Socioeconomic disadvantage, ability to pay and university attendance in Australia

ABSTRACT

Evidence from around the world shows low university participation among young people from low socioeconomic status backgrounds. A common concern is that disadvantaged young people may be unable to afford higher education costs. Using data on government benefits intended to support students from low income households in high school, we identify students at risk of being unable to pay higher education costs. Large differences in university participation rates are observed which are no longer evident after controlling for high school achievement. Results suggest improving high school achievement is an important channel through which disadvantaged student participation may be improved.

KEYWORDS:

• University attendance
• socioeconomic disadvantage
• unobservables

JEL CLASSIFICATION:

• I21
• I22
• I24

Acknowledgments

We are grateful for helpful discussions with and suggestions from Johannes Kunz, Cain Polidano, David Prentice and Joe Vecci and thank participants at the IZA Economics of Education, Higher Education Conference in Bonn, October 2016, the XXVI AEDE (2017) meetings in Murcia, Spain, the PET 2017 meetings in Paris, France and seminars at RMIT Universities and the University of Kent. Cardak gratefully acknowledges the hospitality extended by the Melbourne Institute where part of this work was undertaken.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 Similar intention to undertake post-secondary education survey questions, together with subjective completion beliefs, are used by Kunz and Staub (2020) to estimate a structural model of investment in post-secondary education in Germany.

2 The evidence on university completion and dropout is mixed, with Stinebrickner and Stinebrickner (2008) finding dropout decisions are not affected by ability to pay at Berea College in the US, but Arendt (2013) and Cardak and Vecci (2016) finding that students who might struggle with higher education costs exhibit a greater risk of dropout. For the UK, Johnes and McNabb (2004) show academic achievement and the match with peers at university are important factors influencing completion.

3 Related work by Delaney, Harmon, and Redmond (2011) shows the impact of SES differences on grades declines over time at university but important SES differences in earnings expectations persist, despite university completion.

4 The ENTER score was specific to the state of Victoria. However, each state provided graduating students with an equivalent ranking, for example a Universities Admissions Index (UAI) in New South Wales. These rankings could be used to apply for university places both within and out of the student's home state. We use the term ENTER score as a generic name for these entrance ranks which are calibrated to a common, Australia-wide scale that ranges from 30 to 99.95. Students with a percentile rank of less than 30 are bottom-coded to a rank of 30. Since 2010, these percentile ranks have been renamed Australian Tertiary Admission Rank (ATAR) for all states and territories except Queensland which retained the Overall Position (OP) ranking mechanism until 2020.

5 Admission to university on the basis of ENTER score is the dominant mode for students completing high school in Australia. Other criteria are used for mature aged entrants which form a smaller part of the student body and are not the focus of our analysis, while direct or contextual admissions have also grown in importance.

6 In the year 2000, when many of the students in our sample were attending university, two of the 43 higher eduction institutions in Australia were private, Bond University and University of Notre Dame Australia. While relatively small, the private provision of higher education has since evolved and grown relatively quickly; see Cardak, Brett, and Burt (2022) for details of recent growth in private higher education in Australia.

7 The up front discount applies to the whole university system. The university receives the full tuition with the Commonwealth Government making up the difference to the university. This discount was set at 25% in 1993, reduced to 20% in 2005 and further reduced to 10% in 2012. This differs from the income contingent loan system in England where individual universities may offer different discounts on tuition paid up front.

8 Since 2005, the HECS scheme has been renamed the HECS-Higher Education Loan Program (HECS-HELP).

9 Student performance in ninth grade literacy and numeracy tests were used by Rothman (2002) to construct achievement scales. The individual literacy and numeracy scales were constructed to have a mean of 50 and standard deviation of 10. We use the average of these two scales to reflect individual student achievement. This average has a standard deviation of 8.5. If only one of the literacy and numeracy scales was available, it was used as the achievement score. This affected about 1.9% of observations used in the analysis.

10 The constructed variable is the difference between the student's response (scored from 0 for well below to 4 for well above average) and the conditional mean of similar students' responses, in terms of having similar measured achievement, were of the same gender and who attended schools in the same average student achievement quartile. This difference was then placed on a scale with a mean of 50 and standard deviation of 10. By construction, the scale is independent of own achievement.

