Selection into Vocational Training Programs: Field Experiments from India

It is usually difficult to estimate the causal impact of training programs on welfare outcomes as participation in training programs is often voluntary making it endogenous. While the large literature acknowledges and addresses this selection issue, there is limited information on the kind of selection that takes place and distinguishes program applicants from the non-applicants. The objective of this paper is to combine survey data with experimental data to identify observables and unobservables characteristics that differentiate applicants from non-applicants. This paper adds significant value to the literature as the unique experimental data collected here allows us to give some structure to the error terms about which we know nothing in practice. 

The data collected in this paper corresponds to applicants and non-applicants of a vocational education program administered in selected slum communities located in the South Shahdara region of New Delhi, India. The vocational education program is administered jointly by two NGOs – PDEI (Pratham Delhi Education Initiative) and SATYA (Social Awakening Through Youth Action). A total of 244 applications were received from these slums. A detailed demographic and socioeconomic survey was targeted to all 244 program applicants and 79 non-applicants who were randomly selected from the list of non-applicants to participate in the quasi-laboratory experiments conducted in the field. Of the 244 applicants, we also randomly selected 163 of these program applicants to participate in the quasi-laboratory experiments conducted in June 2010. The quasi-laboratory games were designed to evaluate subjects’ risk preference and intrinsic competitiveness.

We combined the experimental data with the survey data to identify both observable and unobservable differences between program applicants and non-applicants. We find that probability of applying for the training program varies along the following observable characteristics - experience, age and enrollment. Individuals who have no experience in tailoring, younger in age and currently not enrolled in school all have higher probability of applying for the program. All these observables differences between applicants and non-applicants are statistically significant at 5% significance level. We find that competitiveness has no statistically significant impact on explaining application probabilities. As long as applicants and non-applicants differed only along observable characteristics it would sufficient to examine the impact of training programs comparing applicants with non-applicants controlling for all these observable differences. However, we do find that people with higher appetite for risk have higher probability of applying for the program. We find that applicants have much more risky preferences compared to non-applicants. In uncertain environments like the one our population belongs too, venturing into new training programs would be associated with having a higher appetite for risk. The presence of statistically significant unobservables such as differential appetite for risk between applicants and non-applicants calls for the continued use of randomized evaluation strategies, where a certain proportion of the program applicants should be assigned to the treatment group and the remainder of the program applicants to the control group.