Human Capital Investment and the Gender Division of Labor in a Brawn-Based Economy Part 2

Table 1 2SLS Estimates of the (Cobb-Douglas) Body-Mass Production Function, by Survey Population Dependent variable: log weight/height 1981-2b 2001-2 Log individual total calorie consumptiona .136 (3.37) .241 (3.76) Very active occupationa -.0119 (0.23) -.0445 (3.20) Exceptionally active occupationa -.0817 (1.26) -.125 (5.65) Pregnanta .326 (1.34) .0273 (1.33) Lactatinga .513 (4.65) .0339 (1.39) Log age .0987 (1.90) .00804 (9.02) Log age squared .0174 (2.37) -.000092 (8.86) Male -.0578 (1.81) -.00947 (2.89) Male*log age .0687 (4.04) .00116 (3.72) Water drawn from tube well -.0406 (2.10) .00551 (0.88) Water drawn from well -.0693 (3.15) .00118 (0.18) Water drawn from pond -.0649 (2.55) .0216 (2.36) N 1,737 5,750 Input/Survey population H0 : calories, age, age squared, male, exceptionally active, very active, male*age = across populations; ÷2(7) (p) 9.06 (.17) Sources: NSRB 1981-2 and 2001-2. aEndogenous variable: instruments include household head’s age and schooling level, land holdings, and price of all foods consumed interacted with individual age and sex variables, land and head’s schooling and age. b Reproduced from Pitt et al. (1990). Asymptotic tratios in parentheses. Table 2 The Body-mass Endowment and Grip Strength (2007-8), by Gender: Respondents Aged 20-49 in 2002 Dependent variable: Kilograms of pressure Group Men Women Estimation procedure GLS GLLAMM GLS GLLAMM Endowment (ã) 7.41 (5.89) 8.17 (6.03) 1.58 (1.69) 1.93 (1.67) ñ - .907 - .796 N 946 946 1,087 1,087 Source: NSRB 2001-2/2007-8 panel. Absolute values of asymptotic t-ratios corrected for clustering within households in parentheses. All specifications include village fixed effects, age, age squared, landholdings and an indicator of landlessness. Table 6 Are Young Boys Sicker and thus Less Smart than Girls? Children’s Morbidity Symptoms and Raven’s CPM Scores, by Age and Gender Category Diarrhea, Ages 0-5 Diarrhea, Ages 5-10 Fever, Ages 0-5 Fever, Ages 5-10 Raven’s CPM Score, Ages 8-10 Gender Male Female Male Female Male Female Male Female Male Female Incidence (%)/score 6.13 5.24 3.83 3.48 5.97 2.86 4.26 4.74 3.25 3.05 N 636 630 705 718 636 630 705 718 348 325 ÷2 [p] .472 [.49] .122 [.73] 7.28 [.007] .191[.66] - t [p] - - - - 1.50 [.13] Source: NSRB 2001-2. Table 3 Estimated Marginal Effects of the Body-Mass Endowment on the Probability of Attending School, by Gender and Estimation Method: Children Ages 10-15 in 2001-2 Estimation method Multinomial Logita ML Logit-GLLAMMb Group Boys Girls Boys Girls Endowment -.248 (2.37) .0600 (0.68) -.436 (3.60) .0983 (1.04) Household land owned .00670 (3.23) .00135 (1.14) .00674 (3.42) .00135 (1.11) No land owned -.0417 (1.08) -.0463 (1.64) -.0420 (1.24) -.0465 (1.70) Household average endowment of other family members -.0955 (0.62) .0667 (0.87) -.0927 (1.10) .0784 (1.09) Age -.0332 (3.43) -.0243 (2.54) -.0325 (3.23) -.0236 (2.53) 410 353 410 353 N Source: NSRB 2001-2 and ML Estimates reported in Table 4. aAsymptotic t-ratios corrected for clustering at the household level in columns. bBootstrapped t-ratios in parentheses in columns. Table 4 The Body-mass Endowment in 1981-22 and Completed Schooling in 2001-2, by Gender: Respondents Aged 0-15 in 1982 Dependent variable: Completed education (years) in 2001-2 Gender Male Female Estimation procedure