Does It Pay To Work During Study? The Effect Of Student Employment On Later Hiring Chances Stijn Baert, Eddy Omey, Olivier Rotsaert and Dieter Verhaest
Does It Pay To Work During Study? The Effect Of Student Employment On Later Hiring Chances SE Stijn Baert, Eddy Omey, Olivier Rotsaert and Dieter Verhaest
5 questions to answer Why would it (not) pay? What s not cool about the literature? What s cool about our study? What did we do? What did we find?
Why would it (not) pay?
1 Why would it (not) pay? heory Baert, Omey, Rotsaert and Verhaest (2015) 5
1 Why would it (not) pay? Why it may not pay: Human capital (indirect): time-use trade-off between working and studying (Becker, 1964; 1965) heory Baert, Omey, Rotsaert and Verhaest (2015) 6
1 Why would it (not) pay? Why it may pay: Human capital (direct): relevant work experience, practical life skills, labour market abilities and business exposure (Painter II, 2010) Signalling: screening device of intrinsic work motivation (Spence, 1973) Why it may not pay: Human capital (indirect): time-use trade-off between working and studying (Becker, 1964; 1965) Network: market information and relationships (Granovetter, 1973) heory Baert, Omey, Rotsaert and Verhaest (2015) 7
What s not cool about the literature?
2 What s not cool about the literature? 2.1 Literature: SE & later labour market (LM) outcomes European studies: positive effects of SE during secondary and tertiary education on later labour market (LM) outcomes, albeit only for particular groups of student workers. Females in Alam et al. (2013). Individuals with a student job related to their field of study in Geel and Backes-Gellner (2002). Short-run effect in Häkkinen (2006). US studies: positive, zero and negative effects. Positive relationship in Light (1999; 2001) and Molitor and Leigh (2005). Zero effects in Ruhm (1997). Zero and negative effects in Hotz et al. (2002). Canadian studies: zero effect (Parent, 2006; Peng and Yang, 2008). Baert, Omey, Rotsaert and Verhaest (2015) 9
2 What s not cool about the literature? 2.2 Unclear whether SE affects later LM outcomes Doubtful whether results can be given causal interpretation: endogeneity problem. Naively estimated effects may reflect variation in factors such as ability and motivation. Unobservable to the researcher but may influence both the likelihood of SE experience and the probability of later LM success. Mentioned contributions use old school instrumental variables estimation techniques at best. Local unemployment rate as instrument. Problematic since students may already start job search during their last school year(s), and thus LM conditions during education may affect transition to work success at least indirectly via their drop-out decisions and (hence via) their human capital accumulation. One exception: Hotz et al. (2002): semi-structural approach. Identification is achieved from assumption that unobservable determinants of LM outcomes are, after controlling for SES, orthogonal to accumulated schooling at age 13. Baert, Omey, Rotsaert and Verhaest (2015) 10
2 What s not cool about the literature? 2.3 Unclear why and when SE affects later LM outcomes Ruhm (1997): In particular, it is important to better understand the mechanisms by which the [student] employment raises economic attainment, the role of job characteristics of the positions held by inschool youths, and the nature and sources of demographic group differences in the returns to student employment. Anno 2015: still no serious attempts to fill this gap. Why-question. No study has tested empirical power of aforementioned theoretical channels (human capital theory, signaling theory and social network theory). When-question. Little is known about effect heterogeneity by qualitative aspects of SE (e.g., relation to field of study) and of later LM outcomes (e.g., job match quality). Baert, Omey, Rotsaert and Verhaest (2015) 11
What s cool about our study?
3 What s cool about our study? Cool with respect to whether-question. Endogeneity is no issue given design of randomised experiment. Treatment of SE is randomly assigned. Cool with respect to why-question. Only the signalling channel may explain our results. Subjects are, by construction, equal in terms of human capital and network. Cool with respect to when-question. We estimate heterogeneous treatment effects. Qualitative aspects of SE. Qualitative aspects of later employment (offered contracts). Baert, Omey, Rotsaert and Verhaest (2015) 13
What did we do?
4 What did we do? 4.1 Correspondence experiment Fictitious job applications are sent to real job openings. These applications differ only by a ground for discrimination. By monitoring the subsequent callback, unequal treatment is identified. Golden standard to identify unequal treatment in the LM. Employer discrimination is disentangled from supply side determinants of LM outcomes. Selection on unobservable characteristics is not an issue. Recent applications: Kroft et al. (2013, Quarterly Journal of Economics) and Eriksson et al. (2014, American Economic Review). Baert, Omey, Rotsaert and Verhaest (2015) 15
4 What did we do? 4.2 Experimental identities SE, related to study field, during summer No SE Effect of signalling SE SE, not related to study field, during summer SE, not related to field of study, during academic year Baert, Omey, Rotsaert and Verhaest (2015) 16
4 What did we do? 4.3. Experimental data gathering CURRICULUM VITAE TYPE A CURRICULUM VITAE TYPE B CURRICULUM VITAE TYPE C CURRICULUM VITAE TYPE D NO SE (RANDOMISED ASSIGNMENT) NON-RELATED SE, SUMMER (RANDOMISED ASSIGNMENT) RELATED SE, SUMMER (RANDOMISED ASSIGNMENT) NON-RELATED SE, YEAR (RANDOMISED ASSIGNMENT) Call Back Call Back Call Back Call Back Vacancy N 252 x 4 applications to starter jobs in the following occupations: administrative clerk, operator, management assistant and lab analyst. Baert, Omey, Rotsaert and Verhaest (2015) 17
4 Results 4.4 Descriptive analysis: probability of interview 0.09 Positive 0.08 call-back rate sensu stricto 0.07 0.06 6.74% 6.21% 6.35% 5.95% 6.35% 0.05 0.04 0.03 0.02 0.01 0 95% Confidence intervals in dark blue. Baert, Omey, Rotsaert and Verhaest (2015) 18
4 Results 4.4 Descriptive analysis: probability of positive reaction 0.25 Positive call-back rate 0.2 sensu lato 18.65% 16.67% 16.67% 15.08% 18.25% 0.15 0.1 0.05 0 95% Confidence intervals in dark blue. Baert, Omey, Rotsaert and Verhaest (2015) 19
4 Results 4.5 Regression analysis Positive call-back sensu stricto: job interview invitation SE -0.008 (0.016) SE x Relevant SE 0.027 (0.025) SE x SE during year -0.003 (0.023) SE x High-educated -0.055* (0.033) SE x Industrial occupation -0.032 (0.035) SE x Temporary contract -0.022 (0.044) SE x Part-time contract -0.012 (0.023) SE x Male recruiter -0.017 (0.029) Vacancy fixed effects Except for SE, all variables are normalised by subtracting their mean among the population of candidates mentioning SE. Standard errors, corrected for clustering at the vacancy level, are in parentheses. *** (**) ((*)) indicates significance at the 1% (5%) ((10%)) level. Yes Baert, Omey, Rotsaert and Verhaest (2015) 20
What did we find?
Conclusion Why would it (not) pay? Human capital, signalling & social network What s not cool about the literature? No answer to whether, why & when What s cool about our study? Answer to whether, why & when What did we do? Experiment with random assignment of (ir)relevant SE (during year/summer) What did we find? No causal effect of revealed SE on hiring chances at all