-khb- and multiple imputed data

[Kjell Weyde]

Hi!
I am doing mediation analyses with continuous iv and m, and binary dv. Is it possible to use the user-written -khb- (Kohler and Karlson) when you have multiple imputed data? For example, using

mi estimate, cmdok: khb....

Kjell Weyde (Stata 15)

[Marcos Almeida]

I have no experience with this user-written program, but the help files show no option for - mi estimate - command.

[daniel klein]

Conceptionally, I believe that khb can be combined with mi. Technically. it is also possible. However, some of the more useful options, such as disentangle, will not work correctly or will not work at all.

Best
Daniel

[Kjell Weyde]

Thanks, Daniel and Marcos. I ran the code and got the following output:

mi estimate, cmdok: khb logit ADHD_NPR logAs || logTSH_NEO, concomitant(logHg logCd logMn logPb Se_std KJONN MORS_ALDER Maternal_edu Parity MADHD_SS) summary disentangle


Multiple-imputation estimates                   Imputations       =         50
Decomposition                                   Number of obs     =      1,174
                                                Average RVI       =     0.0446
                                                Largest FMI       =     0.1044
DF adjustment:   Large sample                   DF:     min       =   4,527.67
                                                        avg       = 422,896.26
                                                        max       = 659,580.80

------------------------------------------------------------------------------
    ADHD_NPR |      Coef.   Std. Err.      t    P>|t|     [99% Conf. Interval]
-------------+----------------------------------------------------------------
     Reduced |   .0934736   .0899364     1.04   0.299    -.1381878    .3251351
        Full |   .0963671    .089954     1.07   0.284    -.1353395    .3280738
        Diff |  -.0028935   .0045747    -0.63   0.527    -.0146821    .0088951
------------------------------------------------------------------------------

So, it works, but, as Daniel pointed out, the disentangle option does not. According to this Stata web page: https://www.stata.com/support/faqs/s.../cmdok-option/ , one of the criteria for the mi estimate to work properly with an estimation command, is that the estimation command "Store the residual degrees of freedom in the e(df_r) scalar; if this is not applicable for the particular estimator, an e(df_r) scalar should not be returned at all." The -khb- does not seem to do this. Can I still trust the results I get?

Kjell