multiple imputation

[abdelilah arredouani]

Dear All,
I have a question concerning the use multiply-imputed data. i want to d some descriptive statistics by group after multiple imputation but i found out by searching that it is not possible. my question is whether it is possible to take the 10 imputations (say we have only 1 variable) average them and then use the average for descriptive statistics.

thanks in advnace


stata 15.1 Mac

[Carlo Lazzaro]

Abdelilah:
daniel klein 's excellent communiity-contributed -misum- is, in all likelihood, what you're seeking (just type -search misum- from within Stata to spot and install it).

[abdelilah arredouani]

thanks carlo. it works. what if i want to compare means with ttest?
cheers

[Carlo Lazzaro]

Abdelilah:
why not using -regress-?

mi estimate: regress <depvar> i.group

You should get something similar to the the following toy-example:

. use http://www.stata-press.com/data/r15/mhouses1993s30
(Albuquerque Home Prices Feb15-Apr30, 1993)

. mi estimate: regress price i.custom

Multiple-imputation estimates                   Imputations       =         30
Linear regression                               Number of obs     =        117
                                                Average RVI       =     0.0000
                                                Largest FMI       =     0.0000
                                                Complete DF       =        115
DF adjustment:   Small sample                   DF:     min       =     113.05
                                                        avg       =     113.05
                                                        max       =     113.05
Model F test:       Equal FMI                   F(   1,  113.1)   =      51.27
Within VCE type:          OLS                   Prob > F          =     0.0000

------------------------------------------------------------------------------
       price |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
    1.custom |   499.2556   69.72621     7.16   0.000      361.116    637.3951
       _cons |   947.5222   33.49539    28.29   0.000     881.1621    1013.882
------------------------------------------------------------------------------

.

[abdelilah arredouani]

thanks carlos. just to understand how to interpret the table. the mean for 0.custom is 947.52+/-33.49 and the mean for 1.custom is 947.52+499.25 = 1446.77+/-69.7. the p value to use is the top one. am i right? in this case the price mean is significantly different between custom and non custom.

one last question: how do i obtain SD from the table?

regards

[Carlo Lazzaro]

Abdelilah:
your interpretation is correct.
I do not think you can (easiliy) obtain SD from the table, nor I think it's useful, as the standard error is actually the standard deviation of the sample distribution of each coefficient.
That said, you can probably recover SD from -misum-.