GLS GLLAMM GLS GLLAMM Endowment (1981-2) -1.82 (2.05) -2.22 (1.94) 1.00 (0.86) .923 (0.67) Age -.164 (0.76) -.161 (0.77) -.110 (0.49) -.123 (0.57) Age squared .00501 (0.36) .00483 (0.36) -.00908 (0.60) -.00787 (0.54) Household owned land (1981-2) .00711 (5.49) .00711 (5.66) .00467 (3.05) .00468 (3.17) No land owned (1981-2) -1.02 (1.48) -1.03 (1.54) -1.05 (1.55) -1.05 (1.68) Average household endowment (1981-2) .261 (0.12) .270 (0.17) .116 (0.06) .158 (0.08) ñ - .838 - .821 N 311 311 273 273 Source: NSRB 1981-2/2001-2 panel. Absolute values of asymptotic t-ratios corrected for clustering within households in parentheses. All specifications include village fixed effects. Table 5 The Body-mass Endowment and Occupation Choice, by Gender: Respondents Aged 20-49 in 2001-2 Dependent variable: Occupational Energy Expenditure Group Men Women Estimation procedure GLS GLLAMM GLS GLLAMM Endowment 81.7 (14.9) 93.6 (11.6) -5.78 (3.62) -6.78 (2.58) Age .969 (0.89) .957 (0.90) 1.05 (3.23) 1.06 (2.86) Age squared -.00153 (0.10) -.000825 (0.05) -.0132 (2.80) -.0133 (2.63) Household owned land -.0195 (3.13) -.0212 (3.42) -.000269 (0.15) -.000262 (0.18) No land owned 6.62 (2.87) 6.22 (2.67) 2.37 (3.64) 2.38 (3.11) Average household endowment of others (1981-2) -27.0 (4.14) -29.0 (4.53) -2.02 (1.10) -2.02 (.91) ñ - .901 - .828 N 1,236 1,236 1,338 1,338 Source: NSRB 2001-2. Absolute values of asymptotic t-ratios corrected for clustering within households in parentheses. All specifications include village fixed effects. Table 7 The Body-mass Endowment in 1981-2 and Log Wages in 2001-2: Males Aged 0-17 in 1981-2 Dependent variable: Log daily wage in 2002 (tk.) Estimation procedure GLS GLLAMM GLS GLS GLLAMM GLLAMM-IV - - .0261 (2.27) .0277 (2.41) .0311 (3.01) .0432 (2.18) Endowment (1982) .252 (1.21) .327 (2.16) - .303 (1.46) .370 (2.49) .387 (2.39) Age .100 (2.30) .106 (2.73) .0750 (1.90) .0994 (2.33) .103 (2.66) .100 (2.57) Age squared -.00517 (2.08) -.00536 (2.34) -.00356 (1.61) -.00510 (2.08) -.00527 (2.30) -.00518 (2.26) Household owned land (1982) .000404 (2.03) .000412 (2.51) - - - - No land owned (1982) .0528 (0.44) .0607 (0.68) - - - - Average household endowment of others (1982) -.115 (0.27) -.147 (0.45) - - - - ñ - .838 - - .838 .838 Sargan overid. ÷2(2) [p] - - - - - 0.00129 [.99] 225 225 225 225 225 225 Schooling (2002) N Source: NSRB 1981-2/2001-2 panel. Absolute values of asymptotic t-ratios corrected for clustering within households in parentheses in columns 2-6. Bootstrapped t-values in column 7. All specifications include village fixed effects. Instruments for the IV estimates include household owned land, no owned land and the average household endowment of other household members. Table 8 Occupation-Specific Wage Function Estimates, by Gender: Adults Aged 20-49 in 2001-2 Gender Male Female Estimation procedure GLSa GLLAMa GLLAM-IVb GLSa GLLAMa GLLAM-IVb Schooling .0409 (11.6) .0417 (10.1) .334 (2.75) .0487 (2.38) .0467 (2.41) 1.14 (2.05) Schooling x occupation energy expenditure - - -.00256 (2.87) - - -.007 (2.03) Endowment - .0765 (0.84) -1.46 (2.34) - .0895 (0.22) -4.53 (1.28) Endowment x occupation energy expenditure - - .0115 (2.85) - - .0254 (1.52) Age x occupation energy expenditure - - .000401 (1.16) - - .00204 (1.09) ä - - -3.36 (1.67) - - -4.37 (0.44) ë - - - 11.7 (1.46) 12.7 (1.49) 10.3 (0.78) ñ - .889 .867 - .904 .900 Sargan