Lecture 7: Multiple Regression Analysis

with Qualitative Information

Applied Econometrics
Dr. Le Anh Tuan

1
Dummy variables or binary variables

► Qualitative information can be turned into quantitative

information in a straightforward way, using binary coding for
“yes” and “no”.
► “is a certain person in the sample female?”
► yes female → ! = 1 and no not female → ! = 0.
► Examples: employed vs. unemployed, financial crisis vs.
non-financial crisis…

► ! can be called a dummy variable (or indicator variable/binary

variable/ zero-one variable)

► They may appear as the dependent or as independent variables

2
Dummy variables or binary variables
► In general, a dummy variable is a variable that equals 0 or 1.
► The coefficient of a dummy variable is equal to an intercept shift
of size δ when dummy variable = 1. All slope parameters remain
unchanged.
► Example:

Dummy variable:
= the wage gain/loss if the person =1 if the person is a woman
is a woman rather than a man =0 if the person is man
(holding other things fixed)

3
Dummy variables

Alternative interpretation of
coefficient:

i.e. the difference in mean

wage between men and
women with the same level
of education.

Intercept shift

► In order to be able to interpret the coefficients of dummy

variables one has to know the reference/base group. The
reference/base group is given by the group for which the
dummy equals zero.
4
Dummy variables
This model cannot be estimated
► Dummy trap (perfect collinearity)

► When using dummy variables, one category always has to be

omitted:
The base category are men

The base category are women

► Alternatively, one could omit the intercept:

Disadvantages:
1) More difficult to test for differences between the parameters
2) R-squared formula only valid if regression contains intercept
5
Dummy variables
►Gender and Wage

Holding education,
experience, and tenure fixed,
women earn $1.81 less per
hour than men

► Does that mean that women are discriminated against?

► Not necessarily. Being female may be correlated with other
productivity characteristics that have not been controlled for.

6
Dummy variables
►Effects of training grants on hours of training
Dummy variable indicating whether firm
Hours training per received a training grant
employee

► Holding sales, employee, fixed, firms with grants have higher

training per employee by 26.25 hours.
► This is an example of program evaluation
► Treatment group (= grant receivers) vs. control group (= no grant)
► Is the effect of treatment on the outcome of interest causal?

7
Dummy variables
►A new diet program and Weight
! = 12.82 − 1.85'/#0'#1
"#$%ℎ'
(2.12) (0.60)
n=330, R2=15%
►treated is a dummy variable that equals 1 for a person
who joined a new diet program, 0 otherwise.

Is the new diet is effective?

8
Dummy variables
►Using dummy explanatory variables in equations for log(y)

Dummy variable
=1 if a house is of
colonial style.
=0 otherwise
As the dummy for colonial style changes from 0 to
1, the house price increases by 5.4 percentage
points

9
 Using dummy variables for multiple
groups

10
 Several subgroups
► Example: A worker is female or male and married or unmarried
⇒ 4 subgroups:
► female and not married
► female and married
► male and not married
► male and married

► How to proceed:
► (1) Define dummy variables for all subgroups
► (2) Leave out one group, which becomes the base
/reference group.

11
 Several subgroups

Holding other things fixed,

married women earn 19.8% less
than single men (= the base
group)
12
 Incorporating ordinal information
using dummy variables
Municipal bond rate Credit rating from 0-4 (0=worst, 4=best)

This specification would probably not be appropriate as the credit rating only
contains ordinal information. A better way to incorporate this information is to
define dummies:

Dummies indicating whether the particular rating applies, e.g. CR1=1 if CR=1, and
CR1=0 otherwise. All effects are measured in comparison to the worst rating (=
base group).
13
 University ranking
► University ranking from 1-100
► The quality difference between ranks 1 and 2 and ranks 11 and
12, respectively, may be dramatically different.
→ Hence, ranks should not be used as independent var.
► Instead, we have to assign a dummy variable !" for all but one
(the “reference group”) of the universities, inducing several new
parameters which have to be estimated.
► Note: Then, the coefficient of a dummy variable Dj denotes the
intercept shift between university j and the reference university.
► Sometimes there are too many ranks and hence too many
parameters to be estimated. Then it proves useful to group the
data, e.g., ranks 1-10, 11-20, etc.

