[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
Re: st: svrest & svresttest commands
At 03:47 PM 6/14/2004 -0700, you wrote:
I am using the svrmean command to calculate the mean of variable by race and
gender. I then want to see if difference in mean between males and females
among whites is significantly different from the gender difference in other
races. I gather that fro this I need to use the svrest and svresttest
commands. However, I do not know how to access the mean values after the
estimation command and do the test. This is what I am doing (0 and 1 stand
for males and females respectively):
. svrest "svrmean usc, by (smallrace pcpreq)"
error when command executed on original dataset
As author of these programs, I'll take a crack at this.
I think in this case you don't actually need -svrest-, since you can use
-svrmean- to get the basic means and their (co)variances. -svrest- is
designed to let you do replication-based variance estimation for models and
commands that don't have prepackaged svr-based programs already; this is
not the case for means. (Think heckman, or user-written -nl- programs, etc.)
So in this case you can just use -svrmean-, followed by -test-:
. svymean usc, by(smallrace pcpreq)
. test [usc]1-[usc]2-[usc]3+[usc]4=0
(You will need to look at the handy key provided by -svrmean- to get the
mappings between values of your by() variables and the values 1, 2, 3 and 4
displayed in the output).
Note also that -test- now recognizes the fact that the previous command was
a -svy-command, so there is no need for -svytest-; my -svr- commands
masquerade as -svy- commands in this context, so the adjusted test gets done.
Nicholas Winter 607.255.8819 t
Assistant Professor 607.255.4530 f
Department of Government firstname.lastname@example.org e
308 White Hall falcon.arts.cornell.edu/nw53 w
Ithaca, NY 14853-4601
* For searches and help try: