Multiple Imputation

[Jules Herz]

I have a problem running a panel regression. I have data for n=1019 observations and want to carry out a regression analyses with 15 variables. But due to a lot of missing values for some observations and variables, Stata carries out the regression with only n=347 observations. Therefore, I'd like to use multiple imputation with the regression method for missing data but am not quite sure how to implement it.

After specifying my dataset I used the commands

 mi set mlong

mi register var1 var2 var3 ...

 mi impute chained (logit) var1 (logit) var2 (regress) var3 ... , add(15)

But after the last command, Stata shows a lengthy error message:

mi impute logit: perfect predictor(s) detected
Variables that perfectly predict an outcome were detected when logit executed on the observed data. First, specify mi impute's option
noisily to identify the problem covariates. Then either remove perfect predictors from the model or specify mi impute logit's option
augment to perform augmented regression; see The issue of perfect prediction during imputation of categorical data in [MI] mi impute
for details.
error occurred during imputation of var1 var2 var3 ... on m = 1
r(498);

What do I do wrong?

Best,
Jules

[daniel klein]

You did not specify

mi impute chained var1 var2 var3 ... , add(15)

because this is invalid syntax. You did include (logit) somewhere. In this case, this is (probably) not key to giving advice but generally, you are supposed to show what you have typed, exactly.

Did you read the very informative error message? Or the pdf-manual entry to which it links? It tells you what your options are. I recommend reading the respective material as it will provide you with deeper insight into the issue than a quick answer on Statalist would typically do.

[Jules Herz]

daniel klein: thanks for pointing out my mistake in reporting the (stylized) code. I included (logit) or (regress) in front of every variable in the impute statement and changed it accordingly in the code above.

I read the error message but did not understand it properly and the manual was not a big help either.

[Rich Goldstein]

not a clearly written question, but I think you want the "augment" option (here I am assuming that you are using "mi impute chained" or "mi impute logit" - if you are not please show the EXACT command you are using)

[Jules Herz]

I implemented the "augment" option, but now Stata says

convergence not achieved
logit failed to converge on observed data
error occurred during imputation of Static Specific CrossBorder Electricity Capacity Budget Volume Financial Material Ceiling CP_Level
FavTreatment PaB SlidingFIP FixedFIP SupDuration BidCapacity AwPrice AdjAwPrice AwCapacity Penalty on m = 1
r(430);

[Rich Goldstein]

without more info, please read the FAQ and follow its advice, I can not help further but maybe someone else can

[daniel klein]

From a technical point of view, add the option noisily and inspect the model that does not converge. Look for large or missing standard errors and possibly remove the offending predictor from the respective equation. We cannot give substantial advice on building models without further detail; and, even with more detail, this does not seem like my field of expertise.

[Jules Herz]

I added the noisily option and it ran several iterations which seemed to work but in the end Stata showed me this: