As regular readers of this blog will know, I am mildly obsessed with the UK government’s LEO (Longitudinal Education Outcomes) statistical set, which provides vast amounts of easily manipulable statistical data. The set measures what the UK government increasingly considers the key educational outcome of a university education: how much graduates earn. But this is not the post I originally set out to write, and the reason why interests me, and hopefully you.
I wanted to check one of the basic assumptions being made about discrimination in university education (especially racial discrimination): that Oxford and Cambridge graduates earn more than those from other universities and so any discrimination in their intake causes major economic disadvantage right from the start of a career.
The claim about Oxbridge students earning more has tended to be made (especially by the Sutton Trust) either based on Oxbridge over-representation in “top jobs”, e.g. in Leading People 2016 or based on graduate earnings (e.g. Earning by Degrees from 2014). But LEO data would allow us (so I thought) to look in a bit more detail. One crucial thing it’s already demonstrated is regional differences: see for example this chart (Figure 22, p. 46) from Graduate outcomes (LEO): Employment and earnings outcomes of higher education graduates by subject studied and graduate characteristics SFR 15/2018 looking at the home region (address prior to starting an undergraduate degree) of young students attending English universities.
It’s likely that this is due to students wanting to remain in a specific region after graduating, even if doing so harms their earning potential. That’s a problem if you’re comparing graduate outcomes from Oxbridge (in the most prosperous south-east of the UK) with the other Russell Group universities which have a wider geographical spread, or even the Sutton Trust’s highly selective Sutton Trust 13 universities (Birmingham, Bristol, Cambridge, Durham, Edinburgh, Nottingham, Oxford, St Andrews, Warwick, York, Imperial College London, LSE, UCL).
So what I wanted to do instead was compare Oxbridge LEO data by subject against some top London universities: I chose Imperial, King’s College London, LSE and UCL. All of these have above average attendance by ethnic minorities and are in the most prosperous region of the UK. They struck me therefore as a reasonable proxy for graduates who were willing and able to find work in London and the South-East generally.
So far, so good, but then I actually started looking at the data. This is from the most recent data currently available (Graduate outcomes (LEO): subject by provider, 2015 to 2016, https://www.gov.uk/government/statistics/graduate-outcomes-leo-subject-by-provider-2015-to-2016) and shows the median earnings for both sexes, 5 years after graduation:
|Architecture, Building and Planning||Not available||33,900|
|Biological Sciences (excluding Psychology)||Not available||Not available||31,200||Not available||Not available|
|Business & Administrative studies||56,300||67,200||Not available||37,400||Not available||38,400|
|Creative Arts & Design||23,800||26,400||27,000||Not available|
|Economics||Not available||Not available||61,100||Not available|
|Education||Not available||Not available||24,400|
|Engineering and Technology||43,700||44,800||40,700||38,000||40,700|
|Historical & Philosophical Studies||33,700||37,300||28,500||Not available||30,700|
|Mathematical Sciences||Not available||50,300||Not available||37,200||55,400||49,500|
|Medicine and Dentistry||Not available||42,900||45,300||46,400||47,100|
|Physical Sciences||Not available||37,000||38,400||Not available||Not available|
|Social Studies (excluding Economics)||Not available||Not available||Not available||32,300||32,000|
|Subjects Allied to Medicine (excluding Nursing)||Not available||36,300||30,900||34,200||36,300|
I’ve excluded subject categories not studied at either Oxford or Cambridge (Agriculture & Related Subjects, Combined Studies, Mass Communication and Documentation, Nursing, Veterinary Science). In the table, a blank indicates that the subject isn’t studied at that university.
What is striking is how much of the time the median salary data is simply not available. Out of 18 subject categories in this table, there are 4 for which a full comparison of all the relevant data can be made. (I’ve highlighted these in blue on the table). Even if you just restrict it to comparing Oxford with Cambridge, you can only do that for 8 out of 18 subject categories.
I’m not entirely sure why this data is missing: it’s not simply that these represent particularly small cohorts. For example, there are 100 Cambridge mathematics students whose earnings were traced. But I take it that the data is excluded because it isn’t considered by the LEO statisticians to be reliable, and I’m not second-guessing their judgement on that.
So I simply can’t tell, based on this data, whether Oxbridge students do overall earn more than comparable students from top London universities. It might still be possible for expert statisticians working with the full data set to decide that. But what my admittedly crude analysis shows is that the LEO data is not much use for potential students deciding which university to study at.
This is true even in the relatively few cases where we do have a complete set of subject/university group data from LEO. The data is still very noisy. Here’s the five-year data for computer science graduates (both sexes) from the 2008/09 and 2009/10 cohorts:
|Computer Science 2008/09||51,300||43,000||51,800||38,400||43,700|
|Computer Science 2009/10||46,300||55,800||60,000||40,100||41,500|
Which of these universities should you go to improve your earning potential? If you were in the 2008/2009 cohort you’d have done better by choosing Cambridge over Imperial: the median salary is slightly less, but you’d probably have saved enough money on three years living expenses in London for it not to matter. In the 2009/10 cohort the median graduate from Imperial would have earned so much more than a graduate from Cambridge that they could have paid off an entire year’s fees with the difference. From the 2008/09 cohort, the median UCL graduate earned £700 a year more than the median Oxford graduate; for the next cohort, it was £14,300 a year less.
These earnings don’t even move in synchronisation between the two cohorts, as you might expect if they reflect the wider economy: three of the universities have higher median salaries in 2009/10 and two lower. And bear in mind that this is all retrospective data: you’re potentially picking your course to start in 2019 or more likely 2020 based on this data. Which cohort is more typical?
All this suggests that the most that LEO data is likely to give you is some very broad generalisations about which subjects and universities are likely to lead to well-paid jobs. As a specific tool for deciding between a subject at comparable universities, it’s completely inadequate. The problem is, it’s only if you dig deeper into the data, in the way I’m doing, that you’re likely to spot that. I suspect that a lot of spuriously precise salary figures are going to be quoted by the government, individual universities and commentators in the future.