Introduction To Econometric Solutions To Exercises (Part 2)

Download as pdf or txt
Download as pdf or txt
You are on page 1of 58
At a glance
Powered by AI
The document text discusses solutions to empirical exercises involving statistical analyses such as linear regression, time series analysis, unit root tests, cointegration tests, and GARCH modeling.

Linear regression, time series regression, unit root tests, cointegration tests, and GARCH modeling are some of the statistical analyses presented in the document text. These include analyses of relationships between variables like earnings and age/education as well as time series properties of inflation and GDP growth volatility.

Key findings include estimates of how earnings increase with age/education, explanations of small percentages of variability, evidence of persistence in inflation rates but not in inflation rate differences, and cointegration between inflation measures. GDP growth volatility also decreased in the early 1980s according to GARCH modeling.

PART TWO

Solutions to Empirical
Exercises
Chapter 3
Review of Statistics
Solutions to Empirical Exercises
1. (a)
Average Hourly Earnings, Nominal $s
Mean SE(Mean) 95% Confidence Interval
AHE
1992
11.63 0.064 11.5011.75
AHE
2004
16.77 0.098 16.5816.96
Difference SE(Difference) 95% Confidence Interval
AHE
2004
AHE
1992
5.14 0.117 4.915.37
(b)
Average Hourly Earnings, Real $2004
Mean SE(Mean) 95% Confidence Interval
AHE
1992
15.66 0.086 15.4915.82
AHE
2004
16.77 0.098 16.5816.96
Difference SE(Difference) 95% Confidence Interval
AHE
2004
AHE
1992
1.11 0.130 0.851.37
(c) The results from part (b) adjust for changes in purchasing power. These results should be used.
(d)
Average Hourly Earnings in 2004
Mean SE(Mean) 95% Confidence Interval
High School 13.81 0.102 13.6114.01
College 20.31 0.158 20.0020.62
Difference SE(Difference) 95% Confidence Interval
CollegeHigh School

6.50 0.188 6.136.87
Solutions to Empirical Exercises in Chapter 3 109

(e)
Average Hourly Earnings in 1992 (in $2004)
Mean SE(Mean) 95% Confidence Interval
High School 13.48 0.091 13.3013.65
College 19.07 0.148 18.7819.36
Difference SE(Difference) 95% Confidence Interval
CollegeHigh School

5.59 0.173 5.255.93
(f)
Average Hourly Earnings in 2004
Mean SE(Mean) 95% Confidence Interval
AHE
HS,2004

AHE
HS,1992

0.33 0.137 0.060.60
AHE
Col,2004

AHE
Col,1992

1.24 0.217 0.821.66

ColHS Gap (1992) 5.59 0.173 5.255.93
ColHS Gap (2004) 6.50 0.188 6.136.87
Difference SE(Difference) 95% Confidence Interval
Gap
2004
Gap
1992
0.91 0.256 0.411.41
Wages of high school graduates increased by an estimated 0.33 dollars per hour (with a 95%
confidence interval of 0.06 0.60); Wages of college graduates increased by an estimated 1.24
dollars per hour (with a 95% confidence interval of 0.82 1.66). The College High School gap
increased by an estimated 0.91 dollars per hour.
(g) Gender Gap in Earnings for High School Graduates
Year
m
Y
s
m
n
m
w
Y
s
w
n
w
m
Y
w
Y SE(
m
Y
w
Y )
95% CI
1992 14.57 6.55 2770 11.86 5.21 1870 2.71 0.173 2.373.05
2004 14.88 7.16 2772 11.92 5.39 1574 2.96 0.192 2.593.34
There is a large and statistically significant gender gap in earnings for high school graduates.
In 2004 the estimated gap was $2.96 per hour; in 1992 the estimated gap was $2.71 per hour
(in $2004). The increase in the gender gap is somewhat smaller for high school graduates than
it is for college graduates.
Chapter 4
Linear Regression with One Regressor
Solutions to Empirical Exercises
1. (a)

AHE = 3.32 + 0.45 Age


Earnings increase, on average, by 0.45 dollars per hour when workers age by 1 year.
(b) Bobs predicted earnings = 3.32 + 0.45 26 = $11.70
Alexiss predicted earnings = 3.32 + 0.45 30 = $13.70
(c) The R
2
is 0.02.This mean that age explains a small fraction of the variability in earnings across
individuals.
2. (a)

C
o
u
r
s
e

E
v
a
l
u
a
t
i
o
n
Beauty Index
-2 -1 0 1 2
2
3
4
5

There appears to be a weak positive relationship between course evaluation and the beauty index.
(b)

_ Course Eval = 4.00 + 0.133 Beauty. The variable Beauty has a mean that is equal to 0; the
estimated intercept is the mean of the dependent variable (Course_Eval) minus the estimated
slope (0.133) times the mean of the regressor (Beauty). Thus, the estimated intercept is equal
to the mean of Course_Eval.
(c) The standard deviation of Beauty is 0.789. Thus
Professor Watsons predicted course evaluations = 4.00 + 0.133 0 0.789 = 4.00
Professor Stocks predicted course evaluations = 4.00 + 0.133 1 0.789 = 4.105
Solutions to Empirical Exercises in Chapter 4 111


(d) The standard deviation of course evaluations is 0.55 and the standard deviation of beauty is
0.789. A one standard deviation increase in beauty is expected to increase course evaluation by
0.133 0.789 = 0.105, or 1/5 of a standard deviation of course evaluations. The effect is small.
(e) The regression R
2
is 0.036, so that Beauty explains only 3.6% of the variance in course
evaluations.
3. (a)

Ed = 13.96 0.073 Dist. The regression predicts that if colleges are built 10 miles closer
to where students go to high school, average years of college will increase by 0.073 years.
(b) Bobs predicted years of completed education = 13.96 0.073 2 = 13.81
Bobs predicted years of completed education if he was 10 miles from college = 13.96 0.073
1 = 13.89
(c) The regression R
2
is 0.0074, so that distance explains only a very small fraction of years of
completed education.
(d) SER = 1.8074 years.
4. (a)



Growth
Trade Share
0 .5 1 1.5 2
-5
0
5
10

Yes, there appears to be a weak positive relationship.
(b) Malta is the outlying observation with a trade share of 2.
(c)

Growth = 0.64 + 2.31 Tradeshare


Predicted growth = 0.64 + 2.31 1 = 2.95
(d)

Growth = 0.96 + 1.68 Tradeshare


Predicted growth = 0.96 + 1.68 1 = 2.74
(e) Malta is an island nation in the Mediterranean Sea, south of Sicily. Malta is a freight transport
site, which explains its large trade share. Many goods coming into Malta (imports into Malta)
and immediately transported to other countries (as exports from Malta). Thus, Maltas imports
and exports and unlike the imports and exports of most other countries. Malta should not be
included in the analysis.
Chapter 5
Regression with a Single Regressor:
Hypothesis Tests and Confidence Intervals
Solutions to Empirical Exercises
1. (a)

AHE = 3.32 + 0.45 Age


(0.97) (0.03)
The t-statistic is 0.45/ 0.03 = 13.71, which has a p-value of 0.000, so the null hypothesis can be
rejected at the 1% level (and thus, also at the 10% and 5% levels).
(b) 0.45 1.96 0.03 = 0.387 to 0.517
(c)

AHE = 6.20 + 0.26 Age


(1.02) (0.03)
The t-statistic is 0.26/ 0.03 = 7.43, which has a p-value of 0.000, so the null hypothesis can be
rejected at the 1% level (and thus, also at the 10% and 5% levels).
(d)

AHE = 0.23 + 0.69 Age


(1.54) (0.05)
The t-statistic is 0.69/ 0.05 = 13.06, which has a p-value of 0.000, so the null hypothesis can be
rejected at the 1% level (and thus, also at the 10% and 5% levels).
(e) The difference in the estimated |
1
coefficients is
1, 1,

College HighScool
| | = 0.69 0.26 = 0.43. The
standard error of for the estimated difference is SE
1, 1,

( )
College HighScool
| | = (0.03
2
+ 0.05
2
)
1/2
=
0.06, so that a 95% confidence interval for the difference is 0.43 1.96 0.06 = 0.32 to 0.54
(dollars per hour).
2.

_ 4.00 0.13 Course Eval Beauty = +


(0.03) (0.03)
The t-statistic is 0.13/ 0.03 = 4.12, which has a p-value of 0.000, so the null hypothesis can be rejected
at the 1% level (and thus, also at the 10% and 5% levels).
3. (a)

Ed = 13.96 0.073 Dist


(0.04) (0.013)
The t-statistic is 0.073/0.013 = 5.46, which has a p-value of 0.000, so the null hypothesis can
be rejected at the 1% level (and thus, also at the 10% and 5% levels).
(b) The 95% confidence interval is 0.073 1.96 0.013 or 0.100 to 0.047.
(c)

Ed = 13.94 0.064 Dist


(0.05) (0.018)
Solutions to Empirical Exercises in Chapter 5 113



(d)

Ed = 13.98 0.084 Dist


(0.06) (0.013)
(e) The difference in the estimated |
1
coefficients is
1, 1,

Female Male
| | = 0.064 (0.084) = 0.020.
The standard error of for the estimated difference is SE
1, 1,

( )
Female Male
| | = (0.018
2
+ 0.013
2
)
1/2
=
0.022, so that a 95% confidence interval for the difference is 0.020 1.96 0.022 or 0.022 to
0.064. The difference is not statistically different.
Chapter 6
Linear Regression with Multiple Regressors
Solutions to Empirical Exercises
1. Regressions used in (a) and (b)
Model
Regressor
a b
Beauty 0.133 0.166
Intro 0.011
OneCredit 0.634
Female 0.173
Minority 0.167
NNEnglish 0.244
Intercept 4.00 4.07

SER 0.545 0.513
R
2
0.036 0.155
(a) The estimated slope is 0.133
(b) The estimated slope is 0.166. The coefficient does not change by an large amount. Thus, there
does not appear to be large omitted variable bias.
(c) Professor Smiths predicted course evaluation = (0.166 0) +(0.011 0) + (0.634 0) (0.173
0) (0.167 1) (0.244 0) + 4.068 = 3.901
2. Estimated regressions used in question
Model
Regressor a b
dist 0.073 0.032
bytest 0.093
female 0.145
black 0.367
hispanic 0.398
incomehi 0.395
ownhome 0.152
dadcoll 0.696
cue80 0.023
stwmfg80 0.051
intercept 13.956 8.827

SER 1.81 1.84
R
2
0.007 0.279
2
R
0.007 0.277
Solutions to Empirical Exercises in Chapter 6 115
(a) 0.073
(b) 0.032
(c) The coefficient has fallen by more than 50%. Thus, it seems that result in (a) did suffer from
omitted variable bias.
(d) The regression in (b) fits the data much better as indicated by the R
2
,
2
, R and SER. The R
2
and
2
R are similar because the number of observations is large (n = 3796).
(e) Students with a dadcoll = 1 (so that the students father went to college) complete 0.696 more
years of education, on average, than students with dadcoll = 0 (so that the students father did
not go to college).
(f) These terms capture the opportunity cost of attending college. As STWMFG increases, forgone
wages increase, so that, on average, college attendance declines. The negative sign on the
coefficient is consistent with this. As CUE80 increases, it is more difficult to find a job, which
lowers the opportunity cost of attending college, so that college attendance increases. The
positive sign on the coefficient is consistent with this.
(g) Bobs predicted years of education = 0.0315 2 + 0.093 58 + 0.145 0 + 0.367 1 + 0.398
0 + 0.395 1 + 0.152 1 + 0.696 0 + 0.023 7.5 0.051 9.75 + 8.827 = 14.75
(h) Jims expected years of education is 2 0.0315 = 0.0630 less than Bobs. Thus, Jims expected
years of education is 14.75 0.063 = 14.69.
3.
Variable Mean
Standard
Deviation Units
growth 1.86 1.82 Percentage Points
rgdp60 3131 2523 $1960
tradeshare 0.542 0.229 unit free
yearsschool 3.95 2.55 years
rev_coups 0.170 0.225 coups per year
assasinations 0.281 0.494 assasinations per year
oil 0 0 01 indicator variable
(b) Estimated Regression (in table format):
Regressor Coefficient
tradeshare 1.34
(0.88)
yearsschool 0.56**
(0.13)
rev_coups 2.15*
(0.87)
assasinations 0.32
(0.38)
rgdp60 0.00046**
(0.00012)
intercept 0.626
(0.869)
SER 1.59
R
2
0.29
2
R
0.23

