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.

145<br />

5: Study 3 Attrition and non-response in <strong>the</strong> ALS<br />

In this chapter, data loss to <strong>the</strong> sample is examined for <strong>the</strong> <strong>Australian</strong> Longitudinal<br />

Survey data. The <strong>the</strong>ory relating to survey attrition is broadly summarized. Literature on<br />

<strong>the</strong> empirical aspects <strong>of</strong> attrition and effects on estimation are <strong>the</strong>n reviewed. Next, a<br />

practical means <strong>of</strong> investigating and treating attrition in <strong>the</strong> data is presented. Following<br />

this, attrition in <strong>the</strong> ALS is investigated in depth and <strong>the</strong> extent <strong>of</strong> sample reduction is<br />

considered. A univariate analysis <strong>of</strong> <strong>the</strong> scale <strong>of</strong> attrition effects is conducted. The<br />

different sources <strong>of</strong> sample loss are recognized separately. Initially <strong>the</strong> final sample and<br />

those lost from <strong>the</strong> sample are compared. Then, sample loss is examined within <strong>the</strong><br />

SYETP treatment and comparison groups. The non-response to <strong>the</strong> first survey and<br />

sampling design effects are examined to check <strong>the</strong>ir effects on analysis. These are<br />

accounted for with a weight that was supplied with <strong>the</strong> data. Multivariate analysis is <strong>the</strong>n<br />

conducted <strong>of</strong> <strong>the</strong> effects survey attrition on <strong>the</strong> participation model. Attrition is concluded<br />

to be a problem for <strong>the</strong> modelling. Weights are <strong>the</strong>n constructed to repair <strong>the</strong> data. These<br />

attrition weights are <strong>the</strong>n united with <strong>the</strong> sample design/non-response weights. These<br />

combined weights are used in subsequent chapters.<br />

5.1 Examining sample reduction in <strong>the</strong> ALS<br />

A particular concern for <strong>the</strong> analysis <strong>of</strong> longitudinal data is sample attrition. Attrition and<br />

survey non-response are a form <strong>of</strong> non-response usually termed ‘unit non-response’ by<br />

statisticians. The key worry is that attrition may lead to selective samples, which in turn<br />

may make interpretation <strong>of</strong> resulting estimates problematic. One potential problem for <strong>the</strong><br />

analysis in <strong>the</strong> earlier studies is <strong>the</strong> possibility <strong>of</strong> attrition bias due to sample reduction<br />

between <strong>the</strong> first sample survey in 1984 and <strong>the</strong> 1986 survey. An important reason for<br />

properly accounting for attrition is that ignoring attrition can lead to biased parameter<br />

estimates if <strong>the</strong> attrition is related to <strong>the</strong> behaviour being modelled.<br />

The issue <strong>of</strong> attrition was briefly examined in Richardson (1998) by comparing <strong>the</strong><br />

summary statistics <strong>of</strong> <strong>the</strong> final analysis sample (referred to as <strong>the</strong> ‘whole sample’), those

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

Saved successfully!

Ooh no, something went wrong!