Analysis of linked longitudinal data for socially defined population: Methods and examples from Australia
Jul 25, 11:30
As survey data becomes more expensive and difficult to collect, research funders are increasingly looking to augment or replace dedicated longitudinal surveys with linked cross-sectional data from existing surveys or administrative datasets. In Australia, this includes the Australian Census Longitudinal Dataset (ACLD – http://www.abs.gov.au/ausstats/abs@.nsf/mf/2080.0), and linked deaths registrations to Census data (http://www.abs.gov.au/ausstats/abs@.nsf/mf/1351.0.55.058 ).
These datasets provide a rich source of longitudinal data, especially for populations or regions that are traditionally hard to collect using survey data. In particular, they allow us to look at how membership of socially defined populations predicts change in outcomes, and how baseline outcomes predict change in that membership. For example, analysis of the ACLD has looked at changes in identification as Aboriginal or Torres Strait Islander (http://caepr.anu.edu.au/Publications/census-papers/2015CP18.php). One complicating factors with these and similar datasets, however, is that many of the characteristics that we are interested in evaluating are also characteristics that are used to support the individual-level data linkage.
In this paper, we plan to explore the methodological implications of using linked longitudinal data to understand the outcomes of socially defined populations, propose a methodological approach that takes into account the linkage process in that analysis, and discuss the implications for policy conclusions based on the data.