problems when number covariates >33

[jeffmoore]

I started this thread in the other MARK forum (analysis, etc), but figured it's more appropriate to continue here.

I mentioned what seem to be errors in mean/SD estimates (as reported in "real paramters" output) for covariates beyond the 33rd covariate.

I was concerned this could be affecting model selection, so I deleted a bunch variables that I wasn't using so that the number of covariates in my .inp file was reducted to 33 (previously 42).

Indeed, this affected model selection. Example: Variable #42 was previously not selected, but when this became variable #33, it was selected and considered highly important (as I'd suspected).

jeff

[gwhite]

Jeff:
The problem is likely that you have an input file with a very large logical record length. You need to create an input file that has <160 characters in each record. What is happening is that the long record is being broken into pieces at inopportune spots, in this case, probably in the middle of the characters of the 33rd variable.

So, just split up the records in this file, and you'll find it works fine.

Gary

[jeffmoore]

good to know. thanks for the info, Gary.

either way, something for others to be aware of in terms of how they build their .inp files.

best,
jeff