14
 Interactions involving dummy
variables

15
 Interaction terms
► Example: Do returns to schooling differ for men and women?
► Or: is the effect of education on the wage moderated by gender?

educ wage

female

► How do we formulate a model that allows the effect of education

to vary with gender?
► We use a model with interaction terms
log(%&'() = +, + +. (/01 + +2 3(4&5( + +6 3(4&5( ∗ (/01 + 0

16
 Notes
► Consider two models:

(1) log $%&' = )* + ), '-./ + )0 1'2%3' + .

H0: 74 =0, indicating that the whole wage equation is

the same for men and women

4 log $%&' = )* + ), '-./ + )0 1'2%3' + )5 1'2%3' ∗ '-./ + .

H0: 78 =0, indicating that the return to education is the same for men
and women
Or
The impact of education on wage is similar for men and women.

17
 Interaction terms
log(%&'() = +, + +. (/01 + +2 3(4&5( + +6 3(4&5( ∗ (/01 + 0

► Female is a dummy variable that equals 1 for female person, 0

otherwise.
► Expected Wage for:

► Male: +, + +. (/01

► Female:
+, + +. (/01 + +2 + +6 (/01 = (+, + +2 )+(+. + +6 )(/01

18
 Interaction terms

19
 Interaction terms

t-ratio=-0.53→ the coefficient on 9/:;</ ∗ /012 is insignificant → The

impact of education on wage is not different between men and women

"#$!
%&$' =. *++. -./012 − .216'7&"'−. --89/:;</ ∗ /012
(0.12) (0.008) (.17) (.013)
+.005'@A'B + .017tenure
(0.002) (0.003)

This coeffient explains the

Wage is positively associated impact of gender on wage when
with education educ=0. In overall, women
earns less than men.

20
 Example – Investment Inefficiency and Firm
Performance
► We have a proposal with two hypotheses:
H1: Investment inefficiency has a negative impact on firm
performance.
H2: The negative impact of investment inefficiency on firm
performance is stronger for small firms.

Investment H1
Firm performance
Inefficiency
H2

Small firms

21
 Example – Investment Inefficiency and Firm
Performance
► First, to test the first hypothesis, we estimate the model:
9:; = <= + <?;@A:9B;C_DAEFGHBFAH + <ICFEF9;JF + KF;9 LMG + N
"#$
! = .033 − .317$,-#".$/_1-2345.3-5 − 0.042/323"$83
(0.002) (.013) (.002) (.004)
n=76,604, R2=1.38%

The coefficient on abnormal_investment is -0.317 and statistically significant

at 1% level.
→ Investment inefficiency negatively affects firm performance

Support the hypothesis: Firm performance is negatively

associated with investment inefficiency.
→ a 1% increases in abnormal investment is associated with a
decrease in roa by 0.317%.
22
 Example – Investment Inefficiency and Firm
Performance
To test the second hypothesis, we build an interaction terms
model:
Baseline model:
!"# = %& + %(#)*"!+#,_.*/012+0*2 + %3,0/0!#40 + 50#! 671 + 8
Interaction terms model
!"#
= %& + %( #)*"!+#,_.*/012+0*2 + %3 1+#,,_9.!+1
+ %: #)*"!+#,_.*/012+0*2 ∗ 1+#,,_9.!+1 + %< ,0/0!#40 + 50#! 671
+8

► =>?@@_ABC>D is a dummy variable that equals to 1 firms have firm

size is smaller than sample median, 0 otherwise.

► The coefficient on the interaction terms indicates the difference

in the impact of investment inefficiency on firm performance
between small firms and non-small firms.
23
 Example – Investment Inefficiency and Firm
Performance
"#$
! = 0.043 + 0.090$,-#".$/_1-2345.3-5 − 0.0204.$//_81".4
(0.002) (0.020) (.002)
−9. :;9<=>?@A<B_C>DEFGAE>G ∗ FA<BB_IC@AF − .074/323"$K3
(.027) (.004)
n=76,604, R2=4.18%
The coefficient of <=>?@A<B_C>DEFGAE>G ∗ FA<BB_IC@AF is negative (-0.610) and
statistically significant at 1% level.

