Dan Kahan pointed me to a recent post by Robert Lupton and Christopher Hare, who report, based on recent survey data, that “conservatives are more likely to believe that vaccines cause autism.” Lupton and Hare write:

But a recent survey also finds that a partisan gap has emerged since 2009 on the question of mandatory vaccinations. . . . Our results suggest that an erroneous belief in the link between vaccines and autism is most prevalent among conservatives.

More specifically:

In this survey, we asked people whether they agreed or disagreed that “there is a link between childhood vaccinations and autism.” People were also asked to place themselves on a seven-point scale ranging from liberal to conservative, to provide their party identification, and to answer a typical set of demographic questions. . . .

So far, so good. They have the information needed to see whether conservatives are more likely to believe that vaccines cause autism (assuming, as usual, that the survey responses accurately represent people’s views and that the sample is representative of the population).

Here’s what Lupton and Hare report:

And here’s my problem (and Kahan’s). The above graph, and the associated regression analysis, does not directly answer the question by the title of this post. To directly answer that question, you want to make a direct comparison: a simple plot (or even a tabulation) of the percentage of respondents who believe in the vaccines-autism link, as a function of political ideology.

[See P.P.S. below for a graph that shows the direct comparison, sent to me by the authors in response to this post.]

Once you’ve shown the basic comparisons, it makes a lot of sense to do more elaborate analyses.

I have every reason to believe that, if you look directly at the data collected by Lupton and Hare, that you’ll find the pattern they claim, that conservatives are more likely to believe that vaccines cause autism. I just want to see it. More generally, when we claim X, we should show X. This comes up a lot in statistical analysis, that there’s a tendency to do one thing and describe it as another.

None of the above should be taken as a criticism of Lupton and Hare’s research. Rather, it’s an opportunity for them to show the direct comparisons and use these as a baseline to better understand their more elaborate analysis.

It’s not just about showing something simple for the general public; it’s also about understanding regression models by linking them to the numbers.

P.S. Just to clarify: The above post is not a criticism of Lupton and Hare’s research, nor does it represent a disbelief in their claim. Given the analyses they’ve shown, I have every reason to believe their claim that conservatives are more likely to believe that vaccines cause autism. And it’s an important thing to know. That’s why I’d like to see the direct comparison, to confirm that their claim is correct.

P.P.S. Hare and Lupton sent the following graph showing the direct relationship between political ideology and belief in the autism/vaccines link:

So, yes, a direct look at the data reveals a real difference. According to this graph, the percentage of autism/vaccine believers is twice as high among extreme conservatives, as among liberals and moderates.

Hare emphasizes: “the lowess smoother is NOT based on the jittered values, I simply jitter the points to illustrate density.”

And Lupton points to “previous research suggests that low trust in government is related to requests for non-medical exemptions to vaccine-preventable diseases,” for example this paper:

Salmon, Daniel A., Lawrence H. Moulton, Saad B. Omer, Patricia deHart, Shannon Stokley, and Neal A. Halsey. 2005. “Factors Associated with Refusal of Childhood Vaccines among Parents of School-aged Children.” Archives of Adolescent Medicine 159 (5): 470-476.

I’m glad to have had the chance to clear this up. It’s good to see the raw data as well as the more elaborate statistical analysis.