overidentification ÷2 [p] - - 11.09 [.436] - - 0.0033 [.999] 1,094 1,094 1,094 79 79 79 N Source: NSRB 2001-2. All specifications include age age and age squared. aAbsolute values of asymptotic t-ratios corrected for clustering within households in parentheses in column. b Bootstrapped t-ratios in parentheses in column. Figure 1 Appendix Figure 2. School Attendance, by Age, Gender and Survey Year 1 Boys Girls 0.9 0.8 0.7 0.6 0.5 2001-2 0.4 0.3 0.2 1981-2 0.1 0 5 6 7 8 9 10 11 12 13 14 15 16 Appendix Figure 3 Appendix Figure 4 Appendix Figure A. Distribution of Raven’s CPM Test Results by Gender: Number of Correct Answers, Respondents Aged 20-49 0.3 Men (N=1,161) 0.25 Women (N=1,209) 0.2 0.15 0.1 0.05 0 0 1 2 3 4 5 6 7 8 9 Appendix Figure B. National Per-Capita Availability of Food Grains in Bangladesh (Grams per Day), 1980-81 - 1999-00 (Source: Table 9 in Begum and D’Haese, 2010) 600 550 500 450 400 350 300 250 200 150 100 198081 198182 198283 198384 198485 198586 198687 198788 198889 1989- 199090 91 199192 199293 199394 199495 199596 199697 199798 199899 199900 Appendix A Proof of Proposition 1: Assume that UC + θCUCC # 0, then (A1) dS/dθ = -(α(i)CH12UwW/H)Φ22 + (UC + θCUCC)Φ21 > 0 (A2) dC/dθ = (α(i)CH12UwW/H)Φ21 - (UC + θCUCC)Φ11 < 0 (A3) di/dθ = (α(i)CH12UwW/H)Φ32 - (UC + θCUCC)Φ31 > 0 where Φ22 = -p2(πii/α(i)π(i) - πi/(αii)2π(i) - 1/αiπi)/Δ<0 Φ21 = -p(ω + ρ)(πii/α(i)π(i) - πi/(αii)2π(i) - 1/αiπi)/Δ<0 Φ11 = -(ω + ρ)2(πii/α(i)π(i) -πi/(αii)2π(i) - 1/αiπi)/Δ<0 Φ32 = -p2(H1/H)/Δ>0 Φ31 = -p(ω + ρ)(H1/H)/Δ>0 by second-order conditions Proof of Proposition 2: Assume that effective calories do not change, so that UC + θCUCC =0, then: (A4) dS/dθ = -(α(i)CH12UwW/H + γCH1W2UWW(1 - α(i))α(i)/HB - λγωBB1)Φ22 - (γCB1M1/B)Φ32 (A5) di/dθ = -[α(i)CH12UwW/H + γCH1W2UWW(1 - α(i))α(i)/HB - λγωBB1]Φ32 + (γCB1M1/B)Φ33 where Φ22 = -[p - γSωBθM1BM]2(πii/α(i)π(i) - πi/(αii)2π(i) - 1/αiπi)/Δ<0 Φ32 = [p - γSωBθM1BM][γBMαiUWM1θW(ω + ρ)/B + αiUWH1W(p - γSωBθM1BM)/H]/Δ>0 Φ33<0, by second-order conditions Proof of Proposition 3: (A6) dS/dm = (-γB1α(i)H1[(1 - α(i))W/HB + UWWW/H] + λγωBB1)Φ22 - (γB1/B)Φ32 (A7) di/dm = (-γB1α(i)H1[(1 - α(i))W/HB + UWWW/H] + λγωBB1)Φ32 + (γB1/B)Φ33 Table A Characteristics and Performance Scores of Respondents Aged 20-49 in 2001-2 and 2007-8, by Gender Group Statistic Men Mean (SD) Women N Mean (SD) N Characteristics in 2001-2 Schooling (years) Age BMI Height (cm) Occupation energy expenditure Household owned land (hectares x 10-2) No land owned (%) Daily wage (tk.) 3.99 (4.48) 1,264 2.40 (3.57) 1,355 33.7 (8.10) 1,264 32.9 (8.23) 1,355 19.1 (2.66) 1,264 18.8 (2.79) 1,355 162.1 (7.89) 1,264 150.3 (6.97) 1,355 159.2 (38.2) 1,251 143.6 (7.92) 1,346 1,264 81.1 (166.7) 1,355 48.3 (49.9) 1,264 48.4 (49.9) 1,355 94.5 (90.0) 1,094 40.1 (32.5) 79 91.5 (193.0) Assessments in 2007-8 Grip strength (kilograms of pressure) Number of correct answers, abridged Raven’s Matrices Source: NSRB 2001-2 and 2007-8 37.3 (8.36) 946 24.3 (5.66) 1,087 3.66 (1.96) 1,038 2.94 (1.73) 1,200 Table B Multinomial Logit and Logit-GLLAMM Estimates of the Determinants of Children’s Activities: Children Ages 10-15 in 2001-2 (Left out activity = Schooling) Estimation method Activity Multinomial Logita Work ML Logit-GLLAMMb Home Time Work Home Time Boys Endowment 2.33 (2.42) .284 (0.15) 4.09 (3.41) .426 (0.13) Household land owned -.0599 (2.93) -.0515 (1.01) -.0603 (2.94) -.0515 (0.55) No land owned .356 (1.09) .348 (0.45) .359 (1.19) .344 (0.04) Household average endowment .817 (0.57) 1.38 (0.58) .784 (0.58) 1.39 (0.34) Age .261 (2.82) .799 (2.52) .254 (2.53) .800 (0.16) N 410 410 Girls Endowment -5.66 (2.62) 1.46 (0.89) -6.87 (2.84) 1.72 (0.93) Household land owned .0174 (0.84) -.0386 (1.38) .0157 (0.54) -.0389 (1.09) No land owned .705 (1.08) .607 (1.41) .632 (0.88) .605 (1.21) Household average endowment of other family members -1.61 (0.90) -.691 (0.47) -1.94 (1.14) -.642 (0.42) Age -.0178 (0.07) .550 (3.18) -.0994 (0.36) .581 (3.06) N 353 353 Source: NSRB 2001-2. aAsymptotic t-ratios corrected for clustering at the household level in columns. b Bootstrapped t-ratios in parentheses in columns. Appendix B Are body mass and “ability” negatively correlated? To directly assess whether body mass and ability are significantly correlated we estimated the association between a respondent’s body mass endowment and his or her performance on the abridged Raven’s CPM tests that were administered to all adult respondents in 2007-8. Table A reports the GLS and GLLAMM estimates, by gender, of the effects of the body mass endowment on the total number of correct answers (out of nine) for respondents aged 20-49 in 2001-2. As reported in Table A of the Appendix the mean (standard deviation) number of correct answers was 3.66 (1.96) for men and 2.94 (1.73) for women, with the complete distribution of test scores by gender depicted in Figure A in the Appendix. The first two columns of Table C indicate that while larger men perform less well than smaller men, the point estimate is very small - a one standard deviation increase in the body mass endowment reduces the test score (total correct answers) by less than 3 percent (one tenth of a question). It is well recognized, however, that performance on the Raven’s test, despite neither requiring literacy nor numeracy, is affected by schooling, and larger men have less schooling. When schooling attainment is included in the specification in column 3 its coefficient is indeed highly statistically significant and positive. Moreover, the coefficient on the body mass endowment is reduced by more than an order of magnitude to essentially zero. Evidently, larger men, net of schooling, are no less able to carry out mental tasks than are men with less brawn. For women, the reduced-form relationship between the body mass endowment and test score performance is positive and marginally statistically significant when measurement error is taken into account (column 5). However, as for men, schooling and test scores are strongly positively correlated (column 6) and the effect of body mass on the test score is not statistically significant when schooling attainment is included in the specification. Although still positive, the point estimate is also small - a one-standard deviation increase in body mass for women, net of schooling, increases the number of correct answers by less than a tenth (3.1%). Interestingly, when schooling is included in the specification, the association between landholdings and test performance is also eliminated for both men and women - larger landowners do not perform better on the test once their higher level of schooling is taken into account. Table C The Body-mass Endowment and Abridged Raven’s Matrices Test Performance (2007-8), by Gender: Respondents Aged 20-49 in 2002 