The raising of the school leaving age: Returns in later life

a Accepted 12 June 2012 Available online 16 June 2012 JEL classification: I21 I28 J24 Keywords: Returns to education Raising of the school leaving age Earnings Lifecycle 1. Introduction In the past two decades the UK literature on the returns to education has repeatedly made use of compulsory schooling reforms in order to causally identify the returns to education (see inter alia Harmon and Walker, 1995, 1999; Chevalier and Walker, 2002). These studies have generally found that there is a positive and significant return to an additional year of schooling with early studies reporting instrumental variable estimates in the range of 15%–20%. More recent evidence on the matter from Devereux and Hart (2010) and Grenet (forthcoming) appears to suggest smaller causal estimates in the range of 3%–8%, per additional year of schooling. However, data constraints resulted in many of these studies combining multiple waves of data in order to achieve a sufficient sample size for identification purposes.2 One disadvantage of ∗ Corresponding author at: UCD Geary Institute, University College Dublin, Belfield, Dublin 4, Ireland. Tel.: +353 1 716 4621. E-mail addresses: [email protected] (F. Buscha), [email protected] (M. Dickson). 1 Tel.: +44 0 20 7911 5000x66596. 2 Harmon and Walker (1995) further increased sample size by considering two education reforms (in 1947 and 1972) when creating instruments for years of schooling. pooling multiple years of data is that the estimated return to education is an average return over a particular period of the life- cycle. Fig. 1 presents some of this information graphically:3 Fig. 1 illustrates how the average log gross hourly pay varies by age in the 2011 wave of the UKHLS. Notable from Fig. 1 is that: (a) studies which have examined the most recent (1972) school leaving age reform concentrate their analysis on the early and middle sections of the lifecycle; (b) all studies use pooled datasets which averages the estimated return over a wider age range of the lifecycle rather than one particular point; and (c) studies which examined the 1947 schooling reform focus their analysis on later parts of the lifecycle but with a great deal of data pooling. When pooling the data and estimating over a large age range it is necessary that the functional form chosen for age (or preferably experience) accurately fits the lifecycle wage profile or the estimated discontinuity in wages will be inaccurate. Moreover, even in a non-parametric estimation approach with the age distribution of pre- and post-policy samples very similar, the usage of pooled data means that the estimated return to education will be an average of the effects at different points in the lifecycle. Finally, as can be seen by Fig. 1, there is currently no consensus 3 Fig. 1 presents a cross-sectional age–earnings profile which should be viewed as a stylized example of a lifecycle profile. Economics Letters 1 Contents lists available a Economic journal homepage: www.e The raising of the school leaving age: Ret Franz Buscha a,1, Matt Dickson b,c,∗ aWestminster Business School, University of Westminster, 35 Marylebone Road, London, N b UCD Geary Institute, University College Dublin, Belfield, Dublin 4, Ireland c Institute for the Study of Labor (IZA), Germany a r t i c l e i n f o Article history: Received 21 March 2012 Received in revised form 7 June 2012 a b s t r a c t Using the recently released school leaving age in 1972 h 0165-1765/$ – see front matter© 2012 Elsevier B.V. All rights reserved. doi:10.1016/j.econlet.2012.06.018 17 (2012) 389–393 t SciVerse ScienceDirect s Letters lsevier.com/locate/ecolet urns in later life W1 5LS, United Kingdom UK Household Longitudinal Study we examine whether the raising of the d a permanent impact on earnings for individuals in their early 50s. © 2012 Elsevier B.V. All rights reserved. i examined earlier parts of the lifecyclewe canmake some inference on the divergence or convergence of the returns to education over the lifecycle. 