Hard and soft choices? Subject selection by schools and students

ABSTRACT

We present an analysis of A-level subject choices at around age 16 for a cohort of students in English schools who completed their studies in 2014. We examined both the National Pupil Database and a unique rich dataset on the subject preferences and subsequent choices between the ages of 16 and 18 (i.e. GCSE and A-level). We found substantive differences between students’ preferences and actual choices of ‘hard’ and ‘soft’ post-16 subjects (i.e. A-level). These differences were strongly associated with falsification of students’ expectations of examination grades taken at age 16 (i.e. GCSE) in the core subjects of English and mathematics. The sizes of these falsification effects were much larger than other significant associations such as gender, ethnicity, and social class. This suggests that subject choices are not rigidly framed by stable individual preferences and they are therefore open to influence from new information, persuasion, and opportunities.

KEYWORDS:

• Field of study
• subject choice
• economics of education
• grade expectations
• economics education
• business education

Acknowledgments

We acknowledge and thank the Nuffield Foundation for funding the data collection for this project and the University of Birmingham for funding the research and dissemination. We thank seminar participants at the Developments in Economics Education Conference 2015 at the University of Birmingham and at the Work, Pensions and Labour Economics Study Group conference 2015 at the University of Sheffield for their helpful comments. We also thank an anonymous referee for helpful comments and suggestions. All remaining errors are our own.

Ethics approval was obtained via the University of Birmingham under reference number ERN_10-1340 (July 2011). Any opinions expressed herein are those of the authors.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1. This area included all schools teaching pupils aged 16–18 in the postcode areas: AL, B, BA. BR, BS, CH, CR, CV, CW, DE, E, EN, GL, HA, HP, IG, KT, L, LE, LU, M, MK, N, NG, NN, NW, OL, OX, RG, RH, RM, SE, SG, SK, SL, SM, ST, SW, TW, UB, W, WA, WD WR, WS, WV. This area was roughly bounded by the cities of Liverpool, Sheffield, London and Bristol.

2. The ‘average relative variance increase’ statistics in Tables 7 and 8 are all small (though no precise critical values exist) confirming that the missing information did not have a significant impact on the results. The ‘largest fraction of missing observations’ (LFMI) statistics are also quite small, suggesting that 30 imputations is sufficient. Again, there are no precise critical values for LFMI but the rule of thumb is that for the imputations to be sufficient these should exceed 100 times the LFMI and in all regressions this appears to be true. However, 30 imputations is quite high and earlier attempts, with smaller numbers of imputations, did not pass these diagnostic tests, particularly when it came to predict the decision to take business studies. This need for a large number of imputations might arise from the fact that most of the explanatory variables are missing at least a few observations.

3. Appendix Tables A5 and A6 report the same model estimates using the more traditional ‘complete-case cross-section’ method with simple case-wise removal of any student with any missing observations. The Appendix results are similar to those in the main body of the text but with about 1000 fewer observations.

4. We ran a separate multinomial regression omitting the attainment variables (available on request). The associations with student characteristics (particularly socio-economic status) were strengthened indicating that gender, ethnicity, and socio-economic background influences on subject choice operate partly directly and partly through achievement in school.

5. We are grateful to a referee who suggested we examine the relationship between average school performance and the difference between a student’s actual and expected grade (accuracy). We found a small positive association between attending a school with a higher percentage of students gaining five or more GCSE grades A*–C and the difference between the student’s actual and end expected grade in both mathematics and English. This association was slightly attenuated by controlling for other pupil and school characteristics. Once these controls were added we found that one fifth of the residual variance in accuracy in English predictions and one seventh of the residual variance in accuracy in maths predictions was situated at school level. Further analysis of relationships between expected and actual grades in this sample is available in Perry, Davies, and Qiu (2018).’ (This was a late request from the authors when I had to go back to them with a query on a typo (‘month’ on line 867.)

6. It was not possible to include school fixed effects in the multinomial logit regressions. The number of extra parameters (number of schools times number of outcomes) increased the parameter space to such an extent that the maximum likelihood estimator could not achieve convergence.

Additional information

Funding

This work was supported by the Nuffeld Foundation (www.nuffieldfoundation.org) under Grantnumber EDU/39013.

Notes on contributors

Peter Davies

Peter Davies is professor of education policy research at the University of Birmingham and affiliated professor at Stockholm University. He is the author of ‘Paying for education: Debating the price of progress’ (2018) and ‘Markets for schooling: An economic analysis’ (with Nick Adnett) (2002).

Marco G. Ercolani

Marco Ercolani is a senior lecturer in econometrics at the University of Birmingham. He has published research in several academic journals in various areas of social statistics, including: ageing workforces, public transport, sickness absence, stock fund performance, student assessment, student loan repayments, tax evasion, and more.