use of svyset with data from two countries and post-stratification within country - fpc specification

[Michelle Gosse]

Hi,

Sorry, fpc should not be in the topic heading.

I'm relatively new to Stata and new to using the svyset functions. I have Stata/SE 13.1 for Windows.

Very soon I will be receiving a dataset of survey responses from both Australian and New Zealand respondents. The data will be post-stratified within country, at the age*sex*region level. If one does not exist when I receive the data, I will be able to construct a suitable poststrata variable and associated postweight values.

I have read the Stata manual for svyset and have downloaded the poststrata.dta dataset and looked at the coding and set-up commands. However, I can't find an example of how to analyse cross-country data where the weights are country-specific. I have searched online and the Statalist, and watched the videos on YouTube.

The example given for the cat/dog visits for design set-up is: svyset, poststrata(type) postweight(postwgt)

It appears that this type of design specification won't meet my needs (only Australian data contributes to Australian estimates, and only New Zealand data contributes to New Zealand estimates):

svyset, strata(country) poststrata(age*sex*region combination) postweight(the-weights-used)

Do I need to analyse the data using separate datasets for each country (in which case, there is no strata to specify)?

I was hoping to keep the data in the one dataset and undertake the same analysis on a within-country basis simultaneously but the documentation suggests that the values will be produced on the basis of all data in the dataset, so I will need separate datasets.

[Steve Samuels]

The cat/dog example in the Manual is too simple and your original country-specific svyset statement is not correct. You can see a glimpse of the correct poststratification setup if you look at the end of the Stata video on poststratification, linked to on page 55 of the Survey Manual (Version 14).

I assume that there were PSUs ("psu") probability weights ("sampwt"), and sampling strata ("sampstrat") in the original surveys. Stata needs the original sampling weight to create what is called the "the poststratification adjusted sampling weight" and it needs the sampling strata to compute standard errors. The formula for the adjusted weight is on pages 56 of the Version 14 Survey Manual .

The single country survey postratification setup should be:

svyset psu [pwt = sampwt], strata(sampstrat) poststrata(poststratum) postweight(postwgt)

where "poststratum" and "postwgt" are the names for those variables.


To do the combined analysis, first make sure that variable names and poststratum levels are identical in the NZ and Australian data sets. When they are, append one data set to the other to create the combined data. Then:

egen new_sampstrat = group(country sampstrat)
egen new_poststrat = group(country poststratum)
svyset psu [pwt = sampwt], strata(new_sampstrat) poststrata(new_poststrat) postweight(postwgt)

Whether the fpc() option should be in the svyset statements depends on whether you plan a descriptive study or one that is "analytic", i.e. it will test hypotheses, quote p-values. If you wish to investigate this further, please start a new Topic.

[Steve Samuels]

I'll just add that nothing prevents you from doing country-specific analyses with the combined data set.

use combined data, clear
bys country: do something

do something if country == 1
keep if country ==1
do something