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]

st: -khb- updated on SSC

From   Ulrich Kohler <>
To   statalist <>
Subject   st: -khb- updated on SSC
Date   Tue, 12 Apr 2011 18:19:18 +0200

Courtesy of Kit Baum, the package -khb- for the decomposition of effects
in non-linear probability models has been updated on SSC. Users of the
previous version are asked to update their version with

. adoupdate khb, update

Those who like to install -khb- the first time should type

. ssc install khb


(I'm not sure if this is necessary, but if you want to use -khb-
immediately after installation, you might have to update your Mata
libraries first. This is done with 

. mata: mata mlib index

or by restarting Stata.)

Whats new

(1) The main output has changed. In previous versions of -khb-, the
decomposition output was cast in terms of total, direct, and indirect
effects. We now use slightly more general expressions, namely "full" for
the coefficients of X variables controlling for Z, "reduced" for the
coefficients of X not controlling for Z and "diff" for the difference
between the "full" and "reduced" coefs.  This terminology is in line
with that in Karlson, Holm, and Breen (2010).

Second, the standard errors for the difference in effects (the
indirect effect) now fully implements the delta method (as described
in the most recent version of Karlson, Holm, and Breen 2010). The
standard errors also take into account the covariance between the
parameters of auxillary equations (using a system of seemingly
unrelated equations).

Third, the concomitant function has been changed to also include
concomitants in the auxillary equations (those equations which are
used for the residualization of control variables) in addition to
including them in the main equations. This strategy differs from that
in Karlson and Holm (2011), because they included concomitants only in
the main equations. The concomitant function now controls the full
decomposition for the confounding influence of concomitant variables.
In practical terms, this means that concomitants now are treated as
x-variables, whose output is not displayed. If researchers are
interested in learning about the extent to which these concomitants
are mediated by the z-variables, they should enter them as x-variables
in -khb-."

Many regards


*   For searches and help try:

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