Econometric analysis Directory UMM :Data Elmu:jurnal:E:Economics of Education Review:Vol20.Issue2.2001:

173 P. Christie, M. Shannon Economics of Education Review 20 2001 165–180 gap. Eliminating gender differences in the field of study of post-secondary graduates would close the 1990 gap by 455 and raise the earnings ratio from 0.677 to 0.689. This is a bigger effect than that produced by eliminating differences in educational attainment. 14 Projecting future earnings on the assumption that the distribution of fields for 25–34 year-old post-secondary graduates will spread to the older groups, suggests that the gap will fall slightly to 12 428 and the earnings ratio would rise to 0.688 so that future changes in field will have less of an effect than changes in level of attainment. 15 Projecting back- wards by applying the pattern of fields for 55–64 year olds to all those 25 and over, gives a gap of 12 954 and an earnings ratio of 0.663 once again less than the effect of the backward projection for attainment levels. Inter- estingly the projections suggest that, despite having a larger impact than attainment differences on the 1990 gender earnings gap, changes in field of study have not been and will not be as important as changes in attain- ment in producing changes in the earnings gap. The projections suggest that past and future changes in educational attainment and field of study have indi- vidually increased the earnings ratio by 0.046 from 0.650 to 0.696 and 0.025 respectively. To put the size of these changes into perspective, consider that the earn- ings ratio rose by 0.047 over the 1970s and by 0.040 over the 1980s. 16 The implied effects of gender differences in education and changes in educational outcomes are not trivial. The data on earnings and the projections presented in this section of the paper are very simple. They do not allow, for example, for the influence of characteristics other than education on earnings and may incorrectly measure the earnings effects of education. This is addressed in the following sections of the paper by esti- mation of earnings regressions.

4. Econometric analysis

To isolate the effect of education on the gap in wage and salary income, we estimated log-annual earnings equations and adopted the decomposition techniques associated with Oaxaca 1973 and Cotton 1988. The logarithm of annual wage and salary income was speci- 14 The importance of differences in field of study relative to attainment is a bit surprising, given that the post-secondary graduates represent less than half the sample, however segre- gation index measures do suggest that gender differences in field of study are greater than those in educational attainment. 15 This assumes that the share of post-secondary graduates age 25 and over will rise to that of 25–34 year olds. 16 These figures are taken from Statistics Canada 1996 and are based upon Survey of Consumer Finances data. fied to be a linear function of educational attainment, field of study, age and age-squared, personal character- istics marital and family status, language, visible min- ority, and citizenship, region, urban area, hours of work and industry and occupational dummies. The specifi- cation is similar to earlier Canadian studies using Census data with the exception of the detail of the educational variables. A version of the equation without industry and occupation dummies was also estimated to reflect the concern that the effect of education on wages may, in part, be realised through industry and occupation out- comes. These earnings equations were estimated separ- ately for men and women on the sample of 25–64 year old full-timefull-year workers for both census years. Table 7 reports the results of this exercise for 1990 earn- ings when