Interviews were used to gain information about demographics, employment histories, income sources and amounts, clinical symptoms and other relevant information at the beginning of the study and then at 6 month intervals for 24 months. A variety of employment data was gathered such as hours worked, wages, job duties, benefits. Psychiatric symptoms were assessed using the PANSS a semi structured rating scale.
Bivariate analysis were conducted to test for significant differences in study conditions. Two vocational outcome variables were computed for every month of the 24 month follow up period: competitive employment and work for 40 or more hours in a single month. The independent variable was the local unemployment rate for the geographic area surrounding each program. Some demographic and clinical covariates that should be included in multivariate models predicting employment were excluded due to their high correlation with other variables.
Unemployment rates over times were inspected for each of the seven states in which study sites were located. Then hierarchical random regression analysis using time varying and fixed covariates were used to determine the effect of unemployment rate on each of the two employment outcomes. Random effects logistic regression modeling, addressed issues found in longitudinal multi-site data such as missing observations, fixed versus time varying covariates and more.