Last week the Department of Education issued a report on equity students enrolled in research higher degree programs. As those who have read my work over the years know, I think we have significant conceptual and empirical problems in measuring socioeconomic status in higher education. And these are even more significant for higher degree students than they are for undergraduates.
What this means is that even though the report’s overall conclusion, that high SES students are ‘over-represented’ in research degrees, must be true based on other empirical evidence and theory, its statement that ‘this data should … be used with caution’ is a warning that should be heeded.
Problem One: We are only using a geographic proxy indicator for SES, the ABS Index of Education and Occupation. A person is classified as low SES if they live in an area in which the population has relatively low levels of education and relatively high levels of people who are unemployed or work in lower-skill occupations. But people with high levels of education and with high skill jobs live in otherwise low SES areas, and vice-versa.
Problem Two: We define as low SES people living in the lowest 25 per cent of areas by the Index of Education and Occupation. That is too small a share – the next quartile up is sociologically similar.
Problem Three: For research students, are we interested in their current socioeconomic status or their background? Regardless of their background, if they already have a degree (which they almost certainly do if they are in a research degree) and work in a professional job, as is quite likely to to be the case, then they are not going to be classed as low SES by the standard bureaucratic measures. And if they have moved to study and/or to be closer to professional job markets, then they will probably live in high SES areas.
For equity policy purposes, I think we are most interested in the social background of students, and so geographic measures that reflect their current location are not reliable indicators of where they came from. We know from the undergraduate equity data that even pushing address data back a few months, using where they lived when they first applied to university rather than their enrolment permanent home address, increases the number of students identified as low SES. For graduate research students it is at least four years, and might be decades, since they first applied for university while living with their parents.
Problem Four: What is the correct denominator? At the moment the denominator is all research students, producing an enrolment share figure (low SES research students/all research students). Enrolment shares are administratively convenient, as the numerator and denominator are in the same dataset. And if we are interested in socioeconomic diversity on campus, then a campus-level version (if we could improve the numerator quality) would be useful.
But it we are interested in social mobility, or whether potential is being achieved, then the entire student population is not a good denominator. The example below shows how enrolment shares can leave us blind to significant social change. Provided students from all SES backgrounds increase enrolments at the same rate then enrolment shares will not change, even when the participation rate (low SES students/low SES persons) has doubled.
This is not a crazy hypothetical example. Undergraduate participation rates did double between 1990 and 2017, while the low SES enrolment share increased only slightly.
The right denominator for research students is those who have finished honours or some other qualifying criteria for a research degree.
Because we do have some quality numerator and denominator data (using HILDA), we know that low SES young people were significantly less likely to go to university than high SES young people, and therefore significantly less likely to get a degree, and therefore the potential pool of low SES background research students is much smaller than the potential pool of high SES students. This is why I am sure that low SES students must be a low share of a research degree student population, despite the weaknesses of the administrative data also pointing to this conclusion.
There is no easy solution to these problems, but I think there are possibilities within the existing higher education dataset that could help us get a better understanding of whether people from low SES backgrounds are becoming more or less likely to go on to a research degree.
We could improve the numerator by moving the SES measure to parental education. Education is not a perfect proxy for advantage or disadvantage, but this is a personal version of a major driver of the ABS geographic measures (there is more missing data for parental education than student address, but for measuring general trends rather than allocating personal benefits I don’t think that matters).
As we have a common student identification number, the CHESSN, going back to 2005, and the parental education question since 2010, we can also construct relevant denominator populations for domestic students who were educated in Australia.
An example of this would be: of all the people who completed bachelor honours in year X whose parents both had less than a Year 12 education, what proportion had enrolled in a research qualification by some future year? And then compare that with other parental education levels – Year 12, vocational education, bachelor degree and postgraduate.
It is more complex to do than just looking at each year’s enrolment data, but similar to the individual tracking already done for the annual cohort completions analysis.
The current method is just telling us what we already know – that fewer lower SES background people ever got a degree, and that students tend to live near universities, which are usually in high SES areas.
We need something better to understand real low-SES trends.