Chapters 7, 8, and 15: Regression with Qualitative Information,
Heteroskedasticity, and IV
Econometrics for E&BE
Metrics for E&BE: Chapters 7+8+15 1 / 25
Table of Contents
1 Q1: Interaction with Dummy Variables
2 Q2: Interaction with Categorical Variables
3 Q3: Linear Probability Model
4 Q4: Instrumental Variable Approach I
5 Q5: Instrumental Variable Approach II
Metrics for E&BE: Chapters 7+8+15 2 / 25
Q1: Interaction with Dummy Variables Q2: Interaction with Categorical Variables Q3: Linear Probability Model Q4: Instrumental Variable Approach I Q5: Instrumen
Question 1, Sub-point (i), I of II
Using data on high school student performance in the Netherlands, the following equation was
estimated:
G
Ô PA = 5.14 + 0.10 · hsize - 0.01 · hsize2 - 0.26 · f emale
(0.03) (0.02) (0.003) (0.02)
- 0.85 · mi g r at ion + 0.31 · f emal e · mi g r at ion
(0.06) (0.09)
n = 4, 137, R2 = 0.0858. The variables:
GPA is the average grade of the student on a ten-point scale;
hsize is size of the student’s high school graduating class, in hundreds;
female is a gender dummy variable;
migration is a dummy variable equal to one for students with a migration background, and
zero otherwise.
(i) Is there strong evidence that hsize2 should be included in the model? From this equation,
what is the optimal high school size?
Metrics for E&BE: Chapters 7+8+15 3 / 25
Q1: Interaction with Dummy Variables Q2: Interaction with Categorical Variables Q3: Linear Probability Model Q4: Instrumental Variable Approach I Q5: Instrumen
Question 1, Sub-point (i), II of II, Sub-point (ii)
Recall:
G
Ô PA = 5.14 + 0.10 · hsize - 0.01 · hsize2 - 0.26 · f emale
(0.03) (0.02) (0.003) (0.02)
- 0.85 · mi g r at ion + 0.31 · f emal e · mi g r at ion
(0.06) (0.09)
n = 4, 137, R2 = 0.0858.
(ii) Holding hsize fixed, what is the estimated difference in GPA between non-migration back-
ground females and non-migration background males? How statistically significant is this esti-
mated difference?
Metrics for E&BE: Chapters 7+8+15 4 / 25
Q1: Interaction with Dummy Variables Q2: Interaction with Categorical Variables Q3: Linear Probability Model Q4: Instrumental Variable Approach I Q5: Instrumen
Question 1, Sub-point (iii)
Recall:
G
Ô PA = 5.14 + 0.10 · hsize - 0.01 · hsize2 - 0.26 · f emale
(0.03) (0.02) (0.003) (0.02)
- 0.85 · mi g r at ion + 0.31 · f emal e · mi g r at ion
(0.06) (0.09)
2
n = 4, 137, R = 0.0858.
(iii) What is the estimated difference in GPA between non-migration background males and
migration background males? Test the null hypothesis that there is no difference between their
scores, against the alternative that there is a difference.
Metrics for E&BE: Chapters 7+8+15 5 / 25
Q1: Interaction with Dummy Variables Q2: Interaction with Categorical Variables Q3: Linear Probability Model Q4: Instrumental Variable Approach I Q5: Instrumen
Question 1, Sub-points (iv)
Recall:
G
Ô PA = 5.14 + 0.10 · hsize - 0.01 · hsize2 - 0.26 · f emale
(0.03) (0.02) (0.003) (0.02)
- 0.85 · mi g r at ion + 0.31 · f emal e · mi g r at ion
(0.06) (0.09)
2
n = 4, 137, R = 0.0858.
Metrics for E&BE: Chapters 7+8+15 6 / 25
Q1: Interaction with Dummy Variables Q2: Interaction with Categorical Variables Q3: Linear Probability Model Q4: Instrumental Variable Approach I Q5: Instrumen
Question 1, Sub-points (v)
Recall:
G
Ô PA = 5.14 + 0.10 · hsize - 0.01 · hsize2 - 0.26 · f emale
(0.03) (0.02) (0.003) (0.02)
- 0.85 · mi g r at ion + 0.31 · f emal e · mi g r at ion
(0.06) (0.09)
2
n = 4, 137, R = 0.0858.
(v) What is the estimated difference in GPA between females with a migration background and
females without a migration background. What would you need to do to test whether the
difference in (iv) is statistically significant?
