[size=13.000000pt]In this question, we are going to study on how education and other factors affect wages.The dataset [size=13.000000pt]CARD.dta [size=13.000000pt]contains hourly wages, years of schooling, IQ test scores, age,father and mother education level, as well as other potential explanatory factors for asample of 3010 individuals in 1976.
1. One solution to heteroscedasticity is to use robust standard errors which take it intoaccount. If you re‐estimate your model for non‐logged wage using robust standarderrors, what changes?
2. Assuming we have a well‐specified model, another way of dealing withheteroscedasticity is to estimate a Generalized Least Squares (GLS) model. Using(non‐logged) wage as a dependent variable, estimate the coefficients of regressionwith GLS model. How do the coefficient estimates compare to those estimated forOLS? How do the standard errors compare? On this basis why should we prefer GLSestimates over OLS?
3. There are two measures of whether an individual lived close to a college in 1966:nearc2 and nearc4, for whether an individual lived near to a two year and/or fouryear college. If we are going to choose a single instrument for educ, which of thesevariables should we use and why? (Hint: run OLS regressions respectively)
4. Since both indicator variables (nearc2 and nearc4) are correlated individually with education, we could include them as instruments for educ. What happens if you re‐estimate the model using both of these?
5. Do you prefer your answer obtained using nearc2 and nearc4 as instruments, or just using a single instrument?


[size=17.3333px]

[size=17.3333px]麻烦大家附上stata code 谢谢！