The Republican fluency with science

The Audacious Epigone has a post up, Republicans are more scientifically literate than Democrats or independents are, where he reviews pro vs. anti-science attitude by party in the General Social Survey. He concludes that in fact Republicans are more scientifically literate across the issues than Democrats. Jason Malloy saw this trend four years ago in the GSS, and to some extent so have I. One point to keep in mind is that a few specific politicized scientific issues are very much the outliers in exhibiting tight partisan valences in opinion.

So another question: are conservatives more scientifically literate than liberals? If scientific literacy correlates with being Republican, and being Republican correlates with being conservative, shouldn’t scientific literacy correlate with being conservative? Not necessarily. Such correlations are not transitive. Generally what I’ve seen in the survey data is that Republicans tend to be more pro-science than conservatives. I think part of it is the voting by economic position which has become less stark in our culture, but still remains a force. In any case, my table to accompany AE’s is below. I used his variables:

ASTROSCI, SCIBNFTS, EXPDESGN, ODDS1, HOTCORE, RADIOACT, BOYORGRL, LASERS, ELECTRON, VIRUSES, CONDRIFT, EVOLVED, EARTHSUN, SOLARREV, EATGM, ICESHEET, SCITEST5, GRNTEST1, GRNTEST5.

For political ideology, it’s pretty simple: POLVIEWS(r: 1-3 “Liberal” ; 4 “Moderate”; 5-7 “Conservative”)

The percentages given are the correct science answer, or the more pro-science answer. If you want to know my criteria for that, don’t ask, just go to the General Social Survey website and enter in the variables above, and you’ll see the results and understand clearly how I categorized things.

For me the most noticeable trend aside from a few ideological issues which are predictable is that moderates are kind of dumb and ignorant. I think this is just an artifact of the fact that more educated and intelligent people are pretty well sorted ideologically, while the less educated and intelligent are more politically uncommitted because they’re not bright enough to think systematically in terms of ideology. The Audacious Epigone did not control for background variables. For example, is the Democratic anti-science bias simply due to the fact that non-Hispanic whites are more pro-science than other groups, and Republicans have a higher proportion of non-Hispanic whites?

I ran a few logistic regressions. I ran them for all groups, and then limited the sample to non-Hispanic whites for a second run. I replicated the same predictors with different dependents. I transformed the independent variables into dichotomous categorical variables when necessary. The numbers you see are all betas, which indicate the relationship between the predictor and dependent variable. All bold betas are statistically significant at p-value 0.05. Also, since most of you won’t know that in the GSS male is coded = 1 and female = 2, etc., I’ve put in plain English words at the end of the sections what the values really mean for the total sample in regards to the statistically significant betas. I invite readers to replicate, expand, and correct.

A few notes. Some of the tendencies aren’t monotonic. For example, ignorance on many topics often peaks among political moderates. If conservatives/liberals are less ignorant, but one of the two is more ignorant, then the result will be that ignorance may correlate with one of the two antipodal viewpoints, even though moderates are the most ignorant. Second, these groups aren’t homogeneous. The attitude of libertarian conservatives to evolution is going to be different from conservatives as a a whole, while the attitude of liberals who are scientists to nuclear power is going to be different from liberals as a whole. Finally, there’s a tendency for the Democratic coalition to be skewed to those at the top and bottom of the educational qualification ladder (multi-modal). Republicans tend to be bunched in the middle. This means that these sorts of simple correlations don’t take into account the details in the shape of the distribution.