Metrics for E&BE: Chapters 7+8+15 7 / 25
Q1: Interaction with Dummy Variables Q2: Interaction with Categorical Variables Q3: Linear Probability Model Q4: Instrumental Variable Approach I Q5: Instrumen
Question 2, Sub-point (i)
Using data on wages, the following equation was estimated:
log(wa
Û g e) = 0.389 - 0.227 · f emal e + 0.082 · educ - 0.0056 · f emale · educ
(0.119) (0.168) (0.008) (0.0131)
0.029 · e x per - 0.00058 · e x per 2 + 0.032 · t enur e - 0.00059 · t enur e2
(0.005) (0.00011) (0.007) (0.00024)
n = 526, R2 = 0.441. The variables:
wage is the hourly wage rate earned by the person;
female equals one if the person is female, and zero otherwise;
educ are the years of schooling;
exper measures the person’s total work experience (in years);
tenure measures the person’s work experience with their current employer (in years).
(i) What is the effect of experience on wages in this model? Interpret the relevant coefficients.
Metrics for E&BE: Chapters 7+8+15 8 / 25
Q1: Interaction with Dummy Variables Q2: Interaction with Categorical Variables Q3: Linear Probability Model Q4: Instrumental Variable Approach I Q5: Instrumen
Question 2, Sub-point (ii)
Recall:
log(wa
Û g e) = 0.389 - 0.227 · f emal e + 0.082 · educ - 0.0056 · f emale · educ
(0.119) (0.168) (0.008) (0.0131)
0.029 · e x per - 0.00058 · e x per 2 + 0.032 · t enur e - 0.00059 · t enur e2
(0.005) (0.00011) (0.007) (0.00024)
n = 526, R2 = 0.441.
(ii) Interpret the coefficient of f emal e. Comment on the statistical and economic significance of
the variable.
Metrics for E&BE: Chapters 7+8+15 9 / 25
Q1: Interaction with Dummy Variables Q2: Interaction with Categorical Variables Q3: Linear Probability Model Q4: Instrumental Variable Approach I Q5: Instrumen
Question 2, Sub-points (iii)
Recall:
log(wa
Û g e) = 0.389 - 0.227 · f emal e + 0.082 · educ - 0.0056 · f emale · educ
(0.119) (0.168) (0.008) (0.0131)
0.029 · e x per - 0.00058 · e x per 2 + 0.032 · t enur e - 0.00059 · t enur e2
(0.005) (0.00011) (0.007) (0.00024)
n = 526, R2 = 0.441.
Metrics for E&BE: Chapters 7+8+15 10 / 25
Q1: Interaction with Dummy Variables Q2: Interaction with Categorical Variables Q3: Linear Probability Model Q4: Instrumental Variable Approach I Q5: Instrumen
Question 2, Sub-points (iv)
Recall:
log(wa
Û g e) = 0.389 - 0.227 · f emal e + 0.082 · educ - 0.0056 · f emale · educ
(0.119) (0.168) (0.008) (0.0131)
0.029 · e x per - 0.00058 · e x per 2 + 0.032 · t enur e - 0.00059 · t enur e2
(0.005) (0.00011) (0.007) (0.00024)
n = 526, R2 = 0.441.
(iv) What is the gender differential in wages when educ = 0? How about when educ = 12.5
Compare your answers. What do you conclude?
Metrics for E&BE: Chapters 7+8+15 11 / 25
Q1: Interaction with Dummy Variables Q2: Interaction with Categorical Variables Q3: Linear Probability Model Q4: Instrumental Variable Approach I Q5: Instrumen
Question 2, Sub-point (v), I of II
Recall:
log(wa
Û g e) = 0.389 - 0.227 · f emal e + 0.082 · educ - 0.0056 · f emale · educ
(0.119) (0.168) (0.008) (0.0131)
0.029 · e x per - 0.00058 · e x per 2 + 0.032 · t enur e - 0.00059 · t enur e2
(0.005) (0.00011) (0.007) (0.00024)
n = 526, R2 = 0.441.
(v) Consider the alternative model:
log(wa
Û g e) = 0.389 - 0.297 · f emal e + 0.082 · educ - 0.0056 · f emale · (educ − 12.5)
(0.119) (0.036) (0.008) (0.0131)
0.029 · e x per - 0.00058 · e x per 2 + 0.032 · t enur e - 0.00059 · t enur e2
(0.005) (0.00011) (0.007) (0.00024)
n = 526, R2 = 0.441.
Interpret the coefficient for f emal e. Comment on the statistical and economic significance of the
variable. Compare this to the result in (ii).
