Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
From | Austin Nichols <austinnichols@gmail.com> |
To | statalist@hsphsun2.harvard.edu |
Subject | Re: st: Testing whether two estimated survey means are statistically different |
Date | Wed, 19 Oct 2011 11:40:22 -0400 |
Andrew Wade <awade@allenconsult.com.au>: The hypothesis that the mean for a specific sub-population is not different from the mean for the whole population is equivalent to the hypothesis that the mean for a specific sub-population is not different from the mean for its complement. I.e. regress on a dummy for the specific sub-population. The test that the coef on the dummy=0 is the test you want. On Wed, Oct 19, 2011 at 10:45 AM, Andrew Wade <awade@allenconsult.com.au> wrote: > Hello, > > I want to test the hypothesis that the estimated mean of a specific sub-population, is not statistically different from the estimated mean of the whole population. > > The means are generated from a complex survey, using two separate operations of the svy: mean command. > > My reading of the documentation is that this type of test, when using svy: mean, should be conducted using lincom. > > The problem is that lincom can only be applied (it would appear) to the most recent mean estimate. > However, to estimate the means of interest, I first have to estimate the svy: mean for the population, and separately use the svy:mean, over, command to estimate the mean for the specific sub-population of interest. > > Is there a way of 'loading' into the lincom command a previously stored mean and variance estimate, to then test against a separate estimate? > > And presumably, the fact that one of the means of interest is a sub-population of the other mean, is a potential problem? > > Or should I simply use the manual ttesti command for this test? > > The code for this is: > > ttesti 6047 293.9012 9256.260402 79889 767.4463 39570.47905 > > The n values here are the weighted population counts. > As noted above, the 6047 is a subset of the 79889. > > (it is notable here that the CI's for the two means overlap, but that the test doesn't find the two estimates to be statistically difference from each other) > > Or should I assume unequal variances... > > ttesti 6047 293.9012 9256.260402 79889 767.4463 39570.47905 , unequal > > This 2nd ttesti approach does find the estimates to be statistically different to each other, due to a much lower s.e. estimate. > > Keen for any suggestions. > > Regards, > > Andrew > > > Andrew Wade | Principal Consultant > The Allen Consulting Group > > Level 9, 60 Collins Street, Melbourne VIC 3000 > T: 03 8650 6000 | DDL: 03 8650 6043 | F: 03 9654 6363 | M: 0448 367 298 > awade@allenconsult.com.au | www.allenconsult.com.au > * * 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/