st: variation in estimation sample with mi estimate & svy

[Marcus L Britton <britton@uwm.edu>]

Hi, folks.  I'm using the National Education Longitudinal Study (NELS:88) data to predict whether Asian, Black, Hispanic and White respondents live with their parents at age 26.  I used Royston's ice package to impute missing data (m = 20 imputations) on a number of variables in the dataset, including the measures of race/ethnicity and immigrant generation (1.5 generation, 2nd generation, or 3rd or greater generation).  I then imported my data into mi format for Stata 11 using the mi import ice command and used the mi passive: command to generate dummy variables that indicate immigrant family status (0 = 3rd generation or greater; 1 = 1.5 or 2nd generation) and specific combinations of race/ethnicity & immigrant family status (e.g., asianimm = 1 for Asians from immigrant families and 0 otherwise).  

Here's the problem: I am attempting to estimate two sample tests of proportions to assess whether the proportion of respondents living with their parents varies significantly by gender within groups defined by either (1) race/ethnicity (e.g., Asians) or (2) race/ethnicity and immigrant status (e.g., Asians from immigrant families).  I'm using the mi estimate and svy commands, since the NELS:88 was collected using a complex sampling design.  I use commands of the following form to do this:

mi estimate (diff: [_prop_2]male - [_prop_2]female), dots post: svy, subpop(asian): proportion f4liverents, over(female)

mi testtransform diff

where f4liverents is the dummy variable indicating whether the respondent lives with his or her parents and _prop_2 refers to the estimated proportion that did so for either male or female respondents.

The above commands work just fine, even though I have imputed missing values for race/ethnicity, with the result that the number of respondents categorized as Asians varies (slightly) across the m = 20 imputations.  (A "note" does appear in the output indicating that the "number of observations in a subpopulation varies among imputations.")

However, when I use the the same type of command for specific combinations of race/ethnicity and immigrant family status, I get an error message.  For example, the command for estimating the test for Asians from immigrant families is as follows:

mi estimate (diff: [_prop_2]male - [_prop_2]female), dots post: svy, subpop(asianimm): proportion f4liverents, over(female)

mi testtransform diff

which produces the following error message:

Imputations (20): 
 .x
estimation sample varies between m=1 and m=2
r(459);

According to the mi estimate documentation, I should be able to use the esampvaryok option to override this error message, but when I do, I get this error message:

Imputations (20):
 option esampvaryok not allowed
an error occurred when svy executed logit
an error occurred when mi estimate executed svy: logit on m=1

Any ideas on why this is happening or what I should do about it?

Thanks in advance,
Marcus Britton


-- 
Marcus L. Britton
Department of Sociology
University of Wisconsin-Milwaukee
P.O. Box 413
Milwaukee, WI  53211
(414)229-5308
britton@uwm.edu
http://www4.uwm.edu/letsci/sociology/faculty/britton.cfm
*
*   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/