of the survey. The survey covered the capital cities of all States and Territories and major
urban centres such as Newcastle, Wollongong and Cairns. Immigrants settling in these areas
made up 96 percent of the total immigrant population described above.
The selected PAs, together with any accompanying spouse or partner who migrated with
them on the same visa application, were interviewed three times. Wave 1 interviews were
designed to take place approximately 6 months after the PA entered Australia; Wave 2
interviews took place 12 months after the Wave 1 interviews. The third and final wave took
place a further 24 months later. To assist PAs in providing accurate responses, the time
between arrival and the first interview was minimised.
The response variable in this analysis is the employment status of a PA. Survey respon-
dents were shown eleven categories and were asked to indicate which category corresponded
to their employment status. These were further classified into one of three categories: em-
ployed, unemployed and non-participant in the work force. The multinomial variable created
by this categorization is referred to throughout this paper as “employment status.”
Pettitt et al. (2002) and Cobb-Clark (2000) identify several explanatory variables that are
significantly associated with the response variable, employment status. These include age,
gender, English-speaking ability, region of birth, pre-migration visit and visa category. A
stepwise multinomial logistic regression for each wave of the LSIA suggested that the “best”
subset of explanatory variables to explain the variation in employment status contained the
following five variables: age, age
2
, gender, English-speaking ability and pre-migration visit.
As is the case for most large longitudinal social surveys, the data collected from the LSIA
is incomplete. For the variables selected for analysis, data is missing from both the response,
employment status and the explanatory variable, English-speaking ability. One of the ad-
vantages of the Bayesian hierarchical approach to modelling longitudinal multinomial data,
particularly in WinBUGS, is that missing data from both the response and the explanatory
6
Convention