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

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

154<br />

interview date, those who ever entered full-time education, or for whom responses were<br />

missing in 1984, 1985 or 1986 (Richardson (1998): 5). The first <strong>of</strong> <strong>the</strong>se restrictions is<br />

not dealt with here, as it prescribes <strong>the</strong> eligible set, but is returned to subsequently. The<br />

latter <strong>of</strong> <strong>the</strong>se has been dealt with in part, with reference to <strong>the</strong> natural attrition, which<br />

caused missing values. Apart from natural attrition, <strong>the</strong> set discarded for whom responses<br />

were missing in 1984, 1985 or 1986 also includes those cases for whom data on some <strong>of</strong><br />

<strong>the</strong> regressor variables is missing. In particular, when information about <strong>the</strong> family<br />

background is missing, or when <strong>the</strong> proportion <strong>of</strong> time in unemployment is missing, <strong>the</strong>se<br />

cases are dropped from <strong>the</strong> analysis set. Statisticians usually term this form <strong>of</strong> missing<br />

data ‘item non-response’.<br />

Greene (1991) points out <strong>the</strong>re are two main scenarios to consider, depending on why <strong>the</strong><br />

data on <strong>the</strong> regressors is missing. Imputation or discarding <strong>the</strong> cases is <strong>the</strong> usual approach<br />

to this type <strong>of</strong> missing data. Using <strong>the</strong> terminology <strong>of</strong> Griliches (1986) this type <strong>of</strong><br />

incomplete data is described as <strong>the</strong> ignorable case if when using <strong>the</strong> complete<br />

observations <strong>the</strong> data are not missing for self-selection reasons 98 . In <strong>the</strong> ignorable case,<br />

using only <strong>the</strong> complete observations is not problematic if efficiency related to <strong>the</strong><br />

variance is not important. The non-ignorable case, where <strong>the</strong> missing information is<br />

systematically related to <strong>the</strong> variable being modelled, is ano<strong>the</strong>r case <strong>of</strong> <strong>the</strong> sample<br />

selection problem.<br />

98 Ignorable non-response means <strong>the</strong> same as selection on observables.

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

Saved successfully!

Ooh no, something went wrong!