Evaluation of the Australian Wage Subsidy Special Youth ...

Evaluation of the Australian Wage Subsidy Special Youth ... Evaluation of the Australian Wage Subsidy Special Youth ...

02.06.2014 Views

112 4.1 Differences between the treatment and comparison group An important problem for evaluation of programme effects is whether the treated are compared to an adequate reference group. This is the issue of selection distortion, where those treated have a very different profile to those in the comparison group. Selective recruitment onto subsidy would make the treated have a different profile than the larger group of those eligible. Controls for selection distortion can be made on observable characteristics, such as those variables shown in Table 4.1. The Table 4.1 gives the mean and standard deviation from the mean for a set of characteristics, with column 1 and 2 showing these for the SYETP treated group, and columns 3 and 4 showing the comparison group characteristics. The absolute difference between the means is in column 5, with the results of the t-test at one percent level of significance indicated by an asterisk to show statistically significant differences. Richardson (1998) p6 commented that the contrast between those who took part in SYETP and those in the comparison group was “striking”. Indeed, column 1 and column 3 of Table 4.1, of which the means were also presented by Richardson (1998), shows that the profile of the treatment and comparison groups differs strongly. In most cases, means were quite different for the SYETP relative to the comparisons, and in some cases the variation from the mean, as represented by the standard deviation was also very different for each group. As Richardson (1998) highlights, SYETP participants had a different educational attainment profile where post-school qualifications were less common, they were about a year younger, and had poorer labour market experiences than those in the comparison group. However, how different the SYETP group was from the comparisons is related more clearly here by the addition of the test of the statistical significance of the mean difference. Almost all variables have a significant divergence between the means for the SYETP group and the comparisons. Consideration of the many statistically significant differences in Table 4.1 makes the selection distortion for SYETP in the ALS data apparent. The literature review also highlights other evidence that it was generally the case for SYETP entrants to be younger than the eligible group as a whole – mostly teenagers. Thus it is possible that the ALS sample reflects differences that existed between the SYETP and comparison populations.

113 Propensity score matching provides a method of analysis that controls for the lack of correspondence between the treatment and comparison group. As discussed in Chapter 1, evaluation methods seek to maximize the similarity of the comparison and treated groups, in order to return to a quasi-experimental situation where effects can be usefully attributed to the programme. The propensity score matching method is now further discussed and applied. The outcomes of the propensity score matching are then compared to that of the bivariate probit. The relevance of each approach to the case at hand is considered, and the result for the SYETP programme effect is discussed.

113<br />

Propensity score matching provides a method <strong>of</strong> analysis that controls for <strong>the</strong> lack <strong>of</strong><br />

correspondence between <strong>the</strong> treatment and comparison group. As discussed in Chapter 1,<br />

evaluation methods seek to maximize <strong>the</strong> similarity <strong>of</strong> <strong>the</strong> comparison and treated groups,<br />

in order to return to a quasi-experimental situation where effects can be usefully<br />

attributed to <strong>the</strong> programme. The propensity score matching method is now fur<strong>the</strong>r<br />

discussed and applied. The outcomes <strong>of</strong> <strong>the</strong> propensity score matching are <strong>the</strong>n compared<br />

to that <strong>of</strong> <strong>the</strong> bivariate probit. The relevance <strong>of</strong> each approach to <strong>the</strong> case at hand is<br />

considered, and <strong>the</strong> result for <strong>the</strong> SYETP programme effect is discussed.

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

Saved successfully!

Ooh no, something went wrong!