Trust in science & perceptions of GM food risks -- does the GSS have something to say on this?
Thursday, March 16, 2017 at 2:33AM
Dan Kahan

So here’s something fun.

I found it while scraping the bottom of the barrel of a can of GSS data that I had consulted to prepare remarks on the role of “trust in science/of scientists” that I gave at a National Academy of Sciences event a couple of weeks ago.

The GSS has a variety of measures that could be construed as measuring “trust” of this sort. The most famous one is its “Confidence in Institutions” query. It solicits how much “confidence” respondents have in  “those running” various  institutions, including the “science community.” The item permits one of three responses: “hardly any,” “only some,” and “a great deal.”

The wording is kind of odd, but the item is a classic, having been included in every GSS study conducted in 1974.  One can find dozens of studies that use it for one thing or another, including proofs of partisan differences in trust of science.

Well, it turns out that in 2010, the GSS asked this question:

Do you think that modifying the genes of certain crops is:

1. Extremely dangerous for the environment

2. Very dangerous

3. Somewhat dangerous

4. Not very dangerous

5. Not dangerous at all for the environment.

So I decided to see what would happen when one uses trust in since, as measured by the institutional confidence item, to predict responses to the genetically modified crops item. I also included a political orientation measure formed by aggregating responses to the GSS’s 7-point liberal-conservative ideology item and its 7-point party orientation item.

In my analysis, I measured the probability  that a respondent would select a response from 1-3—the ones that evince a perception of danger.

Here’s what I found:



I hadn’t expected partisan identity to matter at all, given that surveys now typically find no meaningful correlation between attitudes toward GM foods and party identity. You can see, though, that there is a bit of a partisan effect here, with those are right-leaning ideologically inclined to find less danger in GM crops as their Always eat your raw data before trying to draw inferences from a regression model. Click it!“confidence” in the “scientific community” increases.  For a "conservative Republican," the estimated difference in the probability of finding GM crops to be environmentally dangerous at the "great deal" vs. "hardly any" response levels is -18% (+/- 15% at 0.95 LOC).

Left-leaning respondents, in contrast, don't budge a centimeter as their science-community “confidence” increases (-3%, +/- 12%). 

What should we make of this, if anything? 

I’m not sure, actually.  I still am inclined to see responses to GM food questions as meaningless—a survey artifact—given how few people are actually aware of what GM foods are. Obviously, here, the level of concern expressed is way out of line with people’s behavior in consuming prodigious amounts of GM food.

We also don’t have any decent validation of the “confidence in science” measure: I’ve never encountered it being used to explain other attitudes in a way that would give one confidence that it really measures trust in science. The same goes, moreover, for all the other “trust” measures in the GSS, which consistently find high levels of trust in science among politically diverse citizens.

But maybe this finding should nudge me in the other direction?

You  tell me what you think & maybe I’ll revise my view!



Update on Thursday, March 16, 2017 at 4:23PM by Registered CommenterDan Kahan

In the raw data, left-leaning subjects budge at least a little s they become more "confident" in "those running" the "science community."  Checking the fit of the ordered logistic regression model, it turns out that the model violated the "parallel lines" test-- that is, that the impact of moving from one increment to the next on the "confidence" variable was not uniform in relation to the outcome variable (perceived risk of GM crops).

So here's a multinomial model -- one that fits the predictor variable to each level of the outcome variable seperately. It at least looks like a better representation of the raw data.

Update on Monday, March 20, 2017 at 8:49AM by Registered CommenterDan Kahan

Digging deeper w/ more GSS science-attitude items here.

Article originally appeared on cultural cognition project (
See website for complete article licensing information.