High quality measures of socio-economic status (SES) are crucial for equitable school funding programs. As low socio-economic background is a recognised source of educational disadvantage, schools enrolling low SES students need to be properly funded to achieve equitable outcomes for their students. In the Australian school education sector, two types of measures are currently used for low SES equity funding for schools: direct measures (using individual student information relating to parents, e.g. parent occupation and education) and area-based measures (using the average characteristics of persons/households in the area in which students live or where a school is located). This study first examines the merits and deficiencies of SES measures using these different sources of data. Using parent background information and family addresses collected from enrolment forms for NSW government students (approximately 700,000), the study finds that school SES measures based on direct parent data are a stronger predictor of average school academic performance than indirect measures derived from student addresses. Further analysis of school drawing areas shows that the extent of bias arising from the source of invalidity (‘ecological fallacy’) for area-based measures differs across areas and across types of schools. At the student level, the misalignment is significant when different sources of data are used to classify students’ SES, with a quarter of all students changing quintile group classification by two or more groups if the alternate source is used. This could affect the number and proportion of students classified as low SES across schools, potentially affecting the school funding distribution.

However, a common issue facing educational systems when using direct parent data to construct SES measures is that parent information is not complete for all students. Furthermore, data is not missing at random. This study illustrates the use of a multiple imputation method (MICE) to reduce bias in the resultant SES measures due to missing parent data. Imputation models were developed using a large number of auxiliary variables such as SEIFA community variables based on student addresses, student test scores, student Aboriginal status and school remoteness, in addition to parent background variables. Following the imputation process, two validation exercises using ABS census data and parent survey data, as well as a simulation analysis, were carried out. These analyses show that multiple imputation has reduced the bias due to missing data in the resultant SES measures, especially for small schools and schools with higher rates of missing data.