Graduate earnings rarely afford good policymaking

The advent of datasets linking graduates’ income to their student records has fuelled calls for certain courses and universities to be excluded from public funding. But, ahead of England’s Augar review of post-18 education, the minister who commissioned the longitudinal education outcomes project, David Willetts, warns against such abuses of the data

April 11, 2019
Montage of man in bowler hat and charts
Source: iStock montage

Which universities’ and subjects’ graduates go on to earn the most – and the least? Those are not unreasonable questions for prospective students to wonder about. They are also very relevant to policymakers – particularly in England, where the government-commissioned Augar review of post-18 funding is due to report imminently.

Until recently, neither students nor policymakers had any firmer basis to answer their questions than anecdote and received wisdom. That is why, as UK minister for universities and science, I commissioned the longitudinal education outcomes project (LEO). This is one of the biggest, most policy-relevant datasets to arrive in Whitehall for years. By linking educational data on students to tax data on their earnings, LEO promises fresh insights into social mobility by tracking specific routes from school to university and out to good jobs. It is a good example of using administrative data for social science. No wonder it is hot.

But it is also dangerous. The idea that we have reached “peak student” is currently fashionable, hovering somewhere between a forecast and a policy preference. And LEO is taken to present an objective means by which student numbers could be reined in, by cracking down on courses that yield low graduate returns. But that, in my view, would be a misuse of the data and a major policy error.

LEO is part of a global trend. Amid widespread concern over the performance of some for-profit providers in the US, President Obama brought in such measures as a means to avoid putting public money into grants and loans for students at those providers with poor employment outcomes. And similar – in some cases, even more elaborate – linked datasets have been generated in Canada and New Zealand.

In the UK, the story begins with Lord Browne’s review of higher education, commissioned by the previous Labour government before the 2010 general election and arriving on my desk, as the new minister, that autumn. Browne wanted to remove the cap on tuition fees to stimulate price competition. But, in a system of income-contingent public loans, he had to stop fees from shooting up to levels where the taxpayer would face a high cost from writing off unpaid debt. His review, therefore, proposed that universities pay a levy on their fees, at an increasing rate as those fees went higher, which would go straight back to the Treasury as pre-payment for future loan write-offs.

It was an ingenious idea. The only trouble was that the intended beneficiaries of uncapped fees, the research-intensive universities, strongly opposed the idea. The levy was going to reach a very high rate, reflecting forecasts of high average loan write-offs across all English universities if fees were very high. But the universities of Oxford and Cambridge, for example, believed that their students' loan write-offs would be much lower than the high levy they would face. We simply did not have the information on likely repayment rates to set the levy university by university and course by course, so Oxbridge were probably right to fear that it would have ended up as an onerous tax on our most prestigious universities.

Illustration of city and money
Getty/iStock montage

It was clear to me that we really needed university-by-university repayment estimates if we were ever to be able to implement a model like Browne’s. It would be useful information for prospective students too. So we set about getting the information from HM Revenue and Customs. The task went to two excellent economists, Neil Shephard (then at Oxford, now at Harvard University) and Anna Vignoles (then at the UCL Institute of Education, now at Cambridge). That led to the first crucial report on graduate earnings, How English-domiciled graduate earnings vary with gender, institution attended, subject, and socio-economic background, which came out in 2016. It found that more than 10 per cent of male graduates from Oxford, Cambridge and the London School of Economics were earning salaries in excess of £100,000 a year 10 years after graduation. Meanwhile, 10 per cent or more of the male graduates of a full 36 universities were earning more than £60,000 after a decade – although for female graduates that statistic held for only 10 universities. At the other end of the spectrum, 23 institutions produced male graduates whose median earnings were less than those of non-graduates 10 years on; for female graduates, the figure dropped to nine.

The research showed that there are wide disparities in graduate earnings university by university, and this would have made it possible to implement a full-blown version of the Browne model. But the research also revealed the actual reasons for the differences in graduate earnings and so raised big doubts about whether this was good grounds for divergence in fees.

