Don’t Know or Won’t Say? Exploring How Colorblind Norms Shape Item Nonresponse in Social Surveys

Abstract

Colorblind norms play an important role in shaping how people discuss race. There is reason to believe that these norms also affect the ways respondents interact with social surveys. Specifically, some respondents may be using nonresponse as a tactic to not discuss race in social surveys. If this is the case, very different demographics of respondents would be most prone to nonresponse, and the phenomenon should also vary on the basis of the interviewer’s race. The author conducted bivariate and multivariate analysis of the Chicago Area Study to examine whether colorblindness may be promoting “don’t know” responses and item refusals. The author finds that nonresponse to a perceived race of interviewer item follows a distinct pattern consistent with previous research on colorblind norms. For example, white respondents have nearly five times the rate of nonresponse compared with blacks and Latinos. Bolstering the colorblindness theory, an interracial interview context nearly triples the nonresponse rate compared with same-race interviews. Findings of this research have important implications for both survey researchers using social surveys to examine race and racial attitudes and race scholars who seek to understand the prevalence of colorblind norms across society.

Keywords

item nonresponse colorblind racism social surveys racial apathy social desirability

Survey researchers go to great lengths to collect accurate and representative data. However, item nonresponse—when a respondent fails to substantively answer a particular survey item (e.g., “don’t know” or answer refusals)—can be a major barrier to constructing accurate representations drawn from surveys. Despite the major effects item nonresponse has on social inquiry, a coherent research agenda concentrating on the phenomenon remains in its “infancy” (Groves et al. 2008). A common explanation for item nonresponse is that marginalized groups are more likely to engage in this behavior because of differences in available resources necessary to form opinions (Berinsky 2002b; Colsher and Wallace 1989). However, because most item nonresponse analysis has focused on survey items that are not race related (Adua and Sharp 2010; Alvik, Haldorsen, and Lindemann 2005; Blom, de Leeuw, and Hox 2011; Chen 2011; Franses, Geluk, and Homelen 1999; Klein et al. 2011; Kupek 1998; Tomaskovic-Devey, Leiter, and Thompson 1995; Tourangeau, Groves, and Redline 2010), it is not known if the processes leading to item nonresponse are the same for race-related items. This is particularly important to understand given the emergence of new racial norms that discourage open discussions of race.

Researchers studying modern racial norms argue that dominant groups tend to be the most likely groups to opt out of talking about racial issues (Bonilla-Silva 2003; Carr 1997). Specifically, a number of scholars have argued that colorblind racism—the belief that race is no longer noticed and/or an important factor—has gained dominance in the United States (Bonilla-Silva 2003; Gallagher 1997; Lewis 2004). According to these scholars, this new form of racism manifests in new norms that encourage certain individuals (particularly whites, well-educated people, young people, and men) to avoid directly talking about race (Apfelbaum, Sommers, and Norton 2008; Bonilla-Silva 2003; Pollock 2005), especially in interracial contexts (Apfelbaum, Sommers, et al. 2008). This perspective suggests that race-related items might be subject to different patterns of item nonresponse than non-race-related items.

There is reason to believe that item nonresponse patterns for race-related items might be influenced by a different set of factors compared with other types of items given that contemporary social norms within U.S. society promote avoidance of the discussion of race altogether.

In this study, I test competing hypotheses about the factors that influence item nonresponse. The marginalized group hypothesis draws on traditional explanations of item nonresponse in survey research and posits that certain groups or individuals have more knowledge resources to draw on during interviews and are thus more likely to respond compared with their less resourced counterparts (Adua and Sharp 2010; Blom et al. 2011; Candido, Kurdyak, and Alter 2011; Klein et al. 2011; Kupek 1998; Watkins and Melde 2007; Wiederman 1993; Wiederman, Weis, and Allgeier 1994). Alternatively, the colorblind hypothesis draws on research in the field of race and ethnic relations claiming that race-related survey items are influenced by the norm of colorblindness and proposing that social correlates associated with nonresponse to non-race-related items will differ from those associated with race-related item nonresponse (Berinsky 1999; Forman 2004). Furthermore, according to this second hypothesis, racially discordant surveys (i.e., interviews in which the interviewer and respondent are of different racial backgrounds) might additionally promote nonresponse to race-related questions.

The Problem of Item Nonresponse in Social Surveys

Item nonresponse poses unique challenges to researchers who use social surveys. Many researchers believe that nonresponse is a problem only when it affects a large proportion of a given sample (e.g., >15 percent). Merely examining the proportion of the sample affected by nonresponse, however, does little to reveal whether it will pose problems. When nonresponse is moderately sized but random, it poses relatively few issues. In many such cases, it could even be appropriate to ignore nonresponders (e.g., listwise deletion). Before taking this step though, researchers must first uncover whether data are missing at random. That is, they must determine a systematic barrier to substantive responses that may bias analyses.¹ In short, knowing whether nonresponse occurs randomly allows a researcher to make informed choices around how to handle the missingness appropriately.

Because so many statistical methods rely on probability estimates, systematically patterned missing data can bias estimates if not properly handled (Schafer and Graham 2002). When patterned nonresponse is erroneously treated as random, it is likely to result in false conclusions and/or misleading estimates (Little and Rubin 2014).² A wide variety of factors could cause nonresponse to race-related social survey items. As such, researchers using social surveys need to be especially thoughtful about examining our data for patterned variation in “don’t knows” and refusals.

Potential Causes of Item Nonresponse

Marginalized Group Explanation

Prior studies of item nonresponse have consistently shown that members of historically disadvantaged groups (e.g., women, the poor, the elderly, nonwhites) are especially vulnerable to engaging in item nonresponse behaviors such as “don’t know” and item refusals (Adua and Sharp 2010; Blom et al. 2011; Candido et al. 2011; Klein et al. 2011; Kupek 1998; Watkins and Melde 2007; Wiederman 1993; Wiederman et al. 1994).

Although most of this research is descriptive in nature, some scholars contend that there is a key mechanism at play: differences in ability to find answers to questions. That is, item nonresponse happens among marginalized groups at higher rates because such respondents are “less easily able to form coherent and consistent opinions” compared with their higher-resourced counterparts (Berinsky 2002b:277). Marginalized respondents, such as those with lesser education and lower incomes, appear to have more difficulty recalling factual information (Kupek 1998). Because respondents tend to not respond unless they can locate an “adequately precise answer” (Groves et al. 2009:188), this may result in an overrepresentation of “don’t knows” among marginalized group members.

Systematic nonresponse patterns are particularly problematic when they underrepresent already marginalized groups. For example, respondents with lower incomes are less likely than respondents with higher incomes to give substantive responses to items about welfare—policies that they would arguably both benefit from and support—because of a lack of access to relevant information (Berinsky 2002b). Because many researchers exclude item nonresponders from analysis, these opinions become invisible to the public at large and advantage high-resourced respondents who respond at higher rates (Berinsky 2002b).

