# re: st: analysis of complex surveys:

 From Steven Samuels To ealhussainy@worldbank.org, statalist@hsphsun2.harvard.edu Subject re: st: analysis of complex surveys: Date Thu, 20 Jul 2006 15:04:12 -0400

Correction: The second test statement should, of course, be:

test Z1=Z2

Sorry,

Steve

1. To pool the results from the different surveys you may treat the country as another stratifying variable. Assume that the name of the within-country stratification variable is
"stratum". Then do the following:

egen superstrat=group(country stratum)

User "superstrat" as your new stratum variable.

2. To make cross-country comparisons of regression coefficients you may also add country indicators to the regression equation. Then you may test country-variable interactions. Alternately, you can use the noconstant option to fit a separate set of regression coefficients for each country. Suppose you have 2 countries, 1 and 2, and your covariates are X and Z

form
I1 = 1 if country =1 =0 if Country = 2
X1= X "
Z1 = Z "

I2 = 1 if country=2 , =0 if country =1
X2= X "
Z2 =Z "

svy: reg y = I1 X1 Z1 I2 X2 Z2, nocons

test X1=X2
test Z1=Z2

etc.

Good luck!

From: ealhussainy@worldbank.org
Subject: st: analysis of complex surveys

Dear all,

I am attempting to analyze a set of complex household surveys that used a
standard questionnaire but were conducted in different countries using different
sampling methods (multistage stratification and clustering). I have household
weights as well as stratification and clustering information for each survey. My
question is two-fold:

1. Is it possible to "pool" or in some way combine data across surveys for a
fixed-effects regression analysis?
2. If I run regressions at the country-level, is there a way to make
cross-country comparisons using regression coefficients or marginal effects?