The key reasons were students’ prior attainment, parental social class and subject studied. For most universities, there was not a strong institutional effect independent of these factors. So a higher fee would be a reward not for educational quality but for selecting students with good A levels from affluent families who want to be bankers or lawyers. It would be the pupil premium in reverse. These arguments are still relevant to today’s debate.

Since 2016, the research programme has developed further, with more important findings published last November. That paper, The impact of undergraduate degrees on early-career earnings, links school, university and tax records for everyone who took their GCSEs in England since 2002. The headline, unadjusted figures are that female graduates earn 50 per cent more than non-graduates with five good GCSEs, compared with a 25 per cent advantage for men. However, after allowing for factors such as prior attainment and social background, indicated by the family being on free school meals, returns fall dramatically to 26 per cent for women and 6 per cent for men. And there are 12 institutions with statistically significant negative returns for the 4 per cent of all male students who go there. Graduates have particularly low earnings from courses in the performing arts, but, overall, 85 per cent of graduates derive significantly positive returns by the age of 29.

The higher return for women is striking. It does make intuitive sense, however, as there are few well-paid stereotypically female non-graduate jobs, whereas non-graduates may have quite high earnings in some predominantly male jobs such as long-distance lorry drivers or plumbers. It suggests that there is a clear economic rationale for higher rates of female university participation.

This is information that certainly ought to be available to students. But now that the Augar review has opened up a wider debate on higher education funding, there are ways that policymakers could be tempted to act upon it, too. Most obviously, they might decide to refuse to provide loans for some courses at the universities with apparently low returns.

However, such a move would be problematic. The initial LEO research project was very well suited to assessing specific policy options around graduate repayments. It used graduate earnings to assess prospects for repaying loans. Since repayment obligations are largely determined on the basis of taxable earnings, the data and the policy question were tied together. Earnings data, however, are not necessarily a guide to wider policy, such as the performance of universities.

For instance, LEO measures annual earnings, with no distinction between part-time and full-time work, so it does not say how much hourly earnings are. Young women with poorer qualifications tend to work part-time; this artificially depresses their earnings, which, in turn, boosts the relative returns to the female graduates.

Furthermore, LEO offers no information on occupation or industry or other employer characteristics, so a university that provides nurse and teacher training will inevitably appear to perform less well than one focused on financial services and City law firms. The original LEO analysis did not include the self-employed (although this is now being rectified), so a university’s strength in fostering entrepreneurial students working for their own start-ups would not show up. And while the data show in which part of the country someone was educated, they do not show where they work. As some graduates stay near where they studied, this penalises universities in areas with lower earnings. So when the data tell us that some non-graduates earn as much as graduates, they could be telling us that a public school dropout working at an upmarket estate agent in Kensington earns as much as a recent graduate working part-time in Bolton.

The dataset stops at age 29 because of limitations on how far back the education data are available. So it fails to capture the evidence that graduate earnings have a better long-term trajectory than non-graduate earnings. This is particularly true of some arts courses. The data favour those occupations where you get to peak earnings early on. They mirror the failures of the British economy, rewarding quick, high returns over longer, slower ones. The researchers themselves confirm that the average graduate premium is likely to grow over the course of a career, especially for men.

Moreover, the post-crash decade covered in LEO has been bad for pay generally, as our research at the Resolution Foundation has shown. It began with graduates trading down into less well paid jobs, displacing non-graduates and putting them at greater risk of being unemployed. Things have got better since then, but any measure that focuses on earnings alone will still fail to capture the higher risks of being unemployed for non-graduates.

Despite that higher risk, current LEO evidence shows a particularly low graduate premium for men with low prior attainment and no science or maths A levels. If you were focused solely on LEO data, those are the people you would exclude from the loans system. But what would that do for social mobility? Likewise, a first-generation university graduate who works part-time in her local area as a nurse does not show up well in the data, but she might still have been on a significant personal journey, with wider social and economic benefits. And what if a university has a programme to take on ex-offenders – socially desirable but lowering its graduate returns?

To be fair, the LEO researchers are aware of these issues and try to capture the effects of social disadvantage. They adjust graduate earnings for socio-economic status, school type and ethnicity, for example. But they recognise that there may be other causal factors that they are unable to allow for, such as parental education, marital status, number of children or health problems.

We have been here before. I can well remember how, during my time as minister in the coalition government, we were very sceptical of the work-based National Vocational Qualifications (NVQs) because research showed low and even negative returns. But since then there has been research that appears to show that after allowing more fully for social disadvantage, the returns for some NVQs are actually higher than previously estimated. So there are dangers in excessive eagerness to crack down on apparently low returns before the evidence has been fully tested.

The other criticism of overreliance on LEO data is that graduate earnings data do not capture all the economic returns of a university education, never mind the wider social benefits. For instance, creative arts graduates contribute to the success of the creative industries. Graduates tend to boost productivity and they may not capture all these returns in their own pay. There are fiscal benefits from improved health and lower levels of crime and offending. That is why in my 2017 book, A University Education, I argue that advocates of higher education need to emulate the advocates of early years education, who have had such an influence in Whitehall by presenting a range of broader arguments for their cause.

Another rather awkward angle is that there seems little correlation between earnings figures and the student satisfaction data that are part of the teaching excellence framework – the other obvious driver of policy direction. This just underlines the point that the LEO data have strongest implications for policy that is most related to earnings and tax. The further you go from the original purpose of the data, the more tenuous the link to the policy conclusion.

It is clear that the LEO data weaken the case for a graduate tax as an alternative to tuition fees. One of the key problems with such a tax is that high earners would face paying far more than the actual cost of their education. I have always disliked this in principle: while it is right to expect graduates to pay back for their own higher education, I do not see why they have a greater obligation than the rest of the population to pay for other graduates’ higher education. The LEO analysis makes it very clear who these higher payers would be, and would introduce strong incentives for English residents contemplating studying economics or law at a prestigious UK university to avoid a lifetime of higher tax by studying overseas instead – or else to pursue a subsequent career abroad, beyond the reach of the UK tax authorities (but not of the Student Loans Company).

As for the idea that LEO data present a strong case for constraining the growth of higher education, I myself doubt the feasibility or desirability of this. In most modern, Westernised societies, participation keeps going up. Swimming against that tide would be, to put it mildly, a bold move. Indeed, anticipating the demographic turning point early in the next decade, when numbers of young adults will start rising again, I would be planning for growth.

Excluding the courses and universities that appear to do badly under LEO from public support would introduce a two-tier system in which affluent parents, whose children do not need public loans, could presumably buy places. The performing arts would become even more middle class. It would also mean that a Whitehall planner has to pronounce on the value of sports science at University X and drama studies at University Y. It would take an interesting new dataset and turn it into a tool of a very significant policy directly constraining the options for prospective students: a role that is quite simply beyond it and a threat to LEO’s long-term credibility and development.

Montage of man with money and briefcase
Getty/iStock montage

Rather than concentrating on the output of a university education – graduate earnings – it has also been suggested that the Augar review may recommend constraining numbers by the input – prior attainment. But setting minimum A-level grades for public support would also create a two-tier system whereby the children of wealthier parents could still go to university even if their grades were very low. It would be a big barrier to mature students and pretty much destroy the rationale of the Open University unless the policy only applied below a maximum age – in which case it would seem tough on younger people. And you might be turning away students from courses that looked a good bet in LEO terms. That is why, when looked at from time to time, this option has always been rejected.

Both of these options would also involve a significant shift from the English model of higher education, in which Whitehall does not specify the courses that universities teach or the students they admit. Many observers regard this high level of university autonomy as one reason for the system’s success. It is, in the best sense of the word, liberal, rejecting attempts at manpower planning. But, as one advocate of restricted access put it to me: “There is not much liberalism around in Whitehall at the moment.”

