I recently read an interesting piece by Kerem Ozan Kalkan at the Monkey Cage regarding anti-Muslim resentment among the American public, and how it continues to be the greatest for any social group. Using various pieces of evidence from survey data and past literature, Kalkan concludes that there are three driving forces behind these attitudes toward Muslims: those that relate to ethnocentrism, partisanship, and cultural portrayals. The first two in this group were of particular interest to me. As I will later show, partisanship is a good predictor of anti-Muslim prejudice. Framing this relationship as the result of elite rhetoric and cues makes it very clear how this resentment arose. Specifically, Kalkan references John Zaller’s seminal work on this subject for his claim, explaining how negative statements about Muslims made by Republican elites has shaped the belief system of right-wing public attitudes on this subject. Importantly, more recent elite rhetoric from figures like Donald Trump marks a shift from less prejudiced cues from past Republican elites, such as George Bush, who had a much more positive tone about Islam and separated the religion from terrorism. The change makes for an important divergence in the flow of elite discourse and information reaching the public. Zaller pointed to this phenomenon as a mechanism that causes public attitudes on an issue split; accordingly, Republicans currently feel much more coldly toward Muslims than Democrats do.
In explaining ethnocentrism as a factor, Kalkan introduces an interesting dimension to it (emphasis mine):
- My research shows that ethnocentrism–generalized dislike toward all out-groups–plays a key role in determining people’s attitudes toward Muslims. Those who feel cold toward blacks, Latinos, Jews, homosexuals, feminists and other minority groups also feel cold toward Muslims. In other words, anti-Muslim prejudice is strongly related to racism, antigay attitudes, sexism and anti-Semitism.
The claim here is that animus toward different marginalized, minority out-groups is interconnected and thus drives Muslim resentment. This made me wonder what forms this dynamic takes–specifically, what kinds of prejudice unrelated to perceptions toward Muslims drives anti-Muslim resentment. To address this question, I use the most recent quality survey data I could find in the 2016 ANES Pilot. The dependent variable in this analysis will be a 0-100 feeling thermometer toward Muslims, which I recoded to make higher values associated with less favorable feelings toward Muslims–a good proxy for prejudice toward this religious group. Using the entire sample in the pilot study (not just whites, for example, as has been done in other studies), I ran three different regression specifications as seen in the table at the very bottom of the article (I’ll be referring to this table throughout this post).
- Regressing Muslim feeling thermometer on only demographic and key political variables
- Same as model 1, but with five other feeling thermometers (reverse coded again) added in as predictors
- Same as model 1, but with five other more general indicators for outgroup prejudices (and excluding feeling thermometers)
Demographics and Key Political Variables
Let’s start with the “demographic”-only model to lay down the foundations for anti-Muslim prejudice, which can be found in column 1 in the below table. For the racial variable, there’s one significant predictor: relative to the baseline group (whites), blacks are 8.4 points less prejudiced toward Muslims using the feeling thermometer scale, an effect that holds even when using party identification as a control (an important thing to mention given the highly correlated nature between race and party). Women are also 4.3 points less prejudiced toward Muslims relative to men, who make up the baseline group. Greater effects appear for effects of education. Using respondents with a high school degree or less as the baseline group in the regression, respondents with some college education are 4.2 points less prejudiced on the thermometer scale, while those have attained a college degree or more are much less prejudiced at 10.3 thermometer points less than high school or less respondents. This pairs well with existing evidence linking the education gap in vote choice in this past election–specifically among whites–to racial resentment. Age is also a significant positive predictor of Muslim prejudice, an unsurprising result given that older people tend to espouse more conservative viewpoints that often encompass this type of prejudice. Partisanship and ideology–measured on a 1-7 scale with 7 as most Republican and a 1-5 scale with 5 as most conservative, respectively–are also positively related with Muslim prejudice at a statistically significant level.
