JASP Process Module Questions: z-Scores and Mean Centered Moderation
Hi there,
I am trying to fully migrate all of my behavioural data analysis to JASP from SPSS (I have been playing around with JASP for a long time but my institution cutting back on SPSS licenses has finally forced me to cut the cord for good) and I am trying to redo some simple moderation analyses using the Process module. I'm pretty sure I have a handle on this module at least insofar as what I need it for right now, but I'm a little confused about two things and hope I could get some clarification.
First, unlike the process macro in SPSS, the one in JASP performs z-tests of each path coefficient whereas the SPSS process macro performed t-tests. Comparing my old results to the current ones the values aren't that different (as I have a very large data set), but I was curious if anyone knows why JASP uses z-tests by default for this and whether there is any way to change it to t-tests (I can't find anything but figured it couldn't hurt to ask).
Under "Options" there is a box for "Mean-centered moderation", and I'm not sure what this means. Is this a box to check if your data are currently not centered, so that the program centers it for you prior to analysis? Or is it a box to check if your data are already centered? I seem to recall always centering my data prior to analysis when using the SPSS process macro that this is based on, but I can't remember whether I had to check this box or not.
Thanks for your help,
Seth W.
Comments
Hi Seth,
1) "I was curious if anyone knows why JASP uses z-tests by default for this and whether there is any way to change it to t-tests": the reason is that the module uses the R library lavaan, which is a general-purpose structural equation modeling library which computes only z tests; as far as I know, no there is no way to get the p-value from a t-test instead (note that the value of the test statistic is the same, but t-test requires degrees of freedom; as degrees of freedom increase, the p-value will drop down to the one you get when you do a z-test)
2) I assume "mean-centered moderation" means the variables involved in the moderation will be centered for you prior to running the analysis; you could check whether this is the case by pre-centering your variables yourself, and running the analysis with the box checked and unchecked: results should remain the same, as your variables are already centered (also it's probably mentioned in the help file if you click to the little "i" button next to the analysis title in the analysis panel)
Edit: from the help file:
Missing Value HandlingisExclude cases listwise, centering is applied only based on complete cases.Also I forgot to mention, the reason you center in a moderation analysis is that the interpretation of the "simple effects" in the output (i.e. the effect of each variable, e.g. x1 and x2) is the effect of that variable when the other variable is 0. Sometimes 0 is not a meaningful value on your variables, so centering your variables means you get the effect of each variable when the other variable is average. Centering does not affect the value or the test of the interaction (x1*x2).
I'll also attend our expert to this!
EJ
Hi, I second patc3's answers. I want to highlight what is stated in the help file: When using list-wise exclusion, you should only center variables based on complete cases. Otherwise variables are based on differing numbers of cases. This can easily happen when manually centering variables before entering them into the analysis which is why we added this option.
Best regards,
Malte (main developer of the Process module)
Thanks so much for your prompt answers! This clears things up.
Seth W.