ABSTRACT
This concluding chapter talks about the arts and ethics of model selection. Novices often feel insecure about the generalized linear mixed effects model framework because the choice of model is all up to them. Two “choice-limiting” approaches are discussed, including what one may call the “cookbook approach” of statistical testing (classical tests such as https://www.w3.org/1998/Math/MathML"> t https://s3-euw1-ap-pe-df-pch-content-public-p.s3.eu-west-1.amazonaws.com/9781315165547/a88e1713-2320-40ca-8338-8f99f9db3fef/content/ineqn072.tif" xmlns:xlink="https://www.w3.org/1999/xlink"/> -tests and ANOVAs), and various automated procedures of model selection, such as stepwise regression. The chapter argues against both of these choice-limiting approaches. When engaging in theory-driven statistical modeling, there simply is no way around exercising researcher judgment and putting one’s domain knowledge (such as knowledge of linguistics or linguistic theories) into one’s models. There simply are no recipes for building models that will work in all circumstances. In the absence of recipes for constructing one’s model, reproducible research practices become all the more important: The process of choosing one’s model needs to be documented so that outsiders can understand what’s going on in an analysis. And it is absolutely essential that data and code are shared so that different modelers can form different opinions.