116 Stock/Watson - Introduction to Econometrics - Second Edition
The coefficient on Rev_Coups is 2.15. An additional coup in a five year period, reduces the
average year growth rate by (2.15/5) = 0.43% over this 25 year period. This means the GPD in
1995 is expected to be approximately .4325 = 10.75% lower. This is a large effect.
(c) The 95% confidence interval is 1.34 1.96 0.88 or 0.42 to 3.10. The coefficient is not
statistically significant at the 5% level.
(d) The F-statistic is 8.18 which is larger than 1% critical value of 3.32.
Chapter 7
Hypothesis Tests and Confidence Intervals
in Multiple Regression
Solutions to Empirical Exercises
1. Estimated Regressions
Model
Regressor a b
Age 0.45
(0.03)
0.44
(0.03)
Female 3.17
(0.18)
Bachelor 6.87
(0.19)
Intercept 3.32
(0.97)


SER 8.66 7.88
R
2
0.023 0.190
2
R 0.022 0.190
(a) The estimated slope is 0.45
(b) The estimated marginal effect of Age on AHE is 0.44 dollars per year. The 95% confidence
interval is 0.44 1.96 0.03 or 0.38 to 0.50.
(c) The results are quite similar. Evidently the regression in (a) does not suffer from important
omitted variable bias.
(d) Bobs predicted average hourly earnings = 0.44 26 3.17 0 + 6.87 0 + 3.32 = $11.44
Alexiss predicted average hourly earnings = 0.44 30 3.17 1 + 6.87 1 + 3.32 = $20.22
(e) The regression in (b) fits the data much better. Gender and education are important predictors of
earnings. The R
2
and
2
R are similar because the sample size is large (n = 7986).
(f) Gender and education are important. The F-statistic is 752, which is (much) larger than the 1%
critical value of 4.61.
(g) The omitted variables must have non-zero coefficients and must correlated with the included
regressor. From (f) Female and Bachelor have non-zero coefficients; yet there does not seem
to be important omitted variable bias, suggesting that the correlation of Age and Female and Age
and Bachelor is small. (The sample correlations are

Cor (Age, Female) = 0.03 and

Cor (Age,Bachelor) = 0.00).


118 Stock/Watson - Introduction to Econometrics - Second Edition

2.
Model
Regressor a b c
Beauty 0.13**
(0.03)
0.17**
(0.03)
0.17
(0.03)
Intro 0.01
(0.06)

OneCredit 0.63**
(0.11)
0.64**
(0.10)
Female 0.17**
(0.05)
0.17**
(0.05)
Minority 0.17**
(0.07)
0.16**
(0.07)
NNEnglish 0.24**
(0.09)
0.25**
(0.09)
Intercept 4.00**
(0.03)
4.07**
(0.04)
4.07**
(0.04)

SER 0.545 0.513 0.513
R
2
0.036 0.155 0.155
2
R
0.034 0.144 0.145
(a) 0.13 0.03 1.96 or 0.07 to 0.20
(b) See the table above. Intro is not significant in (b), but the other variables are significant.
A reasonable 95% confidence interval is 0.17 1.96 0.03 or 0.11 to 0.23.
Solutions to Empirical Exercises in Chapter 7 119

3.
Model
Regressor (a) (b) (c)
dist 0.073**
(0.013)
0.031**
(0.012)
0.033**
(0.013)
bytest 0.092**
(0.003)
0.093**
(.003)
female 0.143**
(0.050)
0.144**
(0.050)
black 0.354**
(0.067)
0.338**
(0.069)
hispanic 0.402**
(0.074)
0.349**
(0.077)
incomehi 0.367**
(0.062)
0.374**
(0.062)
ownhome 0.146*
(0.065)
0.143*
(0.065)
dadcoll 0.570**
(0.076)
0.574**
(0.076)
momcoll 0.379**
(0.084)
0.379**
(0.084)
cue80 0.024**
(0.009)
0.028**
(0.010)
stwmfg80 0.050*
(0.020)
0.043*
(0.020)
urban 0.0652
(0.063)
tuition 0.184
(0.099)
intercept 13.956**
(0.038)
8.861**
(0.241)
8.893**
(0.243)
F-statitistic
for urban and tuition

SER 1.81 1.54 1.54
R
2
0.007 0.282 0.284
2
R
0.007 0.281 0.281
(a) The groups claim is that the coefficient on Dist is 0.075 (= 0.15/2). The 95% confidence for
|
Dist
from column (a) is 0.073 1.96 0.013 or 0.099 to 0.046. The groups claim is included
in the 95% confidence interval so that it is consistent with the estimated regression.
120 Stock/Watson - Introduction to Econometrics - Second Edition

(b) Column (b) shows the base specification controlling for other important factors. Here the
coefficient on Dist is 0.031, much different than the results from the simple regression in (a);
when additional variables are added (column (c)), the coefficient on Dist changes little from the
result in (b). From the base specification (b), the 95% confidence interval for |
Dist
is 0.031
1.96 0.012 or 0.055 to 0.008. Similar results are obtained from the regression in (c).
(c) Yes, the estimated coefficients |
Black
and |
Hispanic
are positive, large, and statistically significant.
Chapter 8
Nonlinear Regression Functions
Solutions to Empirical Exercises
1. This table contains the results from seven regressions that are referenced in these answers.
Data from 2004
(1) (2) (3) (4) (5) (6) (7) (8)
Dependent Variable
AHE ln(AHE) ln(AHE) ln(AHE) ln(AHE) ln(AHE) ln(AHE) ln(AHE)
Age 0.439**
(0.030)
0.024**
(0.002)
0.147**
(0.042)
0.146**
(0.042)
0.190**
(0.056)
0.117*
(0.056)
0.160
(0.064)
Age
2
0.0021**
(0.0007)
0.0021**
(0.0007)
0.0027**
(0.0009)
0.0017
(0.0009)
0.0023
(0.0011)
ln(Age) 0.725**
(0.052)

Female Age 0.097
(0.084)
0.123
(0.084)
Female Age
2
0.0015
(0.0014)
0.0019
(0.0014)
Bachelor Age 0.064
(0.083)
0.091
(0.084)
Bachelor Age
2
0.0009
(0.0014)
0.0013
(0.0014)
Female 3.158*
*
(0.176)
0.180**
(0.010)
0.180**
(0.010)
0.180**
(0.010)
0.210**
(0.014)
1.358*
(1.230)
0.210**
(0.014)
1.764
(1.239)
Bachelor 6.865**
(0.185)
0.405**
(0.010)
0.405**
(0.010)
0.405**
(0.010)
0.378**
(0.014)
0.378**
(0.014)
0.769
(1.228)
1.186
(1.239)
Female Bachelor 0.064**
(0.021)
0.063**
(0.021)
0.066**
(0.021)
0.066**
(0.021)
Intercept 1.884
(0.897)
1.856**
(0.053)
0.128
(0.177)
0.059
(0.613)
0.078
(0.612)
0.633
(0.819)
0.604
(0.819)
0.095
(0.945)
F-statistic and p-values on joint hypotheses
(a) F-statistic on terms
involving Age
98.54
(0.00)
100.30
(0.00)
51.42
(0.00)
53.04
(0.00)
36.72
(0.00)
(b) Interaction terms
with Age and Age
2
4.12
(0.02)
7.15
(0.00)
6.43
(0.00)
SER 7.884 0.457 0.457 0.457 0.457 0.456 0.456 0.456
2
R
0.1897 0.1921 0.1924 0.1929 0.1937 0.1943 0.1950 0.1959
Significant at the *5% and **1% significance level.
122 Stock/Watson - Introduction to Econometrics - Second Edition

(a) The regression results for this question are shown in column (1) of the table. If Age increases
from 25 to 26, earnings are predicted to increase by $0.439 per hour. If Age increases from
33 to 34, earnings are predicted to increase by $0.439 per hour. These values are the same
because the regression is a linear function relating AHE and Age.
(b) The regression results for this question are shown in column (2) of the table. If Age increases
from 25 to 26, ln(AHE) is predicted to increase by 0.024. This means that earnings are predicted
to increase by 2.4%. If Age increases from 34 to 35, ln(AHE) is predicted to increase by 0.024.
This means that earnings are predicted to increase by 2.4%. These values, in percentage terms,
are the same because the regression is a linear function relating ln(AHE) and Age.
(c) The regression results for this question are shown in column (3) of the table. If Age increases
from 25 to 26, then ln(Age) has increased by ln(26) ln(25) = 0.0392 (or 3.92%). The predicted
increase in ln(AHE) is 0.725 (.0392) = 0.0284. This means that earnings are predicted to
increase by 2.8%. If Age increases from 34 to 35, then ln(Age) has increased by ln(35) ln(34) =
.0290 (or 2.90%). The predicted increase in ln(AHE) is 0.725 (0.0290) = 0.0210. This means
that earnings are predicted to increase by 2.10%.
(d) When Age increases from 25 to 26, the predicted change in ln(AHE) is
(0.147 26 0.0021 26
2
) (0.147 25 0.0021 25
2
) = 0.0399.
This means that earnings are predicted to increase by 3.99%.
When Age increases from 34 to 35, the predicted change in ln(AHE) is
(0. 147 35 0.0021 35
2
) (0. 147 34 0.0021 34
2
) = 0.0063.
This means that earnings are predicted to increase by 0.63%.
(e) The regressions differ in their choice of one of the regressors. They can be compared on the basis
of the
2
. R The regression in (3) has a (marginally) higher
2
, R so it is preferred.
(f) The regression in (4) adds the variable Age
2
to regression (2). The coefficient on Age
2
is
statistically significant ( t = 2.91), and this suggests that the addition of Age
2
is important. Thus,
(4) is preferred to (2).
(g) The regressions differ in their choice of one of the regressors. They can be compared on the basis
of the
2
. R The regression in (4) has a (marginally) higher
2
, R so it is preferred.
(h)
Regression Functions
2.4
2.45
2.5
2.55
2.6
2.65
2.7
2.75
2.8
25 27 29 31 33 35
Age
L
n
(
A
H
E
)
(4)
(3)
(2)

Solutions to Empirical Exercises in Chapter 8 123
The regression functions using Age (2) and ln(Age) (3) are similar. The quadratic regression (4)
is different. It shows a decreasing effect of Age on ln(AHE) as workers age.
The regression functions for a female with a high school diploma will look just like these, but
they will be shifted by the amount of the coefficient on the binary regressor Female. The
regression functions for workers with a bachelors degree will also look just like these, but they
would be shifted by the amount of the coefficient on the binary variable Bachelor.
(i) This regression is shown in column (5). The coefficient on the interaction term Female
Bachelor shows the extra effect of Bachelor on ln(AHE) for women relative the effect for men.
Predicted values of ln(AHE):
Alexis: 0.146 30 0.0021 30
2
0.180 1 + 0.405 1 + 0.064 1 + 0.078 = 4.504
Jane: 0.146 30 0.0021 30
2
0.180 1 + 0.405 0 + 0.064 0 + 0.078 = 4.063
Bob: 0.146 30 0.0021 30
2
0.180 0 + 0.405 1 + 0.064 0 + 0.078 = 4.651
Jim: 0.146 30 0.0021 30
2
0.180 0 + 0.405 0 + 0.064 0 + 0.078 = 4.273
Difference in ln(AHE): Alexis Jane = 4.504 4.063 = 0.441
Difference in ln(AHE): Bob Jim = 4.651 4.273 = 0.378
Notice that the difference in the difference predicted effects is 0.441 0.378 = 0.063, which
is the value of the coefficient on the interaction term.
(j) This regression is shown in (6), which includes two additional regressors: the interactions
of Female and the age variables, Age and Age
2
. The F-statistic testing the restriction that the
coefficients on these interaction terms is equal to zero is F = 4.12 with a p-value of 0.02. This
implies that there is statistically significant evidence (at the 5% level) that there is a different
effect of Age on ln(AHE) for men and women.
(k) This regression is shown in (7), which includes two additional regressors that are interactions
of Bachelor and the age variables, Age and Age
2
. The F-statistic testing the restriction that the
coefficients on these interaction terms is zero is 7.15 with a p-value of 0.00. This implies that
there is statistically significant evidence (at the 1% level) that there is a different effect of Age
on ln(AHE) for high school and college graduates.
(l) Regression (8) includes Age and Age
2
and interactions terms involving Female and Bachelor.
The figure below shows the regressions predicted value of ln(AHE) for male and females with
high school and college degrees.
2.25
2.5
2.75
3
3.25
25 26 27 28 29 30 31 32 33 34 35 36
Age
l
n
(
A
H
E
)
Female High School
Male High School
Female BA
Male BA

