Multiple Imputation: generating variables

[Rebecca Shellock]

Hi Members,

This question is in relation to generating variables following Multiple Imputation.

I am using multiple imputation for chained equations (MICE). The data is from a repeat cross-sectional dataset (n=643) with missing data for absolute income (n=345; 53.65%). In my original non-MI model I have used two alternative model specifications, one with log income and the second with the log of equivalised disposable household income.

I am looking to generate the equivalised disposable household variable, following MICE. Log income is entered as an imputed variable and I wanted want to know if it's possible to generate equivalised disposable household income from this variable following MI. Can it be done with mi xeq or mi passive?

For reference, log equivalised disposable household income is generated using the following code:

gen adulthh=householdsize-childrenhousehold
replace adulthh=. if adulthh<=0

gen HHMeqv=1+0.5*(adulthh-1)+0.3*(childrenhousehold)

gen eqv_hhdisposable= disposableincome/HHMeqv
gen eqv_lnhhdisposable= ln(eqv_hhdisposable)

Many thanks
Becca

[Clyde Schechter]

I am looking to generate the equivalised disposable household variable, following MICE. Log income is entered as an imputed variable and I wanted want to know if it's possible to generate equivalised disposable household income from this variable following MI. Can it be done with mi xeq or mi passive?

Yes, it's possible. But you should not do that. You should, instead, generate equivalised disposable household income from the unimputed data and then multiply impute that. That is the theoretically preferred way to do multiple imputation.