industry and occupation dummies are included as regressors. The regression results are quite standard. Wage income rises with age but the effect of an extra year diminishes with age. Wages are lowest in Atlantic Can- ada, Quebec and the Prairies, non-metropolitan areas for those with 30–34 hours per week, visible minorities and non-citizens. Managerial professional and teaching occupations have the highest premiums while the lowest paying occupations are farming, clerical men, service women and transportation women. The pattern for industry coefficients is similar between sexes. Primary industries other than agriculture, communications and utilities and public administration are the best paid. The worst paid are agriculture, services and retail trade. The default group for educational attainment in the earnings equations is less than grade 5 — the lowest level. For men, coefficients are positive and statistically significant on all education variables in both years. The highest return was associated with professional degrees medicine, dentistry, etc., followed by Doctorates, Mas- ter’s and Bachelor’s degrees. Somewhat surprisingly high-school graduates did as well or better than many of those categories with post-secondary education but no university degree. For women, all but one educational attainment coefficient was positive and significant with the largest values for Doctorates, Master’s degrees, then professional degrees. Women with high school degrees typically did as well as those with non-university post- secondary education. Table 8 reports the estimated coefficients on the edu- cation level and field of study dummies for all specifi- cations estimated including those from Table 7. The coefficients on education level for 1990 are much larger for men than women up to Bachelor’s degree when industry and occupational dummies are present as regressors columns 1 and 2. In the 1985 version columns 5 and 6 the size differences between male and female coefficients tend to be smaller than in 1990 for those with less than university education. Furthermore 174 P. Christie, M. Shannon Economics of Education Review 20 2001 165–180 Table 7 Earnings equations estimates for 1990: full-time, full-year, age 25–64 with Industry and Occupation Dummies a Women Men Coefficient t-statistic Coefficient t-statistic Constant 8.653 158.5 8.572 186.9 Urban 0.127 22.7 0.095 20.9 Region: Atlantic 20.123 211.9 20.121 214.4 Quebec 20.087 28.8 20.098 211.8 Prairie 20.145 214.9 20.178 221.8 Alberta 20.081 29.4 20.064 29.0 B.C. 20.017 22.1 0.006 0.9 Occupation: Managerialadministrative 0.233 31.1 0.298 33.4 Natural science 0.208 12.3 0.195 17.8 Social science 0.085 5.7 0.175 9.6 Teaching 0.198 12.7 0.251 14.6 Medicinehealth 0.170 13.1 0.211 9.6 Artsrecreation 0.046 2.0 0.092 4.8 Sales 0.041 3.9 0.108 10.4 Service 20.156 214.7 0.061 5.5 Farming 20.364 210.3 20.150 26.7 Primary occupation 0.145 1.1 0.147 6.5 Processing 20.062 22.7 0.114 8.8 Machining 20.128 28.3 0.058 5.9 Construction 0.105 2.2 0.093 8.1 Transport 20.158 24.3 0.010 0.9 Other 20.102 25.9 0.035 3.2 Industry: Agriculture 20.640 222.5 20.652 233.5 Other primary 0.051 1.8 0.170 11.7 Manufacturing 20.127 211.0 20.037 24.4 Construction 20.188 28.7 20.136 212.1 Transport 20.084 24.3 20.009 20.8 Communications 0.045 3.1 0.061 5.7 Wholesale trade 20.166 211.2 20.092 28.9 Retail trade 20.376 233.1 20.259 226.9 Finance, insurance, real estate 20.141 213.2 20.082 27.4 Business services 20.152 212.1 20.142 213.3 Education services 20.080 25.9 20.137 210.4 Health services 20.197 218.2 20.265 218.8 Accommodation and food 20.441 229.6 20.469 233.2 Other services 20.348 226.1 20.309 225.5 Hours: Hours 35–39 0.222 24.3 0.149 11.1 Hours 40–44 0.214 23.7 0.154 12.0 Hours 45 + 0.236 22.7 0.205 15.8 Education: Grade 5–8 20.011 20.4 0.134 5.8 Grade 9–13 0.124 4.3 0.263 11.9 High-school graduate: no further education 0.205 7.0 0.350 15.6 Trades certificate or diploma, no post-sec. 0.172 5.0 0.266 9.9 Some non-university post-secondary, no certificatediploma, 0.188 5.7 0.302 11.6 not a high school graduate Some non-university post-secondary, no certificatediploma, 0.232 7.6 0.364 15.1 is a high school graduate continued on next page 175 P. Christie, M. Shannon Economics of Education Review 20 2001 165–180 Table 7 continued Women Men Coefficient t-statistic Coefficient t-statistic Non-university post-secondary, trades certificates 0.146 4.3 0.284 10.7 Non-university post-secondary, with certificatediploma 0.217 6.7 0.327 12.5 Some university, no degree, not a high school graduate 0.200 2.6 0.346 5.5 Some University, no degree, high school graduate 0.290 9.3 0.420 17.5 University, with trades certificate 0.201 4.7 0.325 10.3 University, with non-university certificatediploma 0.281 8.2 0.325 11.7 University graduate: below Bachelor’s 0.316 9.3 0.349 12.4 University graduate: Bachelor’s degree 0.432 13.4 0.448 17.5 University graduate: above Bachelor’s 0.506 14.1 0.467 16.3 University graduate: professional degree 0.580 9.0 0.706 16.4 University graduate: Master’s degree 0.573 16.5 0.550 20.5 University graduate: Doctorate 0.629 11.2 0.667 20.7 Field of study: Educationrecreation 0.043 2.6 0.065 3.8 Fine and applied arts 20.044 22.1 20.004 20.2 Social sciences 0.015 0.9 0.128 8.2 Business 0.098 6.1 0.128 8.9 Secretarial 0.059 3.5 0.031 1.0 Agriculturebiology 0.021 1.0 0.018 1.0 Engineeringapplied science 0.063 1.5 0.210 12.8 Engineering technologytrades 0.103 4.5 0.130 8.6 Nursing 0.134 7.3 0.066 1.7 Health non-nursing 0.149 7.1 0.183 7.4 Mathphysics 0.100 3.9 0.140 8.0 All other 0.046 0.6 20.100 21.7 Personal: Single 20.031 24.3 20.154 222.0 Divorced 0.019 2.5 20.039 24.3 Widow 0.021 1.3 20.045 21.5 Children 20.062 211.2 0.057 11.7 Visible minority 20.109 212.3 20.209 226.9 French 20.026 22.1 20.048 24.6 Bilingual 0.019 2.3 0.003 0.4 Other language 20.060 21.8 20.065 22.1 Citizen 0.146 12.8 0.111 11.4 Age 0.041 18.7 0.050 28.0 Age squared 20.0004 216.6 20.0005 223.9 Adjusted R-squared 0.241 0.202 Observations 69945 108323 a Defaults: Occupation: clerical; Industry: public administration; Education: less than grade 5; Field of study: humanities; Region; Ontario. there are more educational levels at which the coef- ficients for women are larger than for men. 17 For field of study of post-secondary graduates humani- ties was the default. Women with health fields did best, ceteris paribus, followed by those in technologytrades 17 Further comparison of the 1985 and 1990 results suggests a fall in returns to education — which was especially large for women. A closer look reveals that the coefficients on the educational attainment dummies have not changed much rela- tive to high school. This suggests that the apparent lower returns and mathematicsphysical sciences. Fine and applied arts and the humanities were the worst paid. The worst paid fields were the same for men as for women. For men, engineering and health paid most. Dropping occupation and industry dummies from the specification columns 3 and 4, 7 and 8 of Table 8 increases the size of the coefficients on education for to education in 1990 reflects improvement in performance of the lowest grouping. 