--- On Sun, 12/4/09, Kanter, Rebecca wrote:
> I am using factor analysis to construct a socio-econonomic
> status (ses) variable for my dataset (and did so), but
> noticed that many households are missing (MAR) information
> on "drainage;" thus I wanted to use multiple
> imputation methods to impute values for drainage (a
> non-ordinal categorical variable) so that there wouldn't
> be thousands of households with a missing ses. I wanted to
> impute the value based on the values each household has for
> the other variables used in the factor analysis (i
> constructed all the categorical variables into binary ones)
> and thought I could do this MI via uvis, below. At first, I
> thought I kept getting the below error because some
> households were missing information on drainage and another
> variable in this list, so I corrected that by removing them
> from the uvis (via the draintag==., if draintag==1 then they
> are missing drainage info and info on another variable
> listed below); but i still get the same error. If anyone can
> help me figure out uvis!
> and/or another way to impute the missing drainage values
> that would be much appreciated. (And just a note, a1-r4 and
> o1-k4 below are a series of household characteristic binary
> variables).
>
> . uvis mlogit drainage a1 a2 r1 r2 r3 r4 radio modcon eapp
> car camioneta vehic tvbw tvcolor refrig gastove othstove
> washmac boiler comp microwave phone blender vcr ventilador
> loghexp nfam crowding o1-k4 if draintag==., gen(md2) boot
> [imputing by drawing from conditional distribution with
> bootstrap]
> [perfect prediction detected: using augmlogit to impute
> drainage]
> equation 0 not found
> r(111);
I have several comments:
First, the error message gave you a suggestion on what the
problem is and on how to solve it: You have perfect
predictions and you should augmlogit instead. So the problem
is not that some other variables also have missing values,
-ice- is explicitly designed for that.
Second, you should never directly use -uvis-, it is just a
helper program used by -ice-.
Third, you should not only include the variables that go
into your factor analysis, but also the variables that
go into the final model of interest, including the depedent
variable of that model.
(Shameless self-promotion warning: I wrote the programs
that I think are more appropriate for creating your ses)
Fourth, factor analysis does not seem like a very
appropriate technique for your problem. I see at least two
problems: a) at least one of your indicators is a
categorical variable and factor analysis is designed to
handle those. b) factor analysis assumes that ses influences
your observed variables, while in many cases it is more
reasonable to assume that it is the other way around. Say
that one of the other indicators is the respondent's
education, and that the respondent improves the drainage of
his house, then the factor analysis model thinks that this
improvement in drainage is the result of an improvement in
ses, as a consequence all other indicators (including
education) should experience a similar improvement.
The alternative model would posit that the observed
indicators influence the ses. So ses is the collection of
economic and social resources available to the individual/
household. So if an individual improves his drainage, than
this will improve the ses, but does not force a similar
improvement in the other indicators, like education. An
added bonus is that these alternative models can naturally
include categorical variables. You can estimate these models
using -sheafcoef- and -propcnsreg-, both available from SSC.
Type in Stata -ssc d sheafcoef- and -ssc d propcnsreg- to
find out more. The helpfile of -propcnsreg- contains a more
detailed describtion of the model and references.
Hope this helps,
Maarten
-----------------------------------------
Maarten L. Buis
Institut fuer Soziologie
Universitaet Tuebingen
Wilhelmstrasse 36
72074 Tuebingen
Germany
http://home.fsw.vu.nl/m.buis/
-----------------------------------------
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