124 Stock/Watson - Introduction to Econometrics - Second Edition

The estimated regressions suggest that earnings increase as workers age from 2535, the range
of age studied in this sample. There is evidence that the quadratic term Age
2
belongs in the
regression. Curvature in the regression functions in particularly important for men.
Gender and education are significant predictors of earnings, and there are statistically significant
interaction effects between age and gender and age and education. The table below summarizes
the regressions predictions for increases in earnings as a person ages from 25 to 32 and 32 to 35
Gender, Education Predicted ln(AHE) at Age
Predicted Increase in ln(AHE)
(Percent per year)
25 32 35 25 to 32 32 to 35
Females, High School 2.32 2.41 2.44 1.2% 0.8%
Males, High School 2.46 2.65 2.67 2.8% 0.5%
Females, BA 2.68 2.89 2.93 3.0% 1.3%
Males, BA 2.74 3.06 3.09 4.6% 1.0%
Earnings for those with a college education are higher than those with a high school degree, and
earnings of the college educated increase more rapidly early in their careers (age 2532). Earnings
for men are higher than those of women, and earnings of men increase more rapidly early in their
careers (age 2532). For all categories of workers (men/women, high school/college) earnings
increase more rapidly from age 2532 than from 3235.
Solutions to Empirical Exercises in Chapter 8 125
2. The regressions in the table are used in the answer to this question.
Dependent Variable = Course_Eval
Regressor (1) (2) (3) (4)
Beauty 0.166**
(0.032)
0.160**
(0.030)
0.231**
(0.048)
0.090*
(0.040)
Intro 0.011
(0.056)
0.002
(0.056)
0.001
(0.056)
0.001
(0.056)
OneCredit 0.635**
(0.108)
0.620**
(0.109)
0.657**
(0.109)
0.657**
(0.109)
Female 0.173**
(0.049)
0.188**
(0.052)
0.173**
(0.050)
0.173**
(0.050)
Minority 0.167*
(0.067)
0.180**
(0.069)
0.135
(0.070)
0.135
(0.070)
NNEnglish 0.244**
(0.094)
0.243*
(0.096)
0.268**
(0.093)
0.268**
(0.093)
Age 0.020
(0.023)

Age
2
0.0002
(0.0002)

Female Beauty 0.141*
(0.063)

Male Beauty 0.141
(0.063)
Intercept 4.068**
(0.037)
3.677**
(0.550)
4.075**
(0.037)
4.075**
(0.037)
F-statistic and p-values on joint hypotheses
Age and Age
2
0.63
(0.53)

SER 0.514 0.514 0.511 0.511
2
R
0.144 0.142 0.151 0.151
Significant at the *5% and **1% significance level.
(a) See Table
(b) The coefficient on Age
2
is not statistically significant, so there is no evidence of a nonlinear
effect. The coefficient on Age is not statistically significant and the F-statistic testing whether the
coefficients on Age and Age
2
are zero does not reject the null hypothesis that the coefficients are
zero. Thus, Age does not seem to be an important determinant of course evaluations.
(c) See the regression (3) which adds the interaction term Female Beauty to the base specification
in (1). The coefficient on the interaction term is statistically significant at the 5% level. The
magnitude of the coefficient in investigated in parts (d) and (e).
(d) Recall that the standard deviation of Beauty is 0.79. Thus Professor Smiths course rating is
expected to increase by 0.231 (2 0.79) = 0.37. The 95% confidence interval for the increase
is (0.231 1.96 0.048) (2 0.79) or 0.22 to 0.51.
126 Stock/Watson - Introduction to Econometrics - Second Edition

(e) Professor Smiths course rating is expected to increase by (0.231 0.173) (2 0.79) = 0.09.
To construct the 95% confidence interval, we need the standard error for the sum of coefficients
Beauty Female Beauty
. | |

+ How to get the standard error depends on the software that you are using.
An easy way is re-specify the regression replacing Female Beauty with Male Beauty. The
resulting regression is shown in (4) in the table. Now, the coefficient on Beauty is the effect of
Beauty for females and the standard error is given in the table. The 95% confidence interval is
(0.090 1.96 0.040) (2 0.79) or 0.02 to 0.27
Solutions to Empirical Exercises in Chapter 8 127
3. This table contains the results from seven regressions that are referenced in these answers. The
Dependent Variable in all of the regressions is ED
(1) (2) (3) (4) (5)

Regressor ED ln(ED) ED ED ED
Dist 0.037**
(0.012)
0.0026**
(0.0009)
0.081**
(0.025)
0.081**
(0.025)
0.110**
(0.028)
Dist
2

0.0046*
(0.0021)
0.0047*
(0.0021)
0.0065*
(0.0022)
Tuition 0.191
(0.099)
0.014*
(0.007)
0.193*
(0.099)
0.194*
(0.099)
0.210*
(0.099)
Female 0.143**
(0.050)
0.010**
(0.004)
0.143**
(0.050)
0.141**
(0.050)
0.141**
(0.050)
Black 0.351**
(0.067)
0.026**
(0.005)
0.334**
(0.068)
0.331**
(0.068)
0.333**
(0.068)
Hispanic 0.362**
(0.076)
0.026**
(0.005)
0.333**
(0.078)
0.329**
(0.078)
0.323**
(0.078)
Bytest 0.093**
(0.003)
0.0067**
(0.0002)
0.093**
(0.003)
0.093**
(0.003)
0.093**
(0.003)
Incomehi 0.372**
(0.062)
0.027**
(0.004)
0.369**
(0.062)
0.362**
(0.062)
0.217*
(0.090)
Ownhome 0.139*
(0.065)
0.010*
(0.005)
0.143*
(0.065)
0.141*
(0.065)
0.144*
(0.065)
DadColl 0.571**
(0.076)
0.041**
(0.005)
0.561**
(0.077)
0.654**
(0.087)
0.663**
(0.087)
MomColl 0.378**
(0.083)
0.027**
(0.006)
0.378**
(0.083)
0.569**
(0.122)
0.567**
(0.122)
DadColl MomColl

0.366*
(0.164)
0.356*
(0.164)
Cue80 0.029**
(0.010)
0.002**
(0.0007)
0.026**
(0.010)
0.026**
(0.010)
0.026**
(0.010)
Stwmfg 0.043*
(0.020)
0.003*
(0.001)
0.043*
(0.020)
0.042*
(0.020)
0.042*
(0.020)
Incomehi Dist 0.124*
(0.062)
Incomehi Dist
2
0.0087
(0.0062)
Intercept 8.920**
(0.243)
2.266**
(0.017)
9.012**
(0.250)
9.002**
(0.250)
9.042**
(0.251)
(a)

(a) Dist and Dist
2
6.08
(0.002)
6.00
(0.003)
8.35
(0.000)
Interaction terms
Incomehi Dist and
Incomehi Dist
2

2.34
(0.096)
SER 1.538 0.109 1.537 1.536 1.536
2
R
0.281 0.283 0.282 0.283 0.283
Significant at the *5% and **1% significance level.
128 Stock/Watson - Introduction to Econometrics - Second Edition

(a) The regression results for this question are shown in column (1) of the table. If Dist increases
from 2 to 3, education is predicted to decrease by 0.037 years. If Dist increases from 6 to 7,
education is predicted to decrease by 0.037 years. These values are the same because the
regression is a linear function relating AHE and Age.
(b) The regression results for this question are shown in column (2) of the table. If Dist increases
from 2 to 3, ln(ED) is predicted to decrease by 0.0026. This means that education is predicted
to decrease by 0.26%. If Dist increases from 6 to 7, ln(ED) is predicted to decrease by 0.00026.
This means that education is predicted to decrease by 0.26%. These values, in percentage terms,
are the same because the regression is a linear function relating ln(ED) and Dist.
(c) When Dist increases from 2 to 3, the predicted change in ED is:
(0.081 3 + 0.0046 3
2
) (0.081 2 + 0.0046 2
2
) = 0.058.
This means that the number of years of completed education is predicted to decrease by 0.058
years. When Dist increases from 6 to 7, the predicted change in ED is:
(0.081 3 + 0.0046 7
2
) (0.081 2 + 0.0046 6
2
) = 0.021.
This means that the number of years of completed education is predicted to decrease by 0.021 years.
(d) The regression in (3) adds the variable Dist
2
to regression (1). The coefficient on Dist
2
is
statistically significant ( t = 2.26) and this suggests that the addition of Dist
2
is important. Thus,
(4) is preferred to (1).
(e)
Regression Functions
14.95
15
15.05
15.1
15.15
15.2
15.25
15.3
15.35
15.4
0 2 4 6 8 10
Distance (10's of Miles)
Y
e
a
r
s

o
f

E
d
u
c
a
t
i
o
n
Regression (3)
Regression (1)

(i) The quadratic regression in (3) is steeper for small values of Dist than for larger values.
The quadratic function is essentially flat when Dist = 10. The only change in the regression
functions for a white male is that the intercept would shift. The functions would have the
same slopes.
Solutions to Empirical Exercises in Chapter 8 129
(ii) The regression function becomes positively sloped for Dist > 10. There are only 44 of the
3796 observations with Dist > 10. This is approximately 1% of the sample. Thus, this part of
the regression function is very imprecisely estimated.
(f) The estimated coefficient is 0.366. This is the extra effect of education above and beyond
the sepearted MomColl and DadColl effects, when both mother and father attended college.
(g) (i) This the coefficient on DadColl, which is 0.654 years
(ii) This the coefficient on MomColl, which is 0.569 years
(iii) This is the sum of the coefficients on DadColl, MomColl and the interaction term. This is
0.654 + 0.569 0.366 = 0.857 years.
(h) Regression (5) adds the interaction of Incomehi and the distance regressors, Dist and Dist
2
.
The implied coefficients on Dist and Dist
2
are:
Students who are not high income (Incomehi = 0)

ED = 0.110Dist + 0.0065 Dist


2
+ other factors
High Income Students (Incomehi = 1)

ED = (0.110 + 0.124) Dist + (0.0065 0.0087) Dist


2
+ other factors
= 0.013 Dist 0.0012Dist
2
+ other factors.
The two estimated regression functions are plotted below for someone with characteristics given
in (5), but with Incomehi = 1 and with Incomehi = 0. When Incomehi = 1, the regression function
is essentially flat, suggesting very little effect of Dist and ED. The F-statistic testing that the
coefficients on the interaction terms Incomehi Dist and Incomehi Dist
2
are both equal to zero
has a p-value of 0.09. Thus, the interaction effects are significant at the 10% but not 5%
signficance level.
Regression Functions
14.8
14.9
15
15.1
15.2
15.3
15.4
15.5
15.6
15.7
0 2 4 6 8 10
Distance (10's of Miles)
Y
e
a
r
s

o
f

E
d
u
c
a
t
i
o
n
Incomehi=0
Incomehi = 1

130 Stock/Watson - Introduction to Econometrics - Second Edition

(i) The regression functions shown in (4) and (5) show the nonlinear effect of distance on years
of education. The effect is statistically significant. In (4) the effect of changing Dist from 20
miles to 30 miles, reduces years of completed education by 0.081 (3 2) + 0.0047
(3
2
2
2
) = 0.0575 years, on average. The regression in (5) shows a slightly effect from non-
high income student, but essentially no effect for high income students.
4. This table contains results from regressions that are used in the answers.
Dependent variable = Growth
Regressor (1) (2) (3) (4) (5)
TradeShare 2.331**
(0.596)
2.173**
(0.555)
1.288*
(0.516)
1.830
(1.341)
5.334
(3.231)
TradeShare
2
7.776
(4.299)
TradeShare
3
2.366
(1.433)
YearsSchool 0.250**
(0.076)

ln(YearsSchool) 1.031**
(0.201)
2.183**
(0.383)
2.404**
(0.653)
2.136**
(0.408)
Rev_coups 2.318*
(0.919)
2.356
(0.924)
2.039*
(0.950)
Assassinations 0.255
(0.323)
0.266
(0.329)
0.102
(0.365)
ln(RGDP60) 1.642**
(0.429)
1.664
(0.433)
1.588**
(0.453)
TradeShare ln(YearsSchool) 0.398
(0.783)