Much of the prior evidence analyzing item nonresponse finds support for the marginalized group hypothesis. For example, nonwhites tend to have higher rates of item nonresponse than their white counterparts (Blom et al. 2011; Klein et al. 2011). One large-scale study of Medicare providers, for example, found that whites were one-third more likely to complete all items substantively in the survey compared with African Americans (Klein et al. 2011).³

Researchers examining item nonresponse by gender and sex have also consistently shown that female respondents exhibit higher rates of “don’t knows” and question refusals compared with male respondents (Adua and Sharp 2010; Candido et al. 2011; Wiederman 1993; Wiederman et al. 1994). Additionally, one of the most consistent patterns in nonresponse literature is that the older a respondent, the more likely he or she is to engage in nonresponse strategies (Adua and Sharp 2010; Blom et al. 2011; Klein et al. 2011; Watkins and Melde 2007; Wiederman 1993). In fact, one study found that far and away, older age was the most predictive variable for nonresponse among all demographic factors (Elliott et al. 2005).

Lower levels of income and education (separately and/or combined into a socioeconomic status variable) also tend to be associated with higher item nonresponse (Adua and Sharp 2010; Alvik et al. 2005; Candido et al. 2011; Wiederman 1993; Wiederman et al. 1994). These effects are particularly apparent in studies of nonresponse to socially sensitive items. For instance, in a study of pregnant women, an item inquiring about alcohol use during pregnancy elicited double the nonresponse among lesser educated women compared with their higher educated counterparts. Similarly, respondents with lower incomes show marked declines when responding to socially sensitive items about sexuality (Wiederman et al. 1994).

Taken as a whole, the research just described shows that more marginalized respondents provide more “don’t knows” than their more resourced counterparts. None of the studies outlined here, however, examined item nonresponse to a race-related survey item. As such, it is unclear whether nonresponse to these items follows the same patterns as non-race-related items. Recent racial attitudes research indicates shifting norms and understandings of race relations that promote an increasing avoidance of talking about racial issues, particularly among whites (Bonilla-Silva 2003; Carr 1997; Forman 2004; Pollock 2005). This new trend may be inhibiting substantive response to race-related social survey items.

Potential Effects of Colorblind Norms on Social Surveys

Racial attitudes in America have undergone a drastic shift over the past century, with traditional prejudice plummeting (Bobo et al. 2012; Schuman et al. 1997). Although on its surface it appears that there has been a drastic reduction in prejudice, many scholars suggest that shifting societal norms have merely transformed its form (Bobo, Kluegel, and Smith 1997; Bobo and Zubrinsky 1996; Bonilla-Silva 2003; Forman 2004). One new type of racial prejudice, colorblind racism, has gained a great deal of attention in scholarly explanations. This new form is characterized by the belief that race is no longer a salient force in American life and thus should be ignored (Bonilla-Silva, Lewis, and Embrick 2004; Forman 2004; Forman and Lewis 2006). It manifests in norms promoting avoidance and ambivalence about racial matters (Forman 2004).

Forman and Lewis (2006) posited that colorblind racism promotes “not knowing” and “not caring” about race (i.e., racial apathy) and that it pushes individuals (particularly whites) to not talk about race at all. For example, during the period between 1964 and 2000, there was more than a threefold increase of whites’ expressing “no interest” to a question asking about attitudes toward school integration (Forman 2004). Although racial attitudes researchers have previously been largely unconcerned with nonresponse to race-related social survey items, increases in the phenomenon and the emergence of new racial norms have fueled a new interest in the topic (Bobo et al. 2012).

Using nonresponse as a strategy to avoid discussing race in social surveys is a relatively recent occurrence. Although in the 1970s, “don’t know” responses to certain items about racial policy tended to signify a respondent’s genuine lack of information or ability to offer an informed answer, “don’t knows” in the modern era are linked to a hesitation to express an opinion on matters of race for fear of social sanction (Berinsky 1999, 2002a). Taken together, this evidence suggests that nonresponse to race-related social survey items may be a function of a new colorblind norm and may follow very different patterns than other explanations for nonresponse.

If nonresponse to race-related items is in fact promoted by a colorblind social norm, I expect that the social correlates associated with it would follow a very different pattern than suggested by the marginalized group hypothesis. From the colorblind perspective, race-related items would exhibit heightened levels of nonresponse among respondents who are white, young, and male and have higher levels of education than their counterparts. Additionally, I expect that nonresponse to race-related items will be more common in racially discordant interview contexts compared with contexts in which the interviewer and respondent are of the same racial background.

Researchers find that colorblindness affects most members of society, but engagement of these norms varies on the basis of social location (Bonilla-Silva 2003; Hartmann, Gerteis, and Croll 2009). For instance, researchers have found that race/ethnicity is a strong determinant of the adoption of colorblind strategies. Whites are nearly twice as likely to claim to be colorblind compared with blacks (Carr 1997) and are much more likely to minimize the salience of race (Bonilla-Silva 2015). Whites are also more likely than nonwhites to deny the existence of institutional discrimination (Hartmann et al. 2009) and adopt explanations for racial inequality that deny the role of discrimination (Bonilla-Silva 2003). Whites also report feeling pressure to disengage from racial discussion altogether (McKinney 2003). This evidence indicates that whites may be much more likely than nonwhites to be opting out of race-related social survey items.

There might also be important effects by age, sex, and education. Younger adults, particularly those of the “millennial generation” (i.e., those who were born between 1982 and 2000), came of age during an era when colorblind attitudes were normative (Bonilla-Silva 2001; Schuman et al. 1997) and may be particularly affected by such norms because they are so deeply embedded. Men are also more prone to adopt colorblind explanations for phenomena (Bonilla-Silva 2003) and may also be more likely to engage in avoidance.

Higher educated respondents have heightened awareness of racial norms and a stronger motivation to avoid appearing prejudiced (Krysan 1998) and thus may additionally be quite prone to enaging in racial avoidance in social surveys. Specifically, college-educated respondents have access to more sophisticated articulations of racial issues that rely on meritocratic explanations compared with their lesser educated counterparts (Jackman and Muha 1984) and may be particularly sensitive to “saying the wrong thing” and avoiding response altogether.

The social pressures around race extend beyond background characteristics and are also affected by context. Laboratory studies show that the discomfort around discussing race is much more common in interracial interactions compared with same-race contexts (Apfelbaum, Sommers, et al. 2008; Norton et al. 2006). If social surveys are affected by colorblind norms, as I suggest, having a different-race interviewer (i.e., a racially discordant survey context) would greatly increase nonresponse to a perceived race of interviewer item compared with scenarios in which respondents are interviewed by members of their own race. That is, respondents will take cues from their environment to determine whether talking about race is more or less appropriate (i.e., they will feel that it is less appropriate to discuss race in an interracial interview context).

