Prac 4: Stata application of IV method 2 The University of Queensland ECON7360, Semester 2
Instructor: Rigissa Megalokonomou
Problem I: IV example: The Effects of Smoking on Birth Weight
Use the data in bwght.dta for the following questions. This data include observations for pregnant women in US.
(i) Describe the variables in the dataset.
(ii) Estimate the model by using OLS and interpret the estimates (hint:
use robust standard errors).
bwght = ¦Â0 + ¦Â1packs + ¦Â2faminc + u (1)
We might worry that packs is correlated with other health factors or the availability of good parental care, so that packs and unobserved u might be correlated. Now, we want to examine whether cigprice can be used as a potential instrumental variable for packs.
(iii) Suppose that we assume that cigprice and u are uncorrelated. Is it a good assumption? Can you somehow test this? (hint: use robust standard errors)
(iv) Draw a scatter plot between packs on cigprice.
(v) Perform a reduced form regression. Regress packs on cigprice (hint:
use robust standard errors). Interpret the regression result.
1
(vi) Implement IV method by 2SLS.
Problem II: Conditional IV example: The Effects of Education on
Wage
Use the data in card.dta for the following questions. This data include men in US during 1976. Card(1995) used wage and education data to estimate the return to education.
(i) Describe the variables in the data.
(ii) Estimate the model by using OLS and interpret the estimates (hint:
use robust standard errors).
ln(wage) = ¦Â0 + ¦Â1educ + ¦Â2exper + ¦Â3exper2 + ¦Â4smsa + ¦Â5south + u (2)
Education is likely to be endogenous. Card proposes that we use near4 as an instrument for the endogenous variable educ.
(iii) Use a simple OLS regression to check whether near4 has an effect on educ (hint: use robust standard errors). Could near4 be correlated with factors in the error term, such as ability?
(iv) Implement the IV method by 2SLS. Interpret the coefficient of educ. Do you find a different coefficient when you use robust standard errors com- pared to when you don¡¯t use robust standard errors?
(v) For a sub-sample of the men in the data set, an IQ score is available. Do IQ scores vary by whether a man grew up near a four-year college? What do you conclude from that?
(vi) Now use OLS to regress IQ on nearc4, smsa66, and 1966 regional dummy variables reg662,…., reg669 (hint: use robust standard errors). Are IQ and near4 related after the geographic dummy variables have been con- trolled for?
2
(vii) What do you conclude about the importance of controlling for smsa66 and the 1966 regional dummies in the wage equation?
(viii) Now implement the IV method by 2SLS again while you control for the regional dummies (hint: use robust standard errors) and interpret the coefficient of educ.
3