Other Outgroup Feeling Thermometers
The significant effects among these demographic and key political identifier variables hold for the most part when moving to the second and third specifications. In the second model, I test the effects of other feeling thermometers. This addresses whether prejudice toward other groups–more negative feelings toward blacks, Hispanics, gays/lesbians, feminists, and transgender people–affects feelings toward a (primarily) distinct outgroup in Muslims. As seen in the second column of the table below, all but one of these five new prejudice variables significantly increases prejudice toward Muslims. For example, a one point increase in prejudice felt toward blacks results in a statistically significant 0.18 point increase in prejudice felt toward Muslims. The coefficient sizes remain relatively small across the board (0.38 for Hispanics, 0.05 for feminists, and 0.21 for transgender people), but the effects are still there and pass for statistical significance.
However, I do suspect it might be problematic to use these new predictors in the model explaining variation in Muslim prejudice. Not only are these independent variable feeling thermometers measured on the same scale as the dependent variable Muslim feeling thermometer, but the questions that have respondents produce these ratings are asked in close proximity of each other within the survey itself. To some degree, respondents likely answer these similarly designed ratings in similar ways, with perhaps a “straightlining” effect coming into play as respondents are placed through a battery of about 20 consecutive thermometer questions. The way responses are recorded on these thermometer ratings are also likely similar, as ratings bunch up at certain values, such as at increments of 10, on a 0-100 scale that supposedly allows for more variation. Model 2 is not meaningless, and it still certainly lends credence to the idea that feelings of prejudice toward social groups like Hispanics and even feminists may affect resentment of Muslims. But it’s still important to keep in mind the aforementioned dynamics related to the survey-taking process, which might partly explain why the R-squared–the predictive power of the model–is about double that of the other two specifications in columns 1 and 3 (i.e. it might have more to do with the relation between the survey devices used in feeling thermometers rather than what the variables here are intending to measure).
Alternative Measures of Prejudice
In order to provide an alternative to these thermometer predictors, I use different measurements for social group prejudice–unrelated to Muslims–in the model in column 3 below. One of these is the popularly used black racial resentment scale, which I calculate as a composite measure of animus toward blacks by averaging across four different questions intended to measure this concept from the ANES pilot dataset. I recode responses so that before averaging across the four questions, the 1-5 value attached to each response goes from least to most racially resentful. Here are the questions that I use:
I also add four more independent variables in this model that measure the level of acknowledgement of discrimination against Hispanics, gays/lesbians, women, and transgender people existing in the U.S. today. Responses to these questions range from a value of 1 for highest amount of acknowledgment of discrimination to a value of 5 for the lowest amount. On its face, acknowledgment of discrimination does not seem to directly indicate prejudice, and might instead appear like an objective assessment of the conditions that these groups face. However, research from Christopher DeSante and Candis Smith, which focused on gauging racial resentment among whites, demonstrates that acknowledgment of discrimination (racism) can captures a dimension of prejudice. In their concluding thoughts of their paper, they write the following (emphasis mine):
- “…the question is not just between principles or prejudice. Instead, the questions researchers of Whites’ racial attitudes must now tackle are how aware Whites might be regarding racial inequality and how much they care about those inequalities.”
As it relates to views on racism, self-described awareness about inequalities such as discrimination that affect outgroups serves as an important proxy for prejudice, which is especially useful in survey settings where social desirability bias often moves respondents to conceal racially resentment beliefs. This finding comes from a study that focuses on attitudes related to race and questions that primarily involve conditions of blacks. While obviously not directly applicable to the acknowledgment variables I use in model 3, I believe this same theoretical insight is transferable when gauging prejudice toward outgroups and marginalized people beyond just blacks, such as women, gays/lesbians, transgender people, and Hispanics. In that sense, I connect the acknowledgement questions to indications of social group prejudice. Given my claim here and for simplicity’s sake, I will just say “prejudice” toward a certain social group in the below explanation, instead of writing out “the lack of acknowledgment” each time.
Out of these four variables concerning acknowledgment of discrimination against outgroups, two attain a level of significance. One of them proves an unexpected result, as prejudice against gays/lesbians is negatively associated with prejudice against Muslims, though only at the p
A much larger effect on Muslim prejudice materializes for the other new predictor in this specification, and one unrelated to discrimination acknowledgment or feeling thermometers: black racial resentment. For every one point increase in the composite metric for racial resentment I created, an 8.3 increase in prejudice toward Muslims results. That effect attains the highest level of significance, and is the largest coefficient of any in model 3, including the demographic/key political variable ones.
