Intepretation of probit coefficients
Message/Author
 Catherine Winsper posted on Tuesday, November 16, 2010 - 1:06 am
I have a path model with dichotomous predictor and outcome variables and dichotomous and three and four category ordinal mediator variables. My problem is the interpretation of the probit coefficients. I wanted to standardise the data before entry, but I believe this is not possible in mplus as it leads to non-integer numbers for the categories. What is the best way to compare the effects of the different mediator scales and particulalry considering the indirect effects? I cannot find the solution to this question in exisitng research, questions/answers so any advice would be gratefully received.
 Linda K. Muthen posted on Tuesday, November 16, 2010 - 9:56 am
You should not standardize categorical variables. You can use weighted least squares and MODEL INDIRECT to obtain indirect effects. If you also ask for the STANDARDIZED option in the OUTPUT command, you will obtain standardized indirect effects.
 shonnslc posted on Thursday, October 17, 2019 - 7:21 am
Hi,

I am running this kind of analysis right now:

usevariables are
x1 x2 x3 x4 z1 male; !male is dichotomous (female = 0, male = 1)

categorical are
x1-z1;

analysis:
estimator = WLSMV;

model:
x by x1-x4;
x on male;
z1 on male;

My questions:
1. If I understand correctly, this is probit regression. But can I call this MIMIC modeling (z1 is a manifest variable) as well?

2. Assume the (raw) coefficient for z1 on male is 0.25. Does that mean the difference in the z-score of z1 between males and females is 0.25?

3. Can I evaluate the magnitude of the raw coefficients using Cohen'd? That is, if the coefficient is 0.25, then the gender difference is considered a small effect. Or should I use STDY or STD to make this interpretation?

Thank you a lot!
 Bengt O. Muthen posted on Thursday, October 17, 2019 - 3:05 pm
1. Yes, it is a form of mimic.

2. No. See our Short Course Topic 2 video and handout for how to interpret probit regression.

3. Cohen's d is not defined for categorical outcomes.