Вы находитесь на странице: 1из 8

Assignment 2 Econometrics

By Naman Khandelwal

A) It is possible that there are omitted variables in the above equation, which are correlated
with smoking. In this case, there could be factors such as stress levels of an individual and
physical activity level of the individual, included in the error term which are known to be
correlated with smoking. Then, cov(smoke,u) != 0 and therefore the estimator will not be
consistent.
When we consider stress level of individuals, then we could expect stress levels and smoking
to be positively correlated as high stress induces more smoking, moreover nicotine addiction
also leads to stress. High stress can also lead to overeating and would therefore increase weight
and bmi. So the sign of the asymptotic bias is positive.
Smoking is correlated with lower levels of physical activity, so their relationship is
negative. Sign of the asymptotic bias in this case is ambiguous as exercise can be used to both
gain as well as lose weight, thereby increasing or decreasing BMI.
B) bmi =24.007 + 0.139smoke -0.7637income + 0.0566incomesq + 0.2967age -
0.0027agesq -1.2female + 1.311preg -0.332educa
Source SS df MS Number of obs = 1476
F( 8, 1467) = 9.39
Model 2305.01052 8 288.126315 Prob > F = 0.0000
Residual 45013.3583 1467 30.6839525 R-squared = 0.0487
Adj R-squared = 0.0435
Total 47318.3688 1475 32.08025 Root MSE = 5.5393

bmi Coef. Std. Err. t P>|t| [95% Conf. Interval]

smoke .1387992 .2968558 0.47 0.640 -.4435078 .7211062


income -.7637054 .3340129 -2.29 0.022 -1.418899 -.1085116
incomesq .0566524 .0340459 1.66 0.096 -.0101313 .1234362
age .2967811 .0495184 5.99 0.000 .1996467 .3939154
agesq -.0027498 .0004977 -5.53 0.000 -.0037261 -.0017735
female -1.200008 .3002962 -4.00 0.000 -1.789063 -.6109518
preg 1.311578 1.119846 1.17 0.242 -.8850916 3.508247
educa -.3320618 .1528099 -2.17 0.030 -.6318111 -.0323126
_cons 24.00737 1.457328 16.47 0.000 21.14871 26.86604

C) The criterion for instrument relevance is cov(z,x) != 0 and cov(z,u) = 0.


The choices for IV for smoke are Cigarette tax, involuntary unemployment and
separate/divorced status. The dependent variable is bmi.
Cigarette tax is the most relevant IV as it is directly related to Smoke (fulfilling cov(z,x)!=0) and
we have no reason to believe that cigarette tax is directly related to the error in bmi of an
individual (fulfilling cov(z,u) = 0).
Separate/divorced status and involuntary unemployment make for poor IV as we cant directly
relate them to Smoke, meaning cov(z,x) = 0. Also unemployment might be related to reduced
bmi, so the second conditions is violated as cov(z,u) != 0.
D)

Source SS df MS Number of obs = 1790


F( 3, 1786) = 13.95
Model 10.2227353 3 3.40757842 Prob > F = 0.0000
Residual 436.195701 1786 .244230515 R-squared = 0.0229
Adj R-squared = 0.0213
Total 446.418436 1789 .249535179 Root MSE = .4942

smoke Coef. Std. Err. t P>|t| [95% Conf. Interval]

cigtax -.0390903 .043157 -0.91 0.365 -.1237338 .0455532


unemployed .085994 .0665281 1.29 0.196 -.044487 .216475
sepdiv .1917839 .0304187 6.30 0.000 .132124 .2514439
_cons .4545614 .0221815 20.49 0.000 .411057 .4980658

The Instruments are jointly insignificant as F<10.


test cigtax unemployed sepdiv
( 1) cigtax = 0
( 2) unemployed = 0
( 3) sepdiv = 0
F( 3, 1512) = 7.12 Prob > F = 0.0001
E) The key difference in the two regressions is the coefficient and standard error of smoke.
In the initial regression, smoke had a coefficient of 0.1388 which decreases to -1.175 in the
subsequent regression in part(e). Furthermore, the standard error of smoke also increases
from 0.2969 to 2.436. There are changes in the coefficients and standard errors of the other
explanatory variables, however these changes are very small in magnitude.
. ivregress 2sls bmi (smoke = cigtax unemployed sepdiv) income incomesq age agesq female preg educa, f

First-stage regressions

Number of obs = 1473


F( 10, 1462) = 10.99
Prob > F = 0.0000
R-squared = 0.0699
Adj R-squared = 0.0635
Root MSE = 0.4838

smoke Coef. Std. Err. t P>|t| [95% Conf. Interval]

income -.0047621 .0293331 -0.16 0.871 -.0623015 .0527773


incomesq -.0001558 .0029993 -0.05 0.959 -.0060392 .0057275
age .0135481 .004453 3.04 0.002 .0048132 .0222831
agesq -.0001183 .0000447 -2.64 0.008 -.000206 -.0000306
female -.1183901 .0260991 -4.54 0.000 -.1695858 -.0671943
preg -.1270225 .0978404 -1.30 0.194 -.3189452 .0649001
educa -.0628967 .0133146 -4.72 0.000 -.0890145 -.0367788
cigtax -.0141404 .0469112 -0.30 0.763 -.106161 .0778801
unemployed .0844418 .0718421 1.18 0.240 -.0564827 .2253664
sepdiv .1566407 .0342948 4.57 0.000 .0893684 .223913
_cons .5181758 .130599 3.97 0.000 .2619943 .7743573

Instrumental variables (2SLS) regression Number of obs = 1473


Wald chi2(8) = 79.08
Prob > chi2 = 0.0000
R-squared = 0.0393
Root MSE = 5.5385

bmi Coef. Std. Err. z P>|z| [95% Conf. Interval]

smoke -1.175366 2.436306 -0.48 0.629 -5.950437 3.599706


income -.7706533 .3350487 -2.30 0.021 -1.427337 -.1139699
incomesq .0522719 .0342925 1.52 0.127 -.0149403 .119484
age .3352436 .0669044 5.01 0.000 .2041134 .4663738
agesq -.0031036 .0006439 -4.82 0.000 -.0043656 -.0018417
female -1.38656 .4121072 -3.36 0.001 -2.194275 -.5788444
preg 1.18685 1.168729 1.02 0.310 -1.103817 3.477516
educa -.369687 .2133194 -1.73 0.083 -.7877853 .0484113
_cons 24.18121 1.825986 13.24 0.000 20.60234 27.76007

Instrumented: smoke
Instruments: income incomesq age agesq female preg educa cigtax unemployed
sepdiv
F) On performing a test for unemployed and cigtax, both the instrument turn out to be jointly
insignificant. Furthermore, when we carry out the regression in part (e), both cigtax and
unemployed have small insignificant coeffecients. So I think that the estimation in part(e) is
incorrect. Sepdiv, on the other hand, has a large coefficient and is also highly significant. Hence,
I believe that the correct estimation would involve using sepdiv as an instrumental variable for
smoke.
. ivregress 2sls bmi (smoke = sepdiv) income incomesq age agesq female preg educa, first

First-stage regressions

Number of obs = 1474


F( 8, 1465) = 13.47
Prob > F = 0.0000
R-squared = 0.0685
Adj R-squared = 0.0634
Root MSE = 0.4838

smoke Coef. Std. Err. t P>|t| [95% Conf. Interval]

income -.0064302 .0292611 -0.22 0.826 -.0638282 .0509678


incomesq -6.13e-06 .0029892 -0.00 0.998 -.0058696 .0058574
age .0131241 .0044362 2.96 0.003 .0044221 .021826
agesq -.0001153 .0000446 -2.59 0.010 -.0002028 -.0000279
female -.1170626 .0260696 -4.49 0.000 -.1682003 -.0659249
preg -.1266797 .0977907 -1.30 0.195 -.3185045 .065145
educa -.0645668 .0132616 -4.87 0.000 -.0905806 -.038553
sepdiv .1559674 .0341937 4.56 0.000 .0888934 .2230413
_cons .5389977 .1277342 4.22 0.000 .2884362 .7895591

Instrumental variables (2SLS) regression Number of obs = 1474


Wald chi2(8) = 73.18
Prob > chi2 = 0.0000
R-squared = 0.0017
Root MSE = 5.6586

bmi Coef. Std. Err. z P>|z| [95% Conf. Interval]

smoke -2.389137 2.56417 -0.93 0.351 -7.414817 2.636543


income -.8061988 .3422499 -2.36 0.018 -1.476996 -.1354014
incomesq .0565152 .0349623 1.62 0.106 -.0120096 .1250401
age .3440183 .0684063 5.03 0.000 .2099444 .4780923
agesq -.0031827 .0006582 -4.84 0.000 -.0044728 -.0018926
female -1.486207 .4244946 -3.50 0.000 -2.318201 -.6542132
preg .9446002 1.198789 0.79 0.431 -1.404982 3.294183
educa -.4846033 .2233514 -2.17 0.030 -.922364 -.0468426
_cons 25.22826 1.907232 13.23 0.000 21.49015 28.96636

Instrumented: smoke
Instruments: income incomesq age agesq female preg educa sepdiv

Вам также может понравиться