www.cmm.bristol.ac.uk/forum

Hello!

I am trying to calculate the VPC for an ordered multinomial model with a logit link function. I used the threshold method (where the level-1-variance is restricted to pi^2/3), but I would also like to use the vpc.txt-macro. However, I alwasy get an error message for line 3:

calc c153=(~c151)*.c1098
pick 1 c153 b2
calc c153=(~c152)*.omega(2)*.c152
pick 1 c153 b4
seed 1
nran 5000 c154
calc c154=alog(c154*b4^0.5+b2)
aver c154 b1 b3 b2
calc c154=c154*(1-c154)
aver c154 b5 b1
calc b8=b2^2/(b1+b2^2)

---------------------------
Error detected by MLN
---------------------------

error while obeying batch file C:\...\mln125F.tmp at line number 3:
calc c153=(~c152)*.omega(2)*.c152

Dî.
---------------------------
OK
---------------------------

I put 1,0,0,0 into variable c151 and 1 into variable 152 (on the basis of the description for binomial models in the MLwiN manual) and I'm pretty sure I'm doing something wrong here. I'd appreciate if you could point me in the right direction.

Thank you very much!
Petra

Hi Petra,
MLwiN fits multinomial models in a somewhat different way to a standard binomial logistic model - in particular it uses an additional level below individual to capture the pattern of the multinomial response (in the ordered case a series of 1s and 0s indicating which threshold the true value is bigger than) and so you cannot simply take the VPC macro as is and translate to the multinomial context.
I hope that helps or at least explains why you are having difficulty.
Best wishes,
Bill.