02.06.2014 Views

Evaluation of the Australian Wage Subsidy Special Youth ...

Evaluation of the Australian Wage Subsidy Special Youth ...

Evaluation of the Australian Wage Subsidy Special Youth ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

244<br />

Table 7.6 shows <strong>the</strong> results <strong>of</strong> <strong>the</strong> new specification <strong>of</strong> <strong>the</strong> probit including only <strong>the</strong>se<br />

variables. The estimated coefficients that are statistically significant are compared to<br />

those in <strong>the</strong> original specification. Age and marital status remain statistically significant<br />

at conventional levels; however partner’s employment loses statistical significance. The<br />

location <strong>of</strong> Western Australia/Tasmania retains a positive influence on participation. This<br />

is consistent with <strong>the</strong> administrative data that showed Western Australia to have a much<br />

higher SYETP placement rate than o<strong>the</strong>r states [see section 2.2.6]. Qualifications lose<br />

<strong>the</strong>ir impact on SYETP participation in <strong>the</strong> reduced form model, whereas in <strong>the</strong> original<br />

model year 12 schooling had a positive impact. A longer qualifying spell <strong>of</strong><br />

unemployment still has a positive impact, and past experience <strong>of</strong> a job held for more than<br />

3 years still has a negative impact on SYETP participation. O<strong>the</strong>r city or country town<br />

kept <strong>the</strong>ir negative impact on SYETP participation in <strong>the</strong> reduced model. The only key<br />

change for variables included in both <strong>the</strong> full and reduced models is that partner’s<br />

employment and highest qualification <strong>of</strong> year 12 lose statistical significance in <strong>the</strong><br />

reduced form.<br />

The predictive power <strong>of</strong> <strong>the</strong> probit is 92 per cent. This is <strong>the</strong> same power as <strong>the</strong> original<br />

model. Thus <strong>the</strong> excluded variables do not change <strong>the</strong> predictive performance <strong>of</strong> <strong>the</strong><br />

probit. This measure <strong>of</strong> predictive power is only very rough however. O<strong>the</strong>r measures <strong>of</strong><br />

fit indicate <strong>the</strong> reduced specification to be slightly better, with <strong>the</strong> AIC lower at 0.53<br />

compared to <strong>the</strong> former models values <strong>of</strong> 0.56 and 0.57 (note <strong>the</strong> sample size has<br />

increased as certain variables dropped had missing cases).<br />

The propensity scores, for both SYETP participants and comparisons, from <strong>the</strong> new<br />

specification are shown in Table 7.7. These are compared to those found earlier [see<br />

section 6.2.2]. This change to <strong>the</strong> specification has quite a strong impact on <strong>the</strong> range <strong>of</strong><br />

<strong>the</strong> distribution. Figure 7.8 shows <strong>the</strong> histograms for <strong>the</strong> propensities estimated. Both<br />

propensities for SYETP and <strong>the</strong> comparison group now have a range less than 0.4. The<br />

distributions have been shifted towards zero. The upper tails have generally disappeared,<br />

with <strong>the</strong> remainder <strong>of</strong> <strong>the</strong> distribution shaped similarly to <strong>the</strong> distribution from <strong>the</strong>

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!