Taken together, the social and contextual correlates expected by the colorblind hypothesis vary greatly from those proposed by the marginalized group hypothesis. New norms may be greatly affecting the ways that respondents engage with social surveys that examine race. Such findings would have very important consequences for survey researchers examining racial attitudes.

Study Hypotheses

On the basis of the previous research examined here, I have developed the following hypotheses that reflect two prevailing understandings of nonresponse, the marginalized group hypothesis and the colorblind hypothesis:

Hypothesis 1: Being nonwhite, female, a senior citizen, foreign born and having less than a college degree will increase nonresponse to race-related and non-race-related items (marginalized group hypothesis).

Hypothesis 2: Being white, a millennial, and college educated and having a racially discordant interview context will increase nonresponse to race-related items (colorblind hypothesis).

Data and Methods

The data for this study are drawn from the 2008 Chicago Area Study (CAS) (Holbrook, Johnson, and Krysan 2008), a random-digit dialing telephone survey conducted to gain insight into the political participation and political attitudes of Chicago residents. Researchers sampled only landline-using households within Chicago city limits and selected respondents 18 years of age or older (Graf and Retzer 2008). I use this data set because it contains a unique set of items (e.g., perceived race of interviewer, racial discordance) that allow rigorous testing of the colorblind hypothesis.

The survey was administered by the University of Illinois at Chicago’s Survey Research Laboratory using computer-assisted telephone interview technology. Each phone number was randomly assigned to a specific racial pool of interviewers (i.e., white, black, Latino, and any other race). Keeping interviewer race by interviewer pools allows a particularly robust measure of racial discordance, because it does not leave room for respondents to actively or passively decline interviewers of particular races, while later agreeing to subsequent interviewers of a different race. The study sample includes a total of 657 completed interviews (616 English-language interviews and 41 Spanish-language interviews) conducted by trained graduate students and professional interviewers in the spring of 2008. The 2008 CAS had a 20.3 percent response rate (33.7 percent cooperation rate). Among the completed interviews, 43 percent of the respondents were white (weighted n = 265), 41 percent of the respondents were black (weighted n = 250), 7 percent of the respondents were Latino (weighted n = 85),⁴ and 13 percent reported some other race or multiple races (weighted n = 46) (the remaining 11 respondents did not report any racial background). The analytic sample for the present study includes respondents who self-identified as white, black, or Latino (unweighted n = 627, weighted n = 598). The data are weighted using both selection weights and poststratification weighting for all analysis in this study. The poststratification weights approximate the expected age, gender, and racial-ethnic categories in the city of Chicago per U.S. census data.⁵

Dependent Variables

Nonresponse to Political Ideology

Respondents were asked, “In general, would you describe your political views as very conservative, conservative, moderate liberal or very liberal?” To reduce nonresponse rates for this item, interviewers probed once if respondents first responded “don’t know” or refused to answer the question. Despite the probe, 4 percent (n = 25) of the respondents within the analytic sample reported not knowing the answer to this item or refused to answer it outright. I create a dichotomous variable to reflect nonresponse to this nonracial item, with 1 representing “don’t know” or refusal to answer this item and 0 representing a coded response from among the choices outlined. I use this item because of its relatively high nonresponse and because researchers have previously noted that respondents have difficulty understanding the meanings of these categories (Luttberg and Gant 1985).

Nonresponse to Income

To determine the category best representing their incomes, respondents were asked a series of up to three branching questions. First, all respondents were asked, “Was your total household income for the year 2007, from all sources, before taxes, more or less than $60,000?” If respondents reported household incomes above $60,000, they were asked if their household incomes were more or less than $80,000. If they reported less, they were categorized as having incomes between $60,001 and $79,999 and asked no further questions about income. If respondents reported incomes above $80,000, they were asked if their household incomes was above or below $100,000. Similarly, if respondents reported “less” to the first item, they were asked a series of branched questions. Respondents were asked between one and three questions about their income levels in total, depending on their answers. I coded respondents who offered “don’t know” responses or refused to answer any of these income items as 1 for this variable. All respondents who gave substantive responses (i.e., “more,” “less,” or “exactly”) to all items asked in their logical branching are coded as 0. More than 12 percent (n = 75) of respondents did not respond substantively to at least one income item. I use this item because it had the highest nonresponse rate among all of the items in the survey. Income items are a common source of nonresponse in social surveys.

Nonresponse to Perceived Race of Interviewer

The race-related dependent variable in this study is perceived race of interviewer. At the end of the survey, respondents were asked the following item, “This last question is just for research purposes. You may not have thought about this but I’d like to ask you to guess my race or ethnicity. Would you guess that I am white, black, Latino, Asian, or some other race?” If respondents did not select any answer, the interviewers were instructed to probe and encourage a guess at least twice, stating that there is no wrong answer and that guessing is encouraged. Respondents are coded 1 if they refused to answer or voluntarily responded “don’t know” to the question. Within the analytic sample, 5.9 percent were nonresponsive to this item. The inclusion of multiple probes makes nonresponse for this item considerably lower than rates reported in other studies. For example, a nationally representative survey of white adults conducted by ABC News and the Washington Post that used only a single probe found 23 percent nonresponse to the same type of question (Blumenthal 2008).⁶ I use this item because it was among the few items in the survey that directly asks about race and because it presents a unique interactional element between the interviewer and respondent.

Independent Variables

Race

A primary independent variable is race. To represent race, I use a dichotomous variable, coded 1 for white and 0 for both Latino and black respondents. Within the analytic sample, 36.4 percent are black (n = 250), 24.4 percent are Latino (n = 85), and 39.2 percent are white (n = 265).

Female

This variable is a self-report coded 1 for female and 0 for male.

Generation

I separate generations into three categories: those reporting a birth year of 1982 or after are coded as millennials (n = 114), those reporting an age of 65 or older are coded as seniors (n = 70), and all other respondents are coded as middle-aged (n = 427). In the multivariate analysis, I construct the models using middle-aged as the reference category.

Education

I separate education into two categories (i.e., college degree or more and less than a college degree). Just above 60 percent of respondents did not report a college degree (n = 380), while the rest reported that they had attained a college degree or more (n = 238). Less than a college degree is the reference category in my multivariate analysis.

Foreign Born

I create a dichotomous variable with 1 representing a respondent reporting birth in another country and 0 for respondents reporting being U.S. born. Twenty-two percent (n = 141) of the respondents in the sample reported being foreign born.

Racial Discordance

Racial discordance measures whether the respondent and interviewer were of different races. CAS interviewers self-reported as having one of the four following backgrounds: white, black, Latino, or “some other race” (including Asian and Asian American interviewers). Among the cases analyzed for this sample, 29 percent were conducted by black interviewers (n = 172), 34 percent by Latino interviewers (n = 198), 28 percent by white interviewers (n = 174), and 9 percent by interviewers of another race (n = 56).

