Hi,
I am estimating a model using survey data. Since my model has a lot of
dummy variables, for which I need exact estimates, I am using devcon
after my survey regression. I then need to test a linear combination
of the coefficients. When I use lincom, I get an error saying: "matrix
has missing values." On the other hand, the test and nlcom commands
work just fine (so I can actually use nlcom to obtain the linear
estimates I need).
This is basically what I'm doing (in short):
svyreg lnincome age agesq female married S1 S2 S3
(where S* are dummy variables & S0 is the left-out reference group.)
devcon, groups(female male, married unmarried, S0 S1 S2 S3)
Now if I do:
test S0 + S1 - S2 - S3 = 0
or
nlcom (_b[S0] + _b[S1] - _b[S2] - _b[S3])
I get an answer with no issues. But if I do:
lincom S0 + S1 - S2 - S3
I get the error saying "matrix has missing values." Lincom works fine
after the regular estimation without using devcon. What exactly is
going on, & why this difference? Aren't lincom and nlcom both
estimating t-statistics? Is this a known issue or bug, or is there
something wrong with what I'm doing?
(Since I can get the estimates I need using nlcom, this is hardly a
pressing matter, but I thought I'd bring it to people's attention just
in case.)
My other minor devcon question: what is the best way to use devcon
with an interaction effect between two dummy variables (other than
generating all the 4 possible cases and using those)? In my case, this
would be the interaction: married*female.
Thanks a ton!
Wafa.
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
