Matched panels: smoking and birthweight. Abrevaya (2006) estimates the effect of smoking on birth outcomes from panel data (i.e., data on mothers with multiple births). Panel data allows the identification of the smoking effect from women who change their smoking behavior from one pregnancy to another. The data set contains 296,218 birth observations with 141,929 distinct mothers (identified by momid3 and idx, an index number of a mother's birth). The data set can be downloaded from the Journal of Applied Econometrics archive web site:
(http://qed.econ.queensu.ca/jae/2006-v21.4/abrevaya/). Birthweight (in grams) is regressed on (i)whether the mother smokes; (ii) the number of cigarettes smoked per day; (iii) the baby's gender; (iv) mother's age and age-squared; (v) whether she is a high-school graduate, had some-college, or is a collegegraduate; (vi) her race and marital status; (vii) adeqcode 2 and adeqcode 3 , which are indicators that the Kessner index \(=2\) or 3 . This measures the adequacy of prenatal care, 2 being intermediate and 3 being inadequate; (viii) a dummy variable for no prenatal visits; (ix) petri2 and petri3, which are indicators that the first prenatal visit occurred in the 2nd or 3rd trimester.

(a) Run the OLS estimates to replicate column 5 of Table IV of Abrevaya (2006). This only includes the dummy variable for smoking but not the number of cigarettes smoked. Be sure to include the dummies for the number of live births, mother's state of residence, and mother's year of birth.

(b) Run the corresponding FE estimates (including the women's fixed effect) and thus wiping out the timeinvariant variables. This should replicate the FE estimates in column 6 of Table IV of Abrevaya (2006).

(c) Add the number of cigarettes smoked and replicate columns 11 and 12 of Table IV of Abrevaya (2006).

(d) comparing the FE and OLS estimates, what do you conclude?

(e) To gauge the degree of incorrect matching and its effect on the estimates reported above, Abrevaya utilized a proxy for correct matches. This dummy variable proxy takes the value 1 if the observed interval since last birth agrees with the record. A value of zero for proxy is extremely strong evidence of an incorrect match since only miscoding of the interval record or birth month could result in proxy \(=0\) for a correct match. Show that the fixed effects estimates for this more reliable sample yield a reduction in birth weight of \(67 \mathrm{~g}\) for smokers which is much smaller than the overall fixed effects estimates based upon the full samples (144 g).