)
No qualifications -.263 (.743) -.992 (.492) ** -.513 (.581) -2.20 (1.06) ** -.595 (.492) Professional .264 (.223) .060 (.153) -.158 (.196) .080 (.211) -.117 (.153)
Manual 1.38 (.393) *** .303 (.217) .433 (.291) .019 (.292) -.137 (.221) Born in UK .159 (.189) .118 (.132) -.245 (.171) .075 (.184) .195 (.131)
Black .317 (.204) -.017 (.146) -.080 (.190) -.026 (.200) -.044 (.145) Personal disc. .171 (.130) .070 (.085) .051 (.111) -.020 (.125) .165 (.086) *
Constant .349 (.245) -.048 (.162) -.205 (.218) -1.06 (.221) *** -.435 (.166) ***
Other/ Don’t Know
Higher education -.261 (.209) .040 (.148) -.098 (.156) -.180 (.132) -.064 (.134) No qualifications -.254 (.692) -.626 (.411) -.712 (.452) -.680 (.373) * -.179 (.400)
Professional .201 (.212) -.045 (.150) -.093 (.154) .023 (.135) -.082 (.134) Manual .959 (.387) ** .142 (.212) .353 (.283) -.040 (.188) .068 (.189)
Born in UK .115 (.179) .043 (.128) -.106 (.134) -.081 (.113) .103 (.115) Black .060 (.197) -.015 (.142) -.030 (.147) .045 (.126) -.057 (.126)
Personal disc. .179 (.122) .101 (.083) .045 (.086) .114 (.073) .126 (.075) * Constant 1.18 (.231) *** .221 (.158) .991 (.173) *** .814 (.141) *** .193 (.145)
N 2008 2008 2008 2008 2008 Pseudo R2 0.01 0.00 0.00 0.00 0.00
Weighted data Cells provide coefficients with robust standard errors in parentheses
*=p<.10, **=p<.05, ***=p<.01 Baseline category is ‘Discrimination only’
41
Appendix Table 5: Robustness checks: Logistic regression results for the likelihood of citing discrimination as an explanation of inequality (expanded dataset with five
observations per respondent) with additional covariate for prior exposure to discrimination
Discrimination only Any discrimination Unemployment question 1.16 (.098) *** 1.03 (.078) ***
MPs question .823 (.101) *** .525 (.082) *** Arts question 1.29 (.097) *** .882 (.079) ***
CEOs question 1.48 (.096) *** 1.27 (.078) *** Higher education .043 (.062) .084 (.054) No qualifications .564 (.177) *** .442 (.161) ***
Professional .032 (.062) .089 (.054) * Manual -.175 (.089) *** -.068 (.077)
Born in UK -.004 (.003) .015 (.047) Black -.006 (.058) -.003 (.051)
Personal discrimination -.101 (.034) *** -.092 (.031) *** Constant -2.28 (.103) *** -1.59 (.081) ***
N 10,040 10,040 Pseudo R2 0.04 0.03
Weighted data Cells provide coefficients with robust standard errors in parentheses
*=p<.10, **=p<.05, ***=p<.01 The omitted question is about educational inequality
42