Intercept 0.370
(0.585)
0.416
(0.468)
11.785**
(3.279)
11.662**
(3.303)
12.904**
(3.168)
F-statistic and p-values on joint hypotheses
Rev_coups and Assasinations 3.38
(0.04)
2.20
(0.12)
TradeShare
2
and TradeShare
3


SER 1.685 1.553 1.389 1.399 1.388
2
R
0.211 0.329 0.464 0.456 0.464
Significant at the *5% and **1% significance level.
Solutions to Empirical Exercises in Chapter 8 131
(a)

G
r
o
w
t
h
Years of School
0 5 10
-5
0
5
10

The plot suggests a nonlinear relation. This explains why the linear regression of Growth
on YearsSchool

in (1) does not fit as the well as the nonlinear regression in (2).
(b) Predicted change in Growth using (1): 0.250 (6 4) = 0.50
Predicted change in Growth using (2): 1.031 [ln(6) ln(4)] = 0.42
(c) See Table
(d) The t-statistic for the interaction term TradeShare ln(YearsSchool) is 0.398/0.783 = 0.51,
so the coefficient is not significant at the 10% level.
(e) This is investigated in (5) by adding TradeShare
2
and TradeShare
3
to the regression. The
F-statistic suggests that the coefficients on these regressors are not significantly different from 0.
Chapter 9
Assessing Studies Based on Multiple Regression
Solutions to Empirical Exercises
1.
Data from 2004
(1) (2) (3) (4) (5) (6) (7) (8)
Dependent Variable
AHE ln(AHE) ln(AHE) ln(AHE) ln(AHE) ln(AHE) ln(AHE) ln(AHE)
Age 0.439**
(0.030)
0.024**
(0.002)
0.147**
(0.042)
0.146**
(0.042)
0.190**
(0.056)
0.117*
(0.056)
0.160
(0.064)
Age
2
0.0021**
(0.0007)
0.0021**
(0.0007)
0.0027**
(0.0009)
0.0017
(0.0009)
0.0023
(0.0011)
ln(Age) 0.725**
(0.052)

Female Age 0.097
(0.084)
0.123
(0.084)
Female Age
2
0.0015
(0.0014)
0.0019
(0.0014)
Bachelor Age 0.064
(0.083)
0.091
(0.084)
Bachelor Age
2
0.0009
(0.0014)
0.0013
(0.0014)
Female 3.158**
(0.176)
0.180**
(0.010)
0.180**
(0.010)
0.180**
(0.010)
0.210**
(0.014)
1.358*
(1.230)
0.210**
(0.014)
1.764
(1.239)
Bachelor 6.865**
(0.185)
0.405**
(0.010)
0.405**
(0.010)
0.405**
(0.010)
0.378**
(0.014)
0.378**
(0.014)
0.769
(1.228)
1.186
(1.239)
Female Bachelor 0.064**
(0.021)
0.063**
(0.021)
0.066**
(0.021)
0.066**
(0.021)
Intercept 1.884
(0.897)
1.856**
(0.053)
0.128
(0.177)
0.059
(0.613)
0.078
(0.612)
0.633
(0.819)
0.604
(0.819)
0.095
(0.945)
F-statistic and p-values on joint hypotheses
(a) F-statistic on
terms involving Age
98.54
(0.00)
100.30
(0.00)
51.42
(0.00)
53.04
(0.00)
36.72
(0.00)
(b) Interaction terms
with Age and Age
2
4.12
(0.02)
7.15
(0.00)
6.43
(0.00)
SER 7.884 0.457 0.457 0.457 0.457 0.456 0.456 0.456
2
R
0.1897 0.1921 0.1924 0.1929 0.1937 0.1943 0.1950 0.1959
Significant at the *5% and **1% significance level.
Solutions to Empirical Exercises in Chapter 9 133
Data from 1992
(1) (2) (3) (4) (5) (6) (7) (8)
Dependent Variable
AHE ln(AHE) ln(AHE) ln(AHE) ln(AHE) ln(AHE) ln(AHE) ln(AHE)
Age 0.461**
(0.028)
0.027**
(0.002)
0.157**
(0.041)

0.156**
(0.041)

0.120*
(0.057)

0.138*
(0.054)

0.104
(0.065)
Age
2
0.0022**
(0.0006)
0.0022**
(0.0007)
0.0015
(0.0010)
0.0020*
(0.0009)
0.0013*
(0.0011)
ln(Age) 0.786**
(0.052)

Female Age 0.088
(0.083)
0.077
(0.083)
Female Age
2
0.0017
(0.0013)
0.0016
(0.0014)
Bachelor Age 0.037
(0.084)
0.046
(0.083)
Bachelor Age
2
0.0004
(0.0014)
0.0006
(0.0014)
Female 2.698**
(0.152)
0.167**
(0.010)
0.167**
(0.010)
0.167**
(0.010)
0.200**
(0.013)
1.273**
(1.212)
0.200**
(0.013)
1.102
(1.213)
Bachelor 5.903**
(0.169)
0.377**
(0.010)
0.377**
(0.010)
0.377**
(0.010)
0.340**
(0.014)
0.340**
(0.014)
0.365**
(1.227)
0.504
(1.226)
Female Bachelor 0.085**
(0.020)
0.079**
(0.020)
0.086**
(0.020)
0.080**
(0.02)
Intercept 0.815
(0.815)
1.776**
(0.054)
0.099
(0.178)
0.136
(0.608)
0.119
(0.608)
0.306
(0.828)
0.209
(0.780)
0.617
(0.959)
F-statistic and p-values on joint hypotheses
(a) F-statistic on
terms involving Age
115.93
(0.00)
118.89
(0.00)
62.51
(0.00)
65.17
(0.00)
45.71
(0.00)
(b) Interaction terms
with Age and Age
2
9.04
(0.00)
4.80
(0.01)
7.26
(0.00)
SER 6.716 0.437 0.437 0.437 0.437 0.436 0.436 0.436
2
R
0.1946 0.1832 0.1836 0.1841 0.1858 0.1875 0.1866 0.1883
Significant at the *5% and **1% significance level.
134 Stock/Watson - Introduction to Econometrics - Second Edition
(a) (1) Omitted variables: There is the potential for omitted variable bias when a variable is
excluded from the regression that (i) has an effect on ln(AHE) and (ii) is correlated with a
variable that is included in the regression. There are several candidates. The most important
is a workers Ability. Higher ability workers will, on average, have higher earnings and are
more likely to go to college. Leaving Ability out of the regression may lead to omitted
variable bias, particularly for the estimated effect of education on earnings. Also omitted
from the regression is Occupation. Two workers with the same education (a BA for example)
may have different occupations (accountant versus 3
rd
grade teacher) and have different
earnings. To the extent that occupation choice is correlated with gender, this will lead to
omitted variable bias. Occupation choice could also be correlated with Age. Because the data
are a cross section, older workers entered the labor force before younger workers (35 year-
olds in the sample were born in 1969, while 25 year-olds were born in 1979), and their
occupation reflects, in part, the state of the labor market when they entered the labor force.
(2) Misspecification of the functional form: This was investigated carefully in exercise 8.1.
There does appear to be a nonlinear effect of Age on earnings, which is adequately captured
by the polynomial regression with interaction terms.
(3) Errors-in-variables: Age is included in the regression as a proxy for experience. Workers
with more experience are expected to earn more because their productivity increases with
experience. But Age is an imperfect measure of experience. (One worker might start his
career at age 22, while another might start at age 25. Or, one worker might take a year off to
start a family, while another might not). There is also potential measurement error in AHE as
these data are collected by retrospective survey in which workers in March 2005 are asked
about their average earnings in 2004.
(4) Sample selection: The data are full-time workers only, so there is potential for sample-
selection bias.
(5) Simultaneous causality: This is unlikely to be a problem. It is unlikely that AHE affects Age
or gender.
(6) Inconsistency of OLS standard errors: Heteroskedastic robust standard errors were used in
the analysis, so that heteroskedasticity is not a concern. The data are collected, at least
approximately, using i.i.d. sampling, so that correlation across the errors is unlikely to be a
problem.
(b) Results for 1988 are shown in the table above. Using results from (8), several conclusions were
reached in E8.1(l) using the data from 2004. These are summarized in the table below, and are
followed by a similar table for the 1998 data.
Results using (8) from the 2004 Data
Gender, Education
Predicted Value of ln(AHE)
at Age
Predicted Increase in ln(AHE)
(Percent per year)
25 32 35 25 to 32 32 to 35
Females, High School 2.32 2.41 2.44 1.2% 0.8%
Males, High School 2.46 2.65 2.67 2.8% 0.5%
Females, BA 2.68 2.89 2.93 3.0% 1.3%
Males, BA 2.74 3.06 3.09 4.6% 1.0%

Solutions to Empirical Exercises in Chapter 9 135
Results using (8) from the 1998 Data
Gender, Education
Predicted Value of ln(AHE)
at Age
Predicted Increase in ln(AHE)
(Percent per year)
25 32 35 25 to 32 32 to 35
Females, High School 2.28 2.42 2.39 2.0 0.9
Males, High School 2.42 2.65 2.70 3.2 1.9
Females, BA 2.64 2.86 2.86 3.3 0.2
Males, BA 2.70 3.01 3.09 4.4 2.6
Based on the 2004 data E81.1(l) concluded: Earnings for those with a college education are
higher than those with a high school degree, and earnings of the college educated increase more
rapidly early in their careers (age 2532). Earnings for men are higher than those of women, and
earnings of men increase more rapidly early in their careers (age 2532). For all categories of
workers (men/women, high school/college) earnings increase more rapidly from age 2532 than
from 3235.
All of these conclusions continue to hold for the 1998 data (although the precise values for the
differences change somewhat.)
2. We begin by discussing the internal and external validity of the results summarized in E8.2.
Internal Validity
1. Omitted variable bias. It is always possible to think of omitted variables, but the relevant question is
whether they are likely to lead to substantial omitted variable bias. Standard examples like instructor
diligence, are likely to be major sources of bias, although this is speculation and the next study on this
topic should address these issues (both can be measured). One possible source of OV bias is the
omission of the department. French instructors could well be more attractive than chemists, and
if French is more fun (or better taught) than chemistry then the department would belong in the
regression, and its omission could bias the coefficient on Beauty. It is difficult to say whether this
is a major problem or not, one approach would be to put in a full set of binary indicators for the
department and see if this changed the results. We suspect this is not an important effect, however
this must be raised as a caveat.
2. Wrong functional form. Interactions with Female showed some evidence of nonlinearity. It would
be useful to see if Beauty
2
enters the regression. (We have run the regression, and the t-statistic on
Beauty
2
is 1.15.)
3. Measurement error in the regressors. The Beauty variable is subjectively measured so that it will
have measurement error. This is plausibly a case in which the measurement error is more or less
random, reflecting the tastes of the six panelists. If so, then the classical measurement error model, in
which the measured variable is the true value plus random noise, would apply. But this model implies
that the coefficient is biased downso the actual effect of Beauty would be greater than is implied by
the OLS coefficient. This suggests that the regressions understate the effect of Beauty.
136 Stock/Watson - Introduction to Econometrics - Second Edition
4. Sample selection bias. The only information given in this exam about the sample selection method
is that the instructors have their photos on their Web site. Suppose instructors who get evaluations
below 3.5 are so embarrassed that they dont put up their photos, and suppose there is a large effect
of Beauty. Then, of the least attractive instructors, the only ones that will put up their photos are those
with particular teaching talent and commitment, sufficient to overcome their physical appearance.
Thus the effect of physical appearance will be attenuated because the error term will be correlated
with Beauty (low values of Beauty means there must be a large value of u, else the photo wouldnt
be posted.) This story, while logically possible, seems a bit far-fetched, and whether an instructor
puts up his or her photo is more likely to be a matter of departmental policy, whether the department
has a helpful webmaster and someone to take their photo, etc. So sample selection bias does not seem
(in my judgment) to be a potentially major threat.
5. Simultaneous causality bias. There is an interesting possible channel of simultaneous causality, in
which good course evaluations improve an instructors self-image which in turn means they have a
more resonant, open, and appealing appearanceand thus get a higher grade on Beauty. Against this,
the panelists were looking at the Web photos, not their conduct in class, and were instructed to focus
on physical features. So for the Beauty variable as measured, this effect is plausibly large.