2. Data and descriptives Wemake use of the recently released UK Household Longitudi- nal Study (UKHLS Wave 1, 2011) which has replaced the now dis- continued British Household Panel Survey. The survey contains ap- proximately 40,000 UK households and is conducted over a period of two years. A big advantage of the UKHLS is that there is detailed earnings information available at the individual level and a com- paratively large sample size. Given the distance, in terms of years, between collection of this data5 and themost recent school leaving age reform (1st Sep. 1972), this data offers the perfect opportunity to provide evidence on the returns to education for individuals in their early 50s. We select all Original Sample Members (OSMs) and drop individuals who were not born in England or Wales due to differences in the education systems in Scotland and Northern 4 The retirement pattern of individuals in the UKHLS indicates that only 2% of individuals in the age range 50 to 54 have retired, therefore our sample should not individuals gained as a result of RoSLA appears to have a positive impact on hourly labour earnings but amuch smaller effect on total hourly income. This conclusion is strengthened when looking at the distribu- tion of log hourly pay for those born in the two years before and the two years after the RoSLA in 1972. Fig. 3 illustrates the right- ward shift of the entire distribution as a result of RoSLA, indicat- ing that the effect is not limited to an increase at the mean, with a formal test of the equality of distributions re-enforcing the visual evidence. 3. Methodology and results We proceed by estimating the first stage effect of RoSLA on educational outcomes: years of schooling, holding of any qualifications and holding of O-level or higher qualifications. We 6 We include individuals born outside of the UK if they moved to the UK before starting secondary school. The selection rules reduce the sample size to 34,509 eligible observations. 7 This figure should only be seen as a general illustrative depiction of the data as different bandwidths/kernels can lead to different estimates. 8 The Raising of the School Leaving Age Order (Statutory Instrument no. 444) was passed in March 1972. It required individuals to remain in school until the end of the academic year in which they turn 16 — a one year increase from the previous 390 F. Buscha, M. Dickson / Econom Fig. 1. Age–earnings profile and on the exact return to either an additional year of schooling or the Raising of the School Leaving Age (RoSLA) reforms and a debate is currently taking place as to whether the ‘old’ estimates of approximately 15%–20% should be revised downwards to more like 3%–8% (Devereux and Hart, 2010; Grenet, forthcoming). In this study we contribute to the debate by examining the newly released UK Household Longitudinal Study (UKHLS). Moreover, due to the large sample size of the UKHLS we focus our attention on a very narrow age range and consider whether the additional human capital gained due to the 1972 education reform translates into persistent returns that remain visible in later life at age 52/53.4 An earnings effect at this later stage would indicate that the additional schooling induced by RoSLA resulted in a permanent increase in earnings with no catch up from non- affected individuals over the working lifecycle. Furthermore, by contrasting the magnitude of such an effect with studies which be affected by any potential differential retirement pattern related to RoSLA. 5 Interviews took place between the 8th January 2009 and the 10th March 2011. cs Letters 117 (2012) 389–393 estimated returns to schooling. Ireland.6 In addition, we select only employed individuals and drop those who are self-employed. We derive hourly pay as the usual function of gross pay permonth and hoursworked, including overtime and remove the top/bottom 1%. We also examine hourly total personal income which comprises labour earnings and in addition non-labour income, benefits and allowances. Finally, we make use of the special licence release of the data which contains information on the exact month of birth of individuals. Fig. 2 shows the evolution of log hourly pay and log hourly personal income over a wide age range,7 with a discontinuity estimated at the point of RoSLA.8 While there is a small and non significant increase in log hourly income for the post- reform cohorts, there is a larger impact on log hourly pay though still not statistically significant. The relative sizes of these observed discontinuities is in line with the predictions of human capital theory — the additional schooling and qualifications that requirement. The law change therefore affected all individuals turning 15 on or after 1st September 1972, i.e. those born from 1st September 1957 onwards. i Fig. 3. Log hourly pay distribution for those born near the education discontinuity. then estimate the reduced form impact of RoSLA on log hourly pay and log total hourly personal income before constructing the 2SLS estimates of the returns to years of schooling and qualifications. In each case we limit the sample to individuals born within +/−24 months9 of 1st