Probit vs. logit
Message/Author
 Anonymous posted on Wednesday, June 01, 2005 - 1:27 pm
When I use WLSMV to estimate the coefficients, everything is fine, but if I switch to ML with numerical integreation, the output gave warning:
Starting Value is wrong, and it only gave estimates without sd and confidence interval. Why is that?
 Linda K. Muthen posted on Wednesday, June 01, 2005 - 2:15 pm
 Silvia S Martins  posted on Wednesday, May 30, 2007 - 1:04 pm
Hi Linda

I recently ran a probit regression model in M-plus and I wanted to know if there's any command in M-plus that already calculates the probabilities from the probit regression coefficients shown in Chapter 13 of the users guide or whether these calculations need to always be done by hand (I have binary and categorical outcomes in my model)?

Thanks,

Silvia Martins
 Linda K. Muthen posted on Monday, June 04, 2007 - 7:27 am
There is no Mplus command to ask for probabilities. If they are not given, you would need to calculate them by hand.
 Cecily Na posted on Friday, February 15, 2013 - 12:38 pm
Hello professors,
I am having trouble interpreting probit thresholds. I have a categorical outcome variable G with 5 categories. The results are briefly listed below:
coefficient = 0.2
threshold 1 = -0.02
threshold 2 = 0.5
threshold 3 = 1
threshold 4 = 2

Questions:
1.Does it mean I will have four equations? 2. How do I interpret the results for each category?
3. When this variable is a mediator and will be regressed on another continuous variable (final outcome), how would I interpret beta coefficient(s)?
I am not able to find answers on any manuals. I would really appreciate some guidance on this.
Thank you!
 Bengt O. Muthen posted on Friday, February 15, 2013 - 2:24 pm
1. No, one regression equation. See ordered polytomous outcomes in our Topic 2 handout and video.

2. See handout.

3. If you use WLSMV it is the M* underlying continuous latent response variable for the mediator which predicts the outcome.