10th Biennial ACSPRI Social Science Methodology Conference in conjunction with the Australian Social Network Analysis Conference (ASNAC)

The Australian Survey of Social Attitudes (AuSSA), Australia's component of the International Social Survey Programme (ISSP), is an important annual survey of Australian citizens enrolled on the Australian electoral roll. Covering a wide range of social, political, and economic topics, the AuSSA provides valuable data for researchers studying Australian public opinion and social issues. However, like many surveys that rely on voluntary participation, the AuSSA exhibits a persistent left-of-centre bias—an overrepresentation of respondents who vote for or identify with parties on the political left. This bias persists even after applying standard post-stratification weights based on official population benchmarks from the Australian Bureau of Statistics.

While multivariate models can often account for such bias by including relevant covariates, this option is not available for simple descriptive estimates such as means or proportions, which are typically adjusted only using standard demographic weights. Given that descriptive estimates are important outputs from survey data, this limitation has practical consequences for the accuracy and credibility of their estimates.

This paper compares three approaches to adjusting AuSSA’s survey weights for partisan bias. Each approach retains the standard post-stratification by sex, age, and education, and adds a raking step to one political benchmark: the 2023 Voice referendum, first-preference votes at the last federal election, or current voting preferences based on the average of national opinion polls. The paper discusses the advantages and disadvantages of each approach and evaluates the extent to which the adjusted weights improve the representativeness of key survey estimates.