Beruflich Dokumente
Kultur Dokumente
Denis Tkachenko
Semester 2 2014/2015
Outline
1.
2.
3.
4.
5.
1 = E(Yi|Xi = 1) E(Yi|Xi = 0)
= population difference in group means
4
Example:
1 if STRi 20
Let Di =
0 if STRi 20
N
238
182
ss2 sl2
19.42 17.92
SE =
=
= 1.8
238
182
ns nl
5
Regression:
References:
Krueger, A. (1993) How Computers Have Changed the Wage Structure: Evidence from Microdata, 19841989, Quarterly Journal of Economics, 108(1), 33-60.
DiNardo, J. and J. Pischke. (1997) The Returns to Computer Use Revisited: Have Pencils Changed the
Wage Structure Too? The Quarterly Journal of Economics, 112(1), 291-303.
8
Regression:
Number of obs
F( 1,
933)
Prob > F
R-squared
Root MSE
lwage
Coef.
educ
_cons
.0598392
5.973063
Robust
Std. Err.
.0060791
.0822718
t
9.84
72.60
=
=
=
=
=
935
96.89
0.0000
0.0974
.40032
P>|t|
0.000
0.000
.047909
5.811603
.0717694
6.134522
10
When?
11
Previous examples
Example: Computer Premium
where
where
2. Is Z correlated with X?
In US, immigrant communities tend to be less affluent and thus
have smaller school budgets and higher STR.
Thus, The OLS estimator is biased. Direction of this bias?
14
n i 1
=
1 1 = i n1
n 1 2
2
(Xi X )
sX
n
i 1
where vi = (Xi X )ui (Xi X)ui. Under Least Squares
Assumption 1,
E[(Xi X)ui] = cov(Xi,ui) = 0.
But what if E[(Xi X)ui] = cov(Xi,ui) = Xu 0?
15
n
n
1 1 i n1
i 1 n
1
1
2
2
(
X
X
)
(
X
X
)
i
i
n i 1
n i 1
Xu u Xu u
2 =
=
Xu ,
X X X u X
where Xu = corr(X,u). If assumption #1 is valid, then Xu = 0,
p
1 1 +
Xu
X
If an omitted factor Z is both:
(1) a determinant of Y (that is, it is contained in u); and
(2) correlated with X,
then Xu 0 and the OLS estimator is biased (and is not
p
consistent).
The math makes precise the idea that districts with few ESL
students (1) do better on standardized tests and (2) have
smaller classes (bigger budgets), so ignoring the ESL factor
results in overstating the class size effect.
Is this actually going on in the CA data?
17
Interpretation of coefficients
Yi = 0 + 1X1i + 2X2i + ui, i = 1,,n
Difference:
So:
1 =
Y
, holding X2 constant
X 1
2 =
Y
, holding X1 constant
X 2
21
22
23
Number of obs
F( 2,
932)
Prob > F
R-squared
Root MSE
lwage
Coef.
educ
iq
_cons
.0391199
.0058631
5.658288
Robust
Std. Err.
.0071301
.0010152
.0943505
t
5.49
5.78
59.97
=
=
=
=
=
935
71.49
0.0000
0.1297
.39332
P>|t|
0.000
0.000
0.000
.025127
.0038709
5.473124
.0531128
.0078554
5.843452
24
25
26
1 1 +
X
p
Xu
28
OVBGuide:Upward/Downwardbiasandover and
underestimationoftheeffectofXonY
(X,Z)
(X,Z)
(Y,Z)
>0
<0
>0
(X,u)>0
UB
(X,u)<0
DB
<0
(X,u)<0
DB
(X,u)>0
UB
UB
DB
(X,Y)>0
Overestimateeffect
Underestimateeffect
(X,Y)<0
Underestimateeffect
Overestimateeffect
Color: Black
1601-2000cc
Automatic transmission
Sedan
BMW and Honda
29
Obs
Mean
price
ryear
coe
omv
mile
115
117
108
116
79
79.39891
2007.889
22241.55
34392.81
79702.82
nowner
bmw
97
117
1.690722
.6324786
Std. Dev.
Min
Max
45.23569
2.058531
20128.93
9869.397
35249.17
24.8
2004
4889
17113
4000
236.888
2013
97000
61322
152000
.7821022
.4842038
1
0
5
1
30
.005
Density
.01
.015
.02
50
100
150
Selling Price
200
250
31
Scatter Plots
32
34
R2 and R
where ESS =
, SSR =
(
)
Y
Y
i
i 1
u
i , TSS =
i 1
2
.
(
Y
Y
)
i
i 1
R2 and R , ctd.
The R 2 (the adjusted R2) corrects this problem by penalizing
you for including another regressor the R 2 does not necessarily
increase when you add another regressor.
n 1 SSR
Adjusted R : R = 1
n k 1 TSS
2
37
SER =
RMSE =
n
1
2
i
n k 1 i 1
1 n 2
ui
n i 1
38
39
Number of obs =
420
F( 1,
418) =
19.26
Prob > F
= 0.0000
R-squared
= 0.0512
Root MSE
= 18.581
------------------------------------------------------------------------|
Robust
testscr |
Coef.
Std. Err.
t
P>|t|
[95% Conf. Interval]
--------+---------------------------------------------------------------str | -2.279808
.5194892
-4.39
0.000
-3.300945
-1.258671
str |
(dropped)
_cons |
698.933
10.36436
67.44
0.000
678.5602
719.3057
-------------------------------------------------------------------------
43
44
1 E ( 1 )
var( 1 )
So too for 2 ,, k
Conceptually, there is nothing new here!
45
X0
1
1
1
Imperfect multicollinearity
Imperfect and perfect multicollinearity are quite different despite
the similarity of the names.
Imperfect multicollinearity occurs when two or more regressors
are highly (but not perfectly) correlated.
Why this term? If two regressors are highly correlated,
then their scatterplot will pretty much look like a straight
line they are collinear but unless the correlation is
exactly 1, that collinearity is imperfect.
50