I recently met Carlos Davidson, a prof at Cal State University. He studies amphibians, with a special interest in why frogs in California are disappearing. He said that he can “predict quite well whether a site will have frogs, based on the pesticide use upwind” and that he thinks that chemical pesticides are a big part of the problem. He thinks that if we switch to other less harmful or even organic insecticides we may see improvements in their numbers. The logic behind this is that the frogs may be eating poisoned insects which poisons them in turn. An insecticide approved for organic use may be less likely to harm the frogs. But he also said that others in his field are far from convinced. What should it take to be convincing? Is there a “statistical” answer to questions like, which is more important: lab work, more field work, more analysis of existing field data (perhaps with more covariates included)?

Some of Davidson’s papers are available at http://bss.sfsu.edu/cdavidson/publications.htm and I read the one called “Declining downwind: Amphibian population declines in California and historic pesticide use.” Davidson looks at 1500 sites where specific species of amphibians are known to have been present in the past. In some locations, they’re still there; they’re absent from others. He performs logistic regressions of the present/absent variable on a bunch of other variables: elevation, degree of nearby urbanization, precipitation, degree of nearby agricultural use, total pesticide use upwind, and a few others. Pesticide use is based on government records of use from 1974-1991. For four of the five species, total pesticide use has one of the biggest effects in the model. Other models look at different types of pesticides, examine cross-terms and autocorrelation, and so on. It seems pretty well done.

It’s really easy to find fault with this sort of analysis: there are always things that the researcher woulda-coulda-shoulda done, but didn’t. (This is true with my own work, too). For example, shouldn’t upSTREAM pesticide use be at least as important as upWIND pesticide use, so shouldn’t it be included? Isn’t there potential bias in looking only at species that other researchers have said may be particularly affected by pesticides, because the reason for that speculation may be that they have noticed this correlation in essentially the same data—if you cherry-pick from a big enough dataset, you can always find some “statistically significant” correlations. What about invasive species, which I thought were implicated in the decline of many amphibian populations; could they be more common near some kinds of agricultural lands and thus a confounder with pesticide use? And so on.

One thought is that it might be worth looking at DOWNwind pesticide use as well. If decline of amphibians is associated with upwind pesticide use but not downwind use, then that’s more suggestive of a causal relationship than if both upwind and downwind use are predictive.

Overall, we have a typical problem in environmental science: there is an observed relationship between a health outcome (local extinction of a species, in this case) and some pollutants. There are other possible explanations for the local extinction. Lab data show that the pollutants do have relevant physiological effects. How should researchers proceed? What does the stats community have to offer?