Trouble Ahead

Here’s the abstract:

Guo, Li, Wang, Cai and Duncan (2015) recently claimed to have provided evidence for a general theory of gene-environment interaction. The theory holds that those who are labelled as having high or low genetic propensity to alcoholuse will be unresponsive to environmental factors that predict binge-drinking among those of moderate propensity. They actually demonstrate evidence against their theory, but do not seem to have understood this.

And here’s the story.

John Levi Martin wrote:

The American Journal of Sociology recently published a definitely wrong piece of work. I submitted a comment, and after revision my comment was rejected after a review by someone who took a methodological perspective different from the authors (long and boring story). If I had a blog, I’d put it there. I don’t, but you do, and if you agree with my assessment, and wanted to comment on their work, I think it would help shift the incentive structure away from shameless production of false positives. And if you were the reviewer who wanted other analyses done (that I can’t do since the data are proprietary) you’d probably write an even more thorough critique. I put my comment here. I sent it to the authors but they didn’t reply.

I replied: No, I was not the reviewer! In any case, I like the abstract [see above]; it is so precise and to the point!

I didn’t read Guo et al.’s original article or Martin’s critique cos that would require, y’know, actual work. But I’m posting it here so that anyone who’s interested can follow up.

P.S. It’s been over a year since Martin sent me that email so maybe the authors have replied to him by now. But I doubt it.

7 thoughts on “Trouble Ahead

  1. “The theory holds that those who are labelled as having high or low genetic propensity to alcoholuse will be unresponsive to environmental factors that predict binge-drinking among those of moderate propensity.”

    This sounds like a pretty fuzzy theory to begin with: How does one define “high or low genetic propensity to alcohol use”? The phrase sounds very prone to multiple different interpretations.

    (I am not able to get to wherever the link points, so am relying just on the quote given above.)

  2. By my lights, Guo et al. did one thing right: They presented a straightforward graphical prediction showing the pattern they’d see in their data if their hypothesis were correct (Fig. 2a). I was expecting a graphical display of their results in the same format. Instead, all we get are tables of slopes for different facets of peer influence. If Guo et al. had plotted graphs, they and everyone else would be able to see at a glance that their hypothesis—not just of any g x e interaction, but a of particular pattern of g x e interaction—is wrong.

  3. I just skimmed the article, but this kind of model always seems particularly problematic to me when using logistic regression since the model specification itself is basically what they are proposing ( pretty flat at the extremes, big slope around the mean). It just feels tautological to break it up. I mean in a way maybe the article is a good example of how such prediction works.

Leave a Reply to jrkrideau Cancel reply

Your email address will not be published. Required fields are marked *