The relative labour market returns to different degrees Research report June 2018 Chris Belfield, Jack Britton, Franz Buscha, Lorraine Dearden, Matt Dickson, Laura van der Erve, Luke Sibieta, Anna Vignoles, Ian Walker and Yu Zhu Institute for Fiscal Studies 5 Contents Executive Summary 1 Introduction 9 2 Existing literature and approaches 11 2.1 Relative returns by subject . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 2.2 Relative returns by institution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 3 Data 15 3.1 Data descriptives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 3.2 Earnings dierences by subject and institution . . . . . . . . . . . . . . . . . . . . . 23 4 Methodology 26 4.1 Controlling for selection into courses . . . . . . . . . . . . . . . . . . . . . . . . . . . 26 4.2 Inverse Probability Weighting rst stage . . . . . . . . . . . . . . . . . . . . . . . . . 30 4.3 Model of earnings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 4.4 Model of employment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33 5 Subject estimates 33 5.1 Heterogeneity in subject estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 6 Higher education institution estimates 44 7 Course (subject-institution) estimates 48 8 Earnings returns and A-level subject mix 55 9 Employment 58 10 Discussion and conclusions 61 Bibliography 63 A Sensitivity checks 65 B Results for women 69 C Coecients on control variables 75 D Other graphs and tables 77 1 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 List of Figures Real earnings by graduation cohort (conditional on sustained employment) . . . . . . . . . . . . 21 Real earnings by time after graduation (conditional on sustained employment) . . . . . . . . . . 21 Real earnings by gender (conditional on sustained employment) . . . . . . . . . . . . . . . . . 21 Real earnings by GCSE maths grade (conditional on sustained employment) . . . . . . . . . . . 22 Real earnings by socio-economic status (conditional on sustained employment) . . . . . . . . . . 22 Real earnings by institution region (conditional on sustained employment) . . . . . . . . . . . . 23 Mean earnings 5 years after graduation by subject for women in sustained employment . . . . . . 24 Mean earnings 5 years after graduation by subject for men in sustained employment . . . . . . . 24 Mean earnings 5 years after graduation by HEI for women in sustained employment . . . . . . . . 25 Mean earnings 5 years after graduation by HEI for men in sustained employment . . . . . . . . . 26 Distribution of UCAS points by university . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 Share of students studying each subject with a maths A- or AS-level, men . . . . . . . . . . . . 28 Share of students studying each subject with a science A- or AS-level, men . . . . . . . . . . . . 28 Subject raw and IPWRA coecients for women . . . . . . . . . . . . . . . . . . . . . . . . 36 Subject raw and IPWRA coecients for men . . . . . . . . . . . . . . . . . . . . . . . . . . 37 Subject IPWRA coecients for women . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38 Subject IPWRA coecients for men . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 Subject IPWRA coecients, men vs women . . . . . . . . . . . . . . . . . . . . . . . . . . 41 Subject IPWRA coecients, high SES vs low SES for men . . . . . . . . . . . . . . . . . . . . 42 Subject IPWRA coecients, high ability vs low ability for men . . . . . . . . . . . . . . . . . 43 Subject IPWRA coecients, academic A-level vs non-academic route for men . . . . . . . . . . 44 HEI raw earnings and IPWRA estimates for women . . . . . . . . . . . . . . . . . . . . . . . 45 HEI raw earnings and IPWRA estimates for men . . . . . . . . . . . . . . . . . . . . . . . . 45 HEI IPWRA coecients for women . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47 HEI IPWRA coecients for men . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47 Course IPWRA estimates for women relative to average course, highlighting Cambridge . . . . . . 50 Distribution of course rank by university group for women . . . . . . . . . . . . . . . . . . . . 50 Course IPWRA estimates for men relative to average course, highlighting Cambridge . . . . . . . 51 Distribution of course rank by university group for men . . . . . . . . . . . . . . . . . . . . . 51 Course IPWRA coecients for women relative to average course, highlighting business . . . . . . 52 Course IPWRA coecients for men relative to average course, highlighting business . . . . . . . . 53 Boxplot of IPWRA course returns by subject for women . . . . . . . . . . . . . . . . . . . . 54 Boxplot of IPWRA course returns by subject for men . . . . . . . . . . . . . . . . . . . . . . 54 Subject returns by proportion of intake with a maths A- or AS-level, men . . . . . . . . . . . . 