![](/web/20060126174351im_/http://www11.hrdc-drhc.gc.ca/edd-img/spacer.gif) |
![](/web/20060126174351im_/http://www11.hrdc-drhc.gc.ca/edd-img/spacer.gif) |
9.0 Applying the Lessons: Choosing Comparison Groups for EBSM
Lessons from previous quasi-experimental evaluations can assist the process of drawing regional
EBSM comparison groups:
- To date, there is little evidence that the matching procedure used makes much difference
to the accuracy of estimates from quasi-experiments. Thus, the simplest course would be
to ensure the comparison group meets the eligibility requirements for the training program.
At the very least the comparison group should be drawn from the same population as the
treatment group. EBSM programs, which are primarily aimed at EI recipients, could use the
HRDC longitudinal file for choice of comparison group subjects, for example. The NESS
database may also be a useful source of comparison group members.
- Individual components of EBSM have specific eligibility criteria that should be taken into
account when selecting a comparison group. For instance, targeted wage subsidies are
largely focused on those who face particular obstacles to employment such as disabilities.
The sample frame for the comparison group should come from the subset of the EI
population who share these obstacles.
- Moving one step beyond simple eligibility matching is advisable in the case of regional EBSM
programs: it seems clear that comparison groups should at least come from the same
region. In the case of Transitional Jobs Fund, the comparison group should be chosen from
the same communities as the treatment group (i.e., those with unemployment rates of 12%
or higher). If desired, and assuming the requisite information is available from the database
used, other potentially important matching variables are labour force status changes, age,
education, marital status, family income, and time period.
- Sampling strategy depends on the matching procedure employed. The easiest approach --
and one that would be no less satisfactory than more complex ones judging by the literature
-- would be to first limit the population to those in the region who were on EI during the
period that the program was in operation (plus any other eligibility factors unique to each
EBSM component). Then a simple random sample could be selected. A more precise
matching strategy would be more difficult to implement, but by no means impossible. For
example, if one wanted to match by sex, age and education, a series of dummy variables
(0-1) could be set up: men=0, women =1; under 30 years old=0, 30+=1; less than high
school=0, high school gradute=1. A three-digit variable could then be computed for each
participant and non-participant with the first digit representing sex, the second age, the third
education (e.g., a 29 year old female high school graduate = 101). The software program
could then be used to select at random, a match (or more than one match) for each
participant.
- As for sample size, standard formulas should be employed, keeping in mind that they
calculate margins of error for final sample size rather than the original sample size. Appendix
D summarizes how to choose an appropriate sample size.
|