Sie sind auf Seite 1von 6

Kevin Sena

Econ 1630 HW4

April 25, 2011

Problem 1
y i 0 1 x i ui
x i 0 1z i v i

z is a binary instrumental variable for x.

y y zi z
szy i1 i

1
szx n x x z z
n

i1

y1

x1

z1

y2
.
.
.

x2
.
.
.

z2
.
.
.

yn

xn

zn

Since z is a binary instrumental variable for x there are n1 items when z = 1 and n0
items when z = 0 (n1 + n0 = n).

n1

numerator

y i y zi z i1y i y i1 y i n1y i1 y i n1
n1

i1

n1

n1

yi yi
n0

i1

n1 n 0

i1

n 0 y i n1 y i
n0
n1
n0
n1 n1
n1
n0
n1
i1
i1
numerator 1
yi
yi
yi
i1 y i

n1 n 0 i1
n1 n 0 i1
n1 n 0 i1
n1 n 0
n1 n 0
Similarly,
n1

n1

denomin ator

i1

n1

n1

xi xi
n0

i1

n1 n 0

i1

n 0 x i n1 x i
n0
n1
n0
n1 n1
n1
n0
n1
i1
i1
denomin ator 1
xi
xi
xi
i1 x i

n1 n 0 i1
n1 n 0 i1
n1 n 0 i1
n1 n 0
n1 n 0
n1

szy
szx

n1
i1

n1

x i x zi z i1x i x i1 x i n1x i1 x i n1
n1

n0

yi

y
x
i1
n

y zi z

i1

x zi z

y1;

n 0 y i n1 y i

n1

n0

i1

i1

n1 n 0

n 0 x i n1 x i
n1

i1

n1 n 0

n0
i1

n0

yi

y 0;

n1
i1

n1

xi

x1;

n0
i1

n0

xi

n0

1
multiply top and bottom by
n 0 n1

i1

n1
i1

n1

n1

i1

n1

x 0
1

y1 y 0
x1 x 0

yi
xi

n0

n0
i1

yi

n0

n0

i1

n0

xi

Problem 2
do "/Users/kevinsena/Desktop/Hw4 Question 2.do"
. //step 1
. drop _all
. quietly set obs 1000
. gen z=round(uniform())
. gen v=invnorm(uniform())
. gen u=0.9*v+sqrt(1-0.9^2)*invnorm(uniform())
. scalar pi=1
. gen x=z*pi+v
. scalar beta=0
. gen y=x*beta+u
.
. //step 2
. ivreg y (x=z)
Instrumental variables (2SLS) regression
Source
|
SS
df
MS
-------------+-----------------------------Model
| 33.277073 1 33.277073
Residual | 970.525568 998 .972470509
-------------+-----------------------------Total
| 1003.80264 999 1.00480745

Number of obs = 1000


F( 1, 998) = 0.09
Prob > F = 0.7588
R-squared = 0.0332
Adj R-squared = 0.0322
Root MSE = .98614

-----------------------------------------------------------------------------y
| Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+---------------------------------------------------------------x
| .0186739 .0607958 0.31 0.759 -.1006284 .1379762
_cons
| .0221472 .0448836 0.49 0.622 -.0659299 .1102244
-----------------------------------------------------------------------------Instrumented: x
Instruments: z
------------------------------------------------------------------------------

. di _b[x]
.01867389
. //step 3
. quietly su y if z==1
. scalar y1=r(mean)
. quietly su y if z==0
. scalar y0=r(mean)
. quietly su x if z==1
. scalar x1=r(mean)
. quietly su x if z==0
. scalar x0=r(mean)
. di (y1-y0)/(x1-x0)
.01867389
The IV estimator and
Problem 3

y1 y 0
both equal 0.01867389.
x1 x 0

a) Stata Command: reg lwage educ exper expersq black south smsa, r

Linear regression

Number of obs
F( 6, 3003)
Prob > F
R-squared
Root MSE

=
=
=
=
=

3010
217.74
0.0000
0.2905
.37419

-----------------------------------------------------------------------------|
Robust
lwage
| Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+---------------------------------------------------------------educ
| .074009 .003642 20.32 0.000 .0668679 .0811501
exper
| .0835958 .0067326 12.42 0.000 .0703948 .0967969
expersq | -.0022409 .0003181 -7.04 0.000 -.0028646 -.0016171
black
| -.1896315 .0174324 -10.88 0.000 -.2238123 -.1554508
south
| -.1248615 .0153508 -8.13 0.000 -.1549606 -.0947625
smsa
| .161423 .0151751 10.64 0.000 .1316683 .1911776

_cons
| 4.733664 .0701577 67.47 0.000 4.596102 4.871226
-----------------------------------------------------------------------------The estimated return to education using OLS is 0.074009 and is significant at
the 5% level (t=20.32>1.96).
b) Stata Command: ivreg lwage (educ=nearc2) exper expersq black south smsa,
r
Instrumental variables (2SLS) regression