56 Subject returns by proportion of intake with a science A- or AS-level, men . . . . . . . . . . . . 56 Economics course returns by proportion of intake with a maths A- or AS-level, men . . . . . . . . 57 Creative arts course returns by proportion of intake with a maths A- or AS-level, men . . . . . . . 58 Subject employment returns 5 years after graduation for women . . . . . . . . . . . . . . . . . 59 Subject employment returns 5 years after graduation for men . . . . . . . . . . . . . . . . . . 59 Institution employment returns 5 years after graduation for women . . . . . . . . . . . . . . . . 60 2 41 Institution employment returns 5 years after graduation for men . . . . . . . . . . . . . . . . . 61 A1 Subject IPWRA coecients for women with and without SA data . . . . . . . . . . . . . . . . 65 A2 Subject IPWRA coecients for men with and without SA data . . . . . . . . . . . . . . . . . 66 A3 Subject IPWRA coecients for women with and with dropouts . . . . . . . . . . . . . . . . . 67 A4 Subject IPWRA coecients for men with and without dropouts . . . . . . . . . . . . . . . . . 68 B1 Subject IPWRA coecients, high SES vs low SES for women. . . . . . . . . . . . . . . . . . . 69 B2 Subject IPWRA coecients, high ability vs low ability for women. . . . . . . . . . . . . . . . 70 B3 Subject IPWRA coecients, academic A-level vs non-academic route for women . . . . . . . . . 71 B4 Share of students studying each subject with a maths A- or AS-level for women . . . . . . . . . . 72 B5 Share of students studying each subject with a science A- or AS-level for women . . . . . . . . . 72 B6 Subject returns by proportion of intake with a maths A- or AS-level for women . . . . . . . . . . 73 B7 Subject returns by proportion of intake with a science A- or AS-level for women . . . . . . . . . 73 B8 Economics course returns by proportion of intake with a maths A- or AS-level for women . . . . . 74 B9 Creative arts course returns by proportion of intake with a maths A- or AS-level for women . . . . 74 3 List of Tables 1 Sample sizes for LEO dataset by cohort . . . . . . . . . . . . . . . . . . . . . . . . . 16 2 Background characteristics of matched and NPD samples . . . . . . . . . . . . . . . 18 3 Subject sizes, dropout rates and self-employment rates . . . . . . . . . . . . . . . . . 19 4 Raw and IPWRA-weighted sample statistics for men used in the nal sample . . . . 31 5 Subject estimates for men and women (in %) . . . . . . . . . . . . . . . . . . . . . . 35 6 Returns by subgroup (in %) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40 C1 Coecients on parameters in subject-level IPWRA estimation . . . . . . . . . . . . . 75 D1 Course aims in sample 2011-12 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77 4 Executive Summary Each year more than 350,000 students start Higher Education (HE) degrees in England at a total cost of around £17 billion paid by graduates in repayments on student loans and the taxpayer (Beleld et al., 2017). This represents a signicant investment and has the potential to have considerable implications for the students later-life outcomes. Students typically make the decision about where and what to study at university at an early age. This decision is inuenced by many things, reecting the various expected benets associated with a higher education, but a key one will be the potential impact on their future employment prospects. However, currently there is little evidence about the dierential impact dierent degrees might have on their medium term earnings or employment prospects. Accurate and timely estimates of the relative value of dierent degree courses are vital to ensure degrees represent value for money for both students and the government. This report is the rst in a series of reports seeking to improve the information available to stakeholders on the value of dierent degree courses. The reports are making use of the new Longitudinal Education Outcomes (LEO) administrative dataset developed by the UK Department for Education (DfE), which tracks English students through school, college, university and into the labour market. The current report will provide estimates of the labour market return - measured by earnings and employment 5 years after graduation - to dierent subjects, institutions and degree courses relative to the average degree. Graduates earnings and employment prospects are aected both by their pre-university char- acteristics, such as their ability level or social background, and by the impact of studying a particular degree. As a result, subjects may have very high average graduate earnings simply because they take high-ability students rather than because of the impact of the degree itself. Raw dierences between courses can therefore be misleading as to the actual return from doing a given