Data and Documentation
The GTSS uses a two-stage sample cluster design. To reflect the complex sample design, there are two sample design variables on your data set named STRATUM and PSU (acronym for Primary Sampling Unit).
The variable STRATUM usually consists of two schools that are paired so that both schools have similar enrollment sizes. However, sometimes a STRATUM may have only one school. For example, if a school has 100% chance of being in the selected school list (due to large enrollment) it will be the only school in that stratum, and we call this type of school a Certainty School.
In most cases, the Primary Sampling Unit represents a school. If the school is a Certainty School then the PSUs are the classes within the school.
The sampling weight variable is named FINALWGT.
Each student in the data set is assigned a sampling weight, which accounts for the following:
- Selection probability of the school
- Selection probability of the class
- Distribution of the population by grade and sex
- Non-responding schools
- Non-responding students
- Non-responding classes
Point estimates and 95% confidence intervals can be calculated using several software packages for statistical analysis of correlated data. Below are sample codes for EPIINFO, SUDAAN and STATA.
EPIINFO Sample Code:
FREQ CR3 STRATAVAR = Stratum WEIGHTVAR=FinalWgt PSUVAR=PSU
SUDAAN Sample Code:
proc sort data = sasdata.dataset;
by stratum psu;
proc crosstab data = sasdata.dataset design = wr;
nest stratum psu/missunit;