To create the racial discordance variable, each case is coded 0 if the respondent’s self-reported race matches the interviewer’s race (i.e., a black respondent with a black interviewer, a Latino respondent with a Latino interviewer, a white respondent with a white interviewer), and respondents who reported a different race from their interviewers (e.g., black respondent and Latino interviewer, white respondent and black interviewer, Latino respondent with some other race interviewer) are coded 1. Nearly 60 percent of the analytic sample (n = 383) is made up of racially discordant cases, while the rest of the respondents in the sample participated in racially concordant interviews. Sixty-nine percent of whites (n = 183) and 59.7 percent of nonwhites (n = 200) had racially discordant interview contexts.⁷ This is an important variable to test colorblindness because variation in nonresponse patterns would indicate that respondents do notice race and are consequently making decisions to respond or not respond on this basis.

Analytic Plan

My analytic plan has two stages. The first stage is considered preliminary and consists of bivariate analyses of the relationships between the independent variables and nonresponse to the non-race-related and race-related items. These findings are reported in Figures 1 and 2 as well as Table 1. To compare the patterns of nonresponse by race and to test both hypotheses, I use Figure 1 to visually characterize the bivariate relationships between respondent race and nonresponse to the race-related and non-race-related dependent variables. In Figure 2, I test the colorblind hypothesis further by outlining bivariate relationship between each dependent variable and racial discordance. In Table 1, I explore the bivariate relationships between select independent variables and the three dependent variables.

Figure 1.

Item nonresponse by race.

Figure 2.

Item nonresponse by racial discordance.

Table 1.

Bivariate Association between Independent Variables and Nonresponse to Non-race-related and Race-related Items.

	Nonresponse to Political Ideology		Nonresponse to Any Income Item		Nonresponse to Perceived Race of Interviewer
Variable	Nonresponse	χ²	Nonresponse	χ²	Nonresponse	χ²
Respondent race		.581		.148		23.710***
Black and Latino (n = 380)	4.47%		11.58%		2.11%
White (n = 246)	3.25%		12.60%		11.38%
Sex		6.669**		.288		.911
Male (n = 300)	2.00%		12.71%		6.69%
Female (n = 327)	6.12%		11.31%		4.91%
Generation		.249		18.402***		2.306
Millennial (n = 114)	4.39%		21.74%		2.63%
Middle-aged (n = 427)	3.51%		7.96%		5.87%
Senior (n = 70)	4.29%		15.71%		7.14%
Education		1.208		.889		12.838**
Less than college degree (n = 379)	4.75%		12.63%		3.17%
College degree or more (n = 237)	2.95%		10.13%		10.13%
Foreign born		5.854*		1.177		.093
U.S. born (n = 496)	3.02%		12.70%		6.05%
Foreign born (n = 130)	7.69%		9.23%		5.34%
Interviewer race		.338		4.466*		7.709**
Same race (n = 255)	4.71%		8.66%		2.75%
Different race (n = 372)	3.76%		14.25%		8.06%

p ≤ .05. **p ≤ .01. ***p ≤ .001.

For the next stage of analyses, reported in Table 2, I examine each hypothesis separately, then together. Specifically, in each model 1 in Table 2, I explore the colorblind hypothesis. The dependent variables (i.e., political ideology, income, perceived interviewer race) are regressed on race and racial discordance to determine the role of each of these independent variables. If the colorblind hypothesis is supported, being white and a racially discordant interview will increase the likelihood of nonresponse to the perceived race of interviewer item.

Table 2.

Multivariate Analysis of Item Nonresponse (Reported as Odds Ratios).

	Nonresponse to Political Ideology			Nonresponse to Any Income Item			Nonresponse to Perceived Interviewer Race
Variable	Model 1	Model 2	Model 3	Model 1	Model 2	Model 3	Model 1	Model 2	Model 3
White	.729	—	1.112	1.024	—	1.314	5.078***	—	4.050**
Racial discordance	.835	—	1.083	1.757*	—	2.020*	2.542*	—	2.544*
Female	—	3.822**	3.830**	—	.949	.915	—	.696	.834
Generation
Millennial	—	1.356	1.353	—	3.070***	3.140***	—	.551	.643
Middle-aged^a	—	—	—	—	—	—	—	—	—
Seniors	—	1.754	1.706	—	1.968*	1.839	—	1.266	1.065
College degree	—	.753	.701	—	1.008	.775	—	3.234**	1.359
Foreign born	—	2.051*	3.165*	—	.788	1.010	—	1.207	1.600
Intercept	.052	.012	.012	.094	.099	.060	.013	.041	.013
Listwise n	626	617	617	626	617	617	626	617	617
Nagelkerke R²	.005	.082	.082	.015	.050	.070	.123	.066	.133
Cox and Snell R²	.001	.024	.024	.008	.026	.036	.044	.024	.048

Omitted category.

p ≤ .05. **p ≤ .01. ***p ≤ .001.

In model 2, I regress the all the independent variables (i.e., political ideology, income, perceived interviewer race) on gender, generation, education, and nativity. Support for the marginalized group hypothesis would be indicated if being a female, older, less educated, and foreign born led to a higher likelihood of item nonresponse for all items. The third models simultaneously measure the effects of all the variables and allow the simultaneous testing of the marginalized group hypothesis and colorblind hypothesis. If the marginalized group hypothesis accurately describes the association, being nonwhite, a female, older, less educated, and foreign born will increase the likelihood of nonresponse on the race-related item. Alternatively, if the colorblind hypothesis is fully supported, being white, younger, and more educated will increase nonresponse rates to the perceived race of interviewer item. Finally, very strong additional support for the colorblind hypothesis would be demonstrated if a racially discordant interview context makes nonresponse to the race-related item more likely net all the other factors.

Results

The following analyses examine competing explanations for item nonresponse. The first hypothesis, drawn from the marginalized group perspective, posits that item nonresponse will be heightened among lesser educated, older, foreign-born individuals, as well as women and nonwhites. In contrast, the colorblind hypothesis, informed by the normative shift toward colorblindness, posits that those who are younger, more educated, and white will be more prone to nonresponse of race-related items. It further posits that interracial interview context will promote nonresponse to race-related items.

Bivariate Results

Figure 1 explores the association with a primary independent variable, race, and all three dependent variables to assess which theoretical perspective is accurate. According to Figure 1, there is no support for the marginalized group hypothesis. Nonresponse to the political ideology and income items are both unaffected by race. However, the results for the race-related item indicate a pattern opposite to that posited by the marginalized group hypothesis. Whites have a nonresponse rate to the race-related item that is more than five times greater than nonwhites (χ² = 23.710, p ≤ .001). This pattern indicates support for the colorblind hypothesis (hypothesis 2), which posits that whites will be more likely than nonwhites to engage in nonresponse to race-related items. These results reveal that the marginalized group hypothesis may not be useful to understanding nonresponse as much of the prior literature suggests, at least when it comes to respondent race.

The findings in Figure 2 reveal further support for the colorblind hypothesis (hypothesis 2). This figure indicates that a racially discordant interview context is associated with increased nonresponse to the race-related item. That is, having an interviewer of a race other than one’s own nearly triples nonresponse to the perceived race of interviewer item (χ² = 7.70, p ≤ .01). This finding clearly indicates that colorblind norms play a key role in the promotion of item nonresponse to this race-related social survey item. Because respondents are selecting whether to answer this question on the basis of the interviewer’s race, they ironically demonstrate a clear awareness of the very information they are purporting not to know.

Racial discordance has no effect on the political ideology item. Interestingly, the income item follows a similar pattern to the race question (though at a somewhat reduced rate). Respondents interviewed by someone of a different race were slightly more likely to not respond to an income item (χ² = 3.949, p ≤ .05) compared with those who had an interviewer of the same race. I explore some potential explanations for this finding in the discussion section.

Table 1 explores additional bivariate relationships between social background factors and item nonresponse. In all, these results provide little support for the marginalized group hypothesis, but they do provide some additional support for the colorblind hypothesis.

The political ideology item shows the most consistency with the marginalized group hypothesis, but evidence for this perspective is limited even here. Although women engage in nonresponse three times more often than men (χ² = 6.669, p ≤ .01) and foreign-born respondents about twice as often as their native-born counterparts (χ² = 5.854, p ≤ .05), the other factors to not exhibit any other significant patterns.

When it comes to nonresponse to the income item and the perceived race of interviewer item, there is virtually no evidence of the marginalized group perspective, except for the finding that seniors are more likely compared with middle-aged respondents to not respond to the income item. Instead, opposite evidence emerges suggesting that the colorblind hypothesis may provide a more valuable explanation for the race-related item.

In addition to the earlier described effects of respondent and interviewer race, education also appears to be a key explanatory factor in race-related nonresponse. That is, nonresponse to the race of interviewer item is three times more prevalent among those with a bachelor’s degree or above compared with respondents without a degree (χ² = 12.838, p ≤ .01). This is consistent with the colorblind hypothesis, which suggests that those who are more educated are more prone to engage in a worldview that minimizes the importance of race. Additionally, it is important to note that this pattern is the opposite of that expected by the marginalized group hypothesis. Sex, generation, and nativity have no significant effect on the race-related item.

In sum, the bivariate associations outlined in Figure 1, Figure 2, and Table 1 indicate very limited support for the marginalized group hypothesis. However, they show moderate support for the colorblind hypothesis. These findings demonstrate that the typical social correlates identified as promoting item nonresponse appear to operate differently for race-related items than they do for non-race-related items. Take for example the relationship between race and item nonresponse: although the marginalized group hypothesis posits that nonwhites are most prone to item nonresponse, this pattern was observed only for the political ideology item and in fact worked inversely for the race-related item. These findings call into question the marginalized group hypothesis and suggest that the colorblind hypothesis better explains item nonresponse to race-related social survey items.

Multivariate Results

Table 2 reports the final stage of analysis, a series of logistic regression results. The first model tests the colorblind hypothesis, while the second model tests the relationships predicted by marginalized group hypothesis not tested in the first model. I also include a third model that tests the effects net all independent variables.

Controlling for all the factors, sex and nativity are the only factors that have a noticeable effect on nonresponse to the political ideology item. As in the bivariate analysis, women and foreign-born respondents are less likely to respond to this item compared with men and native-born respondents, respectively. Null findings for race, generation and education indicate that the marginalized group hypothesis has only limited support.

The logistic regression results exploring nonresponse to the income items show no support for the marginalized group hypothesis. The only significant factors predicting nonresponse to this item are youth (i.e., being a millennial)⁸ and racial discordance. Having an interviewer of a different race more than doubles the likelihood of income item nonresponse (p ≤ .05). Although this finding would not be expected according to either hypothesis, it is possible that respondents are actually reading this item as being race related. If this is true, it may activate the social pressures associated with race-related items, or it may heighten concern of confirming economic status that may confirm economic stereotypes about racial groups (e.g., blacks and Latinos have lower incomes, whites have higher incomes). That is, if a stereotype is known to exist and activated by an interracial interview context, respondents may experience stereotype threat, wherein they have greater difficulty locating a precise response to the item because of the anxiety associated with a fear of reproducing a stereotype (Steele and Aronson 1995).

The logistic regression results for the perceived race of interviewer item (Table 2) indicate nearly no support for the marginalized group hypothesis and some support for the colorblind hypothesis. White respondents are more than four times more likely to not respond to the race-related item than nonwhites in this sample. Further bolstering support for the colorblind hypothesis, racial discordance more than doubles the likelihood of nonresponse to the race-related item compared with same-race interview contexts. This support for the colorblind hypothesis is somewhat limited, however, by the null findings for sex, generation, and age.

In all, the findings from the bivariate and multivariate analyses indicate some very limited support for the marginalized group hypothesis. This support is limited to only non-race-related items. In contrast, there does appear to be modest support for the colorblind hypothesis for the race-related item. For example, net of the effect of other variables, being white is associated with a more than 300 percent increase in race-related item nonresponse, a finding that is opposite the expectation of the marginalized group hypothesis. This indicates that race-related items are likely prone to different types of nonresponse than non-race-related items.

Discussion and Conclusions

In this study, I measure which social correlates are associated with two different types of item nonresponse (race-related and non-race-related). The marginalized group hypothesis posits that respondents with fewer social resources have higher rates of nonresponse, while the colorblind hypothesis suggests that a new colorblind norm pushes people to avoid responding to race-related items, especially in certain interracial contexts.

The analyses show the causes of race-related and non-race-related item nonresponse vary. Specifically, I find that white respondents are much more likely to avoid responding to a race-related survey item, an effect opposite of what is expected by the marginalized group hypothesis.

Overall, the marginalized group hypothesis was not supported by this study. A very limited number of expected effects emerged in the analysis. Although sex and nativity mattered in predicting nonresponse to the political ideology item, this perspective suggests that all questions should have been influenced by at least several of these factors. Additionally, some factors worked opposite to the expected way. For example, being white (a factor that the marginalized group perspective would suggest would promote substantive responses) led to a more than 300 percent increase in nonresponse to the perceived race of interviewer item. This finding, combined with the fact that a discordant interview doubled nonresponse, suggests that colorblindness may be a very valuable tool in understanding race-related item nonresponse in social surveys.

My findings indicate that race-related and non-race-related item nonresponse do have different social and contextual correlates. This is important for two reasons: (1) it may change how we treat data from nonresponders in our statistical analyses, and (2) it opens up a new opportunity for researchers interested in colorblind norms.