September 1957 and estimate for men and women together and separately. We control for ethnicity, region and nationality and a dummy for female in the combined regressions. The first line of Table 1 contains the reduced form effects of RoSLA and confirms the visual evidence of Fig. 1: there is a 9 We present estimates for a bandwidth of 24 months as these provide the best trade-off between sample size and proximity to the discontinuity event. Higher order bandwidths produce similar estimated effects as does using a bandwidth of 12 months, though reducing the sample in this way increases the imprecision. Replacing the single RoSLA dummywith individual dummies for each cohort shows that the impact of RoSLA was approximately the same for the two post-RoSLA significant impact (p = 0.058) of RoSLA on the log of gross hourly pay, the point estimate is an increase of 5.5% for the affected cohorts. For men andwomen separately the estimated coefficients are very similar, 5.1% and 6.0% respectively, though for the sample of men the estimate is less precise. For the measure of total hourly income there are no significant reduced form impacts though the point estimates are all positive and of the order of around 2%. The first stage estimates show the substantial impact that RoSLA had on the education distribution, increasing average years of schooling by 0.23 years for men and 0.36 years for women — implying that just under a quarter of men and just over a third of the women in the affected cohorts were bound by the reform. For both men and women the proportion holding any qualifications increased by approximately 13 percentage points and the proportion gaining O-level qualifications increased by around 10 percentage points. The F-statistics on the RoSLA instrument reveal that it is a strong instrument for qualifications Fig. 2. Earnings over the life-cycle in the UKHLS: Wave 1 (2011). F. Buscha, M. Dickson / Econom cohorts considered and results in no substantive change to the estimates. These additional results are available from the authors on request. cs Letters 117 (2012) 389–393 391 for each sample, and for years of schooling for women and for the combined sample, though less strong for years of schooling formen i t a Notes: using cross sectional weights and restricting the sample to those in the cohorts +/−2 years either side of the reform. Robust standard errors in parentheses. * p < 0.10. ** p < 0.05. *** p < 0.01. and less strong for men and women separately when looking at O- level attainment. Turning to the 2SLS estimates in the lower panel of Table 1, the estimated return to an additional year of compulsory secondary schooling for the combined sample is 18.4% and significant (p = 0.051). For the men only sample the point estimate is higher at 22.6% though in this smaller sample it is imprecisely estimated. For women the return to an additional year of schooling is estimated as 16.9%, (p = 0.061). Similarly, for the combined sample there is a significant causal effect of qualifications on wages, increasing hourly wage by 40.3% (p = 0.051) and O-level qualifications or higher increasing wages by 59.7% (p = 0.055). In each case, the estimated returns to qualifications are higher for women than for men and are more precisely estimated. These are ‘local average treatment effect’ estimates for those who were compelled by the RoSLA to remain in secondary school for an additional year and suggest that there is a long-term causal effect of this education reform on economic outcomes, operating both through the additional time in school and the qualifications gained as a result of sitting ‘high stakes’ exams at age 16 — the new minimum leaving age. 4. Discussion and conclusions We have used the recently released UK Household Longitudinal Study to consider whether the 1972 raising of the school leaving age has had an impact on wages that is still visible when the affected individuals reach their early 50s. Exploiting the large sample size, we focus exclusively on the last two cohorts before the of approximately 5%. This is higher than the reduced form effect estimated in Grenet (forthcoming) using pooled LFS data, where the comparable estimate is 2%. Given that we are looking at the same cohorts of individuals, this disparity suggests that pooling over ages from the mid-thirties to mid-forties — as Grenet’s study does —maymask important differences in the reduced form effect at different points in the lifecycle. Comparing individuals from a narrow band around the 1972 RoSLA suggests that at around the age of 52, the RoSLA has a stronger impact on wages than at earlier