jamovi

Hi everyone,

Does anyone know what the rationale is for automatically mean-centering continuous predictors in mixed-effects models? References would be very much appreciated. Thank you!

In all cases, centering is an optional procedure. The analysis you are using will calculate the solution based on mathematical operations. When the scores in the original variables have either negative or positive values (i.e., not around the zero), any newly created variables (e.g., interaction or quadratic terms) using the original variables will have larger values, which will eventually make the results of the analysis out of the range. Because the mathematical solutions have no limits in values, the results are still correct. However, what we are looking for is the results within the range of our data. Centering keeps the output range around the means of the data helping us understand the results. Not sure about any specific reference for this but Aiken&West's textbook shows this. Or Cohen, Cohen, West, & Aiken's (2002) textbook does too.

I totally agree with @simonmoon. We should add that centering variables greatly improves your chances of convergence of the model. It is one of the first recommendation when the model does not converge.

Thank you for both your insights. Really helpful!