A third option would be to revert to the crude device of simply limiting the number of domestic students per university. That would preserve more autonomy by leaving universities free to decide who to recruit within their fixed totals. The trouble is that this system constrains competition between institutions, making it harder for some to grow and others to shrink. We prided ourselves that the coalition got rid of this system in 2015-16. It meant that the state was no longer requiring universities to turn away young people who wanted to be educated and that a university was willing to educate – particularly important for social mobility given the Scottish evidence that when places are restricted, it is the marginal students from disadvantaged backgrounds who lose out. 

There is one other option. It has the merit of actually tackling the genuine concern of Whitehall about the cost of higher education without getting into the dangerous territory of manpower planning. That is simply to change the graduate repayment terms so that the typical graduate pays back more fully for the cost of their higher education.

The current system’s high repayment threshold of £25,000 means that too high a proportion of the loans is written off. Predictably, this has opened up the whole question of the treatment of the write-offs in the public accounts, leading to their proposed reclassification as public spending. It is not even politically popular because, combined with the high interest rate on some outstanding debt, many graduates now see their debt rising every year, which understandably upsets them.

So I propose a package of abolishing the interest rate and lowering the repayment threshold back down to its original £21,000 – which virtually nobody ever complained about. One might add a few extra years to the repayment period as well. That would make the system both financially sustainable and more politically acceptable without having to constrain the autonomy of universities.

As for LEO, the data should be part of the increasing mix of information available to prospective students and their parents, but we need to understand them more fully before wider lessons can be drawn. The best way to extract more value from the dataset would be for more researchers to be able to access it – with the necessary privacy protections, of course. We at the Resolution Foundation are keen to analyse the raw data, and so are others.

If we are going to overturn the time-honoured and highly successful liberal settlement in English higher education policy, we had better be very sure that the case for it bears every scrutiny.

David Willetts is chair of the Resolution Foundation and a visiting professor at King’s College London. He was UK minister of state for universities and science from 2010 to 2014. His book, A University Education, is published by Oxford University Press.


Print headline: You are what you earn?

Register to continue

Why register?

  • Registration is free and only takes a moment
  • Once registered, you can read 3 articles a month
  • Sign up for our newsletter
Please Login or Register to read this article.

Related articles

Reader's comments (4)

What is being proposd here is tinkering around at the edges of a student loan system that is fundamentally unfit for purpose. Willetts seems to want his cake in 2019, having eaten it as a minister in 2010-2014. Providing better information to students and parents is part of the myth of choice required to sustain his market based approach to policy making. The data is not the real issue here, it is the intention, and Willetts remains as confused now as when he was in office. What precisely does he intend to achieve?
Thus we see the danger in collecting large amounts of data without a clear idea of what you intend to do with them. The whole thing is flawed anyway. The value of a university education is not measured in how much you can earn after receiving it, unless you are some narrow-minded ideology-driven politician. Consider the inflated salaries paid in London. If you don't happen to work there, you earn less even if you are doing the same kind of work. Has that been considered in these rather feeble attempts at data science? Consider the inflated salaries paid in certain trades: financial services, the law, medicine (although that's a slower build)... have these been taken into account? Pure earning power of graduates is a poor metric on which to judge a university. I suppose it appeals to politicians because it is simple.
notably silent on the more fundamental underlying questions about the (non-monetary) 'value' of a university education and the corrosive effects of 'marketisation' on university culture and practice
I do find it rather jarring to see Willetts, who was a fairly enthusiastic opponent of market values in HE clutching his pearls when, unsurprisingly, this is taken to its logical conclusion by using the only data the market really takes any notice of. And it is just one more example of the tyranny of the metric, now not just used for benchmarking and observation, but employed (usually poorly) as a blunt instrument of policy. As at least one the poster has already said, earning data is likely to be a poor indicator of job outcomes, and even then is job outcome the only thing hat is seen as important in the obtaining of a degree? In these rather instrumental times it appears so, and is especially galling when a great number of the politicians making such policy decisions appear to be bordering on the functionally innumerate. The Two Cultures is still alive and surviving