ages.10 The implications of these findings are important for estimation of the returns to education and also for policy. Our findings suggest that pooling successive cross sections of survey data and averaging treatment effects across various points in the lifecycle will mask potentially important heterogeneity in the impacts. A corollary of this is that the RoSLA may have changed the gradient of the age–earnings profile aswell as the intercept— that is, age–earnings profiles are not parallel in education levels, therefore any estimate of the return to education should be defined with respect to a specific point in the lifecycle. For example, assuming there is a positive and unobserved gradient in Grenet’s (forthcoming) estimates and extrapolating this to our estimates suggest that the return to education slightly increases over the lifecycle. This could be explained by lower skill depreciation over time amongst individuals affected by RoSLA, though this question remains for future research. 10 Although we favour this interpretation some caution is advised. The size of our standard errors does not imply that our estimates are statistically significantly different fromGrenet’s (forthcoming).We acknowledge this, though the robustness 392 F. Buscha, M. Dickson / Econom Table 1 Estimates of the reduced form impact of the 1972 RoSLA and 2SLS es Log of hourly p All M Reduced form estimates Dummy: post-RoSLA cohort 0.055* 0 (0.029) ( 2SLS estimates First stage (a) Years of schooling 0.296*** 0 (0.066) ( F-statistic on instrument 20.44 4 (b) Highest qualification= any 0.135*** 0 (0.023) ( F-statistic on instrument 33.89 1 (c) Highest qualification= O-levels or higher 0.091*** 0 (0.027) ( F-statistic on instrument 11.26 5 Second stage (a) Years of schooling 0.184* 0 (0.094) ( (b) Highest qualification= any 0.403* 0 (0.207) ( (c) Highest qualification= O-levels or higher 0.597* 0 (0.311) ( Controls Dummy: female Yes N Ethnicity Yes Y Region Yes Y Nationality Yes Y N 1262 5 reform and the first two cohorts affected by the reform and find a reduced form ‘‘intention to treat’’ impact of RoSLA on later wages cs Letters 117 (2012) 389–393 imates of the returns to education. y Log of hourly income en Women All Men Women .051 0.060* 0.021 0.030 0.017 0.048) (0.035) (0.029) (0.047) (0.035) .225** 0.356*** 0.301*** 0.233** 0.356*** 0.107) (0.079) (0.066) (0.108) (0.079) .45 20.45 20.72 4.66 20.35 .144*** 0.127*** 0.135*** 0.142*** 0.128*** 0.036) (0.030) (0.023) (0.036) (0.030) 5.76 17.86 33.63 15.35 18.00 .103** 0.078** 0.091*** 0.104** 0.076** 0.043) (0.035) (0.027) (0.043) (0.035) .65 4.93 11.25 5.83 4.67 .226 0.169* 0.071 0.128 0.048 0.211) (0.090) (0.092) (0.192) (0.095) .353 0.475* 0.159 0.210 0.135 0.319) (0.265) (0.208) (0.317) (0.269) .496 0.769* 0.235 0.287 0.225 0.446) (0.464) (0.301) (0.422) (0.446) o No Yes No No es Yes Yes Yes Yes es Yes Yes Yes Yes es Yes Yes Yes Yes 07 755 1261 507 754 of our point estimates to using different bandwidths does suggest that focusing on ages close to 52 the returns to RoSLA are higher than those at earlier ages. i F. Buscha, M. Dickson / Econom From a policy perspective, the evidence presented is positive, suggesting that increasing the education of those at the lower end of the distribution led to a substantial and sustained increase in wages that remains visible into the later stages of the lifecycle. Acknowledgements Matt Dickson is funded by EC Framework FP7 programme, grant no. PIEF-GA-2010-275964. The UKHLSwave 1 special licence access data (SN: 6931)was kindly provided by theUKData Archive, University of Essex, Colchester, UK. cs Letters 117 (2012) 389–393 393 References Chevalier, A., Walker, I., 2002. Further estimates of the returns to education in the UK. In: Harmon, C., Walker, I., Westergard-Nielsen, W. (Eds.), The Returns to Education Across Europe. Edward Elgar. Devereux, P., Hart, R., 2010. Forced to be rich? Returns to compulsory schooling in Britain. The Economic Journal 120, 1345–1364. Grenet, J., 2012. Is it enough to increase compulsory education to raise earnings? Evidence from French and British compulsory schooling laws, Scandinavian Journal of Economics (forthcoming). Harmon, C., Walker, I., 1995. Estimates of the economic return to schooling for the United Kingdom. American Economic Review 85 (5), 1278–1286. Harmon, C., Walker, I., 1999. The marginal and average returns to schooling in the UK. European Economic Review 43, 879–887. The raising of the school leaving age: Returns in later life Introduction Data and descriptives Methodology and results Discussion and conclusions Acknowledgements References