Error message when trying to run MI

[Amal Khanolkar]

Hi All

I don't have too much experience running multiple imputation, but I've been running the following imputation on the same dataset for a while now and I never received this error message before. In fact, I've imputed the same dataset with the same set of variables, and it's worked smoothly. So, no idea why I received the error message, and what the possible reason could be. Any ideas?

Code:

mi set mlong
mi register imputed socialc42 employ42 partner42 home42 voting42 edu42 child42 health42 illness42 smoke42 alcol42 fsocial zrutter5all zmal9all zrutter5xall zmal5all ///
lifesatis42 bmi42 malcont42 wdmental42 sex86
mi impute chained (logit) socialc42 employ42 partner42 home42 voting42 edu42 child42 health42 illness42 smoke42 alcol42 ///
(mlogit) fsocial (regress) zrutter5all zmal9all zrutter5xall zmal5all lifesatis42 bmi42 malcont42 wdmental42 = sex86, add(50) rseed (53421) savetrace(trace1,replace) force

Error message:

Performing chained iterations ...
mi impute: VCE is not positive definite
The posterior distribution from which mi impute drew the imputations for zrutter5xall is not proper when the VCE estimate from the observed data is not positive definite. This may happen, for example, when the number of parameters exceeds the number of observations. Choose an alternate imputation model.
error occurred during imputation of socialc42 employ42 partner42 home42 voting42 edu42 child42 health42 illness42 smoke42
alcol42 fsocial zrutter5all zmal9all zrutter5xall zmal5all lifesatis42 bmi42 malcont42 wdmental42 on m = 1
r(498);

Thanks!
/Amal

[Marcos Almeida]

Starting by Stata's output, I wonder whether a) the variable zrutter5xall was checked concerning weird numbers, missing, etc. b) the number of parameters exceeded the number of observations.

That said, if you used - force - option - , missing values will increase. According to the Stata Manual:

force allows string variables to be appended to numeric variables and vice versa. The results of such type mismatches are, of course, missing values. Default behavior is to issue an error message rather than append datasets with such violently differing types.