Estimating the impact of selected programs on participants' subsequent welfare dependence and employment in British Columbia
This thesis is concerned with estimating the impacts of employment and training programs without the use of random assignment. It reviews the literature on the American CETA programs that led researchers to conclude that random assignment was needed to produce useful estimates. It reports an investigation of selection bias that yields four findings. It is possible to test for selection bias in the absence of random assignment. Pre-program tests for selection bias are not valid. Selection bias due to unmeasurables is small. Controlling for changes in major explanatory variables such as pre-program employment is crucial. It shows that the CETA data were inadequate to control for these changes. This thesis reports the finding that 25% non-response in a survey can lead to qualitative changes in the estimates of program impact. It illustrates the way in which undetected non-linearities can bias estimates. It reports estimates of the impacts of a range of programs. Wage subsidies with the private sector have a large (ten percentage points) sustained beneficial impact on subsequent welfare dependence and employment. On-the-job training in community projects (make-work) has no long-term impact on welfare dependence or employment. Within classroom training, upgrading (adult basic education) has no impact on subsequent welfare dependence. Vocational training has a large (15 to 20 percentage points), and sustained beneficial impact on subsequent welfare dependence. The job club studied in this thesis had a significant short term beneficial impact, but no long term impact.