Ordered logistic regression coefs int...
Message/Author
 Albrecht Küfner posted on Wednesday, October 28, 2015 - 7:26 am
Dear Drs Muthén,

I ran a ordered logistic regression (DV with 3 outcome levels) and constraints on some IVs. The model worked very well and I now want to use the coefficients to calculate/predict (in Excel) the probabilities for each respondent and each level of the DV. This should work well using the following generalized formulae (with c being the threshold):

P_level_1 = 1-exp(-c + a*x1 + b*x2)/1+exp(-c + a*x1 + b*x2)

P_level_2 = 1-exp(-c + a*x1 + b*x2)/1+exp(-c + a*x1 + b*x2) - P_level_1

P_level_3 = 1 - P_level_1 - P_level_2

In general this worked, when I ran easier models (in another stats package). However I am getting wrong results with the MPlus generated Coefficients/Thresholds.

Is there anything MPlus does (I always call it magic), which could cause this issue. Basically the level_1 probabilities are too large.

Thank you so much in advance!
Best,
Albrecht
 Bengt O. Muthen posted on Wednesday, October 28, 2015 - 2:49 pm
See top of page 493 in the UG, where your see how the c thresholds are different in different places.
 Albrecht Küfner posted on Thursday, October 29, 2015 - 6:37 am
Dear Bengt, thank you very much.

I am sorry for not being specific - of course I used the different thresholds (c1 and c2, in this case) for the three levels. But still there are different outcome probabilities.
I also tried specifying the levels as mentioned in the UG.

Thanks again!
 Bengt O. Muthen posted on Thursday, October 29, 2015 - 8:33 am
If you like, you can send the output with your specific question on why you think the probability of the lowest category is too large to Support along with your license number.
 Albrecht Küfner posted on Thursday, November 05, 2015 - 7:30 am
Dear Bengt,

thanks again so much. It appeared it was a mere rounding error in Excel (yes, you can hit me now).

Thanks!