Number of obs = 3010


F( 6, 3003) = 46.32
Prob > F = 0.0000
R-squared =
.
Root MSE = .65464

-----------------------------------------------------------------------------|
Robust
lwage
| Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+---------------------------------------------------------------educ
| .3497636 .2022577 1.73 0.084 -.0468141 .7463413
exper
| .1966903 .0841514 2.34 0.019
.03169 .3616906
expersq | -.0024452 .0006216 -3.93 0.000 -.003664 -.0012265
black
| .0887244 .2078812 0.43 0.670 -.3188794 .4963283
south
| -.0304149 .0737579 -0.41 0.680 -.1750361 .1142062
smsa
| .0190063 .1084208 0.18 0.861 -.1935802 .2315929
_cons
| .0925577 3.405536 0.03 0.978 -6.584861 6.769976
-----------------------------------------------------------------------------Instrumented: educ
Instruments: exper expersq black south smsa nearc2
-----------------------------------------------------------------------------The estimated return to education using nearc2 as an instrument for educ is
0.3497636 and is not significant at the 5% level (t = 1.73 <1.96).
c) Stata Command: ivreg lwage (educ=nearc4) exper expersq black south smsa,
r
Instrumental variables (2SLS) regression

Number of obs = 3010


F( 6, 3003) = 131.70
Prob > F = 0.0000
R-squared = 0.2252
Root MSE = .39103

-----------------------------------------------------------------------------|
Robust
lwage
| Coef. Std. Err. t P>|t| [95% Conf. Interval]

-------------+---------------------------------------------------------------educ
| .1322888 .0485779 2.72 0.007 .0370396 .2275381
exper
| .107498 .0211375 5.09 0.000 .0660525 .1489434
expersq | -.0022841 .0003467 -6.59 0.000 -.0029639 -.0016042
black
| -.1308019 .0515112 -2.54 0.011 -.2318027 -.0298011
south
| -.1049005 .0229264 -4.58 0.000 -.1498535 -.0599475
smsa
| .1313237 .029803 4.41 0.000 .0728872 .1897601
_cons
| 3.752781 .8177012 4.59 0.000 2.14947 5.356092
-----------------------------------------------------------------------------Instrumented: educ
Instruments: exper expersq black south smsa nearc4
-----------------------------------------------------------------------------The estimated return to education using nearc4 as an instrument for educ is
0.1322888 and is significant at the 5% level (t =2.72 > 1.96).
d) . reg educ nearc2 exper expersq black south smsa, r
Linear regression

Number of obs
F( 6, 3003)
Prob > F
R-squared
Root MSE

=
=
=
=
=

3010
606.80
0.0000
0.4720
1.947

-----------------------------------------------------------------------------|
Robust
educ
| Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+---------------------------------------------------------------nearc2
| .122237 .0734467 1.66 0.096 -.021774 .266248
exper
| -.4095854 .03193 -12.83 0.000 -.4721923 -.3469784
expersq | .0006992 .001701 0.41 0.681 -.0026361 .0040345
black
| -1.017135 .0880085 -11.56 0.000 -1.189698 -.8445716
south
| -.326916 .0787159 -4.15 0.000 -.4812586 -.1725734
smsa
| .4968801 .0815703 6.09 0.000 .3369409 .6568194
_cons
| 16.78535 .1428128 117.53 0.000 16.50533 17.06537
-----------------------------------------------------------------------------. test nearc2
( 1) nearc2 = 0
F( 1, 3003) = 2.77
Prob > F = 0.0962
. reg educ nearc4 exper expersq black south smsa, r

Linear regression

Number of obs
F( 6, 3003)
Prob > F
R-squared
Root MSE

=
=
=
=
=

3010
608.02
0.0000
0.4745
1.9425

-----------------------------------------------------------------------------|
Robust
educ
| Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+---------------------------------------------------------------nearc4
| .3373208 .0806044 4.18 0.000 .1792753 .4953663
exper
| -.4100081 .0319791 -12.82 0.000 -.4727112 -.3473049
expersq | .0007323 .0017029 0.43 0.667 -.0026067 .0040713
black
| -1.006138 .0877849 -11.46 0.000 -1.178263 -.8340136
south
| -.291464 .0784454 -3.72 0.000 -.4452761 -.1376519
smsa
| .4038769 .0849467 4.75 0.000 .2373173 .5704364
_cons
| 16.65917 .1467408 113.53 0.000 16.37145 16.9469
-----------------------------------------------------------------------------. test nearc4
( 1) nearc4 = 0
F( 1, 3003) = 17.51
Prob > F = 0.0000
Among the 3 different approaches, using nearc4 as an instrument for
educ is the best method to estimate the return on education. The OLS
estimation performed in part A suffers from endogenity because the
estimation performed ignores this endogenity, which implies that the OLS
estimator of the return to education is inconsistent. Nearc2 is a weak
instrument because it fails the relevant condition, which in its most generic
form states that x and z must be very correlated, which implies that 1 is
significantly different from 0. The test that nearc2 is significantly different
from 0 yielded an F value of 2.77, which is less than the 5% significance level
critical value of 3. This shows that nearc2 is not significantly different from 0
and is therefore a weak variable. On the other hand, the F value for the test
that nearc4 is significantly different from 0 yielded an F value of 17.51, which
is much greater than 3. This shows that nearc4 is significantly different from
0 and is therefore a stronger instrument than nearc2. Therefore estimating
the model using near4 as the instrument for educ is the bet approach to
estimate the return on education.

Das könnte Ihnen auch gefallen