In the next few weeks we will start rolling out access to a large database of projections we have been compiling: 200+ questions, monthly, modeled to 1,000+ demographic combinations, at all 435 house districts, 50 states, 1 nation. Included in this is 10 “value-frame” or psychometrics (along with 11 issues clusters, 2 economic clusters, approval, vote choices, etc.), that attempt to map people’s underlying beliefs. Our hope is that these values-frames will help us better understand the electorate and the decisions they make. For instance, our racial resentment score works well in predicting voting above and beyond traditional data sources.

These value-frames are constructed as a combination of a bunch of questions, voter file data, and additional behavioral data. The idea is that you cannot accurately assess something like racial resentment, by asking people directly (and it is always better to have a few questions to minimize any measurement error from a single question). Respondents may not understand the terminology and it is socially undesirable to be racist (well, in some parts of the US) so many people with high racial resentment would lie and say they were not. Thus, we ask a series of questions that help us model latent racial resentment, without asking about it directly.

Here is our modeled values for racial resentment. On the left we show racial resentment by geographic type: racial resentment increases as you move from the densest urban area to the most sparsely populated rural areas. Of course, this is correlated with decreases in the percentage of minorities, who have much, much lower rates of racial resentment than White people. So, we can also pull out White people urbanicity: Dense Urban (54% on our racial resentment scale), Sparse Urban (64%), Dense Suburban (71%), Sparse Suburban (74%), Dense Rural (77%), and Sparse Rural (78%). Same monotonic increase, but a higher base-rate. On the right we show the wide gulf between Republicans and Democrats.


Figure 1: Images PredictWise’s forthcoming Insight Engine

It may surprise you that Democrats are still at 46 percent and dense urban areas at 42 percent: the scores are providing a mean for that group, where everyone in that group is scored from 0 to 100. There is no attempt to normalize a mean of 50, while some of our scores have means near 50, some are higher and some are lower. 50 represents the idealized moderate response pattern for the cluster of questions we use to create the score. So you could imagine a person getting a score of 50 if they answer half the questions in one direction and half in the other, or answered all questions in the middle.

We thought it would be helpful to show the questions that compromise the racial resentment score, and you may be surprised by some of the answers. For instance: “Most racial minorities who receive welfare could get along without it if they tried.” Only 22 percent of Republicans disagreed with this statement, but only 40 percent of Democrats did as well. A less surprising question response is to: “Overall, the police have displayed racism when dealing with racial minorities.” where only 33 percent of Republicans agree, but 66 percent of Democrats.


Figure 2: Images PredictWise’s forthcoming Insight Engine

Americans have a lot of racial resentment. President Trump tapped into the same vein Republicans have been using for decades and Democrats before that. Understanding exactly where it is strongest (and it nuances) among the hundreds of thousands of demographic and geographic clusters in this country is critical to both mapping public opinion and understanding how it can change.