11 In the data used here, the variable has a mean of 41 and a standard deviation of 22.

12 The conditional means shown in figures in this paper were estimated using the lowess or mlowess commands in STATA.

13 Ideally we would use data on parental income and wealth to identify those students who might struggle to afford higher education costs. However the LSAY data do not contain such information.

14 They are also likely to be additional earners in their households and therefore face greater opportunity costs of study and are anticipated to have a lower willingness to incur debt.

15 The probability of receipt of the payment fell with parental SES, having a parent with a university degree, living in a suburb where fewer people received government payments, where both parents were present in the household, living in a metropolitan area and where students had fewer siblings.

16 Recall that ENTER scores are bottom-coded to a value of 30 for students ranked below the 30th percentile.

17 Tests of the inclusion of interactions between higher order ENTER score terms and risk group indicators did not reject the null hypothesis that these effects were zero.

18 Alternative specifications included other background characteristics such as number of siblings, living in a single parent household, parental SES as a variable in its own right and various postcode-based SES and income measures. In all cases, the results were similar to those presented here. The group indicators were always insignificant after the inclusion of ENTER score while its marginal effect was unchanged at around 0.012.

19 Using data from the Youth in Focus project, Cobb-Clark and Gorgens (2014) studied parental financial support provided to 18 and 20 year-old young people in Australia. The support studied involved monetary payments from parents to children and co-residence. They found around 80 per cent of 18 year-old and two-thirds of 20 year-old Australians lived with their parents while there was no evidence that any lack of parental support contributed to the socioeconomic gradient in participation in post-secondary study also apparent in their data.

20 Repeating the analysis for the more recent but smaller LSAY 06 cohort leaves our results and conclusions unchanged. Analysis of the LSAY 09 cohort shows end of school achievement continues to dominate ninth grade achievement in determining university attendance (Productivity Commission 2019, Table B16).

21 The model is estimated using the heckprob command in STATA.

22 Standard errors are estimated using 50 bootstrapped replications to deal with the inclusion of a generated regressor in the university attendance equation.

23 The approach does not fit the Heckman selection framework because the attendance equation is estimated over those who planned to attend university as well as those who did not. Some 17% of those observed to attend university indicated in year 9 they did not intend to do so, implying the plans equation cannot be treated as a selection equation.

24 Notwithstanding concerns mentioned above, inclusion of the actual plans rather than the residual did not change our results qualitatively. The group indicators were significantly different from zero before ENTER score was included but not after, and the marginal effect of the ENTER score is similar to that reported in the preferred specification in column (5) of Table 3.

25 We varied this approach to capturing unobservables by utlilising answers to questions about whether students intended to go to university after leaving school, given in each school year from year 9 to year 12. We distinguished students who always answered they did (value 2) from students who always answered they did not (value 0) and those who sometimes answered ‘yes’ and sometimes ‘no’ (value 1). Each group comprised about one-third of the population. Using the residuals from a regression of the resulting variable on the same set of explanatory variables as Equation (3(3) $\widehat{{ϵ}_i}=P_i-\mathrm{\Phi }(\widehat{{\psi }_0}-A_i\widehat{{\psi }_1}-X_i^{\mathrm{\prime }}\hat{\psi }),$(3) ) to represent unobservables, the model from column (3) of Table A.1 of Online Appendix A was re-estimated. We found our key results to be qualitatively very similar to those already presented.

26 This analysis was extended by the inclusion of interactions between school type and ENTER score. These interactions were statistically insignificant and conclusions regarding the risk group indicators were qualitatively unchanged.

27 We also experimented with group definitions in a similar manner to that described here but using other alternatives for the SES measure, predominantly involving postcode-based measures of where students lived and where schools drew their students from. The results were similar in that the group indicator measures were never significant once ENTER scores were included.

28 As for results reported in Section 4.2, the results for the two group indicator variations that included only private school students in the no risk group were not driven by that aspect of the definition. When relaxed to include public school students too, the results were qualitatively unchanged with the group indicators statistically insignificant when ENTER score was also part the estimated model, consistent with findings presented in Section 4.2.

29 Concerned about growing costs, the government wound back the ‘demand-driven’ system from 2018, limiting places to 2017 levels plus population growth.