The data presented here indicate that researchers examining race-related items should carefully consider the cause of nonresponse in their data. Newer colorblind norms that push respondents to avoid talking about race in interracial interactions may be encouraging systematic nonresponse to race-related items. As such, researchers must work to use appropriate strategies to address nonresponse in our analyses and reports. In particular, we must be careful to not treat this type of nonresponse as random without evidence that it is. Doing so may result in gross underrepresentations of some of the groups whose attitudes we are most interested in examining.

Choosing to exclude nonresponders from analyses could potentially greatly distort the makeup of racial attitudes if substantial numbers of nonresponders exist and are not randomly distributed. Take for example a 2013 poll conducted by the Pew Research Center exploring racial attitudes across the United States. According to my own analysis of these data, 38.1 percent of whites report that they believe police treat blacks unfairly compared with whites, while 49.4 percent report believing that blacks are treated equally in this area of life (Pew Research Center 2013). The remaining 12.5 percent of whites in the sample either refused to answer the question or said they did not know (Pew Research Center 2013). If researchers analyzing these data assume that nonresponse is randomly distributed and exclude these cases from their analysis, their analysis would show that 43.5 percent of respondents believe that police treat blacks unfairly and 56.5 percent think they are treated fairly (Pew Research Center 2013). On the other hand, if the researchers use imputation to predict responses, they could theoretically find great shifts in the attitudinal makeup of the public. For example, in this case, attitudes could theoretically show splits that vary between 38.1 percent and 61.9 percent and between 51.6 percent and 49.4 percent. These two possibilities are radically different characterizations of white racial attitudes. Furthermore, conducting multivariate analyses using these items would present additional challenges. Unfortunately, without further research on the distribution of nonresponse and efforts to understand its causes, it would be exceedingly difficult to provide more “accurate” reads of white racial attitudes.

Although this study suggests that researchers should be more careful about race-related item nonresponse as they produce estimates of attitudes, I also argue that researchers interested in colorblindness may want to use nonresponse analysis as a new research tool. This type of analysis offers race scholars a mechanism to further explore and understand the emerging social norms of colorblindness and racial avoidance. Some social researchers purport that social surveys fail to provide adequate nuance to measure modern racial phenomena because respondents merely seek to express the most socially appropriate responses (Bonilla-Silva 2003). My research, however, demonstrates the possibility that social survey data can be used to creatively capture the expression of colorblindness and racial apathy. Using this methodological approach offers an opportunity to learn more about how respondents use colorblindness with the added benefits of survey research (e.g., the ability to capture trends, measure prevalence and change over time and explore correlates). This type of research presents opportunities to use survey data to learn more about the types of people who engage colorblindness in the contexts that promote it.

In particular, one of my findings presents a particularly interesting case for reconsidering the ways we believe different groups are affected by colorblind norms. Although the finding does not achieve statistical significance, I observe that the youngest cohort of respondents is actually least likely to engage in nonresponse to the perceived race of interviewer item. This goes against much of the previous research on colorblindness and my own hypothesis and may be providing an important clue about the ways youth engage racial norms. Similarly, in her study of economically advantaged white youth, Margaret Hagerman (2015) found that despite extended pleas and lessons from their parents, young whites are actually quite comfortable and open in discussing race, often in very taboo ways. Her qualitative finding coupled with the low rates of nonresponse among youth in this study make a strong suggestion that youth, particularly white youth, may not be engaging in colorblindness in the ways researchers typically assume. The distribution of nonresponse may hold additional important clues when exploring further.

The findings I present here also share important information about how racial norms continue to shape our broader social world. Specifically, by using a race-of-interviewer experiment (i.e., randomization of interviewer race to find patterned difference) to examine racial avoidance in a systematic fashion, I show that engagement of colorblindness varies by the racial context of a given situation. Because respondents offered responses less often when their interviewers were of a different race (indicating that they actually were hazarding a guess inside their minds and choosing not to share it with different-race interviewers), this research provides an important example of how individuals may be using avoidance as a routine technique in interracial interactions. Among this sample of more than 600 people, respondents were much more likely to avoid talking about race when they were in interracial contexts compared with same-race contexts. This finding bolsters those of qualitative researchers and researchers working in laboratory settings who have described similar patterns.

These conclusions also have important implications for researchers using data drawn from social surveys that ask about racial issues. If response rates to particular items systematically vary by group to this extent (e.g., a 300 percent increase for whites compared with blacks and Latinos, as found in this analysis), great care must be taken when generalizing from findings drawn from these items. Further research should be conducted to see if similar systematic patterns in other studies and using other race-related survey items. If similar findings are drawn from other data, this presents strong evidence that nonresponse is a useful tool for researchers interested in normative behaviors related to race. That is, by understanding predictors of race-related item nonresponse in the social surveys, we can potentially better understand who is engaging in racial avoidance and why. Understanding this is crucial to understanding our modern racial landscape.

Study Limitations

My findings are limited by a relatively small sample size, a focus on a specific geographic location (the city of Chicago), and a limited number of dependent variables. The small sample size in particular limits the statistical power of the results. I recommend the development of additional studies to confirm these patterns across bigger samples and more varied question types. The geographic specificity poses a minimal problem relative to the sample size. Because much of the research on colorblindness draws on respondents who live in suburban, predominately white areas (Apfelbaum, Pauker, et al. 2008; Carr 1997; Forman and Lewis 2006; Perry 2001), I expect that my analysis likely underestimates the effect of colorblindness on race-related item nonresponse among whites. Still, more work needs to be done to expand knowledge in this area.

Future Research Directions

Because this is an exploratory study, there are a host of follow-up studies that could and should be conducted that will confirm and build on these findings. Rates of nonresponse are growing, with some studies finding item nonresponse to similar race-related items as high as 25 percent (Blumenthal 2008). As such, there are many opportunities to further explore this phenomenon. I describe two potential directions for future research below.

Researchers could further examine the social aspects of nonresponse by exploring the ways respondents engage different types of survey instruments. Because I use only one mode of administration in this study (i.e., a telephone survey), it is unclear what role the mode of administration (e.g., in-person, telephone, mail-in, online) plays in producing nonresponse to race-related social survey items. As researchers increasingly take advantage of online platforms to collect data, it is important to understand how respondents interact with these instruments differently than more traditional survey formats. Previous work has shown that respondents express their racial attitudes differently on the basis of the level of privacy provided by the survey instrument (Krysan 1998). By using an experimental survey design that randomizes mode of administration, researchers could clarify whether nonresponse is being used to conceal responses that would otherwise be expressed in more private contexts.

Second, I recommend that researchers explore whether there are any identifiable links between racial prejudice and nonresponse. That is, are those respondents who exhibit more prejudice more or less likely to not respond to race-related social survey items? More specifically, might nonresponders be concealing their prejudice through nonresponse? Or are they less prejudiced and merely more attuned to the social desirability of talking about race? The research presented here does not address the role of prejudice in nonresponse, but many readers will likely be left with questions about potential connections.

