Professional Documents
Culture Documents
(0.0031)
(0.0211)
(0.0002)
t5%, n-k, where 5% is the level of significance, n-k is the degrees of freedom (n being the
number of observations and k the number of regression coefficients)
t5%, 548= 1.964
Since the computed t statistic ( 1.5639 is less than critical t (1.964), we fail to reject
the null hypothesis and conclude that mothers years of schooling does not affect wages.
c) Since this is a test for joint significance, the most appropriate test is the F test (F
distribution)
Hypotheses
H0: 1= 2=0 ( own and mothers years of schooling are jointly insignificant)
HA: 1 2 0 (own and mothers years of schooling are jointly significant)
To perform this we need we need both the restricted and the unrestricted model, where
the unrestricted model is the one in which all the independent variables are present and
the restricted model is one in which the coefficients of independent variables under test
have been put to zero.
We have the restricted model as
log ( wagei ) = 0 + 3 exper i+ i ,i=1, , n
The estimated model is
log ( wagei ) =0.10+0.25 exper i
(0.002)
(0.0013)
R2=0.422, RSS=176.23
First, we compute the calculated F statistic using the following formula (RSS means
Residual sum of squares, subscripts ur and r represent unrestricted and restricted models
respectively, and q the number of restrictions.
(RSS r RSSur )/q
F=
RSSur /(nk1)
(176.23135.463)/ 2
F=
135.463 /(55241)
20.3835
F=
0.2476
F=82.3243
At 5% level of significance critical F2, 547=3.0121
Since F calculated is greater than F critical, we reject the null hypothesis and conclude
that education background of an individual does affects their income.
Alternative approach to calculating F
( R2urR 2r )/q
F=
(1R2ur )/( nk 1)
(0.630.422)/2
F=
(10.63)/(55241)
0.104
F=
0.000676416
F=153.7515
Since this values is also greater than F, we uphold the earlier decision to reject the null
hypothesis and conclude that education background of an individual does affect their income.
Question 2
Given this form of F statistic, we can derive the RSS form as follows
( R2urR 2r )/q
F=
(1R2ur )/( nk 1)
By definition
2
R=
But,
ESS
TSS
ESS+ RSS=TSS
RSS
TSS
to the RHS
ES S
RSS
=1
TSS
TSS
R 2=
Therefore,
ESS
RSS
=1
TSS
TSS
model as follows
R2ur=1
RSSur
TSS
, and
R2r =1
RSSr
TSS
RSSur
)/(nk 1)
TSS
RSSur
RSSr
(1
)(1
)/q
TSS
TSS
F=
1(1
RSSur
RSSr
)(1
)
TSS
TSS
q
RSSur
RSS r
1+
TSS
TSS
q
Rearranging
RSSr RSSur
TSS
TSS
q
Since they components in the numerator have the same denominator, we can combine them to
form
RSSr RSSur
TSS
q
Next, we simplify the denominator
1(1
RSS ur
)/( nk 1)
TSS
1(1
(nk 1)
11+
This translates to
RSSur
TSS
(nk1)
Combining the simplified numerator and denominator
RSS rRSSur
TSS
q
F=
RSSur
TSS
(nk 1)
Since TSS is present in both the numerator and the denominator, it cancels out.
Therefore,
F=
(RSSr RSSur) /q
, hence proven.
RSSur /(nk1)
Question 3
wagei =0 + 1 educ i+ 2 IQ+ 3 exper i+ 4 sibsi + 5 meduci + 6 feduc i+ i ,i=1, , n
. regress wage educ IQ exper sibs meduc feduc
Source
SS
df
MS
Model
Residual
20838854.5
99235939.5
6
715
3473142.42
138791.524
Total
120074794
721
166539.243
wage
Coef.
educ
IQ
exper
sibs
meduc
feduc
_cons
48.38163
4.48491
19.64647
-.789886
6.861842
11.65737
-556.0551
Std. Err.
8.329356
1.148545
3.666309
6.607232
6.21868
5.389282
146.3361
t
5.81
3.90
5.36
-0.12
1.10
2.16
-3.80
Number of obs
F( 6,
715)
Prob > F
R-squared
Adj R-squared
Root MSE
P>|t|
0.000
0.000
0.000
0.905
0.270
0.031
0.000
=
=
=
=
=
=
722
25.02
0.0000
0.1735
0.1666
372.55
64.73455
6.739835
26.84449
12.18201
19.0709
22.23808
-268.7553
wagei =556.0551+ 48.38163educ i + 4.48491 IQ+19.64647 exper i0 .789886 sibsi +6.861842 meduci +11.6573
a)
Coefficien
t statistic
t
Intercept
Educ
IQ
Exper
sibs
meduc
feduc
-3.80
5.81
3.90
5.36
-0.12
1.10
2.16
The coefficient of IQ is also statistically significant at 5% since its p-value (0.000) is less
than 5% (0.05).
Similarly, the coefficient of years of experience (exper) is statistically significant at 5%
level because the associated p-value (0.000) is less than 5% (0.05).
The coefficient of the number of siblings (sibs) is not statistically significant given that pvalue (0.905) is greater than 5% (0.05).
The coefficient of mothers education (meduc) is also not statistically significant because
its p-value (0.270) is greater than 5% (0.05).
Lastly, the coefficient of fathers education (feduc) is statistically significant because the
associated p-value (0.031) is less than 5% (0.05).
c) By default, STATA conducts an F-test, testing the null hypothesis that all the coefficients
on the independent variables are equal to zero.
Given this null hypothesis, then the restricted model is wagei =556.0551 . The F
statistic provided by STATA is F( 6, 715) = 25.02, and the P-value for the F test is
Prob > F = 0.0000. Given the P-value of 0.0000, we can reject the null hypothesis at 5%
because this value is less than 0.05 (5%). We can also reject the null at 1% because the pvalue is less than 0.01.
d) To test the joint significance of the parents education, we put the coefficients of of
meduc and feduc to zero hence our restricted model:
wagei =0 + 1 educ i+ 2 IQ+ 3 exper i+ 4 sibsi + i , i=1, ,n
Using STATA we can estimate this model to get
. regress wage educ IQ exper sibs
Source
SS
df
MS
Model
Residual
25036148.4
127680020
4
930
6259037.1
137290.344
Total
152716168
934
163507.675
wage
Coef.
educ
IQ
exper
sibs
_cons
56.81433
4.795294
17.1867
-8.148281
-467.7095
Std. Err.
7.105801
.9582977
3.117501
5.526098
126.3738
t
8.00
5.00
5.51
-1.47
-3.70
Number of obs
F( 4,
930)
Prob > F
R-squared
Adj R-squared
Root MSE
P>|t|
0.000
0.000
0.000
0.141
0.000
=
=
=
=
=
=
935
45.59
0.0000
0.1639
0.1603
370.53
70.7596
6.675971
23.30486
2.696786
-219.6985