External Validity
The question of external validity is whether the results for UT-Austin in 20002002 can be generalized to,
say, Harvard or Cal-State University Northridge (CSUN) in 2005. The years are close, so the question
must focus on differences between students and the instructional setting.
1. Are UT-Austin students like Harvard (or CSUN) students? Perhaps Beauty matters more or less
to Harvard (CSUN) students?
2. Do the methods of instruction differ? For example, if beauty matters more in small classes (where
you can see the instructor better) and if the distribution of class size at UT-Austin and Harvard
(CSUN) were substantially different, then this would be a threat to external validity.
Policy Advice
As an econometric consultant, the question is whether this represents an internally and externally valid
estimate of the causal effect of Beauty, or whether the threats to internal and/or external validity are
sufficiently severe that the results should be dismissed as unreliable for the purposes of the Dean.
A correct conclusion is one that follows logically from the systematic discussion of internal and external
validity.
We would be surprised if the threats to internal and external validity above are sufficiently important,
in a quantitative sense, to change the main finding from E8.2 that the effect of Beauty is positive and
quantitatively large. So our advice, solely econometric consultants, would be that implementing a policy
of affirmative action for attractive people (all else equal, higher the better-looking) would, in expectation,
improve course evaluations.
This said, a good econometric policy advisor always has some advice about the next study. One thing that
next study could do is focus on institutions like yours. (UT-Austin students and professors might be
different from students at Harvard or CSUN), and collect data on some potential omitted variables
(department offering the course, etc.).
A very different study would be to do a randomized controlled experiment that would get directly at the
policy question. Some department heads would be instructed to assign their most attractive teachers to
the largest introductory courses (treatment group), others would be instructed to maintain the status quo
Solutions to Empirical Exercises in Chapter 9 137
(control group). The study would assess whether there is an improvement in evaluation scores (weighted
by class size) in the treatment group. A positive result would indicate that this treatment results in an
increase in customer satisfaction.
Finally, some thoughts that were out of bounds for this question, but would be relevant and important
to raise in the report of an econometric consultant to the Dean. First, the Course Evaluation score is just
a student evaluation, not a measure of what students actually learned or how valuable the course was;
perhaps an assessment of the value of the course, five years hence, would produce a very different effect of
Beauty, and that is arguably a more important outcome than the end-of-semester evaluation (this could be
thought of as a threat to external validity, depending on how one defines the Deans goal). Second, academic
output is not solely teaching, and there is no reason at all that the results here would carry over to an analysis
of research output, or even graduate student advising and teaching (the data are only for undergrad courses);
indeed, the sign might be the opposite for research. Third, the econometric consultant could raise the
question of whether Beauty has the same moral status as gender or race, even if it does not have the same
legal status as a legally protected class; answering this question is outside the econometric consultants area
of expertise, but it is a legitimate question to raise and to frame so that others can address it.
3. (a) (1) Omitted variables: This is potentially important. For example, students from wealthier
families might live closer to colleges and have higher average years of completed education.
The estimated regression attempts to control for family wealth using the variables Incomehi
and Ownhome, but these are imperfect measures of wealth.
(2) Misspecification of the function form: This was investigated in the Chapter 8 empirical
exercise.
(3) Errors-in-variables: It has already been noted that Incomehi and Ownhome are imperfect
measures of family wealth. Years of completed education may also be imprecisely measured
as the data description makes clear.
(4) Sample Selection: This is a random sample of high school seniors, so sample selection within
this population is unlikely to be a problem. However, when considering external validity,
these results are not likely to hold for the general population of high school students, some
of which may drop out before their senior year.
(5) Simultaneous causality: The argument here would be that parents who want to send their
children to college may locate closer to a college. This is possible, but the effect is likely
to be small.
(6) Inconsistency of standard errors: Heteroskedasticity-robust standard errors were used. The
data represent a random sample so that correlation across the error terms is not a problem.
Thus, the standard errors should be consistent.
(b) The table below shows the results for the non-West regions analyzed earlier along with the
results for the West.
The sample from the West contains 943 observations, compared to 3796 in the non-West sample.
This means that the standard errors for the estimated coefficients in the West will be roughly
twice as large as the standard errors in the non-West sample. (The ratio of the standard errors will
be roughly

.)
Non West
West
n
n

Because the samples are independent, the standard errors for the estimated difference in the
coefficients can be calculated as
2 2

( ) ( ) ( ) .
Non West West Non West West
SE SE SE | | | |

= +
For example, the standard error for the difference between the non-West and West coefficients on
Dist is
2 2
(0.028) (0.045) 0.053 + = .
138 Stock/Watson - Introduction to Econometrics - Second Edition
The coefficients on Dist and Dist
2
in the West are very similar to the values for the non-West. This
means that the estimated regression coefficients are similar. The interaction terms Incomehi Dist
and Incomehi Dist
2
look different. In the non-West, the estimated regression function for high
income students was essentially flat (E8.3(h)), while the estimated regression coefficient in the
West for students with Incomehi = 1 is very similar to the regression function for students with
Incomehi = 0. However, the coefficients on the interaction terms for the West sample are
imprecisely estimated and are not statistically different from the non-West sample. Indeed, the only
statistically significant coefficient across the two samples is the coefficient on Bytest. The
difference is 0.093 0.073 = 0.20, which has a standard error of
2 2
0.003 0.006 0.0067. + =
Solutions to Empirical Exercises in Chapter 9 139
Regession Results for Non-Western and Western States
Non-West West
Dist 0.110**
(0.028)
0.092*
(0.045)
Dist
2
0.0065*
(0.0022)
0.0041
(0.0031)
Tuition 0.210*
(0.099)
0.523*
(0.242)
Female 0.141**
(0.050)
0.051**
(0.100)
Black 0.333**
(0.068)
0.067**
(0.182)
Hispanic 0.323**
(0.078)
0.196**
(0.115)
Bytest 0.093**
(0.003)
0.073**
(0.006)
Incomehi 0.217*
(0.090)
0.407*
(0.169)
Ownhome 0.144*
(0.065)
0.199*
(0.127)
DadColl 0.663**
(0.087)
0.441**
(0.144)
MomColl 0.567**
(0.122)
0.283**
(0.262)
DadColl MomColl 0.356
(0.164)
0.142
(0.330)
Cue80 0.026**
(0.010)
0.045**
(0.023)
Stwmfg 0.042*
(0.020)
0.031*
(0.044)
Incomehi Dist 0.124*
(0.062)
0.005*
(0.090)
Incomehi Dist
2
0.0087
(0.0062)
0.0000
(0.0057)
Intercept 9.042**
(0.251)
9.227**
(0.524)
F-statistics (p-values) and measures of fit
(a) Dist and Dist
2
8.35
(0.000)
2.66
(0.070)
(b) Interaction terms
Incomehi Dist and
Incomehi Dist
2

2.34
(0.096)
0.01
(0.993)
SER 1.536 1.49
2
R
0.283 0.218
n 3796 943
Significant at the *5% and **1% significance level.
Chapter 10
Regression with Panel Data
Solutions to Empirical Exercises
1.
(1) (2) (3) (4)
shall 0.443
**

(0.048)
0.368**
(0.035)
0.0461
*

(0.019)
0.0280
(0.017)
incar_rate 0.00161
**

(0.00018)
0.00007
(0.00009)
0.0000760
(0.000090)
density 0.0267
(0.014)
0.172
**

(0.085)
0.0916
(0.076)
avginc 0.00121
(0.0073)
0.00920
(0.0059)
0.000959
(0.0064)
pop 0.0427
**

(0.0031)
0.0115
(0.0087)
0.00475
(0.0079)
pb1064 0.0809
**

(0.020)
0.104
**

(0.018)
0.0292
(0.023)
pw1064 0.0312
**

(0.0097)
0.0409
**

(0.0051)
0.00925
(0.0079)
pm1029 0.00887
(0.012)
0.0503
**

(0.0064)
0.0733
**

(0.016)
Intercept 6.135
**

(0.019)
2.982
**

(0.61)
3.866
**

(0.38)
3.766
**

(0.47)
State Effects No No Yes Yes
Time Effects No No No Yes
F-Statistics and p-values testing exclusion of groups of variables
State Effects 210.38
(0.00)
309.29
(0.00)
Time Effects 13.90
(0.00)
2
R
0.09 0.56 0.94 0.95
(a) (i) The coefficient is 0.368, which suggests that shall-issue laws reduce violent crime by 36%.
This is a large effect.
(ii) The coefficient in (1) is 0.443; in (2) it is 0.369. Both are highly statistically significant.
Adding the control variables results in a small drop in the coefficient.
Solutions to Empirical Exercises in Chapter 10 141
(iii) Attitudes towards guns and crime. Quality of schools. Quality of police and other crime-
prevention programs.
(b) In (3) the coefficient on shall falls to 0.046, a large reduction in the coefficient from (2).
Evidently there was important omitted variable bias in (2). The 95% confidence interval for |
Shall

is now 0.086 to 0.007 or 0.7% to 8.6%. The state effects are jointly statistically significant,
so this regression seems better specified than (2).
(c) The coefficient falls further to 0.028. The coefficient is insignificantly different from zero. The
time effects are jointly statistically significant, so this regression seems better specified than (3).
(d) This table shows the coefficient on shall in the regression specifications (1)(4). To save space,
coefficients for variables other than shall are not reported.
Dependent Variable = ln(rob)
(1) (2) (3) (4)
shall 0.773
**

(0.070)
0.529**
(0.051)
0.008
(0.026)
0.027
(0.025)
F-Statistics and p-values testing exclusion of groups of variables
State Effects 190.47
(0.00)
243.39
(0.00)
Time Effects 12.39
(0.00)
Dependent Variable = ln(mur)
shall 0.473
**

(0.049)
0.313**
(0.036)
0.061
*

(0.027)
0.015
(0.027)
F-Statistics and p-values testing exclusion of groups of variables
State Effects 88.22
(0.00)
106.69
(0.00)
Time Effects 9.73
(0.00)
The quantative results are similar to the results using violent crimes: there is a large estimated
effect of concealed weapons laws in specifications (1) and (2). This effect is spurious and is due
to omitted variable bias as specification (3) and (4) show.
(e) There is potential two-way causality between this years incarceration rate and the number of
crimes. Because this years incarceration rate is much like last years rate, there is a potential
two-way causality problem. There are similar two-way causality issues relating crime and shall.
(f) The most credible results are given by regression (4). The 95% confidence interval for |
Shall

is +1% to 6.6%. This includes |
Shall
= 0. Thus, there is no statistically significant evidence that
concealed weapons laws have any effect on crime rates. The interval is wide, however, and
includes values as large as 6.6%. Thus, at a 5% level the hypothesis that |
Shall
= 0.066
(so that the laws reduce crime by 6.6%) cannot be rejected.
142 Stock/Watson - Introduction to Econometrics - Second Edition
2.
Regressor (1) (2) (3)
sb_useage 0.00407
***

(0.0012)
0.00577
***

(0.0012)
0.00372
***

(0.0011)
speed65 0.000148
(0.00041)
0.000425
(0.00033)
0.000783
*
(0.00042)
speed70 0.00240
***

