Dunn et al. (1999) discuss issues in analyzing two-phase data to estimate prevalence of psychiatric disorders. Participants in a phase I sample were given the General Health Questionnaire (GHQ) and classified into three strata based on their GHQ score. The stratification was used to take a stratified random sample of 250 persons for the phase II sample; the Composite International Diagnostic Interview (CIDI), considered to be a more accurate diagnostic tool, was administered to each person in the phase II sample. The CIDI score was used to classify the phase II sample members as having at least one psychiatric disorder (case) or having no psychiatric disorder (non-case). The results are given in the following table. The counts of cases and non-cases are from the phase II sample.

a. Calculate the phase II sampling weight w (2) i for each stratum.
b. Use the two-phase sample to estimate the percentage of persons with at least one psychiatric disorder, along with its standard error. Since we do not know the population size N, use a relative phase I weight of wi (1) = 1.

Stratum Th m Non-case Case MA GHQ 3 (low) GHQ = 4, 5 (medium) GHQ 2 6 (high) 1049 237 272 60 48 142 14 23 27 34 119 Total 1558 250