For data sets that contain many possible predictors that could predict overlapping variance, it can be useful to build a model in a data-driven way.

This example script walks through how to build a series of mixed models in Julia where one predictor is added at a time, and then evaluated for contribution to model fit. In doing so, it explores 2 properties of Julia: its lightning speed and the ease with which one can modify the model formula to fit new models.

The test case explored here is what cues speakers use in order to synchronise their speech production with an interlocutor: for timely responding in dialogue, do properties like the last turn’s onset or offset matter, and do they account for more variance than speaker-internal properties such as recent reaction time? See Brehm & Meyer, Psychonomics 2019 talk.

This repo contains:

Dependencies are in general, common to the rest of the RePsychLing tutorials, but I’ve also used Distributions (and include code for downloading it).

For more specific instructions on running Julia code: https://repsychling.github.io/intro.html