Wednesday, November 29, 2017

Polarization of Controversial Science and Limitations of Science Literacy

I recently reviewed a PNAS article titled "Individuals with greater science literacy and education have more polarized beliefs on controversial science topics". This was interesting along a couple of dimensions.

For one this is similar to research I was working on during my masters degree examining interrelationships between religous, science, and other beliefs and perceptions of genetically engineered foods. However this work had much better data.

The primary finding was that when it comes to certain beliefs about things like climate change, stem cell research, or evolution there are differences influenced largely by politics or religion. This polarization is actually increased as levels of education and science literacy are raised - "Individuals with greater education, science education, and science literacy display more polarized beliefs on these issues."

The authors looked at these effects in relation to genetically modified foods, but results were non-significant. This may be explained by the fact that the data source was ten years old (2006) and perhaps this issue had not become quite as polarized in a political sense like current data may suggest. This begs the question or the hypothesis proposed in the conclusion that "Understanding these mechanisms can guide science communication so that the evidence gets through before it no longer matters" i.e. perhaps before it becomes polarized by political priors.

This is really significant in an age where science literacy is a concern (right that's one reason we just had a march for science). This will ultimately influence the types of products we see on the shelves and the potential for healthier and more environmentally sustainable solutions to challenging worldwide problems.

From an econometric sense,  the polarization was captured in a quantitative sense via interaction terms in logistic regression. Predicted or fitted values were used to plot and visualize the interactions, sidestepping the hassle of attempting to discuss or interpret the estimates from the interaction terms in a statistical sense. Output was provided in the paper and an appendix.

I also would be interested to know more about any related work that may be looking at some of these factors as latent constructs in a structural equation context. 

See also: Voter Irrationality and Systematic Bias