Increasing the price of a university degree does not significantly affect enrolment if income contingent loans are available: evidence from HECS in Australia
IIncreasing the price of a university degree does not significantly affect enrolment ifincome contingent loans are available: evidence from HECS in Australia
Fabio I. Martinenghi a,b a School of Economics, University of New South Wales, Australia. b Australian Research Council Centre of Excellence for Children and Families over the Life Course, Australia.
Abstract
I provide evidence that, when income-contingent loans are available, student enrolment in university courses is notsignificantly affected by large increases in the price of those courses. I use publicly available domestic enrolment datafrom Australia. I study whether large increases in the price of higher education for selected disciplines in Australia in2009 and in 2012 was associated with changes in their enrolment growth. I find that large increases in the price of acourse did not lead to significant changes in their enrolment growth for that course.
JEL classification:
I21, I23, I28.
Keywords:
Educational finance, Income contingent loans, Demand for higher education
1. Introduction
In Australia, the federal government provides financialaid for its undergraduate domestic students (hereafter re-ferred to as “students”) in several ways. A complete expo-sition of how this system works can be found in Chapman(2006) and a summary in Chapman and Leigh (2009). Iwill only outline the essential features of this system. Un-der the
Higher Education Support Act 2003 (The Parlia-ment of Australia, 2003), the government pays their de-grees upfront and in full. Part of this government paymentis a subsidy, while the remainder is a loan. This latterpart is called “student contribution” and it is the totalamount that students will pay for their university degree.The student contribution is repaid via the Higher Edu-cation Contribution Scheme (HECS). Under this scheme,students will gradually repay the government only afterthey earn a sufficiently high income. In short, the gov-ernment gives prospective students an income-contingentloan . Moreover, HECS is progressive so that graduateswith higher incomes will pay the loan back at higher rates.The federal government sets the maximum student con-tribution that Australian universities can charge for eachdiscipline, or, equivalently, the highest amount that stu-dents will have to pay for studying a certain discipline. Email address: [email protected] (Fabio I.Martinenghi)
URL: https://fabitmart.github.io/ (Fabio I. Martinenghi) On income-contingent loans, see for instance (Stiglitz et al.,2014) and (Britton et al., 2019).
Each discipline is assigned to one of three bands, whichare associated with different maximum student contribu-tions. The
Higher Education Support Act 2003 (The Par-liament of Australia, 2003) introduced a fourth band, the“National Priorities”. This band is assigned to those dis-ciplines that the Commonwealth government is seeking topromote, including by reducing their maximum studentcontribution. Between 2005-2009,
Nursing and
Education were declared national priorities. Between 2009-2012, allthe Natural and Physical Sciences (which include Mathe-matical Sciences) were declared national priorities.After 2009, when Nursing and Education were no longerconsidered a national priority, their maximum student con-tributions increased by approximately 25%, from $ $ $ $ a r X i v : . [ ec on . GN ] F e b he price of a degree to encourage or discourage enrolmentin it would not work.In this note, I test whether the event of removing a dis-cipline from the National Priority band – hence increasingits maximum student contribution – is associated which achange in the student enrolment for that discipline. Thechange in student enrolment is measured as a change inthe slope of the student enrolment trend after the event.I find that moving a discipline from the National Priorityband had no significant impact on student enrolment.
2. Methods
I collect and combine publicly available data on com-mencing domestic undergraduate students and on the An-nual Course Contribution Value (ACCVAL) tables be-tween 2004-2018. The ACCVAL tables provide infor-mation on maximum student contribution by discipline.Years 2004 and 2005 have been dropped from the analy-sis due to inconsistencies in how cross-institutional under-graduate students were categorised. This categorisationdiffers from the rest of the sample and creates uncertaintyaround the yearly number of commencing undergraduatestudents. The dataset and the R code to reproduce it areavailable on my website (link). While commencing students data is available for all dis-ciplines, the sample used for estimation does not includedisciplines that were never in the National Priority band.Given this data availability, an obvious candidate method-ology is difference-in-differences (DID). However, the par-allel trends assumption does not seem to hold when com-paring the time series of the treatment and candidate con-trol groups (not shown). Different disciplines often displaymarkedly different trends which—in absence of richer andmore disaggregated data—will lead to biased estimates.Moreover, the “no spillover effects” assumption does notseem to hold either. This is because, by design, the disci-plines chosen for the control group would have to be similarto the ones in the treatment group. This very similarityimplies that a student might respond to an increase inthe price for a degree in the treatment group by choosinganother similar degree in the control group. This wouldviolate the “no spillover effects” assumption.Instead of a difference-indifferences approach, I chooseone close to an event study. I define the event as “discipline i is removed from the National Priority band (or, equiva-lently, it experiences an increase in student contribution)”. This includes an automated download of all the data and theirmanipulation.
