EMMANUEL NKANSAH ASSIGNMENT ON INSTRUMENTAL VARIABLE AND PANEL ESTIMATION MODEL I use data from Wooldridge's book called the bwght (birthweight data). This is cross-sectional individual data on birth weights and it looks at how cigarette smoking affects the weight of newborns controlling for other factors. The number of observations is 1388 but the final data used is 1191. (I & II) The dependent variable is the log of birth weight (log of bwght) and the explanatory variables are a log of family income (lfaminc), father's years of education (fatheduc), mother's years of education (motheduc), birth order of child (parity), male (1 if a male child), white (1 if White), cigarette tax (cigtax), packs smoked per day when a mother is pregnant (packs)- is the endogenous variable, price of cigarettes (cigprice). (IV) The instrument is the average price of cigarettes (cigprice) in each woman’s state of residence and the endogenous variable is packs smoked per day when a mother is pregnant. Table 1: First stage regression Dependent variable: Packs For an instrument to be valid, it must be significant in the first stage and insignificant in the second stage. From Table 1, the instrument used, the average price of the cigarettes is not significant hence not a valid instrument. Accordingly, cigprice fails as an IV for packs because cigprice is not partially correlated with packs. Also, the p-value (0.7653) is above (p-value of 0.05), hence the IV is not significant. Can the instrument be excluded from the second stage? Table 2: Regression Results The test shows that the p-value of the instrument (cigprice) is 0.8199 which is above 0.005. The instrument is insignificant which satisfies the second assumption that the IV should be insignificant in the second regression equation. Thus, the instrument can be excluded from the second-stage outcome regression. Table 3: Regression Results (VI) The p-value (0.8217) shows that the OLS estimate (on the potentially endogenous covariate) is not statistically different from the IV estimate. Assignment on Fixed Effects and Random Effects This data comes from the Panel Study of Income Dynamics (PSID), which covers the years 1976 to 1982 and includes details on the demographics and earnings of 595 people. Id = Unique identifier for each survey respondent T = time (1 through 7) Wks = Weeks worked in the past year lwage = Natural logarithm of earnings in the past year Ms = binary indicator for whether respondent is married (1 = married) Occ = binary indicator for whether a respondent is a blue-collar (= 0) or white-collar (= 1) worker. Ind = binary indicator for whether respondent works in manufacturing (= 1) South = binary indicator for whether the respondent lives in the South (= 1) Smsa = binary indicator for whether the respondent lives in a standard metropolitan area (SMSA; = 1) Fem = binary indicator for whether the respondent is female (= 1) blk = binary indicator for whether the respondent is African-American (= 1) ed = years of education exp = years in the workforce. Random Effects Using the Hausman test, the fixed effects is different from the random effect The null hypothesis is that the preferred model is random effects; The alternate hypothesis is that fixed effects is the preferred model. Since the p-value (0.000) is small (less than 0.05), we go with the fixed effects and conclude that it is the preferred model The fixed Effects