Metrics for E&BE: Chapters 7+8+15 12 / 25
Q1: Interaction with Dummy Variables Q2: Interaction with Categorical Variables Q3: Linear Probability Model Q4: Instrumental Variable Approach I Q5: Instrumen
Question 2, Sub-point (v), II of II
Metrics for E&BE: Chapters 7+8+15 13 / 25
Q1: Interaction with Dummy Variables Q2: Interaction with Categorical Variables Q3: Linear Probability Model Q4: Instrumental Variable Approach I Q5: Instrumen
Question 3
The variable smokes is a binary variable equal to one if a person smokes, and zero otherwise. We
estimate the equation:
Ú = 0.656
smokes - 0.069 · log(ci g pr ice) + 0.012 · log(income) - 0.029 · educ
(0.855) (0.204) (0.026) (0.006)
[1.357] [0.307] [0.038] [0.012]
+ 0.020 · a g e - 0.00026 · a g e2 - 0.101 · r est aur n - 0.026 · nonmi g r
(0.006) (0.00006) (0.039) (0.052)
[0.011] [0.00013] [0.055] [0.085]
n = 807, R2 = 0.062. The variables:
cigprice is the per-pack price of cigarettes (in cents);
income is the annual income;
educ are the years of schooling;
age measures how old each individual is (in years);
restaurn is a binary indicator equal to 1 if the person resides in a region with restaurant
smoking restriction and zero otherwise;
nonmigr equals one if the respondent does not have a migration background, and zero
otherwise;
Both the usual and heteroskedasticity-robust standard errors (in square brackets) are reported.
Metrics for E&BE: Chapters 7+8+15 14 / 25
Q1: Interaction with Dummy Variables Q2: Interaction with Categorical Variables Q3: Linear Probability Model Q4: Instrumental Variable Approach I Q5: Instrumen
Question 3, Sub-points (i) and (ii)
Recall:
Ú = 0.656
smokes - 0.069 · log(ci g pr ice) + 0.012 · log(income) - 0.029 · educ
(0.855) (0.204) (0.026) (0.006)
[1.357] [0.307] [0.038] [0.012]
+ 0.020 · a g e - 0.00026 · a g e2 - 0.101 · r est aur n - 0.026 · nonmi g r
(0.006) (0.00006) (0.039) (0.052)
[0.011] [0.00013] [0.055] [0.085]
n = 807, R2 = 0.062.
(i) Consider the nature of the dependent variable smokes. What does this imply about our model?
Metrics for E&BE: Chapters 7+8+15 15 / 25
Q1: Interaction with Dummy Variables Q2: Interaction with Categorical Variables Q3: Linear Probability Model Q4: Instrumental Variable Approach I Q5: Instrumen
Question 3, Sub-point (ii)
Recall:
Ú = 0.656
smokes - 0.069 · log(ci g pr ice) + 0.012 · log(income) - 0.029 · educ
(0.855) (0.204) (0.026) (0.006)
[1.357] [0.307] [0.038] [0.012]
+ 0.020 · a g e - 0.00026 · a g e2 - 0.101 · r est aur n - 0.026 · nonmi g r
(0.006) (0.00006) (0.039) (0.052)
[0.011] [0.00013] [0.055] [0.085]
n = 807, R2 = 0.062.
(ii) Explain why our model must suffer from heteroskedasticity and how to deal with this?
Metrics for E&BE: Chapters 7+8+15 16 / 25
Q1: Interaction with Dummy Variables Q2: Interaction with Categorical Variables Q3: Linear Probability Model Q4: Instrumental Variable Approach I Q5: Instrumen
Question 3, Sub-points (iii), (iv), and (v)
Recall:
Ú = 0.656
smokes - 0.069 · log(ci g pr ice) + 0.012 · log(income) - 0.029 · educ
(0.855) (0.204) (0.026) (0.006)
[1.357] [0.307] [0.038] [0.012]
+ 0.020 · a g e - 0.00026 · a g e2 - 0.101 · r est aur n - 0.026 · nonmi g r
(0.006) (0.00006) (0.039) (0.052)
[0.011] [0.00013] [0.055] [0.085]
n = 807, R2 = 0.062.
(iii) Are there any important differences between the two sets of standard errors?
(iv) Holding other factors fixed, if education increases by four years, what happens to the estimated
probability of smoking?
Metrics for E&BE: Chapters 7+8+15 17 / 25
Q1: Interaction with Dummy Variables Q2: Interaction with Categorical Variables Q3: Linear Probability Model Q4: Instrumental Variable Approach I Q5: Instrumen
Question 4
You have data from 935 men with completed education that are employed in 2022. The data
contains the following variables:
wage: the hourly wage earned by person i ;
exper: the total number of years spent in the workforce by person i ;
tenure: the total number of years spent with their current employer by person i ;
educ: the total number of years spent in formal education by person i ;
married: a dummy variable taking value 1 if person i is married, and 0 otherwise;
union: a dummy variable taking value 1 if person i is in a labor union, and 0 otherwise;
south: a dummy variable taking value 1 if person i is living in the South of the US, and 0
otherwise;
black: a dummy variable taking value 1 if person i is black, and 0 otherwise.
