Complex Survey Software
From ICE Primer: A Tobacco Control Research Methodology Primer
[This page is under construction]
Software for analyzing data from complex surveys is continuously under development.
Currently (September 2008), SUDAAN, SAS, SPSS, Stata and R can all analyze data from complex surveys in at least the following circumstances:
(i) The sampling design is stratified multistage sampling with at least two PSUs sampled per stratum
(ii) There is a large number of PSUs in each stratum, compared to the number of PSUs sampled in the sample
(iii) The survey weights are approximately the reciprocals of the inclusion probabilities.
Under these circumstances the estimation of population totals and means is correct, and by extension the estimation of solutions of estimating equations is correct.
Generally at this time these survey packages do not take into account post-stratification or calibration of the weights to known totals, and to do that precisely requires more programming or the use of bootstrap weights. However, for many practical purposes the packages are adequate, because they do account for variability of the reciprocals of inclusion probabilities, and because they do account for the clustering in the sampling design. Since SAS assumes “with replacement” first stage sampling, so as to be able to ignore what happens at the second and later stages, condition (ii) is important for SAS. (The survey packages of SPSS, Stata and R will also handle some departures from (ii), as will SUDAAN. At the same time, if condition (ii) is not satisfied, the error estimates assuming “with replacement” first stage sampling may well be conservative.)
In each case it is necessary to declare the variable which indicates the stratum, the variable which indicates the PSU (often termed cluster), and the weight variable. In SPSS, R and Stata we construct a survey design object which specifies these three variables, and then we can refer to it in the analysis code. In SAS, these declarations are part of the set of commands for the procedure.
Some useful websites are as follows:
