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.

140<br />

As would be expected where <strong>the</strong> same caliper widths were used, <strong>the</strong> same numbers <strong>of</strong><br />

SYETP cases were matched. But <strong>the</strong> number <strong>of</strong> comparisons that could be matched from<br />

within <strong>the</strong> radius is much greater for every caliper, growing from 462 matched to <strong>the</strong> 87<br />

SYETP in <strong>the</strong> 0.001 caliper, to 920 comparisons matched to <strong>the</strong> 104 SYETP in <strong>the</strong> 0.05<br />

caliper. The good common support for <strong>the</strong> treated and <strong>the</strong> great size <strong>of</strong> <strong>the</strong> comparison<br />

group contribute to <strong>the</strong> large number <strong>of</strong> comparison that could match within each caliper.<br />

The standardised bias is smaller than found for <strong>the</strong> same caliper when pair matching was<br />

used, but not dramatically. Once again, all <strong>the</strong> mean differences in employment are<br />

statistically significant. The difference in employment varies in size more across <strong>the</strong><br />

different caliper widths than for pair matching. The difference in employment found for<br />

<strong>the</strong> 0.001 caliper varies from 0.18 for pair matching to 0.14 for <strong>the</strong> all-in-radius matching.<br />

However, for <strong>the</strong> 0.005 caliper <strong>the</strong> difference is very small, with 0.18 for pair matching to<br />

0.17 for <strong>the</strong> all-in-radius matching. The mean difference in employment for <strong>the</strong> 0.05<br />

caliper (0.15) is more similar to that for <strong>the</strong> 0.001 caliper (0.14). The difference in<br />

employment does not <strong>the</strong>n grow generally with <strong>the</strong> greater caliper. This suggests that<br />

similar propensity scores do not smoothly relate to similar employment outcomes.<br />

Comparing <strong>the</strong> estimate employment effects <strong>of</strong> SYETP in Table 4.8 to those <strong>of</strong> Table 4.7<br />

also gives variable results. At <strong>the</strong> caliper width <strong>of</strong> 0.001, ‘all in radius’ PSM gives a net<br />

employment effect <strong>of</strong> 0.14 percentage points while nearest neighbour PSM gives 0.18.<br />

However for calipers 0.005 to 0.02, <strong>the</strong> estimates are very similar in size, but diverging<br />

again for caliper width 0.05. Generally <strong>the</strong> ‘all in radius’ impact estimates are smaller.<br />

Overall, it is difficult to judge which PSM protocol is preferable. However, all-in-radius<br />

matching is not found used in <strong>the</strong> literature very <strong>of</strong>ten, and most literature apply pair<br />

matching, although some newer literature also applies kernel matching. From <strong>the</strong>se<br />

results, where <strong>the</strong> differences are usually slight, <strong>the</strong> gains to be had from all-in-radius<br />

matching appear to be small relative to pair matching for this data. As such, pairmatching<br />

is concluded to be <strong>the</strong> most useful algorithm and also conforms to <strong>the</strong> great<br />

majority <strong>of</strong> <strong>the</strong> literature. This is <strong>the</strong>n applied in <strong>the</strong> later analyses in following chapters.

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

Saved successfully!

Ooh no, something went wrong!