176 P. Christie, M. Shannon Economics of Education Review 20 2001 165–180 Table 8 Education Coefficient Estimates, 1990 and 1985 1990 1985 With industry and Without industry and With industry and Without industry and occupation occupation occupation occupation Women Men Women Men Women Men Women Men 1 2 3 4 5 6 7 8 Educational attainment a GR58 20.011 c 0.134 20.010 c 0.146 0.141 0.189 0.130 0.183 GR913 0.124 0.263 0.201 0.304 0.262 0.327 0.322 0.350 HS 0.205 0.350 0.348 0.413 0.347 0.414 0.471 0.458 TRD1 0.172 0.266 0.327 0.267 0.314 0.270 0.454 0.221 SPNHS 0.188 0.302 0.327 0.358 0.290 0.356 0.400 0.388 SPHS 0.232 0.364 0.401 0.444 0.334 0.418 0.485 0.481 TRD2 0.146 0.284 0.285 0.286 0.307 0.306 0.444 0.244 PSNCD 0.217 0.327 0.409 0.378 0.371 0.363 0.553 0.363 SUNHS 0.200 0.346 0.439 0.444 0.397 0.453 0.556 0.520 SUHS 0.290 0.420 0.504 0.527 0.442 0.469 0.629 0.555 TRD3 0.201 0.325 0.384 0.372 0.395 0.367 0.580 0.353 UNUC 0.281 0.325 0.499 0.394 0.421 0.371 0.636 0.395 UNLTB 0.316 0.349 0.554 0.414 0.464 0.379 0.702 0.405 BACH 0.432 0.448 0.723 0.558 0.574 0.497 0.871 0.557 BACH + 0.506 0.467 0.833 0.585 0.581 0.520 0.921 0.576 UPROF 0.580 0.706 0.830 0.729 0.637 0.803 0.893 0.784 MASTER 0.573 0.550 0.890 0.665 0.722 0.581 1.036 0.635 DOCT 0.629 0.667 1.017 0.788 0.788 0.658 1.154 0.717 Field b EDREC 0.043 0.065 0.111 0.105 20.003 c 0.063 0.068 0.105 ARTS 20.044 20.004 c 20.144 0.008 c 20.029 c 0.053 20.152 0.094 SOCSC 0.015 c 0.128 0.018 c 0.180 0.002 c 0.128 20.007 c 0.186 BUS 0.098 0.128 0.128 0.215 0.046 0.127 0.041 0.242 SECR 0.059 0.031 c 0.069 0.109 0.011 c 0.052 c 20.006 c 0.149 AGRBIO 0.021 c 0.018 c 20.067 20.032 c 0.018 c 0.037 c 20.079 0.049 ENGIN 0.063 c 0.210 0.130 0.309 0.040 c 0.201 0.053 c 0.314 ENGTR 0.103 0.130 0.155 0.201 0.100 0.139 0.119 0.229 NURS 0.134 0.066 c 0.182 0.043 c 0.154 0.123 0.152 0.060 c HEALTH 0.149 0.183 0.170 0.192 0.202 0.252 0.168 0.255 MATHPH 0.100 0.140 0.148 0.212 0.116 0.134 0.130 0.225 OTHER 0.046 c 20.100 c 0.029 c 20.066 c 20.030 c 0.042 c 20.089 c 0.068 c a The default category is less than grade 5. Acronyms are defined in Table 1. b The default category is humanities. Acronyms are defined in Table 3. c The coefficient is not statistically significant at 1. both sexes but much more for women in both years. In fact, education coefficients are larger for women than men at most levels when the industry and occupation dummies are absent. This change suggests that expansion of the range of occupation and industry of work may be a particularly important source of extra returns to edu- cation for women. 18 Removing industry and occupational 18 Apparently the payoffs to men of the same education level are more equal across industryoccupations than for women. dummies also tends to increase the absolute value of the coefficients on the field of study variables. Does the extra information on education available in the Census add anything to our understanding of wage determination? An examination of the coefficient esti- mates in Table 8 suggests that aggregation hides substan- tial variation in returns. For example, most Canadian microdata surveys include a single university degree variable. Our estimates suggest a range of returns between degree types of from 0.43 to 0.63 for women in 1990 in the specification with industry and occupation 177 P. Christie, M. Shannon Economics of Education Review 20 2001 165–180 dummies. Similarly coefficients on Grades 5–8 and Grades 9–13 have quite different size coefficients across specifications. In many Canadian surveys these categor- ies would be aggregated into a single “some secondary education” grouping. The restrictions required to sim- plify our 19 category educational variable to a seven cat- egory variable, similar to that found in the Labour Mar- ket Activity Survey LMAS, were tested and rejected for both sexes and each year. 19 Tests of the joint signifi- cance of the field of study variables also suggest that they contribute to explaining variation in earnings. 20 Apparently, the greater detail adds significantly to the ability of the regressions to explain variation in earnings.

5. Decomposition of the wage gap