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

[Ole Dahl Rasmussen <odr@dca.dk>]

Thanks for the answers and suggestions. Permutation seems very relevant.

Let me provide a few more details, in particular on the issues raised by Stas Kolenikov.

The population I want information on is the 7100 households living in the 46 villages. I do not want to generalized beyond this area. Using lists of the 7100 households, I've sampled 1775 households for interviews. Since 1775 is a reasonable share of 7100, I think I need to do FPC.

The sample was done randomly within the 90 strataIDs. All Strata were included in the survey and thus all villages were included (whereas only half of the villages were included in the treatment). That's why I have no sampling units - I do not sample units, just households in strata. I do, however, use the sampling unit feature of svyset because I want Stata to cluster standard errors at the village level. 

You write that I should omit the -fpc()- in the first stage. How can I then cluster SEs at the village level?

It seems that I should look other's work on the combination of randomized treatment and stratified random sampled surveying. 

Thanks again.

Ole


      pweight: weights
          VCE: linearized
  Single unit: certainty
     Strata 1: <one>
         SU 1: vid
        FPC 1: one
     Strata 2: strataID
         SU 2: <observations>
        FPC 2: f


 

Ole Dahl Rasmussen
Rådgiver indenfor mikrofinans og evaluering. Ph.d.-studerende på Syddansk Universitet
Adviser on Evaluation and Microfinance. PhD-student at University of Southern Denmark
 
Nørregade 15
1165 København K / 1165 Copenhagen K
www.noedhjaelp.dk / www.danchurchaid.org
 
odr@dca.dk
 
P +45 33152800
M +45 29699145
SKYPE odrdca
 
VI TROR PÅ ET LIV FØR DØDEN
 
 


*
*   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/