Multiple Imputation style

[Senor Edward]

Withing multiple imputation, you can set your style as wide, mlong etc. I select which one to use based on my data set's format. Once selected, you register your imputed and regular variables and I run the chained multiple imputation before the final analysis..
i.e.
mi set wide
mi register imputed marriage_Stat
mi register regular age sex cancer_Found
mi impute chained (mlogit) marriage_Stat = age sex cancer_Found, add(10) rseed augment
mi estimate: logit cancer_Found age i.sex i.marriage_Stat

However I was having doubts on which data set format to use i.e. wide vs mlong.
I then read that once the multiple imputation command has started, you can switch between formats using: mi convert which I was mainly looking at for curiosity.

However, when I change data set format, all of my values from analysis change. This hypothetically makes sense, but I'm trying to figure out why the results would be different. I'm fairly certain that my style is actually wide, but I wonder if I can change format after the declaration, why would the results from wide and mlong be different?

[daniel klein]

This is the second post today, claiming that results differ between the styles. My intuition here, without seeing results, that there is something fundamentally wring with the data. The style should mostly be a matter of taste and certainly not affect the obtained results. Why do you think that different results would "hypothetically make sense"? And, what are the results that differ?

Best
Daniel

[Senor Edward]

I say hypothetically because under mlong, new observations are added to the data for the imputed where as under wide, new variables are added to the data. I was assuming that this would have an impact, but after reading up more on imputation, the technique knows which are the imputed values and factors for them anyway.

Is mlong and wide supposed to give the same results and the only difference is the manner in which the new imputed values are stored in the data set?

[daniel klein]

I
Is mlong and wide supposed to give the same results and the only difference is the manner in which the new imputed values are stored in the data set?

Yes. There are exceptions for so-called super-varying variables that can only be stored in flong or flong sep.

Best
Daniel