Dependent variable: Number of correct answers Group Men Women Estimation procedure GLS GLLAMM GLLAMM GLS GLLAMM GLLAMM Endowment -.355 (1.05) -.371 (0.96) -.0251 (0.07) .605 (2.14) .724 (1.86) .347 (0.96) - - .151 (9.92) - - .180 (9.18) .000905 (2.52) .000896 (2.55) .0000958 (0.26) .000561 (1.67) .000561 (1.73) -.000127 (0.38) No land owned -.365 (2.63) -.380 (2.77) -.168 (1.31) -.328 (2.83) -.329 (2.80) -.128 (1.13) Age -.0502 (0.73) -.0475 (0.67) -.0316 (0.47) -.0671 (1.16) -.0682 (1.15) -.0265 (0.48) .000110 (0.11) .000069 (0.07) -.000031 (0.03) .000429 (0.51) .000561 (0.52) .000112 (0.14) ρ - .872 .872 - .805 .805 N 1,038 1,038 1,038 1,200 1,200 1,200 Schooling Household owned land Age squared Source: NSRB 2002-2007 panel. Absolute values of asymptotic t-ratios corrected for clustering within households in parentheses. All specifications include village fixed effects. Table D The Body-mass Endowment in 1982 and Age at Menarche Reported in 2001-2, Girls Aged 0-13 in 1982 Dependent variable: Whether age at menarche occurred after age 13, as reported in 2002 Specification (1) Estimation procedure (2) GLLAMM-Logit Logit Endowment (1982) GLLAMM-Logit Logit -2.23 (2.11) -2.81 (1.97) -2.22 (2.00) -2.72 (1.85) Age - - -.132 (0.69) -.119 (0.62) Age squared - - .00710 (0.48) .00586 (0.39) Household owned land (1982) - - -.000785 (0.87) -.000784 (0.87) No land owned (1982) - - .0966 (0.20) .107 (0.23) ρ - .740 - .740 N 216 216 216 216 Source: NSRB 1981-2/2001-2 panel. Absolute values of asymptotic t-ratios corrected for clustering within households in parentheses. All specifications include village fixed effects. Appendix C Procedure and Estimates for Correcting for Selectivity in the Female Wage Equation The estimates of the wage equation (17) for the sample of women are selectivity corrected using the approach of Lee (1982). The specification of the selection equation is the same as that for the occupational choice equation (13), reproduced below, yj = Zjζ + bmj + εj, where now yj is a binary indicator of whether or not a women is working for a wage, mj = the production function residual; the Zj = a vector of exogenous control variables; and εj=an error term. We estimate the selection equation by logit GLLAMM that corrects for measurement error in the endowments with the replication sub-sample. The Logit and GLLAMM estimates are reported in Appendix Table E. The distributions εj and ξj , the error of the wage equation, are allowed to be correlated. As Lee points out, only the marginal disturbances of εj and ξj needs to be specified, and not their joint bivariate distribution. Let Φ(.) be the standard normal distribution function, and Φ-1(.) be its inverse function. With the completely specified marginal distributions G(ξj) and F(εj) of εj and ξj, respectively, each of these errors can be transformed into a standard normal random variable N(0, 1). Let ε*j = J1 (εj) = Φ-1(F(εj)) and ξ*j = J2 (ξj) = Φ-1(G(ξj)). Both the transformed random variables ε*j and ξ*j are standard normal variables with zero means and unit variances. If F(.) and G(.) are standard normal distributions, then the term to add to the censored wage equation to correct for selection is the well-known inverse Mills ratio, n(Zjζ + bmj)/ Φ(Zjζ + bmj), evaluated at the estimated values of ζ and b. If G(.) is normal and F(.) is logistic, the selection correction term becomes (A1) n(J1(Zjζ + bmj)/ Φ(J1(Zjζ + bmj)). The implementation of this selection correction method is complicated by the presence of measurement error in the endowments mj. We use quadrature methods to compute posterior means in order to net out covariate measurement error in constructing the estimated (Zjζ + bmj) obtained using GLLAMM. Finally we include the selection correction term (A1), net of covariate measurement error, in the estimation of the wage equation (15) for women reported in the last three columns of Table 8. Standard errors are constructed by bootstrapping. Table E Determinants of Working for a Wage, Women Aged 20-49 in 2001-2 Estimation procedure Logit GLLAMM Logit -.00437 (2.32) -.00441 (2.33) No land owned .283 (0.86) .278 (0.84) Endowment 2.27 (4.30) 2.79 (4.32) Household average endowment of other family members .0533 (0.06) .0186 (0.02) Age .288 (2.02) .290 (2.04) -.00392 (1.88) -.00393 (1.88) ρ - .829 N 1,348 1,348 Household owned land Age squared Source: NSRB 2001-2. Absolute values of asymptotic t-ratios corrected for clustering within households in parentheses in column. All specifications include village fixed-effects. Table F First-Stage Estimates for the Occupation-Specific Wage Function: Men Aged 20-49 in 2001-2 Schooling (years) Schooling x occupational energy expenditure Endowment x occupational energy expenditure Age x occupational energy expenditure Log occupational energy expenditure Household owned land .00779 (8.98) .943 (7.93) -.00568 (2.87) -1.32 (4.07) -.000237 (4.76) No land owned -.815 (2.89) -92.6 (2.40) 2.54 (3.82) 226.9 (2.52) .0373 (2.29) Endowment -1.57 (2.48) -105.4 (1.22) 174.6 (113.5) 3028 (12.8) .534 (14.6) Household average endowment of other family members 2.39 (3.08) 208.8 (1.96) -7.27 (3.93) -876.2 (3.53) -.140 (3.12) F(19, 1022) 15.3 13.0 176.2 82.6 20.6 F(3,1022): land, no land, average household endowment 33.1 27.0 18.4 20.1 17.0 F(16,1022): land, no land, average household endowment + village fe 15.3 14.0 6.35 8.00 7.83 N 1,094 1,094 1,094 1,094 1,094 Dependent variable: Source: NSRB 2001-2. Absolute values of asymptotic t-ratios corrected for clustering within households in parentheses in column. All specifications include village fixed-effects, age and age squared. Table G First-Stage Estimates for the Occupation-Specific Wage Function: Women Aged 20-49 in 2001-2 Schooling (years) Schooling x occupational energy expenditure Endowment x occupational energy expenditure Age x occupational energy expenditure Log occupational energy expenditure Household owned land .0207 (1.96) 1.99 (1.22) -.0352 (1.65) -5.37 (3.11) -.00116 (3.66) No land owned -.0163 (0.01) -111.7 (0.54) -3.77 (1.87) -349.4 (1.79) .0781 (1.97) Endowment 1.58 (0.54) -199.9 (0.47) 167.9 (25.9) 317.7 (0.89) .0588 (0.70) Household average endowment of other family members -146.0 (1.53) -12595 (0.90) -52.8 (0.69) 36107 (3.24) 8.08 (3.71) λ 2964 (1.58) 2602.4 (0.94) 926.5 (0.63) -72269 (3.38) -161.6 (3.86) F(19, 75) 7.90 6.25 860.4 1161 136.7 F(3,75): land, no land, average household endowment 4.02 2.42 1.44 7.21 9.26 F(16,75): land, no land, average household endowment + village fe 5.00 3.83 1.99 6.78 14.9 79 79 79 79 79 Dependent variable: N Source: NSRB 2001-2. Absolute values of asymptotic t-ratios corrected for clustering within households in parentheses in column. All specifications include village fixed-effects, age and age squared.