There are two instances of this event: one in 2009 and asecond in 2012, as detailed above. Furthermore, I restrictthe sample to (i) the ever-treated disciplines, i.e. the dis-ciplines that were in the National Priority band at leastonce; (ii) a time window spanning between 3 years beforeeach event and 5 years after. The size of the window is de-termined by how many periods are available around bothinstances of the event. I use this sample to test whether,under an income-contingent loan regime as HECS, drasti-cally increasing the maximum student contribution for adiscipline—i.e. the cost to study it—is followed by signif-icant changes in enrolment trends in that discipline.
To test this hypothesis, I estimate the following fixedeffect equation for discipline i and time t : ln ( Y i,t ) = α i + β T t + β D i,t + δD i,t ∗ T t + (cid:15) i,t (1)where ln ( Y i,t ) is the natural logarithm of the number ofcommencing students by discipline and year, T t is a yearlyevent time variable , D i,t takes a value of one after thediscipline is no longer in the National Priority band andzero otherwise and D i,t ∗ T t is their interaction term. Thecoefficient of interest is δ , which can be interpreted as theyearly percentage change in enrolments following a sharpincrease in student contribution.
3. Results
Figure 1 shows the number of commencing students en-rolled by disciplines and year. The disciplines shown arethose which have been in the National Priority band atleast once. Panel (a) shows the number of yearly com-mencing students who benefited from the National Prior-ity band between 2005 and 2009 - where the year 2009is marked by a red vertical line. The National Priorityband between these years included disciplines related toNursing and Education. Panel (b) shows the number ofyearly commencing students who benefited from the Na-tional Priority band between 2009 and 2012 – 2012 alsobeing marked in red. The National Priority band dur-ing these years included all Physical and Natural Sciences,plus Mathematics.An inspection of Figure 1 suggests that commencing stu-dents were rather insensitive to the large changes in stu-dent contribution marked by the end of the National Pri-ority programmes. In other words, the demand for those An event time variable is a variable equal to the number of peri-ods before/after the event. It takes negative values before the event,positive values after the event and it equals to zero in the year of theevent. able 1. Changes in enrolments after an increase in student contri-bution (1) (2) (3) (4)T 0.0523 0.0632 ∗ ∗∗ ∗∗ [0.03,0.10] [0.02,0.09](0.02) (0.02)Event × T -0.0616 -0.0616 -0.0363 -0.0363[-0.15,0.03] [-0.15,0.03] [-0.12,0.04] [-0.12,0.04](0.04) (0.04) (0.03) (0.03)Constant 8.206 ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ [8.10,8.31] [8.12,8.34] [8.37,8.52] [8.39,8.55](0.05) (0.05) (0.03) (0.04)Obs. 90 90 81 81
95% confidence intervals in brackets, robust s.e. in parentheses ∗ p < . ∗∗ p < . ∗∗∗ p < . Note.
Column 1 estimates of Equation 1. A linear model with disciplinefixed effects is fit to data on the logarithm of yearly commencing studentsby discipline. The specification controls for a linear event time trend, whichis interacted with the treatment dummy variable. A discipline is defined astreated after it is no longer in the National Priority band. Column 2 restrictsEquation 1 to the case of no change in the intercept after the event (i.e. β = 0).Columns 3 and 4 estimate the same equations as column 1 and 2 respectivelybut excluding “Other education”. This is done in order to show how “Othereducation” alone doubles the size of the (negative) coefficient. disciplines seems inelastic to changes in price. The onlydiscipline not displaying an uninterrupted linear trend is Teacher Education . The number of Teacher Educationcommencing students peaks in 2012, after which a decreas-ing trend starts. This change in trend occurs several yearsafter the change in student contribution, hence the two areunlikely to be related.
Column 1 in Table 1 shows the estimates for Equation 1.As anticipated by the graphical analysis, on average thereis no significant change in the trend of enrolments after asharp increase in student contribution ( δ = − .
