Chapter 8 Using simulation to understand your model
Data analysis is often taught as if the goal is to work out the p-value and make a decision: reject or fail to reject the null hypothesis. However, understanding the long-run properties of one’s experiment design and statistical model under repeated sampling requires more work and thought. Specifically, it is important to understand (a) what one’s model’s power and Type I error properties are, and (b) whether the model we plan to fit to our data can, even in principle, recover the parameters in the model.
In order to study these properties of one’s model, it is necessary to learn to simulate data that reflects our experimental design.
Let’s think about how to simulate data given a Latin-square \(2\) condition repeated measures design. We begin with our familiar running example, the Grodner and Gibson (2005) English relative clause data.