Coping with Prosperity: A Double Hurdle Approach to Parents’ Earnings and Coping with Prosperity: The Response of Parents’ Child Care Time in the Earnings ATUS Child Care Time UseUse to Rising James M. Payne Calhoun Community College November 23, 2015 1 Coping with Prosperity Demand for human capital in children Demand for human capital in children Higher earnings More time spent with children But where is the . . . 2 Coping with Prosperity The unobserved substitution effect ? market services b a { I own-time services 3 Coping with Prosperity Data & model Data American Time Use Survey (ATUS) & Current Population Survey (CPS) 2003 – 2010 n = 45,716 parents Model Cragg’s (1971) Double Hurdle 4 Coping with Prosperity Missing data Challenge # 1: Missing Data — Biased Census hot-deck imputation 15.6% missing earnings data — Single-imputation methods underestimate uncertainty 5 Coping with Prosperity Multiple imputation Multiple Imputation (MI) (Rubin 1987) • • • • • Fill blanks with “neutral” values Repeat m times → m data sets Preserve variation in the data Model each set (imputation) separately Combine estimates using Rubin’s Rule 6 Coping with Prosperity Sample selection and endogeneity Challenge # 2: Sample Selection Bias Challenge # 3: Endogeneity 7 Concurrent Heckman-IV approach Coping with Prosperity Concurrent approach* 1st step probit (Heckman) for WORKING = 1 Double hurdle model with x = earnings 2nd step OLS (Heckman) for earnings Instrument for endogeneity (METRO†) *Millimet, 2001; Amemiya, 1985 † π½ππΈππ π = 2.838; π‘ = 20.32 8 Coping with Prosperity Heteroscedasticity with zeroes Challenge # 4: Heteroscedasticity with zero values in the data 9 Coping with Prosperity Inverse hyperbolic sine Inverse Hyperbolic Sine transformation • Burbidge, et al. (1988) sinh-1 • Defined for all real numbers ln • Equivalent to a log-linear model ππππ−π (π) = ππ π + ππ + π 10 Time use definitions Coping with Prosperity } Purchases + BEHALFTIME market services b a { I FACETIME own-time services 11 Coping with Prosperity Theoretical model The issue: How will a higher hourly wage (w ) affect input prices and time use? • pfacetime : FACETIME consists only of time, so pfacetime = w πππππππ‘πππ & =1 ππ€ • pservices = γppurchases + τpbehalftime (weighted average of components) π πΎπππ’ππβππ ππ • = 0 (prices of purchases are orthogonal to w ) ππ€ π ππππβππππ‘πππ πππππππ‘πππ • = τ (since = 1, if pbehalftime = pfacetme ) ππ€ ππ€ πππππππ‘πππ π ππ πππ£ππππ • So = τ <1< ππ€ ππ€ a higher wage reduces pservices relative to pfacetime , and thus BEHALFTIME will be substituted for FACETIME 12 Coping with Prosperity Propositions Propositions • Higher wages will lead to: • higher levels of FACETIME • greater use of market services and hence of BEHALFTIME • a higher level of BEHALFTIME relative to FACETIME } Greater demand for human capital in children substitution effect 13 Substitution of BEHALFTIME for FACETIME Coping with Prosperity Sensitivity analysis—marginal effects from bivariate probit model Age of youngest child (CHILDAGE ) Hourly earnings (2003$) (rEARNHRhat)* $2.69 $5.92 $9.61 $15.18 $18.55 $21.73 P 10 Q1 Median Q3 P 90 P 95 -0.080 0 P 10 1.476 -0.05 -0.095 2 Q1 1.359 Median 0.891 Q3 0.305 P 90 Cell contents: -0.019 1.444 -0.081 0.976 -0.159 0.390 0.085 1.658 0.05 0.070 1.541 0.05 0.007 1.073 0.01 -0.071 0.488 -0.77 -0.41 -0.15 -0.2831 15 0.00 -0.18 -0.08 -0.236 11 1.561 -0.07 -0.01 -0.158 6 -0.003 -0.0457 † -0.2061 0.0393 -0.1179 0.1367 -5.25 -0.86 MEBEHALFTIME MEFACETIME MEBEHALFTIME /MEFACETIME 0.218 1.805 0.12 0.202 1.688 1.220 0.635 0.220 1.308 0.141 0.722 0.20 0.2834 0.05 1.776 0.17 0.10 0.0149 0.282 0.16 0.11 0.062 1.893 0.16 0.12 0.140 0.298 0.0944 0.3713 0.25 0.374 1.978 0.19 0.359 1.861 0.19 0.296 1.392 0.21 0.218 0.807 0.27 0.1710 0.4559 0.38 n = 26,963 women 14 Double hurdle results Coping with Prosperity Double hurdle estimates: Salient results, 1st hurdle probit (n = 26,963 women) FACETIME marginal marginal VARIABLE effect* BEHALFTIME p -value effect* p -value EARNINGS 0.010 <0.001 0.010 <0.001 CHILDAGE -0.041 <0.001 -0.006 <0.001 SINGLEPARENT -0.033 0.168 0.057 0.031 WORKHOURS -0.003 <0.001 -0.001 0.005 *at means of regressors 15