Analysis of pooled data with different survey strata

[Anne Golla]

I want to compare means & do regression analysis on 2 waves of househlold survey data (2 separate years, LSMS) using svy commands; issue is that in 2008 4 general regions were used for strata, in 2012 they split these regions into 24 specific strata (12 prefectures X rural/urban). The strata in 2012 are subdivisions of the strata in 2008. The psu in each case was the household, as I understand I need to recode that so each is unique in the pooled data, and also adjust hh & ind weights by dividing by 2 - but how do I deal with the strata? Thanks!

[Steve Samuels]

You don't need to recode the households; PSUs are identified by the household stratum combination, so that the same household ID used in two different strata will identify two unique PSUs. (I assume from your statement about "unique" households that none appeared in both years. If that's not the case, please describe the overlap.)

To handle the strata, you will renumber to create a a "super-stratum" with 28 unique values (1-4 from 2008, 5-28 from 2012). Assuming that strata are numbered 1-4 in 2008 and 1-24 in 2012:

gen sstrat = stratum if year ==2008
replace sstrat = stratum +5 if year==2012

[Steve Samuels]

Of major concern, I know of no household surveys of large areas in which individual households are the PSUs). Please provide details of the sampling design. If necessary, link-to or attach the relevant study document.

Of minor concern: For estimating all parameters except totals, division of the weights by 2 is unnecessary, since it leaves the paramters unchanged. For estimating yearly totals, division of the weights by 2 is obviously incorrect. If you estimate a total for the combined sample after dividing the weights by 2, you must interpret it as the average total per year.