Two problems with this literature None has established causality a Economists have ignored a very important aspect: the Chinese education system
6 Two problems with this literature ◼ None has established causality ◼ Economists have ignored a very important aspect: the Chinese education system
Our contributions a Estimate the causal effect of education using twins data we collected a Investigate how the Chinese education system may have affected the return to education
7 Our contributions ◼ Estimate the causal effect of education using twins data we collected ◼ Investigate how the Chinese education system may have affected the return to education
Empirical specification: OLS One equation log(earnings of person i y=Xa+B,; +ZB2+u+8 (1) y: log of monthly earnings Edu: years of schooling X: observable family variables Z: observable individual variables(age, gender job tenure) unobservable determinants of earnings lybbthesis: Edu has a positive effect on earnings Difficult to identify this effect because of endogeneity
8 Empirical specification: OLS ◼ One equation: log(earnings) of person i ◼ y: log of monthly earnings ◼ Edu: years of schooling ◼ X: observable family variables ◼ Z: observable individual variables (age, gender, job tenure) ◼ : unobservable determinants of earnings ◼ Hypothesis: Edu has a positive effect on earnings ◼ Difficult to identify this effect because of endogeneity (1) i Xi 1 Edui Zi 2 i i y = + + + +
Empirical specification Within-twin-pair estimations Twins: log(earnings) of twins in family Vi=X,+B,Edui+Z1 B2+u:+a y2=Xa+BEu2+22B2+1+E2 (3) Taking the within-twin difference we have y1-y21=B(Eh1-E2)+(21-Z2)B2+E1-E2(4) Note that the unobservables u has been removed
9 Empirical specification: Within-twin-pair estimations ◼ Twins: log(earnings) of twins in family i ◼ Taking the within-twin difference, we have ◼ Note that the unobservables, has been removed (3) (2) 2 1 2 2 2 2 1 1 1 1 2 1 i i i i i i i i i i i i y X Edu Z y X Edu Z = + + + + = + + + + ( ) ( ) (4) 1i 2i 1 Edu1i Edu2i Z1i Z2i 2 1i 2i y − y = − + − + −
Empirical specification: GLS Following ashenfelter and Krueger( 1994) 11=yEha1+yEdh2+16+22+X。+ (5) r measures the selection effect, relating family effect to education Assume correlations between the family effect and education of each twin are the same y1=X1(a+6)+BEh1+y(Edl1+Eah21)+Z1(B2+)+El(6) y2=X(a+6)+BEdh2+y(Edl1+Edh21)+Z2(B2+6)+62(7) B,measures the return to education r measures the selection effect 10
10 Empirical specification: GLS ◼ Following Ashenfelter and Krueger (1994) ◼ measures the selection effect, relating family effect to education ◼ Assume correlations between the family effect and education of each twin are the same ◼ measures the return to education ◼ measures the selection effect ( ) ( ) ( ) ' (7) ( ) ( ) ( ) ' (6) 2 1 2 1 2 2 2 2 1 1 1 1 2 1 2 1 i i i i i i i i i i i i i i y X Edu Edu Edu Z y X Edu Edu Edu Z = + + + + + + + = + + + + + + + (5) i Edu1i Edu2i Z1i Z2i Xi i = + + + + + 1