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.

245<br />

original model, but with <strong>the</strong> peaks at lower points in <strong>the</strong> distribution. The central peak <strong>of</strong><br />

<strong>the</strong> SYETP propensities has been changed most. The mean <strong>of</strong> <strong>the</strong> propensities for <strong>the</strong><br />

treatment SYETP has been shifted much lower down than for <strong>the</strong> original specification,<br />

while <strong>the</strong> mean <strong>of</strong> <strong>the</strong> comparisons has hardly changed. Figure 7.9 shows <strong>the</strong> kernel<br />

densities <strong>of</strong> <strong>the</strong> SYETP and comparison distribution <strong>of</strong> propensities overlaid. The peak <strong>of</strong><br />

<strong>the</strong> SYETP is raised higher than before. This is likely due to <strong>the</strong> bunching effect <strong>of</strong><br />

reducing <strong>the</strong> range. The peak <strong>of</strong> <strong>the</strong> distribution <strong>of</strong> propensities for <strong>the</strong> comparison is<br />

lower than before, but <strong>the</strong> upper tail is slightly fatter, so <strong>the</strong> reduction in range for <strong>the</strong><br />

comparisons has a more diffuse impact. The change to <strong>the</strong> specification has brought <strong>the</strong><br />

peaks <strong>of</strong> <strong>the</strong> SYETP and comparisons closer in height than before. The different locations<br />

<strong>of</strong> <strong>the</strong> propensity peaks for <strong>the</strong> SYETP and comparisons indicate that <strong>the</strong> participation<br />

model is still producing an adequate model that distinguishes participation.<br />

The effects <strong>of</strong> <strong>the</strong> new specification on <strong>the</strong> matching results are shown in Table 7.5. The<br />

number <strong>of</strong> SYETP matched is raised to 104. No cases have been dropped due to<br />

application <strong>of</strong> <strong>the</strong> common support rule. The number <strong>of</strong> cases <strong>of</strong> comparisons matched to<br />

<strong>the</strong> SYETP is higher than before at 93 cases, while more are also used with replacement.<br />

The mean difference in propensities is no different to that found before. 165 The mean<br />

standardised bias after matching is now worse than before however, as it has risen from<br />

11.04 to 14.48. However, in comparison to <strong>the</strong> o<strong>the</strong>r specification change, where CEP<br />

referrals was excluded from <strong>the</strong> original model, shown in column 2 Table 7.5, <strong>the</strong> bias<br />

has improved. As <strong>the</strong> bias was much worse where CEP referrals was excluded than <strong>the</strong><br />

original model, CEP referrals seems to have helped balance <strong>the</strong> bias.<br />

D’Agostino and Rubin (2000) consider <strong>the</strong> issue <strong>of</strong> missing data in item non-response,<br />

and note that most propensity score methods are based on complete data. They suggest<br />

that missing-ness needs to be controlled in PSM. This suggests <strong>the</strong> new specification may<br />

also benefit from <strong>the</strong> reduction in missing data due to item non-response, which may help<br />

explain <strong>the</strong> great gain in size and statistical significance <strong>of</strong> <strong>the</strong> measured employment<br />

effect. In <strong>the</strong> new specification, only <strong>the</strong> proportion <strong>of</strong> time unemployed contains missing<br />

165 Although rounding hides that <strong>the</strong> mean is very slightly smaller at 0.00016.

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

Saved successfully!

Ooh no, something went wrong!