06 with ap-value of 0.16). In Column 2, Equation 1 is estimated inthe restricted case where β = 0. In other words, in Col-umn 2 the intercept is not allowed to vary after the event.Column 3 and 4 estimate the same equations as columns1 and 2, respectively, but dropping the field “Other Edu-cation” from the sample. While the coefficient of interestis never significantly different from zero, Columns 3 and4 show that “Other Education” is responsible for half ofthe size of the coefficient of interest, δ . When “Other Ed-ucation” is dropped, δ reduces to less than 0 .
04 (in bothColumns 3 and 4) and the p-values double to 0 .
32 (alsoin both columns). The leverage of “Other Education” canalso be noticed upon inspecting Figure 2, which visualisesthe sample after trimming and centering the outcome vari-able, ln ( Y i,t ).As an additional test of δ = 0 and following (Steiger, Table 2. Testing δ = 0 using 90% confidence intervals (1) (2) (3) (4)Event × T -0.0616 -0.0616 -0.0363 -0.0363[-0.14,0.01] [-0.13,0.01] [-0.10,0.03] [-0.10,0.03]Obs. 90 90 81 81
90% confidence intervals in brackets ∗ p < . ∗∗ p < . ∗∗∗ p < . Note.
I construct and inspect 90% confidence intervals for δ , using robuststandard errors. This is for testing whether δ = 0. The model specificationsare identical to the ones associated with Table 1 and are identically ordered. Iam again unable to reject the null hypothesis of δ = 0. δ (Table 2). The model specifications are identicalto the ones associated with Table 1. In each specifica-tion, zero is included within the 90% confidence interval.In Columns (3) and (4), whose specifications exclude theoutlier discipline, the confidence interval is even more cen-tered around zero.
4. Discussion
These results are consistent with the hypothesis thatincome-contingent loans, such as the one featuring in theHECS system, make the demand for a discipline inelasticto price increases. Even when the price for studying adiscipline (the student contribution) increases sharply, thetrend in new enrolments does not significantly decrease onaverage. Therefore, HECS seems to be meeting its goalof allowing students to make decisions about their highereducation independently its price.At the same time, these results suggest that if studentshave access to income-contingent loans, they will not re-spond to policies that increase the price of a discipline todisincentivise them from choosing it. In light of the ev-idence provided, the economic argument supporting thiskind of policies seems weak, leaving a justification basedon moral grounds as the only avenue for a government tosupport them . Declaration of Competing Interest
I have no conflicting interest to declare
Acknowledgements
I thank Andrew Norton for his help in navigating theAustralian education policy world and for sharing his data,which allowed me to start working on this project quickly. (Chapman, 1997) calls this issue “the issue of whether or notit is appropriate for the government to charge individuals based onwhat the expected direct benefits on average from a course are”. c o mm e n c i n g s t ud e n t s (a) Nursing and Education (b) Physical and Natural Sciences Figure 1. Termination of National Priority
I thank Oliver Maclaren, Dani¨el Lakens and Guy Prochilofor providing guidance and references on how to provideevidence of absence of an effect rather than an absence ofevidence.
References
Britton, J., van der Erve, L., and Higgins, T. (2019). Income con-tingent student loan design: Lessons from around the world.
Eco-nomics of Education Review , 71:65 – 82. Higher Education Fi-nancing: Student Loans. [Cited on page 1.]Chapman, B. (1997). Conceptual Issues and the Australian Experi-ence With Income Contingent Charges for Higher Education.
TheEconomic Journal , 107(442):738–751. [Cited on page 3.]Chapman, B. (2006).
Government Managing Risk: Income contin-gent loans for social and economic progress , volume 40. Routledge,London. [Cited on page 1.]Chapman, B. and Leigh, A. (2009). Do very high tax rates inducebunching? implications for the design of income contingent loanschemes*.
Economic Record , 85(270):276–289. [Cited on page 1.]Steiger, J. (2004). Beyond the f test: Effect size confidence inter-vals and tests of close fit in the analysis of variance and contrastanalysis.
Psychological methods , 9:164–82. [Cited on page 3.]Stiglitz, J. E., Higgins, T., and Chapman, B. (2014).
Income Contin-gent Loans: Theory, Practice and Prospects . Palgrave Macmillan.[Cited on page 1.]The Parliament of Australia (2003). Higher education support act2003. . [Cited on page 1.]
Figure 2. Event window - . . C e n t e r e d l n ( C o mm e n c i n g s t ud e n t s ) -3 -2 -1 0 1 2 3 4 5Event time Note.