Concluding Thoughts
In sum, it’s clear that different prejudices shown toward different social groups–often marginalized or minority ones–are connected. Whether using attitudinal measures built up from feeling thermometers or Likert scale questions, prejudice toward blacks, Hispanics, feminists, and transgender people all have positive relationships with prejudice toward Muslims, and at statistically significant levels. The strongest effect (as measured by coefficient size) comes from racial resentment toward blacks on anti-Muslim prejudice. This all goes to say that while Muslims are seen least favorably of all the prominent social groups in this country, animus toward them goes hand in hand with prejudice expressed toward other vulnerable social groups.
Here is the regression table with the three models that I have been referring to throughout this post:
| Dependent variable: | |||
| Muslim Prejudice (Feeling Thermometer) | |||
| (1) | (2) | (3) | |
| Black Race (Relative to White) | -8.439*** | -6.190*** | -1.463 |
| (2.759) | (2.256) | (2.751) | |
| Hispanic Race (Relative to White) | -2.916 | 4.826** | -2.260 |
| (2.853) | (2.374) | (2.768) | |
| Other Race (Relative to White) | 1.831 | 2.037 | 1.457 |
| (3.644) | (2.858) | (3.497) | |
| Female (Relative to Male) | -4.306*** | -0.165 | -4.799*** |
| (1.626) | (1.296) | (1.574) | |
| Some College (Relative to High School or Less) | -4.160** | 2.196 | -2.923 |
| (1.969) | (1.564) | (1.889) | |
| College Degree or More (Relative to High School or Less) | -10.325*** | -4.291*** | -7.030*** |
| (1.999) | (1.590) | (1.942) | |
| Age | 0.168*** | 0.229*** | 0.126*** |
| (0.049) | (0.039) | (0.047) | |
| Partisanship (Least to Most Republican) | 2.836*** | 1.867*** | 2.057*** |
| (0.490) | (0.394) | (0.489) | |
| Ideology (Least to Most Conservative) | 7.258*** | 1.404 | 3.445*** |
| (1.009) | (0.854) | (1.036) | |
| Black Prejudice (Feeling Thermometer) | 0.181*** | ||
| (0.038) | |||
| Hispanic Prejudice (Feeling Thermometer) | 0.377*** | ||
| (0.037) | |||
| Gay/Lesbian Prejudice (Feeling Thermometer) | 0.044 | ||
| (0.035) | |||
| Feminist Prejudice (Feeling Thermometer) | 0.058** | ||
| (0.029) | |||
| Transgender Prejudice (Feeling Thermometer) | 0.208*** | ||
| (0.037) | |||
| Hispanic Discrimination Lack of Acknowledgement | -0.674 | ||
| (1.038) | |||
| Gays/Lesbians Discrimination Lack of Acknowledgement | -2.120* | ||
| (1.243) | |||
| Women Discrimination Lack of Acknowledgement | -0.103 | ||
| (1.032) | |||
| Transgender Discrimination Lack of Acknowledgement | 2.542** | ||
| (1.129) | |||
| Black Racial Resentment | 8.307*** | ||
| (0.883) | |||
| Constant | 21.241*** | -1.226 | 9.427** |
| (3.913) | (3.234) | (4.609) | |
| Observations | 1,061 | 1,061 | 1,059 |
| R2 | 0.226 | 0.528 | 0.299 |
| Adjusted R2 | 0.220 | 0.521 | 0.289 |
| Residual Std. Error | 26.095 (df = 1051) | 20.436 (df = 1046) | 24.929 (df = 1044) |
| F Statistic | 34.178*** (df = 9; 1051) | 83.517*** (df = 14; 1046) | 31.754*** (df = 14; 1044) |
| Note: | *p<0.1; **p<0.05; ***p<0.01 | ||
Agadjanian Politics 