****** *** missreg.do --- DO file that provides the regressions for Table 10 in Abrevaya and Donald (2014, working paper) ****** * wls-data.dta contains data from Wisconsin Longitudinal Study * educ = years of education, iq = IQ score, bmirating = standardized BMI rating (0 if missing), bmimissing = 1 if bmirating is missing use wls-data, clear *** Regressions for men * complete-data regression regr educ bmirating iq if male & !bmimissing, robust * dummy-variable method regr educ bmirating iq bmimissing if male, robust * optimal GMM estimator (this command does the iterated GMM (using igmm option)); * to get standard two-step optimal GMM estimator, simply drop the igmm option gmm ((1-bmimissing)*(educ - {beta_0} - {alpha}*bmirating - {beta_iq}*iq)) /// ((1-bmimissing)*(bmirating - {gamma_0} - {gamma_iq}*iq)) /// (bmimissing*(educ - ({gamma_0}*{alpha}+{beta_0}) - ({gamma_iq}*{alpha}+{beta_iq})*iq)) if male, /// instruments(1: bmirating iq) /// instruments(2: iq) /// instruments(3: iq) /// winitial(identity) igmm *** Regressions for women * complete-data regression regr educ bmirating iq if !male & !bmimissing, robust * dummy-variable method regr educ bmirating iq bmimissing if !male, robust * optimal GMM estimator (this command does the iterated GMM (using igmm option)); * to get standard two-step optimal GMM estimator, simply drop the igmm option gmm ((1-bmimissing)*(educ - {beta_0} - {alpha}*bmirating - {beta_iq}*iq)) /// ((1-bmimissing)*(bmirating - {gamma_0} - {gamma_iq}*iq)) /// (bmimissing*(educ - ({gamma_0}*{alpha}+{beta_0}) - ({gamma_iq}*{alpha}+{beta_iq})*iq)) if !male, /// instruments(1: bmirating iq) /// instruments(2: iq) /// instruments(3: iq) /// winitial(identity) igmm * wls-data-adulbmi.dta contains data from Wisconsin Longitudinal Study * adultbmi = BMI as an adult, educ = years of education, iq = IQ score, bmirating = standardized BMI rating (0 if missing), bmimissing = 1 if bmirating is missing use wls-data-adultbmi, clear *** Regressions for men * complete-data regression regr adultbmi bmirating iq educ if male & !bmimissing, robust * dummy-variable method regr adultbmi educ bmirating iq bmimissing if male, robust * optimal GMM estimator (this command does the iterated GMM (using igmm option)); * to get standard two-step optimal GMM estimator, simply drop the igmm option gmm ((1-bmimissing)*(adultbmi - {beta_0} - {alpha}*bmirating - {beta_iq}*iq - {beta_educ}*educ)) /// ((1-bmimissing)*(bmirating - {gamma_0} - {gamma_iq}*iq - {gamma_educ}*educ)) /// (bmimissing*(adultbmi - ({gamma_0}*{alpha}+{beta_0}) - ({gamma_iq}*{alpha}+{beta_iq})*iq - ({gamma_educ}*{alpha}+{beta_educ})*educ)) if male, /// instruments(1: bmirating iq educ) /// instruments(2: iq educ) /// instruments(3: iq educ) /// winitial(identity) igmm *** Regressions for women * complete-data regression regr adultbmi bmirating iq educ if !male & !bmimissing, robust * dummy-variable method regr adultbmi educ bmirating iq bmimissing if !male, robust * optimal GMM estimator (this command does the iterated GMM (using igmm option)); * to get standard two-step optimal GMM estimator, simply drop the igmm option gmm ((1-bmimissing)*(adultbmi - {beta_0} - {alpha}*bmirating - {beta_iq}*iq - {beta_educ}*educ)) /// ((1-bmimissing)*(bmirating - {gamma_0} - {gamma_iq}*iq - {gamma_educ}*educ)) /// (bmimissing*(adultbmi - ({gamma_0}*{alpha}+{beta_0}) - ({gamma_iq}*{alpha}+{beta_iq})*iq - ({gamma_educ}*{alpha}+{beta_educ})*educ)) if !male, /// instruments(1: bmirating iq educ) /// instruments(2: iq educ) /// instruments(3: iq educ) /// winitial(identity) igmm