I'm reading a paper and I'm unsure how valid their methodology is, and I'm trying to find some information related to that.

Basically, they're doing a meta-analysis of swallows and PCB exposure. They're using several different log response ratios (for the effect size) that are related to reproductive success (number of nestlings, number of eggs, successful fledglings, etc). They make a GLMM using the log response ratios as the dependent variable, the rest of their GLMM makes sense. But is it valid to use response ratios across multiple studies in this way? I can't find any example of another meta-analysis being performed this way. None of the books that they cited in the paper, or any of the other studies they cited, go into detail about using GLMM and log response ratios to perform a meta-analysis with an information-theoretical approach across multiple studies. They also vaguely mention using a weighting system for the response ratios (they don't go into detail), but none of their citations explain the validity behind this decision within the context of GLMM and model selection; the citations only explain waiting schemes for traditional meta-analyses.

Do you know of any examples of this approach to a meta-analysis, or any books or anything like that? It might be something that is done frequently, but has a special name that I don't know about or something. I don't have a strong math background.

  • $\begingroup$ When you say response ratio what exactly do you mean. For instance for number of nestlings what is tat divided by to get a ratio? $\endgroup$ – mdewey Jun 21 '17 at 12:35
  • $\begingroup$ These studies were all field studies, so there were various levels of exposure reported for each study. So they divided the response of the most exposed group by the response of the least exposed group for each study (or the other way around, I don't have my notes with me). Their fixed effects and interaction terms all made sense and accounted for every everything in the model. It just seems odd to me to use response ratios across studies and then to do a model selection like this, the only reason I'm uncomfortable with it because I'm unfamiliar with it. $\endgroup$ – Wen Baldron Jun 21 '17 at 13:18
  • $\begingroup$ And they used a meta-analysis weighting system that they don't go into detail about. I see no reason why a weighting system is even necessary to do this, but maybe it reduces bias like the weighting in pairwise meta-analysis. What they claim makes logical sense, but I'd like to see something more formal. $\endgroup$ – Wen Baldron Jun 21 '17 at 13:19
  • $\begingroup$ It is a well- crafted question. It appears that different studies are working on different measures of reproductive success. It is not a good- idea to work on a meta analysis unless and until there is a large number of studies and use of only one of the measures of response by an individual study. $\endgroup$ – Subhash C. Davar Jun 26 '17 at 7:46

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.