(0.00047)
0.00123
***

(0.00033)
0.000804
**

(0.00034)
ba08 0.00192
***

(0.00036)
0.00138
***

(0.00037)
0.000822
**

(0.00035)
drinkage21 0.0000799
(0.00099)
0.000745
(0.00051)
0.00113
**

(0.00054)
lninc 0.0181
***

(0.0011)
0.0135
***

(0.0014)
0.00626
(0.0039)
age 0.00000722
(0.00016)
0.000979
**

(0.00038)
0.00132
***

(0.00038)
State Effects No Yes Yes
Year Effects No No Yes
0.544 0.874 0.897
(a) The estimated coefficient on seat belt useage is positive and statistically significant. One the face
of it, this suggests that seat belt useage leads to an increase in the fatality rate.
(b) The results change. The coefficient on seat belt useage is now negative and the coefficient is
statistically significant. The estimated value of |
SB
= 0.00577, so that a 10% increase in seat belt
useage (so that sb_useage increases by 0.10) is estimated to lower the fatality rate by .000577
fatalities per million traffic miles. States with more dangerous drving conditions (and a higher
fatality rate) also have more people wearing seat belts. Thus (1) suffers from omitted variable
bias.
(c) The results change. The estimated value of |
SB
= 0.00372.
(d) The time effects are statistically significant the F-statistic = 10.91 with a p-value of 0.00.
The results in (3) are the most reliable.
(e) A 38% increase in seat belt useage from 0.52 to 0.90 is estimated to lower the fatality rate by
0.00372 0.38 = 0.0014 fatalities per million traffic miles. The average number of traffic miles
per year per state in the sample is 41,447. For a state with the average number of traffic miles, the
number of fatalities prevented is 0.0014 41,447 = 58 fatalities.
(f) A regression yields

_ sb useage = 0.206 primary + 0.109 secondary +


(0.021) (0.011)

(speed65, speed70, ba08, drinkage21, logincome, age, time effects, state effects)

Solutions to Empirical Exercises in Chapter 10 143
where the coefficients on the other regressors are not reported to save space. The coefficients on
primary and secondary are positive and significant. Primary enforcement is estimated to increase
seat belt useage by 20.6% and secondary enforcement is estimated to increase seat belt useage by
10.9%.
(g) This results in an estimated increase in seatbelt useage of 0.2060.109 = 0.094 or 9.4% from (f).
This is predicted to reduce the fatality rate by 0.00372 0.094 = 0.00035 fatalities per million
traffic miles. The data set shows that there were 63,000 million traffic miles in 1997 in New
Jersey, the last year for which data is available. Assuming the same number of traffic miles in
2000 yields 0.00035 63,000 = 22 lives saved.
Chapter 11
Regression with a Binary Dependent Variable
Solutions to Empirical Exercises
1.
(1) (2) (3)
Linear
Probability
Linear
Probability Probit
Smkban 0.078**
(0.009)
0.047**
(0.009)
0.159**
(0.029)
Age 0.0097**
(0.0018)
0.035**
(0.007)
Age
2
0.00013**
(0.00002)
0.00047**
(0.00008)
Hsdrop 0.323**
(0.019)
1.142**
(0.072)
Hsgrad 0.233**
(0.013)
0.883**
(0.060)
Colsome 0.164**
(0.013)
0.677**
(0.061)
Colgrad 0.045**
(0.012)
0.235**
(0.065)
Black 0.028
(0.016)
0.084
(0.053)
Hispanic 0.105**
(0.014)
0.338**
(0.048)
Female 0.033**
(0.009)
0.112**
(0.028)
Intercept 0.014
(0.041)
1.735**
(0.053)
F-statistic and p-values on joint hypotheses
Education
indicators
140.09
(0.00)
464.90
(0.00)
Significant at the 5% * or 1% ** level.
Solutions to Empirical Exercises in Chapter 11 145
(a) Estimated probability of smoking (mean of smoker)

p SE

( ) p
All Workers 0.242 0.004
No Smoking Ban 0.290 0.007
Smoking Ban 0.212 0.005
(b) From model (1), the difference in 0.078 we a standard error of 0.009. The resulting t-statistic
is 8.66, so the coefficient is statistically significant.
(c) From model (2) the estimated difference is 0.047, smaller than the effect in model (1).
Evidently (1) suffers from omitted variable bias in (1). That is, smkban may be correlated with
the education/race/gender indicators or with age. For example, workers with a college degree are
more likely to work in an office with a smoking ban than high-school dropouts, and college
graduates are less likely to smoke than high-school dropouts.
(d) The t-statistic is 5.27, so the coefficient is statistically significant at the 1% level.
(e) The F-statistic has a p-value of 0.00, so the coefficients are significant. The omitted education
status is Masters degree or higher. Thus the coefficients show the increase in probability
relative to someone with a postgraduate degree. For example, the coefficient on Colgrad is 0.045,
so the probability of smoking for a college graduate is 0.045 (4.5%) higher than for someone
with a postgraduate degree. Similarly, the coefficient on HSdrop is 0.323, so the probability
of smoking for a college graduate is 0.323 (32.3%) higher than for someone with a postgraduate
degree. Because the coefficients are all positive and get smaller as educational attainment
increases, the probability of smoking falls as educational attainment increases.
(f) The coefficient on Age
2
is statistically significant. This suggests a nonlinear relationship between
age and the probability of smoking. The figure below shows the estimated probability for a
white, non-Hispanic male college graduate with no workplace smoking ban.

146 Stock/Watson - Introduction to Econometrics - Second Edition
2 (a) See the table above.
(b) The t-statistic is 5.47, very similar to the value for the linear probability model.
(c) The F-statistic is significant at the 1% level, as in the linear probability model.
(d) To calculate the probabilities, take the estimation results from the probit model to calculate , z
and calculate the cumulative standard normal distribution at , z i.e., . ( ) ( ) Prob smoke z = u The
probability of Mr. A smoking without the workplace ban is 0.464 and the probability of smoking
with the workplace bans is 0.401. Therefore the workplace bans would reduce the probability of
smoking by 0.063 (6.3%).
(e) To calculate the probabilities, take the estimation results from the probit model to calculate , z
and calculate the cumulative standard normal distribution at , z i.e., . ( ) ( ) Prob smoke z = u The
probability of Ms. B smoking without the workplace ban is 0.143 and the probability of smoking
with the workplace ban is 0.110. Therefore the workplace bans would reduce the probability of
smoking by .033 (3.3%).
(f) For Mr. A, the probability of smoking without the workplace ban is 0.449 and the probability
of smoking with the workplace ban is 0.402. Therefore the workplace ban would have a
considerable impact on the probability that Mr. A would smoke. For Ms. B, the probability
of smoking without the workplace ban is 0.145 and the probability of smoking with the
workplace ban is 0.098. In both cases the probability of smoking declines by 0.047 or 4.7%.
(Notice that this is given by the coefficient on smkban, 0.047, in the linear probability model.)
(g) The linear probability model assumes that the marginal impact of workplace smoking bans on
the probability of an individual smoking is not dependent on the other characteristics of the
individual. On the other hand, the probit models predicted marginal impact of workplace
smoking bans on the probability of smoking depends on individual characteristics. Therefore,
in the linear probability model, the marginal impact of workplace smoking bans is the same for
Mr. A and Mr. B, although their profiles would suggest that Mr. A has a higher probability of
smoking based on his characteristics. Looking at the probit models results, the marginal impact
of workplace smoking bans on the odds of smoking are different for Mr. A and Ms. B, because
their different characteristics are incorporated into the impact of the laws on the probability of
smoking. In this sense the probit model is likely more appropriate.
Are the impacts of workplace smoking bans large in a real-world sense? Most people might
believe the impacts are large. For example, in (d) the reduction on the probability is 6.3%.
Applied to a large number of people, this translates into a 6.3% reduction in the number of
people smoking.
(h) An important concern is two-way causality. Do companies that impose a smoking ban have
fewer smokers to begin with? Do smokers seek employment with employers that do not have a
smoking ban? Do states with smoking bans already have more
or fewer smokers than states without smoking bans?
Solutions to Empirical Exercises in Chapter 11 147
3. Answers are provided to many of the questions using the linear probability models. You can also
answer these questions using a probit or logit model. Answers are based on the following table:
Dependent Variable
Insured Insured Insured Healthy Healthy Healthy Any
Limitation
Regressor
(1) (2) (3) (4) (5) (6) (7)
selfemp 0.128**
(0.015)
0.174**
(0.014)
0.210**
(0.063)
0.010
(0.007)
0.020*
(0.008)
0.015
(0.008)
0.010
(0.012)
age 0.010**
(0.003)
0.001
(0.001)
0.0006
(0.0017)
0.002
(0.002)
0.003
(0.002)
age
2
0.00008*
(0.00003)
0.010**
(0.003)
0.00003
(0.00002)
0.000
(0.000)
0.000
(0.000)
age selfemp 0.000
(0.000)

deg_ged 0.151**
(0.027)
0.151**
(0.027)
0.045*
(0.020)
0.061*
(0.024)
deg_hs 0.254**
(0.016)
0.254**
(0.016)
0.099**
(0.012)
0.012
(0.012)
deg_ba 0.316**
(0.017)
0.316**
(0.017)
0.122**
(0.013)
0.042**
(0.014)
deg_ma 0.335**
(0.018)
0.335 **
(0.018)
0.128**
(0.015)
0.078**
(0.018)
deg_phd 0.366**
(0.026)
0.366 **
(0.025)
0.138**
(0.018)
0.084**
(0.027)
deg_oth 0.288**
(0.020)
0.287**
(0.020)
0.115**
(0.014)
0.049**
(0.017)
familysz 0.017**
(0.003)
0.017**
(0.003)
0.001
(0.002)
0.016**
(0.002)
race_bl 0.028*
(0.013)
0.028*
(0.013)
0.022*
(0.009)
0.035**
(0.010)
race_ot 0.048*
(0.023)
0.048**
(0.023)
0.029
(0.015)
0.046
(0.016)
reg_ne 0.037**
(0.012)
0.037**
(0.012)
0.006
(0.008)
0.046**
(0.011)
reg_mw 0.053**
(0.012)
0.053**
(0.012)
0.012
(0.008)
0.008
(0.011)
reg_so 0.003
(0.011)
0.004
(0.011)
0.001
(0.008)
0.007
(0.010)
male 0.037**
(0.008)
0.037**
(0.008)
0.015**
(0.005)
0.005
(0.007)
married 0.136**
(0.010)
0.136 **
(0.010)
0.001
(0.007)
0.017**
(0.009)
Intercept 0.817
(0.004)
0.299**
(0.054)
0.296**
(0.054)
0.927**
(0.003)
0.953**
(0.031)
0.902**
(0.035)
0.071
(0.044)
Significant at the 5% * or 1% ** level.
148 Stock/Watson - Introduction to Econometrics - Second Edition
(a) Probability of being insured

p SE

( ) p
All Workers 0.802 0.004
Self Employed 0.689 0.014
Not Self Employed 0.817 0.004
The self-employed are 12.8% less likely to have health insurance. This is a large number.
It is statistically significant: from (1) in the table the difference is significant at the 1% level.
(b) From specification (2), the result is robust to adding additional control variables. Indeed, after
controlling for other factors, the difference increases to 17.4%
(c) See specification (2). There is evidence of nonlinearity (Age
2
is significant in the regression). The
plot below shows the effect of Age on the probability of being insured for a self-employed white
married male with a BA and a family size of four from the northeast. (The profile for others will
look the same, although it will be shifted up or down.) The probability if being insured increases
with Age over the range 2065 years.

