A new study by Andy Feng and Georg Graetz, research students at the Centre for Economic Performance, London School of Economics, finds that university degree classification matters for initial job outcomes.1
The degree classification is a system of categorizing performance on university degree programmes. The highest distinction for an undergraduate is the First Class honours followed by the Upper Second and Lower Second degrees (we do not consider Third Class and below in this study). A sizeable fraction of employers in the UK report using the classification system in hiring decisions and universities often use degree class to screen applicants to postgraduate programmes. However, it is not obvious that the classification system is useful if degree transcripts provide more accurate information.
Using survey and administrative data from the London School of Economics and Political Science (LSE), we find significant effects of degree class on initial labour market outcomes (six months after graduation). An Upper Second earns 7 per cent higher wages compared to a Lower Second while a First Class earns 3 per cent higher wages compared to an Upper Second.
Identifying these effects is complicated by the fact that a naïve comparison of, say, students who received a First Class against students who received an Upper Second could be biased because the former have higher ability or worked harder for the degree. To isolate the pure ‘sheepskin effect’ we employ a regression discontinuity design. Essentially we mimic a randomized experiment by comparing two similar students one of whom was lucky on a critical exam and scored 70 (thereby receiving a First Class), while the unlucky student just missed by scoring 69 (thereby receiving an Upper Second). We attribute the difference in their initial labour market outcomes to the effects of degree class alone.
LSE degree classification rules
Undergraduates in the LSE take nine courses over three years. Every course is graded out of 100 marks and fixed thresholds are used to map the marks to degree class. For example, a First Class honours requires either 5 marks of 70 or 4 marks of 70 with aggregate marks of at least 590. This mapping applies across all departments and years (Table 1).
These rules mean that the fourth highest mark for any student is critical in determining the eventual degree class. A student whose fourth highest mark is 70 (60) is much more likely to receive a First Class (Upper Second), everything else equal. This can be seen by plotting a graph of the fraction of students who receive a given degree on the y-axis against the marks on the fourth highest mark on the x-axis (see Figure 1). There is a clear jump in the probability of receiving a First Class at the 70-mark threshold. A similar story is seen for the Upper Second at the 60-mark threshold.
This jump in degree class ‘treatment’ is ideal for a regression discontinuity (RD) design where the fourth highest mark plays the role of the assignment variable. We argue that whether a student receives a 70 or 69 on the critical exam is down to random luck and this generates randomized assignment to the First Class or Upper Second ‘treatment’. In practice we employ a fuzzy RD where a dummy variable indicating the crossing of the relevant marks threshold is an instrument for the degree class.
Data and empirical strategy
We use two datasets. First, LSE student records gives us (fully anonymised) administrative data on age, gender, nationality and country of domicile information. It also includes course history and eventual degree class. The course history information allows us to find the critical fourth highest exam for each student.
The second dataset is the Destination of Leavers from Higher Education Survey (DLHE) that reports employment circumstances of students six months after graduation. From the DLHE we construct the relevant labour market outcome variables. We have a dummy variable for full-time employment and two-digit SIC industry codes. To construct our wage measure we merge Labour Force Survey (LFS) data at the industry by year by gender level into the DLHE survey. One concern is that the industry average wage is not representative of the earnings facing undergraduates. To address this we also calculate mean log wages conditional on university education and various experience levels.
Our interpretation of the wage outcomes is that these measure the industry’s ‘rank’ compared to other industries. This is a relevant measure because students form expectations of future wages on the basis of industry differences. The unconditional and conditional (on education and experience) industry wage measures are highly correlated and our results are comparable across different outcomes.
The details of our empirical strategy are explained in the working paper. We employ a fuzzy RD design using the marks received on the critical fourth highest exam as the assignment variable. A dummy variable indicating whether this critical mark crosses the 70 (60) mark threshold is used as an instrument for receipt of a First Class (Upper Second) degree. Our benchmark model uses linear regressions for samples of students close to the thresholds. In the results reported below, we always control for gender, age, nationality, year of award and department dummies (including year and department interactions).
Our first result compares the returns to receiving a First Class against an Upper Second degree. A First Class is ‘worth’ roughly 3 per cent in starting wages which translates into £1,000 per annum. An Upper Second is worth more — 7 per cent in starting wages which is roughly £2,040. When we look at men and women separately, men receive a statistically significant 6 per cent return to a First Class compared with a statistically insignificant — 2 per cent for women. For the Upper Second, men receive 8 per cent while women receive 5 per cent — however both effects are statistically insignificant.
These results are robust to a battery of specification checks. The combination of quasi-experimental variation and robustness of results strongly suggests that degree grades matter for labour market outcomes.
However there are important caveats. First, this is a highly selected sample because LSE is a specialized institution that consistently ranks within the top of the national distribution. Thus, the external validity of our results could be an issue. Second, we do not have data on students’ work experience during the course of their studies and this may affect their future wages regardless of degree class. Third, we cannot follow these graduates to see if the initial wage differences persist or fade. This is important because lifetime earnings are more relevant in welfare calculations.
We propose four areas for future work. First, replicating this methodology in other institutions to test the external validity of our results. Second, more work can be done to understand how employers make hiring decisions. It appears that informational frictions exist and employers use degree class in their decisions but it is less clear how exactly this is done. Third, on the supply side an interesting area of research would be to understand how students perceive the returns to degrees. To the extent that degree majors vary in their distribution of awards, the expected returns to degree class may influence programme choice. Finally, it would be interesting to see how these initial differences change over the careers of graduates.
Broader discussion and conclusion
Our results show that degree class matters for initial job outcomes. These results speak to the broader policy discussion regarding whether the UK should move to alternative award systems like the grade point average (GPA) system that is adopted in the US. To the extent that our findings show that degree class results in sharp differences in outcomes, a more graduated system of awards like the GPA may be attractive. On the other hand, if full transcripts (which provide greater detail on student achievement) are already available to employers, it is not clear that moving to any other system of awards would lead to an improvement. In the end, we think that more work in this area needs to be done to understand the process involved in hiring workers and in understanding the informational frictions that exist in the graduate labour market and hope that our contribution provides some insight into this under-studied field.
1. A fuller version of this paper can be downloaded at http://cep.lse.ac.uk/_new/publications/abstract.asp?index=4246