Page 1 of 1

Adding and specifying terms

Posted: Mon Oct 12, 2020 7:07 am
by SabineKatzdobler
Good day,

apologies for asking again and again. MLwiN is full of challenges, but so far I have been very happy and grateful regarding the solutions of the support team. Generally, my goal involves testing a multilevel ordered logistic regression analysis including a within-between random effects specification (level 1: response indicator; level 2: occasion; level 3: individual). At the moment, I have two related issues regarding adding terms in MLwiN via the "Specify Term" window:
  • When I try to add a term as "separate coefficent" one-by-one after the constant-variable, I noticed in the majority of my trials that MLwiN does either not react at all nor does converge finally. As a consequence, I add the relevant variable as "common coefficient". This, however, does not allow to test the proportional-odds assumption in the same way as the example in Module 9 p. 10, 11 suggests. Thus, I have the impression that I am forced to include several variables (haben vorher in stata die PPO-assumption nicht erfüllt) only as common coefficients. Is my situation problematic?
  • Specifically, I have difficulty adding the "within-term" (deviation from the individual group mean). I had to include them as common coefficients (see first question). Finally, the model resulted in coefficients of 0 for every within-variable. This is a strange result and I did not expect that. What have I done wrong? How can I add the within variables properly?
Remark ref. within-variable - Example:
  • Individuals´ Age: 50 / 54 / 59 ;
  • Age av: (50+54+59) divided by 3 time points > 54,333 ;
  • Age dev. (within): -4.333 (50 - 54,3339 / -0,333 (54 - 54,333) / 4,667 (59 - 54,333)
Kind regards,

Sabine

Re: Adding and specifying terms

Posted: Tue Oct 13, 2020 3:22 pm
by ChrisCharlton
If you were able to send an example worksheet demonstrating the problem and/or screenshots of the model to myself and Bill (seehttps://www.bristol.ac.uk/cmm/team/ for contact details) we should be able to check whether there are any obvious problems with the model/data.