critique of a recent ajs genetics paper: levi-martin v. guo, li, wang, cai and duncan

John Levi-Martin has written a comment on a recent paper by Guo, Li, Wang, Cai, and Duncan  claiming that the social contagion of binge drinking associated with a medium genetic propensity. Levi-Martin claims that GLWCD having simply misread their data:

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 alcohol use 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.
The main claim is that GLWCD are testing against nulls rather than properly estimating a U-shaped effect:
This is consequential because of the way that choose to examine their data. Although
the verbal description of the swing theory here refers to the comparison of magnitudes  (“more likely”), the methods used by GLWCD involve successive tests of the null hypothesis across three subsets formed by partitioning the sample by level of what is termed genetic propensity. If we denote these three subsets L, M and H, standing for low, medium and high propensity, then, for the kth predictor, they estimate three slopes, bLk, bMk, and bHk. Because the swing theory does not require that any particular predictor have an effect, but only that if it does, it does not in the extreme propensity tiers, this theory holds that for any k, bLk≈bHk≈ 0.
Publishing note: The comment is on SocArXiv for all to read. If the criticism holds water, it’s a shame that it is not in a journal, preferably the AJS. If journals simply aren’t interested in error correction, then they simply aren’t into science.
50+ chapters of grad skool advice goodness: Grad Skool Rulz ($2!!!!)/From Black Power/Party in the Street

Written by fabiorojas

November 21, 2016 at 3:29 am

One Response

Subscribe to comments with RSS.

  1. Hey Fabio, glad that you are posting this. It is difficult to have timely debate on methodological issues in the print world. Let me state, though, for the record, that it actually isn’t even that I think that they should have tested the differences as opposed to successive tests of the null. It’s that their claim was that successive tests of the null should show no effects at the high and low propensity ranges. They actually show more significant effects of independent variables at the high and low ranges. That includes one of the two variables that can be considered to be an experimental treatment. That is, it’s like they have a model that says, “coins fall heads,” toss ten coins in the air, and get seven tails. It’s just that they ignore the ones that they aren’t interested in to focus on one in particular. The models are also constructed in a way that makes it impossible to correctly adjudicate between “this is genetics having an effect” and “this is socially ascribed statuses having an effect.” Finally, so far as I can tell, models are presented misleadingly so that it appears that only the interaction they are interested in occurs. They do this by making constraints across models for the coefficients they prefer not to vary across propensity tiers. That’s why these look like they aren’t varying. The pairs of constrained models are reported as if they were independent.



    November 21, 2016 at 2:47 pm

Comments are closed.

%d bloggers like this: