Daniel posted on Wednesday, March 23, 2005 - 6:19 am
Hi, I'm trying to run the following model, but I'm getting an error message that I do not understand. For sake of parsimony, I only present code for 9th grade. Can you please provide some assistance?
My code, excluding the data statement and list of variables. missing is .; idvariable is id; usevariables=smoker9-smoker12 u9-u12 y9-y12; categorical are smoker9-smoker12; define: u9=1; if (smoker9 eq 0) then u9=0; if (smoker9 eq .) then u9=.; y9=smoker9; if (smoker9 eq 0) then y9=.; analysis: type=basic;
I am trying to fit a two-part growth curve model using repeated measure data from 20 time points. The dependent variable has a lot of zeros and missing values due to mortality during follow-up. As the missing due to mortality is likely to be missing not at random, I would like to combine the two-part model with the non-ignorable models. So far I fitted the data by two-part model (MAR) and two-part pattern-mixture model. When fitting the Diggle & Kenward’s two-part growth model, the computation is extremely slow (in fact, it is still running for days). The sample I am running is only 10% of our original sample (around 15,000 out of 150,000).
I think the extremely slow computing is related to the fact that I have too many dropout indicators (19 in total) in the model. If possible, is there a way to speed up the computation? Or there may be something wrong in my model, for example, trying to combine two-part model with the selection model?
Check how many dimensions of integration you have (top of TECH8 screen output). If more than 3-4 you should switch to Montecarlo integration using fewer points. For instance, 4 dimensions gives 15*15*15*15 = 50625 points using regular integration. Here you can instead use Montecarlo integration with 5000 points.
Also, start with a smaller sample - perhaps 5% instead of 10.