Can the Multiple Imputation Control Panel produce a single imputed dataset from a dataset with missing values?

[Kelly Hoffman]

Hi. I'm confused as to what type of output the Multiple Imputation Control Panel is capable of producing. I am using Stata/SE 15.1 for Windows.

I have 11 Likert questions in a larger survey that I want to imput missing values for. Roughly 20% of 355 survey respondents did not answer these 11 questions. I have tried to use Statistics > Multiple Imputation > Multiple Imputation Control Panel > Setup where I indicated 'Marginal long' for MI data style, 'Imputed' for registered variable type, and specified the variables to be imputed and the number of imputations (20).

From what I understood of MI, I thought Stata would perform as many imputations as I specified then produce a single, complete dataset with no missing values. However, once I go to the Impute tab on the Multiple Imputation Control Panel, I'm given a list of regressions to perform with the imputed data.

I imagine it is my inexperience and I apologize up front for that, but I don't want to perform a regression at present. I simply want to imput the missing values and have a single, clean dataset to work with. Is that not an option and I have to perform a regression, or am I missing something obvious?

Thank you in advance for any help provided.

[Clyde Schechter]

The imputations in multiple imputation are done by regressing the missing variables on other variables and using predicted values from those regressions plus a random piece sampled from the residual distribution of the regression. Stata is asking you to choose which type of regression you want to use for this process.

These are not the regressions that you will ultimately analyze your data with--those come later (in the Estimate phase) and are different.

[Kelly Hoffman]

Thanks very much for that answer, Clyde. I'll research the various types of regressions listed.

Have a nice day/night.

[daniel klein]

.
From what I understood of MI, I thought Stata would perform as many imputations as I specified then produce a single, complete dataset with no missing values.

No. Stata will keep the original dataset with missing values and additionally produce multiple complete datasets; the imputed values will be different in each completed dataset. The intuition behind multiple imputation is that no matter how good your guess about the missing values is, it is still a guess and the "true" values remain unknown. You want to reflect the uncertainty about those imputed values in the later analyses; you do this by introducing random variation in the imputed values.

Best
Daniel

[Kelly Hoffman]

Daniel, thanks for the explanation. I see my understanding of multiple imputation was very flawed.