Metrics for E&BE: Chapters 7+8+15 18 / 25
Q1: Interaction with Dummy Variables Q2: Interaction with Categorical Variables Q3: Linear Probability Model Q4: Instrumental Variable Approach I Q5: Instrumen
Question 4
The following model is estimated using OLS:
log(wa
Û g e) = 5.248 + 0.063 · educ + 0.007 · a g e + 0.196 · mar r ied − 0.188 · black
− 0.091 · south + 0.183 · ur ban + 0.011 · e x per + 0.011 · t enur e.
(i) Explain why you cannot interpret the coefficient for educ in the multiple regression model
as a causal effect.
Metrics for E&BE: Chapters 7+8+15 19 / 25
Q1: Interaction with Dummy Variables Q2: Interaction with Categorical Variables Q3: Linear Probability Model Q4: Instrumental Variable Approach I Q5: Instrumen
Question 4
Recall:
log(wa
Û g e) = 5.248 + 0.063 · educ + 0.007 · a g e + 0.196 · mar r ied − 0.188 · black
− 0.091 · south + 0.183 · ur ban + 0.011 · e x per + 0.011 · t enur e.
(ii) You obtain a variable for the score of an intelligence test IQ for each observation in your
data. The following estimates are obtained:
log(wa
Û g e) = 4.987 + 0.051 · educ + 0.003 · IQ + 0.008 · a g e + 0.196 · mar r ied − 0.141 · black
− 0.079 · south + 0.181 · ur ban + 0.011 · e x per + 0.011 · t enur e.
Make an argument for including IQ in the regression. Comment on the change of the
coefficient for educ .
Metrics for E&BE: Chapters 7+8+15 20 / 25
Q1: Interaction with Dummy Variables Q2: Interaction with Categorical Variables Q3: Linear Probability Model Q4: Instrumental Variable Approach I Q5: Instrumen
Question 4
You obtain a variable for the birth order of the individual compared to their siblings br thor d for
each observation in your [Link] following two-stage least squares estimates are obtained:
Õ = 11.089 − 0.199 · br thor d + 0.191 · a g e − 0.353 · mar r ied − 0.624 · black
educ
− 0.101 · south + 0.328 · ur ban − 0.289 · e x per + 0.011 · t enur e
log(wa
Û g e) = 4.450 + 0.134 · educ
Õ − 0.005 · a g e + 0.220 · mar r ied − 0.159 · black
− 0.062 · south + 0.178 · ur ban + 0.030 · e x per + 0.010 · t enur e
(iii) Explain the model that is estimated above. Discuss the relevance and exogeneity of
br thor d . The standard error of the estimate for br thor d is 0.039.
Metrics for E&BE: Chapters 7+8+15 21 / 25
Q1: Interaction with Dummy Variables Q2: Interaction with Categorical Variables Q3: Linear Probability Model Q4: Instrumental Variable Approach I Q5: Instrumen
Question 5, Sub-point (i)
(i) Suppose you wish to measure the impact of smoking on the weight of newborns. You are
planning to use the following model,
lo g(bw) = β0 + β1 mal e + β2 ci g + β3 l o g(inc)
where bw is the birth weight, male is a dummy variable assuming the value 1 if the baby is a boy
or 0 otherwise, cig is the average weekly amount of cigarettes per day smoked during pregnancy.
What could be the problem in using OLS to estimate the above model?
Metrics for E&BE: Chapters 7+8+15 22 / 25
Q1: Interaction with Dummy Variables Q2: Interaction with Categorical Variables Q3: Linear Probability Model Q4: Instrumental Variable Approach I Q5: Instrumen
Question 5, Sub-point (ii)
(ii) Suppose you have data on the average price of cigarettes in the place of residence. Would
this information help to identify the true parameters of the model?
Metrics for E&BE: Chapters 7+8+15 23 / 25
Q1: Interaction with Dummy Variables Q2: Interaction with Categorical Variables Q3: Linear Probability Model Q4: Instrumental Variable Approach I Q5: Instrumen
Question 5, Sub-points (iii)
Metrics for E&BE: Chapters 7+8+15 24 / 25
Q1: Interaction with Dummy Variables Q2: Interaction with Categorical Variables Q3: Linear Probability Model Q4: Instrumental Variable Approach I Q5: Instrumen
Question 5, Sub-points (iv)
Metrics for E&BE: Chapters 7+8+15 25 / 25