Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: mi impute chained, interactions, and the force option

From (Wes Eddings, StataCorp)
Subject   Re: st: mi impute chained, interactions, and the force option
Date   Mon, 07 May 2012 17:51:42 -0500

Jonathan Bartlett ( received a "missing imputed
values" error after using -mi impute chained-:

> To do this using Stata 12's mi impute chained command I have tried the
> following syntax:
> mi impute chained (reg, include((x2*y))) x1 (reg, include((x1*y))) x2 = y,
> add(10)
> This results in the following error being displayed:
> x2: missing imputed values produced
>     This may occur when imputation variables are used as independent variables
>     or when independent variables contain missing values.  You can specify
>     option force if you wish to proceed anyway.
> If I use the force option the command runs, and the estimates I obtain appear
> to be similar to those I get using ice, suggesting it is fitting my desired
> imputation models correctly. However, I am concerned that I have had to use
> the force option, and am wondering whether I should be using a different
> syntax in order to specify my desired imputation models without needing to use
> the force option? If anyone can shed any light on this I would be most
> grateful.

The "missing imputed values" error that Jonathan received often occurs when a
right-hand-side variable, such as the variable y in Jonathan's example, contains
missing values.  One can use the -force- option, as Jonathan did, to suppress
the error, but then the observations with the missing values of y will be
excluded from the imputation, and the corresponding missing values of x1 and x2
will not be imputed.  If excluding those values of y seems unreasonable, it
would be best to impute y along with x1 and x2.

More generally, "missing" imputed values may also arise when the estimates of
the model parameters underlying the imputation model are such that the
corresponding draws of the imputed values result in missing values.  This may
often happen when the observed-data estimates of the model parameters are
unreliable or approach their boundary values.  In Jonathan's example, another
possibility for "missing imputed values" is large standard errors for some of
the coefficients.  This can be visually checked by specifying -mi impute
chained-'s option -noisily-.

If the y variable doesn't have missing values and the intermediate regression
results displayed by -noisily- appear reasonable, we would ask Jonathan to send
his data and do-file to to determine the cause of this
error message.


*   For searches and help try:

© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index