A host of creative mixed-method approaches using surveys and in-depth interviews could produce some very important insights into modern racial attitudes. Given that nonresponse is notoriously difficult to study, researchers must be very flexible and innovative in their future research designs. These two suggestions are merely starting points for what could potentially become a thriving area of research within racial and ethnic relations research that would contribute greatly to our field.

Footnotes

Appendix

Table A1.

Demographic Distributions by Age, Race/Ethnicity, and Sex for the 2000 Census and Sampled Cases, City of Chicago.

Demographic Characteristics	2000 Census	Sample Distributions with Selection Weights
Age (years)
18–34	40.1%	25.3%
35–44	20.3%	22.7%
45–59	20.9%	30.6%
≥60	18.7%	21.4%
Race/ethnicity
Non-Hispanic white	36.7%	38.3%
Non-Hispanic black	33.8%	35.2%
Hispanic/Latino	22.8%	15.2%
Other	6.7%	11.3%
Sex
Male	47.7%	41.3%
Female	52.3%	58.7%

Source: Chicago Area Study methodological report.

Acknowledgements

I thank Tyrone Forman, Margaret Hagerman, and Michelle Manno for their feedback on an earlier draft of this article, as well as Maria Krysan, Allyson Holbrook, and Timothy Johnson for granting access to the 2008 CAS data. Writing of this article was supported by a fellowship from the Laney Graduate School at Emory University.

Notes

Author Biography

Elizabeth C. Alexander is a PhD candidate in the Department of Sociology at Emory University.

References

Adua

Lazarus

Sharp

Jeff S.

2010. “Examining Survey Participation and Response Quality: The Significance of Topic Salience and Incentives.” Survey Methodology 36(1):95–109.

Alvik

Astrid

Haldorsen

Tor

Lindemann

Rolf

. 2005. “Consistency of Reported Alcohol Use by Pregnant Women: Anonymous versus Confidential Questionnaires with Item Nonresponse Differences.” Alcoholism: Clinical and Experimental Research 29(8):1444–49.

Apfelbaum

Evan P.

Pauker

Kristin

Ambady

Nalini

Sommers

Samuel R.

Norton

Michael I.

2008. “Learning (Not) to Talk about Race: When Older Children Underperform in Social Categorization.” Developmental Psychology 44(5):1513–18.

Apfelbaum

Evan P.

Sommers

Samuel R.

Norton

Michael I.

2008. “Seeing Race and Seeming Racist? Evaluating Strategic Colorblindness in Social Interaction.” Journal of Personality and Social Psychology 95(4):918–32.

Berinsky

Adam J.

1999. “The Two Faces of Public Opinion.” American Journal of Political Science 43(4):1209–30.

Berinsky

Adam J.

2002a. “Political Context and the Survey Response: The Dynamics of Racial Policy Opinion.” Journal of Politics 64(2):567–84.

Berinsky

Adam J.

2002b. “Silent Voices: Social Welfare Policy Opinions and Political Equality in America.” American Journal of Political Science 46(2):276–87.

Blom

Annelies G.

de Leeuw

Edith D.

Hox

Joop J.

2011. “Interviewer Effects on Nonresponse in the European Social Survey.” Journal of Official Statistics 27(2):359–77.

Blumenthal

Mark

. 2008. “Bradley-Wilder: New Data on Interviewer Race.” Retrieved March 17, 2016 (http://www.pollster.com/blogs/bradleywilder_new_data_on_inte.php?nr=1).

10.

Bobo

Lawrence D.

Kluegel

James R.

Smith

Ryan A.

1997. “Laissez-faire Racism: The Crystallization of a Kinder, Gentler, Antiblack Ideology.” Pp. 15–42 in Racial Attitudes in the 1990s: Continuity and Change, edited by Tuch

Steven A.

Martin

Jack K.

New York: Praeger.

11.

Bobo

Lawrence D.

Zubrinsky

Camille L.

1996. “Attitudes on Residential Integration: Perceived Status Differences, Mere In-group Preference, or Racial Prejudice?” Social Forces 74(3):883–909.

12.

Bobo

Lawrence D.

Charles

Camille Z.

Krysan

Maria

Simmons

Alicia D.

Fredrickson

George M.

2012. “The Real Record on Racial Attitudes.” Pp. 38–83 in Social Trends in American Life: Findings from the General Social Survey Since 1972. Princeton, NJ: Princeton University Press.

13.

Bonilla-Silva

Eduardo

. 2001. White Supremacy and Racism in the Post–Civil Rights Era. Boulder, CO: Lynne Reinner.

14.

Bonilla-Silva

Eduardo

. 2003. Racism without Racists: Color-blind Racism and the Persistence of Racial Inequality in the United States. Lanham, MD: Rowman & Littlefield.

15.

Bonilla-Silva

Eduardo

2015. “More Than Prejudice: Restatement, Reflections and New Directions in Critical Race Theory.” Sociology of Race and Ethnicity 1(1):73–87.

16.

Bonilla-Silva

Eduardo

Lewis

Amanda

Embrick

David G.

2004. “‘I Did Not Get That Job Because of a Black Man . . .’: The Story Lines and Testimonies of Color-blind Racism.” Sociological Forum 19(4): 555–81.

17.

Borg

Ingwer

Braun

Michael

Baumgaertner

Miriam K.

2008. “Attitudes of Demographic Item Non-respondents in Employee Surveys.” International Journal of Manpower 29(2):146–60.

18.

Brown

Anna

. 2015. “The Unique Challenges of Surveying U.S. Latinos.” Retrieved December 30, 2016 (http://www.pewresearch.org/2015/11/12/the-unique-challenges-of-surveying-u-s-latinos/).

19.

Candido

Elisa

Kurdyak

Paul

Alter

David A.

2011. “Item Nonresponse to Psychosocial Questionnaires was Associated with Higher Mortality after Acute Myocardial Infarction.” Journal of Clinical Epidemiology 64(2):213–22.

20.

Carr

Leslie

. 1997. Colorblind Racism. Thousand Oaks, CA: Sage.

21.

Chen

Pu-Shih Daniel

. 2011. “Finding Quality Responses: The Problem of Low-quality Survey Responses and Its Impact on Accountability Measures.” Research in Higher Education 52(7):659–74.

22.

Colsher

Patricia L.

Wallace

Robert B.

1989. “Data Quality and Age: Health and Psychobehavioral Correlates of Item Nonresponse and Inconsistent Responses.” Journal of Gerontology 44(2):45–52.

23.

Elliott

Marc N.

Edwards

Carol

Angeles

January

Hambarsoomians

Katrin

Hays

Ron D.

2005. “Patterns of Unit and Item Nonresponse in the CAHPS^® Hospital Survey.” Health Services Research 40(6p2):2096–2119.

24.

Forman

Tyrone A.