degree. This report tries to disentangle the impact of a degree on earnings and employment outcomes from that of student characteristics, providing estimates of the impact of dierent degrees on graduate earnings. The LEO dataset provides a unique opportunity to do so by allowing us to account for dierences in background and prior attainment between graduates who take dierent degrees. Findings The labour market returns to dierent degrees vary considerably even after accounting for the considerable dierences in student composition. Both the subject of degree and institution attended make a considerable dierence to graduates earnings. All these estimates refer to dierences in earnings 5 years after graduation (or expected graduation for dropouts). We know there is wide variation in the earnings between graduates from dierent degrees. Medicine, maths and economics graduates all typically earn at least 30% more than the 5 average graduate, while creative arts graduates earn around 25% less on average. A large proportion of these dierences in raw earnings can be explained by dierences in the char- acteristics of students taking these degrees. However, after accounting for these, signicant dierences in the relative returns to dierent subjects remain. Once these dierences have been controlled for, medicine and economics degrees have returns around 20% greater than the average degree, and business, computing and architecture degrees all oer relative earn- ings premia in excess of 10% above the average earnings for graduates. Creative arts - which enrols more than 10% of all students - still has very low returns: around 15% less than the average degree. These dierences in returns are large. By comparison, after conditioning on all other char- acteristics, degree subject and institution, graduates from independent schools and the top quintile earn around 7% to 9% more than those graduates from the lowest SES backgrounds. Similarly, adding an extra A at A-level increases earnings by around 3%. These gures represent the average returns based on the students who take these subjects. There is no reason to expect the returns will be the same for all students; for example, lower ability students would be unlikely to be able to achieve the high returns that we observe for medical degrees (even if they were able to gain access to the course). Indeed we do nd evidence that degrees have a dierent impact on dierent types of students. Medicine, pharmacology and English have relatively higher returns for females than males. Computer science by contrast is more benecial for males. Medicine and education have higher returns for students from lower socio-economic backgrounds, while economics and history have higher returns for students from higher socio-economic backgrounds. Social care and creative arts have a relatively higher return for students with lower levels of ability, as measured by their prior achievement. There is also considerable variation in raw earnings across institutions. High-status universi- ties, such as the Russell Group and universities established before 1992, typically have higher- earning graduates. These universities however also typically take the highest-ability students. Once dierences in the student composition between universities have been accounted for, the variation in returns is considerably reduced, but signicant dierences remain. Even after controlling for these dierences, the traditionally high-status universities such as the Russell Group still provide the highest returns. This analysis cannot distinguish whether these dier- ences result from the dierences in the economic value of the skills provided by the universities or the signalling value of having attended a prestigious university. However, recent evidence on the issue of signalling vs. human capital eects of university education has suggested that the latter is important (Arteaga, 2018). Further, from a student choice perspective, this distinction might be less important. One of the key contributions of the LEO data including the full population of students is that we are able to estimate the returns to specic courses (a specic subject at a given university). Some of the estimates are imprecise due to small sample sizes. Nonetheless, the variation across courses is striking. The top-earning courses attract a 100% premium over 6 average graduate earnings, whilst the lowest-earning courses attract earnings that are around 40% below average graduate earnings. These ndings imply that studying the same subject at a dierent institution can yield a very dierent earnings premium. For example, the best business studies degrees have returns in excess of 50% more than the average degree while the worst business degrees have below average returns. These are considerable dierences in graduat