A while back, I announced on twitter that I wanted to put together a book showing how to use strategies from the generalized linear mixed model (GLMM) to analyze experimental data. This is the public-facing home for that project.
Welcome.
I should unpack a couple terms. I’m using GLMM in an expansive way to include not only single- and multilevel models using familiar likelihoods like the Gaussian and binomial, but I’m also thinking in terms of distributional models where all parameters in the likelihood function can have a linear model and of distinctly non-linear model types including splines. By experimental data, I mean fully randomized true experiments (between persons, within them, and both) and also quasi-experimental variants, where some of that control is lost for various pragmatic reasons. I plan to aim this textbook at students and researchers within the social sciences. Given my background in clinical psychology, the content will bend in that direction. However, I am keen to bring in examples from other areas within psychology and perhaps other social sciences, as well.
Books of this kind often use examples from the scientific literature, and this will too. However, I intend to go further and want to provide students with data sets from or based on the examples. If you’re going to learn how to analyze data from a randomized controlled trial, then why not practice with data resembling one of those trials? Which brings me to my main point:
You can help me write this book!
The primary purpose of this repository is to collect suggestions for studies with data you would want to learn how to analyze. I am particularly eager to include studies whose authors used open science practices, such as open data. Yet since these practices aren’t widespread in the social sciences, studies using older closed practices are also of interest.
A collaborator and I are in the early-stages of a study that will use an encouragement design and will leverage the instrumental variable approach in the analytic strategy. Kristoffer Magnusson has written interesting things on the perils of ignoring therapist effects. Which is all to say, I’m open to including studies that range from methodically simple to intimidating.
Just like with real-world data, the examples will contend with complications like
Other interesting topic areas are okay, too. For any of these topic areas, the studies could be gold-star examples of their method, or they could have substantial flaws. They could have large effect sizes, or they could show little difference among conditions. They could be citation classics, or something you just published last month.
I'm leaning towards including light discussions of the potential-outcomes framework. I'm also thinking about using DAGs to (we'll see).
I’m not necessarily saying these things are bad. They’re just not going to be the focus of this book.