(d) Specification (3) adds an interaction of Age and selfemp. Its coefficient is not statistically
significant, and this suggests that the effect of selfemp does not depend on Age. (Note: this
answer is specific to the linear probability model. In the probit model, even without an
interaction, the effect of selfemp depends on the level of the probability of being insured, and this
probability depends on Age.)
(e) This is investigated in specifications (4)(7). The effect of selfemp on health status or Any
Limitation is small and not and statistically significant. This result obtains when the regression
controls for Age or for a full set of control variables.
Solutions to Empirical Exercises in Chapter 11 149
There are potential problems with including healthy on the right hand side of the model
because of adverse selection problems. It is possible that only those less healthy
individuals pursue health insurance, perhaps through their employer. This causes a self-
selection problem that more healthy individuals might (a) choose to be self-employed or
(b) choose not to obtain health insurance. While the evidence suggests that there might
not be a strong correlation between health status and self-employment, the adverse
selection concerns still exist.
Chapter 12
Instrumental Variables Regression
Solutions to Empirical Exercises
1. This table shows the OLS and 2SLS estimates. Values for the intercept and coefficients on Seas are
not shown.
Regressor OLS 2SLS
ln(Price) 0.639
(0.073)
0.867
(0.134)
Ice 0.448
(0.135)
0.423
(0.135)
Seas and intercept Not Shown Not Shown

First Stage F-statistic 183.0
(a) See column the table above. The estimated elasticity is 0.639 with a standard error of 0.073.
(b) A positive demand error will shift the demand curve to the right. This will increase the
equilibrium quantity and price in the market. Thus ln(Price) is positively correlated with the
regression error in the demand model. This means that the OLS coefficient will be positively
biased.
(c) Cartel shifts the supply curve. As the cartel strengthens, the supply curve shifts in, reducing
supply and increasing price and profits for the cartels members. Thus, Cartel is relevant. For
Cartel to be a valid instrument it must be exogenous, that is, it must be unrelated to the factors
affecting demand that are omitted from the demand specification (i.e., those factors that make up
the error in the demand model.) This seems plausible.
(d) The first stage F-statistic is 183.0. Cartel is not a weak instrument.
(e) See the table. The estimated elasticity is 0.867 with a standard error of 0.134. Notice that the
estimate is more negative than the OLS estimate, which is consistent with the OLS estimator
having a positive bias.
(f) In the standard model of monopoly, a monopolist should increase price if the demand elasticity
is less than 1. (The increase in price will reduce quantity but increase revenue and profits.) Here,
the elasticity is less than 1.
2. (Results using full dataset)
Estimation Method Regressor
OLS IV IV
Morekids 5.387
(0.087)
6.313
(1.275)
5.821
(1.246)
Additional Regressors Intercept Intercept Intercept, agem1,
black, hispan, othrace

First Stage F-Statistic 1238.2 1280.9
Solutions to Empirical Exercises in Chapter 12 151
(a) The coefficient is 5.387, which indicates that women with more than 2 children work 5.387
fewer weeks per year than women with 2 or fewer children.
(b) Both fertility and weeks worked are choice variables. A women with a positive labor supply
regression error (a women who works more than average) may also be a woman who is less
likely to have an additional child. This would imply that Morekids is positively correlated with
the regression error, so that the OLS estimator of |
Morekids
is positively biased.
(c) The linear regression of morekids on samesex (a linear probability model) yields

morekids = 0.346 + 0.066samesex


(0.001) (0.002)
so that couples with samesex = 1 are 6.6% more likely to have an additional child that couples
with samesex = 0. The effect is highly significant (t-statistic = 35.2)
(d) Samesex is random and is unrelated to any of the other variables in the model including the error
term in the labor supply equation. Thus, the instrument is exogenous. From (c), the first stage
F-statistic is large (F = 1238) so the instrument is relevant. Together, these imply that samesex
is a valid instrument.
(e) No, see the answer to (d).
(f) See column (2) of the table. The estimated value of |
Morekids
= 6.313.
(g) See column (3) of the table. The results do not change in an important way. The reason is that
samesex is unrelated to agem1, black, hispan, othrace, so that there is no omitted variable bias
in IV regression in (2).
3. (Results using small dataset)
Estimation Method Regressor
OLS IV IV
Morekids 6.001
(0.254)
6.033
(3.758)
5.781
(3.645)
Additional Regressors Intercept Intercept Intercept, agem1, black,
hispan, othrace

First Stage F-Statistic 143.2 150.9
(a) The coefficient is 5.387, which indicates that women with more than 2 children work 5.387
fewer weeks per year than women with 2 or fewer children.
(b) Both fertility and weeks worked are choice variables. A women with a positive labor supply
regression error (a women who works more than average) may also be a woman who is less
likely to have an additional child. This would imply that Morekids is positively correlated with
the regression error, so that the OLS estimator of |
Morekids
is positively biased.
(c) The linear regression of morekids on samesex (a linear probability model) yields

morekids = 0.344 + 0.067samesex


(0.004) (0.006)
so that couples with samesex = 1 are 6.7% more likely to have an additional child that couples
with samesex = 0. The effect is highly significant (t-statistic = 12.0)
152 Stock/Watson - Introduction to Econometrics - Second Edition
(d) Samesex is random and is unrelated to any of the other variables in the model including the error
term in the labor supply equation. Thus, the instrument is exogenous. From (c), the first stage
F-statistic is large (F = 143) so the instrument is relevant. Together, these imply that samesex is
a valid instrument.
(e) No, see the answer to (d).
(f) See column (2) of the table. The estimated value of |
Morekids
= 6.033.
(g) See column (3) of the table. The results do not change in an important way. The reason is that
samesex is unrelated to agem1, black, hispan, othrace, so that there is no omitted variable bias
in IV regression in (2).
4. (a) |
2

= 1.26, SE( |
2

) = 0.44, the 95% confidence interval is 0.30 to 2.02


(b) F-statistic = 4.3, which suggests a weak instrument problem.
(c) These are the values of | that are less than 3.84 in the figure below. The 95% confidence interval
is 2.46 < | < 1.68.
F-Statistic from regression of Y |X onto Z.

(d) The confidence interval in (a) is not reliable because of the weak instrument problem.
The confidence interval in (c) is reliable even when instruments are weak.
Chapter 13
Experiments and Quasi-Experiments
Solutions to Empirical Exercises
1. The following table provides answers to (a)(c)
Dependent Variable = Call_Back
Regressor (1) (2) (3) (4)
Black 0.032**
(0.008)
0.038**
(0.012)
0.023*
(0.011)
Female Black 0.008
(0.012)

High 0.014
(0.008)
0.023
(0.012)
High Black 0.018
(0.016)
Intercept 0.097**
(0.006)
0.097**
(0.006)
0.073**
(0.005)
0.084**
(0.008)
Significant at the 5% * and 1% ** level.
(a) From (1) in the table, the call-back rate for whites is 0.097 and the call-back for blacks is
0.097 0.032 = 0.065. The difference is 0.032 which is statistically significant at the 1% level.
These number implies that 9.7% of resumes with white-sounding names generated a call back.
Only 6.5% of resumes with black-sounding names generated a call back. The difference is large.
(b) From (2) in the table, the call-back rate for male blacks 0.097 0.038 = 0.059, and for female
blacks is 0.097 0.038 + 0.008 = 0.067. The difference is 0.008, which is not significant at the
5% level.
(c) From (3) in the table, the call-back rate for low-quality resumes is 0.073 and the call-back rate
for high-quality resumes is 0.073 + 0.014 = 0.087. The difference is 0.014, which is not
significant at the 5% level. From (4) the (high-quality)(low-quality) difference for whites is
0.023 and for blacks is 0.023 0.018 = 0.005; the black-white difference is 0.018 which is not
statistically significant at the 5% level.
(d) The following table shows estimated means of other characteristics for black and white sounding
names. There are only two significant difference in the mean values: the call-back rate (the
variable of interest) and computer skills (for which black-named resumes had a slightly higher
fraction that white-named resumes). Thus, there is no evidence of non-random assignment.
154 Stock/Watson - Introduction to Econometrics - Second Edition

Black-Sounding Names White-Sounding Names Black-White Difference
Variable n X
se ( ) X
n X
se ( ) X

b w
X X ( ) se
b w
X X
t-stat
ofjobs 2435 3.658 1.219 2435 3.664 1.219 0.006 0.035 0.18
yearsexp 2435 7.830 5.011 2435 7.856 5.079 0.027 0.145 0.18
honors 2435 0.051 0.221 2435 0.054 0.226 0.003 0.006 0.45
volunteer 2435 0.414 0.493 2435 0.409 0.492 0.006 0.014 0.41
military 2435 0.102 0.303 2435 0.092 0.290 0.009 0.008 1.11
empholes 2435 0.446 0.497 2435 0.450 0.498 0.004 0.014 0.29
workinschool 2435 0.561 0.496 2435 0.558 0.497 0.003 0.014 0.20
email 2435 0.480 0.500 2435 0.479 0.500 0.001 0.014 0.06
computerskills 2435 0.832 0.374 2435 0.809 0.393 0.024 0.011 2.17
specialskills 2435 0.327 0.469 2435 0.330 0.470 0.003 0.013 0.21
eoe 2435 0.291 0.454 2435 0.291 0.454 0.000 0.013 0.00
manager 2435 0.152 0.359 2435 0.152 0.359 0.000 0.010 0.04
supervisor 2435 0.077 0.267 2435 0.077 0.267 0.000 0.008 0.00
secretary 2435 0.333 0.471 2435 0.333 0.471 0.000 0.014 0.03
offsupport 2435 0.119 0.323 2435 0.119 0.323 0.000 0.009 0.00
salesrep 2435 0.151 0.358 2435 0.151 0.358 0.000 0.010 0.00
retailsales 2435 0.168 0.374 2435 0.168 0.374 0.000 0.011 0.00
req 2435 0.787 0.409 2435 0.787 0.409 0.000 0.012 0.00
expreq 2435 0.435 0.496 2435 0.435 0.496 0.000 0.014 0.00
comreq 2435 0.125 0.331 2435 0.125 0.331 0.000 0.009 0.00
educreq 2435 0.107 0.309 2435 0.107 0.309 0.000 0.009 0.00
compreq 2435 0.437 0.496 2435 0.437 0.496 0.000 0.014 0.03
orgreq 2435 0.073 0.260 2435 0.073 0.260 0.000 0.007 0.00
manuf 2435 0.083 0.276 2435 0.083 0.276 0.000 0.008 0.00
transcom 2435 0.030 0.172 2435 0.030 0.172 0.000 0.005 0.00
bankreal 2435 0.085 0.279 2435 0.085 0.279 0.000 0.008 0.00
trade 2435 0.214 0.410 2435 0.214 0.410 0.000 0.012 0.00
busservice 2435 0.268 0.443 2435 0.268 0.443 0.000 0.013 0.00
othservice 2435 0.155 0.362 2435 0.155 0.362 0.000 0.010 0.00
missind 2435 0.165 0.371 2435 0.165 0.371 0.000 0.011 0.00
chicago 2435 0.555 0.497 2435 0.555 0.497 0.000 0.014 0.00
high 2435 0.502 0.500 2435 0.502 0.500 0.000 0.014 0.00
female 2435 0.775 0.418 2435 0.764 0.425 0.011 0.012 0.88
college 2435 0.723 0.448 2435 0.716 0.451 0.007 0.013 0.51
call_back 2435 0.064 0.246 2435 0.097 0.295 0.032 0.008 4.11
Solutions to Empirical Exercises in Chapter 13 155
2. (a) (i) A person will trade if he received good A but prefers good B or he received good B and
prefers good A. 50% receive good A, of these (100-X)% prefer good B; 50% receive good B,
of these X% prefer good A. Let x = X/100. Thus Expected Fraction Traded = 0.5 (1 x) +
0.5x = 0.5.
(ii) Use X = 100%;
(iii) Use X = 50%
(b)(d)
Answers are based on the following table
Dependent Variable = Trade
All Traders Dealers Non-Dealers
Regressor (1) (2) (3) (4) (5) (6) (7) (8)
Goodb 0.018
(0.078)
0.021
(0.117)
0.564
(0.100)

