[The following is a supplement to a New York Times Op-Ed by Christopher Federico, Howard Lavine, and Christopher Johnston, appearing on Tuesday, September 11th, 2012]
For the very first time in its history, the American National Election Study in 2008 included in its design a measure of implicit racial attitudes. Greenwald and Banaji (1995; page 8 ) define an implicit attitude in the following way:
Implicit attitudes are introspectively unidentified (or inaccurately identified) traces of past experience that mediate favorable or unfavorable feeling, thought, or action toward social objects.
Implicit attitudes are the traces of favorability or unfavorability stored in memory as a result of past experiences with a given object. While scholars have utilized diverse methods to assess implicit attitudes, the 2008 ANES utilizes one in particular: the Affect Misattribution Procedure (hereafter AMP).
The logic of the AMP is that people will misattribute feelings arising from an unconscious source to an object of conscious evaluation. To be more concrete, in the AMP included in the 2008 ANES, respondents were seated at a computer, and were told that their task was to judge whether a pictograph (a Chinese character) presented on the screen was “pleasant” or “unpleasant.” Prior to the appearance of each pictograph, respondents were exposed to an image of either a white male face or a black male face. Importantly, the face images appeared on the screen for a very brief interval of time (~75ms), making the images themselves barely perceptible. The theory is that respondents with predominantly negative feelings toward blacks (whites) will be more likely to state that pictographs following a black (white) face are unpleasant, and vice versa. Put another way, the implicit attitude “activated” by the brief picture flash is attributed to the pictograph, because the individual is unaware of the true source of the experienced feeling of positivity or negativity.
In the 2008 ANES, each respondent completed 48 trials, 24 with white male faces, and 24 with black male faces. We first calculated the proportion of “pleasant” responses for each person across all trials. We then calculated the proportion positive for the black trials. We then subtracted the proportion for black trials from the proportion for all trials. This gives a measure of implicit relative black negativity. It also corrects for individual differences in the tendency to respond with a disproportionate number of pleasant or unpleasant responses. The distribution for white respondents in the 2008 ANES is below.
As would be expected, there is a negative implicit bias toward whites within the white sample.
More interesting, for our purposes at least, is the association between implicit black negativity and preferences over economic policy – in this case, healthcare reform. The 2008 ANES asked respondents about their preferences over government-provided health insurance for all Americans. Respondents recorded their preferences on a 1 to 7 scale, with higher values indicating greater opposition to government health insurance. We rescaled this variable to range from zero to one for easier interpretation. Zero is maximum support for government-provided health insurance, and one is maximum opposition. Moreover, we distinguish between respondents with and without a college degree. As per the NYT article, we expected that the association between implicit relative black negativity and opposition to government healthcare would be much larger for college-educated whites. The key results, taken from an OLS model with controls for age, gender, income, and employment status, are shown in the table and figure below.
We can interpret these results as follows. For whites without a college degree, there is no relationship at all between implicit negativity and healthcare attitudes (the blue line in the figure). For college-educated whites, by contrast, the association is quite strong. Moving from the 5th to the 95th percentile of negativity, we expect opposition to government healthcare to increase from about .35 to about .60, all else equal.
There is an obvious potential here for the OLS regression to be a misleading representation of the underlying data (given the distribution of the implicit measure). For those who are interested, we thus provide two additional graphs below, the first for whites without a college degree, and the second for those with a college degree. Each is a lowess (i.e., locally weighted) fit to the bivariate data, constraining the range of the implicit measure to fall between 5% and 95%. The key conclusion is maintained with respect to the moderating role of education, with the caveat that the relationship appears to be non-linear for the college-educated.