I have a random intercept multilevel logistic regression model with cross-level interactions. The final sample size is 14489 with 25 groups (average size of 590) and an ICC of .109. I have used group mean centered level-1 variables and grand mean centered group means. To interpret the impact of a variable at both levels (age here), I see two ways of using the margins command:
1.
Code:
margins , pred(pr) at (gm1_c=(-2.073019 -1.264275 -1.089293 -.5425472 -.091465 2.29e-07 .6355743 .9508991 1.047394 1.649328)) over(agegroup) vsquish post nose
Array
2.
Code:
margins , pred(pr) at (gm1=(25.67114 26.47989 26.65487 27.20161 27.65269 27.74416 28.37973 28.69506 28.79155 29.39349)) over(v013) vsquish post nose
Array
Looking at them side by side, the difference between the two graphs is probably due to the difference in scale, if I am assuming right. This will mean that both graphs are right.
Array
However, the clusters represented at the left end of graph 1 (where average cluster age is lower than average age across clusters i.e. relatively younger clusters) are the same as those represented at the left end in graph 2 (clusters with low average age i.e. relatively younger clusters). This means that the interpretation changes, though it should not.
I will request your guidance in clarifying this issue. What am I doing wrong?
Thanks
DG
0 Response to Interpretation of grand mean centered versus original form level-2 variables in multilevel logistic regression
Post a Comment