Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: Finite population correction with clustering of SE at a different level than the strata


From   Steve Samuels <[email protected]>
To   [email protected]
Subject   Re: st: Finite population correction with clustering of SE at a different level than the strata
Date   Sun, 10 Jun 2012 22:18:58 -0400


I should have added: In Stata, you can do the randomization version of any test with -permute-.  Among the tests to consider are the two-sample t-test or a rank test like -kwallis-.   You can compute the CI for a parameter by inverting the test as discussed (briefly) at www.epibiostat.ucsf.edu/biostat/sen/statgen/permutation.html. With 23 villages per group, you will probably need to simulate the p-values for each parameter, and that web page recommends 10,000 replicates for high precision. Also  Roger Newson's -somersd- (from SSC) will compute CIs for Hodges-Lehmean median difference (which is not the same as the difference in medians).


Steve
[email protected]


Inference can also be based on the treatment randomization (e.g.http://www.ncbi.nlm.nih.gov/pubmed/8804140), no superpopulation needed. Randomization-based hypothesis tests would be done on village-level estimates, and confidence intervals calculated by inverting the tests.  

There is, of course, extra information about individual households that can be mined. The village-to-village contamination that Ole reports is a problem for which I have no advice. 

Steve
[email protected]

On Jun 6, 2012, at 7:00 AM, Ole Dahl Rasmussen wrote:

Dear Austin

Thanks for your response. Let me try to clarify.

I'm running a randomized trial. 23 villages do not get treatment. So there's not complete correlation between interested and treatment. 23 villages do get treatment, which is an offer to participate in a microfinance intervention. Interest was elicited in all village prior to randomization. The regression is supposed to give the intent-to-treat estimate and I could IV for the effect of treatment on the treated, except that I have contamination.

Best,
Ole





© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index