Alternative statistical models for birth parity counts
A few days ago, my VID Working Paper was published on alternatives to the Poisson and Negative-Binomial distribution for modelling parity counts, in other words: numbers of births that women experience. It’s a brief technical note rather than a full-length research paper, and should be of interest to anyone who conducts parametric statistical analyses of fertility.
This paper has once again raised the question, both in principle and in conversation with some colleagues, how far studies with a methodological focus should be expected to go to demonstrate their practical relevance. One position seems to be that there should be an application included in the methodological study itself that shows how it affects substantive conclusions. Personally, I think that is asking too much, as well as defeating the academic division of labour. Substantive analyses are already hard to make watertight on their own terms, so doing it two different ways, in addition to introducing the methodological innovation, would never allow me to do justice to the issue. Disagreements about the specifics of the example application would distract from the main methodological point. For this reason I have simply shown conclusively that these alternative distributions certainly do make a difference in general. If it turns out that the potential users do need to be walked through a more elaborate example of how a specific study would be affected, I might return to this.