Years_trade > 10 0.046
(0.128)
0.297
(0.140)
Trades Per Month >
8


Intercept 0.338
(0.039)
0.329
(0.057)
0.445
(0.058)
0.457
(0.085)
0.230
(0.049)
0.200
(0.079)
0.220
(0.055)
0.169
(0.050)
(b) From (1) the fraction of trades is 0.338; the t-statistic for H
o
: p = 0.5 is t = (0.338 0.5)/0.039 =
4.15, so the fraction is statistically significantly different from p = 0.5. From (2) the fraction of
recipients of good A who traded for good B was 0.329 and the fraction of recipients of good B
who traded for good A was 0.329 + 0.018 = 0.347. Both are statistically different from 0.5 at the
1% level. The fraction of good B recipients who traded was not statistically significantly different
from the fraction of good A recipients.
(c) The story is different for dealers (see (3) and (4)). The fraction of trades in 0.446, which is not
statistically significantly different from 0.5. This is true for recipients of good A and good B.
(d) Specification (5)(8) use data on non-dealers. (5)(6) repeat the analysis from parts (b) and (c).
Specification (7) adds an indicator variable that is equal to 1 if the trader has been active in the
market for more than 10 years (approximately 25% of the traders). Specification (8) adds an
indicator variable that is equal to 1 if the trader reports making more than 8 trades per month
(approximately 25% of the traders). Long-term traders (7) are no different than short-term
traders: the coefficient on Years_trade > 10 is small and not statistically significant. Participants
who engage in more than 8 trades per month are different from those who dont: the coefficient
on Trades Per Month > 8 is large and statistically significant. The fraction of these traders who
traded their endowment was 0.297 + 0.169 = 0.466, which is not statistically significantly
different from 0.5.
Chapter 14
Introduction to Time Series Regression and Forecasting
Solutions to Empirical Exercises
1. (a)(c)
Mean Standard Deviation
Quarterly Growth Rate Unscaled
[ln(GPD
t
/GDP
t 1
]

0.0083

0.0092
Quarterly Growth Rate
Percentage Points at an annual rate
[400 ln(GPD
t
/GDP
t 1
]

3.30

3.68
(d) Estimated Autocorrelations (unit free)
Lag Autocorrelation
1 0.29
2 0.17
3 0.03
4 0.02
2. (a)

t
Y A = 0.0058 + 0.301AY
t 1
,
2
R = 0.086, SER = 0.0088
(0.0010) (0.076)
(b)

t
Y A = 0.0052 + 0.272AY
t 1
+ 0.096AY
t 2
,
2
R = 0.090, SER = 0.0088
(0.0010) (0.081) (0.086)
(c) Minimized value shown in BOLD
Lag BIC AIC
1 9.4234 9.4564
2 9.4063 9.4557
3 9.3834 9.4494
4 9.3598 9.4423
3. Regressing AY
t
on Y
t 1
, AY
t 1
, time trend, and intercept yields a t-statistic on Y
t 1
that is
t = 2.51. The 10% critical value is 3.12, so that DF t-statistic is not significant at the
10% level.
4. The QLR F-statistic is 1.26 (maximized value at 1966:01). This is less than the 10% critical value
of 5.00, so the null of stability is not rejected.
Solutions to Empirical Exercises in Chapter 14 157
5. (a)

t
Y A = 0.0060 + 0.270AY
t 1
+ 0.0018AR
t 1
0.0037AR
t 2
+ 0.0098AR
t 3
0.0030AR
t 4

(0.0010) (0.081) (0.0009) (0.0010) (0.0007) (0.0008)


2
R = 0.175, SER = 0.0084
The
2
R has increased from 0.086 to 0.175.
(b) The F-statistic is 6.93 with a p-value of 0.00.
(c) The QLR F-statistic is 4.80 (maximum in 1974:3). This is larger than the 1% critical value of
4.53 suggesting instability in the ADL(1,4) model.
6.
Selected Pseudo Out-Of-Sample Forecast Results
(percentage points at an annual rate)
Model Forecast Error Mean (SE) RMSFE
AR(1) 0.22
(0.26)
2.06
ADL(1, 4) 0.51
(0.29)
2.29
Naive 0.07
(0.28)
2.18
The AR and ADL models show a negative bias, but neither is statistically significant at the 5% level.
The AR model has the smallest RMSFE.
7. (a) Table 14.3 Extended Dataset (Sample Period 1932:12002:12)
Regressors (1) (2) (3)
Excess Return
t 1
0.098
(0.061)
0.102
(0.061)
0.099
(0.058)
Excess Return
t 2
0.040
(0.057)
0.029
(0.054)
Excess Return
t 3
0.098
(0.054)
Excess Return
t 4
0.006
(0.046)
Intercept 0.524
(0.181)
0.543
(0.186)
0.590
(0.199)
F-statistic on all coefficients (p-value) 2.61
(0.11)
1.51
(0.22)
1.41
(0.23)
2
R
0.009 0.009 0.016
158 Stock/Watson - Introduction to Econometrics - Second Edition

(b)
(1) (2) (3)
Estimation Period 1932:1 2002:12 1932:1 2002:12 1932:1 1982:12
Regressors
Excess Return
t 1
0.093
(0.135)
0.109
(0.124)
0.128
(0.07)
Excess Return
t 2
0.088
(0.153)

Aln(dividend yield
t 1
) 0.005
(0.132)
0.007
(0.119)

Aln(dividend yield
t 2
) 0.048
(0.129)

ln(dividend yield
t 1
) 0.020
(0.11)
Intercept 0.526
(0.203)
0.559
(0.228)
6.759
(3.623)
F-statistic on all coefficients (p-value) 1.34
(0.26)
0.81
(0.52)

2
R
0.007 0.007 0.022
(c) The ADF statistic from the regression using 1 lagged first difference and a constant term is
2.78. This is smaller (more negative) than the 10% critical value, but not more negative than
the 5% critical value.
(d)
Model RMSFE
Zero Forecast 4.28
Constant Forecast 4.25
ADL(1, 1) 4.29
(e) No. The in-sample regressions in Tables 14.3 and 14.7 suggest that the coefficients on lagged
excess returns and lags of the first difference of the dividend yield are insignificant. The dividend
yield is persistent, and this makes statistical inference in (3) of Table 14.7 difficult. In the
pseudo-out-of sample experiment the ranking of the forecasts (Constant, Zero, ADL(1,1)) is the
same as reported in the box.
Chapter 15
Estimation of Dynamic Causal Effects
Solutions to Empirical Exercises
1. (a) Mean = 0.27; Standard Deviation = 0.94
(b) O
t
is the greater of zero or the percentage point difference between oil prices at date t and their
maximum value during the past year. Thus O
t
> 0, and O
t
= 0 if the date t is not greater than the
maximum value over the past year.
1952 1957 1962 1967 1972 1977 1982 1987 1992 1997 2002
0.00
0.05
0.10
0.15
0.20
0.25
0.30

(c) m was chosen using 0.75T
0.33
rounded to the nearest integer; m = 6 in this case. The estimated
coefficients and 95% confidence intervals are shown in the figure in part (e).
(d) The F-statistic testing that all 19 coefficients are equal to zero is 1.78, with a p-value 0.02;
the coefficients are significant at the 5% but not the 1% level.
(e) The cumulative multipliers show a persistent and large decrease in industrial production
following an increase in oil prices above their previous 12 month peak price. Specifically
a 100% increase in oil prices is leads to an estimated 15% decline in industrial production
after 18 months.
160 Stock/Watson - Introduction to Econometrics - Second Edition
Dynamic Effect of Oil on IP Growth
(a) Estimated Dynamic Multipliers and 95% Confidence Interval
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
-7.5
-5.0
-2.5
0.0
2.5
5.0
(b) Estimated Cumulative Multipliers and 95% Confidence Interval
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
-28
-24
-20
-16
-12
-8
-4
0
4

(f) In this case O
t
is not exogenous and the results summarized in (e) are not reliable.
2. (a) Mean of t
CPI
= 4.10. Mean of t
PCED
= 3.67
(b) Mean of Y = 0.44. The mean of Y is the difference in the means because Y = t
CPI
t
PCED
.
(c) Y = t
CPI
t
PCED

, so E(Y) = E(t
CPI
) E(t
PCED
).
(d) Y
t
= |
0
+ u
t
, so E(Y
t
) = |
0
+ E(u
t
) = |
0
because E(u
t
) = 0.
(e)
0

0.44 | = and SE(


0

| ) = 0.09, so a 95% confidence interval is 0.44 1.96 .09. m was chosen


using 0.75T
0.33
rounded to the nearest integer; m = 6 in this case.
(f) Yes. |
0
represents the difference in the mean inflation rates (see (c) and (d)), and (e) suggests that
|
0
is between 0.26 and 0.62 percentage points.
Chapter 16
Additional Topics in Time Series Regression
Solutions to Empirical Exercises
1. (a) The Granger-causality F-statistics and p-values are:
(i) lags of AR in AY equation: 6.93 (0.00)
(ii) lags of AR in AY equation: 4.05 (0.00)
Thus, the null of no-Granger-causality is rejected in both equations.
(b) The table below shows the values of the BIC and AIC computed using equation (16.4) in the text
Minimized Value Shown in Bold
Lag BIC AIC
1 10.0044 10.1034
2 10.0709 10.2358
3 10.0546 10.2854
4 10.0038 10.3006
5 10.0405 10.4034
6 9.9434 10.3722
7 9.8791 10.3739
8 9.7976 10.3583
BIC suggests using 2 lags. AIC suggests using 5 lags.
2. The table below summarizes the pseudo-out-of sample forecasting performance
Summary statistics for forecast errors 1990:22004:4
Model Mean (SE) Standard Deviation RMSFE
Nave 0.06
(0.31)
2.35 2.35
AR 0.32
(0.27)
2.11 2.13
VAR 0.73
(0.30)
2.27 2.38
Both the AR and VAR have negative mean forecast errors over the out-of-sample period. The VAR
bias is statistically significant. The AR model has the smallest RMSFE.
162 Stock/Watson - Introduction to Econometrics - Second Edition
3. ADF statistic (from E14.3) is t = 2.51, which was not significant at the 10% level. The DF-GLS
statistic is t = 2.24, which can be compared to the 10% critical value (from Table 16.1) of 2.57.
Again, the statistic is not significant at the 10% level.
4. (a) and (b)
ADF t-statistic DF-GLS t-statistic
t
CPI
2.57
+
1.97*
t
PCE
2.27 2.27*
t
CPI
t
PCE
5.01** 5.00**
+
Significant at the 10% level
* Significant at the 5% level
** Significant at the 1% level
The levels of inflation t
CPI
and t
PCE

are quite persistent. The ADF statistic for t
CPI

has a p-value
of 0.10, and the p-value for t
PCE

is larger than 0.10. The more-powerful DF-GLS tests yield p-
values less than 0.05 but greater than 0.01. The difference in the inflation rates t
CPI
t
PCE

is far
less persistent: the p-values for both tests is less than 0.01.
(c) t
CPI
and t
PCE

can be viewed, at least approximately, as I(1) processes (which not rejected at the
1% level in (b) although it is rejected at the 5% level using the DF-GLS test). On the other
hand t
CPI
t
PCE

seems to be well described as an I(0) process (the I(1) null rejected in (b)). Thus
t
CPI
and t
PCE

can be viewed as cointegrated with a value of u that is equal to one.
(d) The EG-ADF test yields a t-statistic of 5.56, which is more negative than the 1% critical value of
3.96. Thus the null of no-cointegration is rejected. Estimating the cointegrating coefficient by
DOLS yields

u = 1.13 with a standard error of 0.05 (using a lag truncation parameter of m = 6 for
the Newey-West HAC estimator). This value is slightly greater than 1.0, the value imposed above.
5. (a) The estimated model is

t
Y A = 0.006 + 0.319AY
t 1
,
(0.001) (0.072)

2

t
o = 0.000001 + 0.141
2
1 t
u

+ 0.848
2
1 t
o


(0.000002) (0.083) (0.080)
(Note: your estimates may differ slightly from those presented above depending on the software that
you used to estimate the model.)
Solutions to Empirical Exercises in Chapter 16 163
(b)
GARCH(1,1) Bands For GDP Growth
1955 1959 1963 1967 1971 1975 1979 1983 1987 1991 1995 1999 2003
-0.03
-0.02
-0.01
0.00
0.01
0.02
0.03
0.04

(c) The GARH standard deviations bands narrow considerably in the early 1980s, providing
evidence of a decrease in volatility.

You might also like