Multiple imputation for missing values - error

[Guest]

I am trying to apply multiple imputation method to account for missing values in my dataset. Holdings is a continuous variable, convenience is a categorical variable and credit is a dummy variable.

(1) When I use the following commands, it states “option augument() incorrectly specified" when I type the code: below. How do i resolve this please?

mi set wide
mi register imputed holdings convenience credit
mi register regular age
mi impute chained (regress) holdings (logit) credit (mlogit) convenience  = age add(20) rseed(1234) augment

Also:
(2) How can i set a constraint values are between 0 and 1 for my continuous variable
(3) After using these commands, I understand that the dataset introduces new variables e.g. _1_holdings up to 20. Where do i go from here to ensure the missing values are filled in?

Thanks very much!

[Weiwen Ng]

I am trying to apply multiple imputation method to account for missing values in my dataset. Holdings is a continuous variable, convenience is a categorical variable and credit is a dummy variable.

(1) When I use the following commands, it states “option augument() incorrectly specified" when I type the code: below. How do i resolve this please?

mi set wide
mi register imputed holdings convenience credit
mi register regular age
mi impute chained (regress) holdings (logit) credit (mlogit) convenience = age, add(20) rseed(1234) augment

Also:
(2) How can i set a constraint values are between 0 and 1 for my continuous variable
(3) After using these commands, I understand that the dataset introduces new variables e.g. _1_holdings up to 20. Where do i go from here to ensure the missing values are filled in?

Thanks very much!

first, add a comma after age.

second, take a look at the available imputation models in the manual. I believe the manual suggests that the truncated regression model may be acceptable in this case. Also, predictive mean matching may be another alternative. In principle, what you describe sounds like a fraction, and if mi impute had a fractional logistic model, you would probably be well justified in using that - except that it doesn’t.

I’m not sure what you mean by the last point. Stata will return an error if it imputes some missing values, if that’s what you are getting at.

[Guest]

first, add a comma after age.

second, take a look at the available imputation models in the manual. I believe the manual suggests that the truncated regression model may be acceptable in this case. Also, predictive mean matching may be another alternative. In principle, what you describe sounds like a fraction, and if mi impute had a fractional logistic model, you would probably be well justified in using that - except that it doesn’t.

I’m not sure what you mean by the last point. Stata will return an error if it imputes some missing values, if that’s what you are getting at.

Thanks for your response Weiwen.

When I perform the -impute command-, it creates 20 new variables. My question is: how do I perform a regression after multiple imputation? Is there a way of averaging these 20 variables into one variable where there are no longer any missing values?

Also, do i perform the truncated regression separately? Or do I include it in a -mi impute chained- command?

[Vince Vo]

Hi Guest,

I had the same problem that you did when it came to imputation. I tried both mi impute before the regression command and only the regression command. It showed two different outputs that I could not find a reasonable explanation for my data. Therefore, back to basics, do you have to impute the dataset, in my case it is over 10% for some variables and it is quite acceptable under some literature. Then I decided to run without the imputation and probably will report the sensitivity analysis for with and without imputation approach, if time is sensitive to you now.

Vince.

[Weiwen Ng]

Thanks for your response Weiwen.

When I perform the -impute command-, it creates 20 new variables. My question is: how do I perform a regression after multiple imputation? Is there a way of averaging these 20 variables into one variable where there are no longer any missing values?

Also, do i perform the truncated regression separately? Or do I include it in a -mi impute chained- command?

If you want to use truncated regression for the variable holdings, you'd do it within mi impute chained. I believe this syntax should work:

mi impute chained (truncreg, ll(0) ul(1)) holdings (logit) credit (mlogit) convenience = age, add(20) rseed(1234) augment

You'd then use the mi estimate prefix to analyze the data.

mi estimate: regress ...