Education, Employment and Earnings of Secondary School-Leavers ...
Education, Employment and Earnings of Secondary School-Leavers ... Education, Employment and Earnings of Secondary School-Leavers ...
where φ(·) denotes the probability density function for the standard normal.These selection terms are analogous to those computed using the more conventionaltwo-step Heckman procedure. The relevant selection terms are then added to the x ivector in the interval regression models. Instruments are required to identify theparameters of the selection effects in the earnings equation. These need to be chosensuch that they shift the probability of sectoral attachment but not the level of earningswithin the sector. A number of instruments have been used to assist in identificationand they appear adequate for the task at hand. 2424 This could be construed as a crude way to correct for selectivity bias but more adequate procedureshave not been developed for correcting interval regression models for selectivity bias when there aremultiple labour market outcomes to be treated.17
5. Empirical ResultsThe approach adopted is to first estimate a basic model of earnings determination thatincludes controls for gender, religion, the individual’s school-leaving cohort, joblocation and a variety of human capital controls. These latter controls includemeasures designed to capture both general human capital (e.g., educationalqualifications) and job specific human capital (e.g., job tenure). This basic model,motivated by a Mincerian approach, is then augmented in turn by controls for jobbranch sector, father’s educational background and finally a set of school controls(designed to proxy for school quality and potential labour market network effects).The primary motivation for adopting this approach is to determine how returns to thegeneral human capital measures alter with the inclusion of variables that proxy forparental background and schooling quality.We estimate separate earnings equations for the employed and self-employedsamples. The fact that the earnings measures are subject to different coding intervalsacross the two employment types and relate to different measurement periods vitiatesany sensible pooling exercise. However, as noted in the methodology section,correction terms for sample or self-selection into these two distinct employment typesare inserted into the earnings equations. The selection measures are computed fromestimates derived from a multinomial logit model estimated for a pooled samplecomprising employed, the self-employed, the unemployed, and those still in training.In order to conserve space estimates from the sectoral attachment model are neitherreported nor the subject of separate discussion here. 25 Our primary focus is anexamination of the earnings equation estimates for the employees and the selfemployed.25 A number of identifying variables were used in the multinomial logit sectoral attachment modelincluding a set of cohort interaction terms, the education level of the respondent’s spouse, time spentunemployed and self-employed since completing junior secondary school and the division achieved onthe Form IV secondary school examination. These variables generally had a significant impact on thecurrent labour force activity of respondents but not on their labour market earnings. The results of themultinomial logit model for sectoral attachment are available from the authors upon request.18
- Page 1 and 2: Education, Employment and Earnings
- Page 3 and 4: The structure of the paper can now
- Page 5 and 6: Despite large absolute increases in
- Page 7 and 8: nine studies reviewed for sub-Sahar
- Page 9 and 10: It should be noted, however, that e
- Page 11 and 12: history, current activity and incom
- Page 13 and 14: In order to understand how the mode
- Page 16 and 17: freedom for this test. In regard to
- Page 20 and 21: 5.1 Earnings Equation - The Employe
- Page 22 and 23: 4.8% per annum in the first three s
- Page 24 and 25: confirm to some extent that unobser
- Page 26 and 27: The ceteris paribus earnings for th
- Page 28 and 29: [Table 7]The rate of return for a u
- Page 30 and 31: 6. ConclusionsA key finding of our
- Page 32 and 33: that the rates of return to educati
- Page 34 and 35: Knight, J.B. and Sabot, R.H. (1981)
- Page 36 and 37: Table 1: Highest level of education
- Page 38 and 39: Table 4 (Cont’d)SelectionCorrecti
- Page 40 and 41: Table 5 (cont’d)Selection Correct
- Page 42 and 43: Table A1: Variable DescriptionVaria
5. Empirical ResultsThe approach adopted is to first estimate a basic model <strong>of</strong> earnings determination thatincludes controls for gender, religion, the individual’s school-leaving cohort, joblocation <strong>and</strong> a variety <strong>of</strong> human capital controls. These latter controls includemeasures designed to capture both general human capital (e.g., educationalqualifications) <strong>and</strong> job specific human capital (e.g., job tenure). This basic model,motivated by a Mincerian approach, is then augmented in turn by controls for jobbranch sector, father’s educational background <strong>and</strong> finally a set <strong>of</strong> school controls(designed to proxy for school quality <strong>and</strong> potential labour market network effects).The primary motivation for adopting this approach is to determine how returns to thegeneral human capital measures alter with the inclusion <strong>of</strong> variables that proxy forparental background <strong>and</strong> schooling quality.We estimate separate earnings equations for the employed <strong>and</strong> self-employedsamples. The fact that the earnings measures are subject to different coding intervalsacross the two employment types <strong>and</strong> relate to different measurement periods vitiatesany sensible pooling exercise. However, as noted in the methodology section,correction terms for sample or self-selection into these two distinct employment typesare inserted into the earnings equations. The selection measures are computed fromestimates derived from a multinomial logit model estimated for a pooled samplecomprising employed, the self-employed, the unemployed, <strong>and</strong> those still in training.In order to conserve space estimates from the sectoral attachment model are neitherreported nor the subject <strong>of</strong> separate discussion here. 25 Our primary focus is anexamination <strong>of</strong> the earnings equation estimates for the employees <strong>and</strong> the selfemployed.25 A number <strong>of</strong> identifying variables were used in the multinomial logit sectoral attachment modelincluding a set <strong>of</strong> cohort interaction terms, the education level <strong>of</strong> the respondent’s spouse, time spentunemployed <strong>and</strong> self-employed since completing junior secondary school <strong>and</strong> the division achieved onthe Form IV secondary school examination. These variables generally had a significant impact on thecurrent labour force activity <strong>of</strong> respondents but not on their labour market earnings. The results <strong>of</strong> themultinomial logit model for sectoral attachment are available from the authors upon request.18