[daniel klein]

Before, you said the error code is 2000; I assume you have fixed that. Now look up option augment, which will help with perfect predictions.

Best
Daniel

[Neeraj Kumar]

Thank you so much Mr. Daniel for advice and help. I sort out that issues.

[Neeraj Kumar]

I am having some issue in estimating MI estimate. I want to use order probit for my estimation. kindly suggest me the commands. I am using following commands:

mi estimate: oprobit V22 _20_education  _20_employ1 _20_V143 _20_V142 _20_V138 _20_V137 _20_V136

it always showing error that variable _20_education not found and an error occurred when mi estimate executed oprobit on m=1. one more question if I have to calculate chi square for that also i have to use mi estimation in that case what ould be the command? kindly help me with commands.

[Neeraj Kumar]

Kindly help me with above problem, which is mentioned in #18. And thank for your suggestions Mr. Daniel

[daniel klein]

There is little more I can say. If Stata tells you

_20_education not found

then that variable is obviously not in your dataset.

However, I wonder whether you are supposed to refer to this variable at all. The _#_varname scheme might hold the imputed value in the # complete dataset. This would be the case in mi style wide. You are not supposed to refer to the imputed values. If in your non-imputed dataset, you would type

oprobit V22 education ...

then in your imputed dateset(s) you are supposed to type

mi estimate : oprobit V22 education ...

You refer to the original variable names.

More generally, I have the feeling that you have not read the MI manual entries that come as a pdf with your Stata. I encourage you to invest one or two hours and read the manual. There are plenty of examples that explain all the steps involved in MI.

Best
Daniel

[Neeraj Kumar]

Thanks you so much for your's constant and prompt reply Mr. Daniel. As I used M=20 in wide style. It generate 20 variable of one variable (_1_education, _2_education....... _20_education). I was using the last variable of all my imputed variable (_20_education, _20_108 and so on) for my analysis. I just change the name of these variable from _20_education to edu and i did the same thing with other variable. and the error issue is get solved, which I mentioned in #18. Mr. Daniel I go through with estimation command support the mi estimate. There is no such chi square command for supporting mi estimate. May I used this command for chi?

tab V22 V108, chi

and V108 is imputed variable. if I have to make dummy out of any categorical variable, may i used this command?

tab v106, gen (vvv)

Commands which we usually use in the absence of MI estimation.

I really appreciate your efforts. Once again Thanks you so much.

[daniel klein]

There is no such chi square command for supporting mi estimate.

And there might be a good reason for the absence of such a command. See this recent discussion. While I agree that mi test, suggested in #8 is safe to use, I would not fully agree with the final advice to combine association measures. If you believe that those measures represent population values and (approximately) follow a normal distribution, that approach might be fine. I am not sure whether a Chi-square value qualifies in this respect.

Best
Daniel

[Neeraj Kumar]

Thanks for the reply once again. It's clear my doubt regarding chi square. Last question what about making dummy? May I make it out of imputed variable (_20_edu), using the command mention in #21?.

[daniel klein]

Last question what about making dummy? May I make it out of imputed variable (_20_edu), using the command mention in #21?.

I would not even use this command in non-imputed data; use factor variable notation (see help fvrevar) instead.

Best
Daniel

[Neeraj Kumar]

So, instead of that I should go with factor variable notation.. Thank you so much for clearing my doubts and for your constant and prompt reply Mr. Daniel.