Here are some simple data analyses that reflect how a wider range of GSS science-attitude variables relate to perceptions that GM crops harm the environment, and how that relationship is affected by partisanship.
I’d say they tell basically the same story as my initial analysis of CONSCI, the item that measures “confidence” in “those running” the “scientific community”: basically, that higher, pro-science scores on these measures is associated with less concern with GM crops. This is so particularly among right-leaning respondents; indeed, left-leaning ones don’t really move at all when one looks at risk perceptions in relation to the composite “proscience” scale.
There is also a small zero-order correlation (r (1189) -0.12, p < 0.01) between GENEGEN—the GSS’s 2010 GM risk perception item—and the composite left-right scale that I constructed and that is coded so that higher scores denote greater conervatism.
All of this is out of keeping with the usual finding of a lack of partisan influence on GM food risks. I have reported many times that there is no partisan effect when GM food risks are measured with the Industrial Strength Risk Perception measure. Surveys conducted by other opinion analysts using different measures have shown the same thing.
So what’s going on?
One possibility, suggested by loyal listener @Joshua, is that the GSS’s GM-concern item looks at people’s anxiety about the impact of GM crops on “the environment” as opposed to the safety of consuming GM foods. The “environmental risk” cue is enough information for the public—which is otherwise pretty clueless (“cueless”?) about GM risks—to recognize how the issue ought to cohere with their political outlooks.
Seems persuasive to me . . . but what do you—the 14 billion daily readers of this blog—think?!
Oh, one more thing: I did a quick search and found only one paper that addresses partisanship and the GSS’s “GENEGEN” item. If others know of additional ones, please let me & all the readers know.
Oh, one more “one more” thing. Here are the raw data:
Because of the number of observations (i.e., people) in the “Confidence in science community” & the “Proscience scale” graphs, it’s difficult to discern the relative proportions of “< avg” & “> avg” (i.e., below the mean on left_right scale & above it) along various points on the x- & y axes.
One way to try to deal with that is by using transparencies, which vary the intensity of the colors as observations pile up on top of each other. These differences convey information on the density of observations of right- and left-leaning respondents at different x-/y-axis coordinates.
I did that for “Confidence in science community” & “Proscience.” (I also jittered–added a little random noise–to the observations to spread them out a bit, a technique I used yesterday, too.)
That’s a little better, right? But the proportions of red and blue at 1.0 and 0.0 are still hard to detect, so I would for sure still include the locally weighted regression lines, as I did here.
Indeed, one might reasonably argue for dropping the scatterplot & going only with loess plots. Locally weighted regression, in my view, is definitely the best way to enable observation of the “raw” data, particularly when there is an over-plotting problem of this sort.