2004. “Color-blind Racism and Racial Indifference: The Role of Racial Apathy in Facilitating Enduring Inequalities.” Pp. 43–66 in The Changing Terrain of Race and Ethnicity, edited by Krysan

Maria

Lewis

Amanda E.

New York: Russell Sage.

25.

Forman

Tyrone A.

Lewis

Amanda E.

2006. “Racial Apathy and Hurricane Katrina: The Social Anatomy of Prejudice in the Post–Civil Rights Era.” Du Bois Review: Social Science Research on Race 3(1):175–202.

26.

Franses

Hans

Geluk

Irma

Van Homelen

Paul

. 1999. “Modeling Item Nonresponse in Questionnaires.” Quality & Quantity 33(2):203.

27.

Gallagher

Charles A.

1997. “White Racial Formation: Into the Twenty-first Century.” Pp. 6–11 in Critical White Studies: Looking beyond the Mirror, edited by Delgado

Richard

Stefancic

Jean

. Philadelphia, PA: Temple University Press.

28.

Graf

Ingrid

Retzer

Karen

. 2008. “2008 Chicago Area Study: Final Methodological Report.” Chicago: Survey Research Laboratory, University of Illinois Chicago and University of Illinois Urbana-Champaign, The College of Urban Planning and Public Affairs.

29.

Groves

Robert M.

Fowler

Floyd J.

Jr. Couper

Mick P.

Lepkowski

James M.

Singer

Eleanor

Tourangeau

Roger

. 2008. Survey Methodology. Hoboken, NJ: John Wiley.

30.

Hagerman

Margaret Ann

. 2015. “Reproducing and Reworking Colorblind Racial Ideology Acknowledging Children’s Agency in the White Habitus.” Sociology of Race and Ethnicity 2(1):58–71.

31.

Hartmann

Douglas

Gerteis

Joseph

Croll

Paul R.

2009. “An Empirical Assessment of Whiteness Theory: Hidden from How Many?” Social Problems 56(3):403–24.

32.

Holbrook

Allyson

Johnson

Timothy

Krysan

Maria

. 2008. “2008 Chicago Area Study: Racial-ethnic Differences in Issue-specific Political Participation and Race of Interviewer Effects.” Chicago: Survey Research Laboratory, The College of Urban Planning and Public Affairs and the Department of Sociology, University of Illinois at Chicago.

33.

Jackman

Mary R.

Muha

Michael J.

1984. “Education and Intergroup Attitudes: Moral Enlightenment, Superficial Democratic Commitment, or Ideological Refinement?” American Sociological Review 49(6): 751–69.

34.

Klein

David J.

Elliott

Marc N.

Haviland

Amelia M.

Saliba

Debra

Burkhart

Edwards

Carol

Zaslavsky

Alan M.

2011. “Understanding Nonresponse to the 2007 Medicare CAHPS Survey.” Gerontologist 51(6):843–55.

35.

Krysan

Maria

. 1998. “Privacy and the Expression of White Racial Attitudes: A Comparison across Three Contexts.” Public Opinion Quarterly 62(4):506–44.

36.

Kupek

Emil

. 1998. “Determinants of Item Nonresponse in a Large National Sex Survey.” Archives of Sexual Behavior 27(6):581–94.

37.

Lewis

Amanda E.

2004. “‘What Group?’ Studying Whites and Whiteness in the Era of Color-blindness.” Sociological Theory 22(4):623–46.

38.

Little

Roderick J. A.

Rubin

Donald B.

2014. Statistical Analysis with Missing Data. Hoboken, NJ: John Wiley.

39.

Luttberg

Norman R.

Gant

Michael M.

1985. “The Failure of Liberal/Conservative Ideology as a Cognitive Structure.” Public Opinion Quarterly 49(1):80–93.

40.

McKinney

Karyn D.

2003. “‘I Feel “Whiteness” When I Hear People Blaming Whites’: Whiteness as Cultural Victimization.” Race and Society 6(1):39–55.

41.

Norton

Michael I.

Sommers

Samuel R.

Apfelbaum

Evan P.

Pura

Natassia

Ariely

Dan

. 2006. “Color Blindness and Interracial Interaction.” Psychological Science 17(11):949–53.

42.

Perry

Pamela

. 2001. “White Means Never Having to Say You’re Ethnic: White Youth and the Construction of ‘Cultureless’ Identities.” Journal of Contemporary Ethnography 30(1):56–91.

43.

Pew Research Center. “2013 Social and Demographic Trends Project: 2013 Martin Luther King, Jr. Race Survey (August 1–11)” [Computer file]. Retrieved January 4, 2016 (http://www.pewsocialtrends.org/2014/11/19/martin-luther-king-jr-race-survey/).

44.

Pickery

Jan

Loosveldt

Geert

Carton

Ann

. 2001. “The Effects of Interviewer and Respondent Characteristics on Response Behavior in Panel Surveys—A Multilevel Approach.” Sociological Methods & Research 29(4):509–23.

45.

Pollock

Mica

. 2005. Colormute: Race Talk Dilemmas in an American School. Princeton, NJ: Princeton University Press.

46.

Schafer

Joseph L.

Graham

John W.

2002. “Missing Data: Our View of the State of the Art.” Psychological Methods 7(2):147.

47.

Schuman

Howard

Steeh

Charlotte

Bobo

Lawrence D.

Krysan

Maria

. 1997. Racial Attitudes in America: Trends and Interpretations. Cambridge, MA: Harvard University Press.

48.

Steele

Claude M.

Aronson

Joshua

. 1995. “Stereotype Threat and the Intellectual Test Performance of African Americans.” Journal of Personality and Social Psychology 69(5):797–811.

49.

Stocké

Volker

. 2006. “Attitudes toward Surveys, Attitude Accessibility and the Effect on Respondents’ Susceptibility to Nonresponse.” Quality & Quantity 40(2):259–88.

50.

Tomaskovic-Devey

Donald

Leiter

Jeffrey

Thompson

Shealy

. 1995. “Item Nonresponse in Organizational Surveys.” Sociological Methodology 25(1):77–110.

51.

Tourangeau

Roger

Groves

Robert M.

Redline

Cleo D.

2010. “Sensitive Topics and Reluctant Respondents.” Public Opinion Quarterly 74(3):413–32.

52.

Watkins

Adam M.

Melde

Chris

. 2007. “The Effect of Self-control on Unit and Item Nonresponse in an Adolescent Sample.” Journal of Research in Crime and Delinquency 44(3):267–94.

53.

Wiederman

Michael W.

1993. “Demographic and Sexual Characteristics of Nonresponders to Sexual Experience Items in a National Survey.” Journal of Sex Research 30(1):27–35.

54.

Wiederman

Michael W.

Weis

D. L.

Allgeier

E. R.

1994. “The Effect of Question Preface on Response Rates to a Telephone Survey of Sexual Experience.” Archives of Sexual Behavior 23(2):203–15.