→ The negative impact of investment inefficiency on firm performance is

stronger/more pronounced for small firms.
→ In other words, compared to large firms (non-small firms), the negative impact of
investment inefficiency on firm performance is stronger/more pronounced for small
firms.

In terms of economic significance, a one unit increase in the abnormal investment,

the performance of small firms is 0.610 unit lower than that of non-small firms.

24
 Example – Investment Inefficiency and Firm
Performance

25
 Example – Corruption and Firm Performance
► We have a proposal with three hypotheses:
H1: Firm performance is negatively associated with corruption.
H2: The negative impact of corruption on firm performance is
stronger for young firms.
H3: The negative impact of corruption on firm performance is
stronger in financial crisis.

Financial crisis
H3
H1
Corruption Firm performance
H2

Young firms

26
 Example – Corruption and Firm Performance
► First, to test the first hypothesis, we estimate the model:
"#$ = ;< + ;= -#""./01#2 + ;> 415$ + ;? 072819:$ + .
"#$
! = .156 − .033-#""./01#2 + .005415$ − .057072819:$
(0.024) (0.006) (.002) (.017)
n=2,597, R2=2.61%

The coefficient on corruption is -.033 and statistically significant at 1%

level.
→ Corruption negatively affects firm performance

Support the hypothesis: Firm performance is negatively

associated with corruption.
→ a one-unit increases in corruption is associated with a
decrease in roe by 0.033 unit (3.3%).
27
 Example – Corruption and Firm Performance

To test the second hypothesis, we build an interaction terms

model:
Baseline model:
!"# = %& + %( )"!!*+,-". + %/ 0-1# + %2 ,3.4-56# + *
Interaction terms model
!"#
= %& + %( )"!!*+,-". + %/ 7"*.489:;< + %2 )"!!*+,-".
∗ 7"*.4_?-!@0 +%A 0-1# + %B ,3.4-56# + *

► C"*.4_?-!@0 is a dummy variable that equals to 1 firms have

firm age is smaller than sample median, 0 otherwise.

► The coefficient on the interaction terms indicates the difference

in the impact of corruption on firm performance between young
firms and non-young firms.
28
 Example – Corruption and Firm Performance

"#$
! = .113 − .002,#""-./0#1 + .0624#-15_70"89
(.032) (.014) (.027)
−.039,#""-./0#1 ∗ 4#-15_70"89 − 00490A$ − .056/C150DE$
(.016) (.002) (.018)

The coefficient of ,#""-./0#1 ∗ 4#-15_70"89 is negative (-0.039) and statistically

significant at 5% level.

→ The negative impact of corruption on firm performance is stronger/more

pronounced for young firms.
→ In other words, compared to mature firms (non-young firms), the negative impact
of corruption on firm performance is stronger/more pronounced for young firms.

In terms of economic significance, a one unit increase in the corruption, the

performance of young firms is 0.039 unit lower than that of non-young firms.

29
 Example – Corruption and Firm Performance

To test the third hypothesis, we build an interaction terms

model:
Baseline model:
!"# = %& + %( )"!!*+,-". + %/ 0-1# + %2 ,3.4-56# + *
Interaction terms model
!"#
= %& + %( )"!!*+,-". + %/ )!-0-0 + %2 )"!!*+,-".
∗ )!-0-0 +%8 0-1# + %9 ,3.4-56# + *

► Crisis is a dummy variable that equals to 1 for years of 2007,

2008, 2009, 0 otherwise.

30
 Example – Corruption and Firm Performance

"#$
! = .153 − .029.#""/012#3 + .004."2626
(.026) (.008) (.027)
−.011.#""/012#3 ∗ ."2626 + .00562>$ − .0571?3@2AB$
(.011) (.002) (.018)

The coefficient of .#""/012#3 ∗ ."2626 is -.011 and statistically insignificant, even at

10% level.

→ The negative impact of corruption on firm performance is similar during the

financial crisis and non-crisis periods.
→ In other words, compared to before and after the crisis, the negative impact of
corruption on firm performance is similar in the financial crisis period.

31
 Summary
Consider a model:
y = #$ + #& '1 + )*+,-*./ + 0 (model 1)
y = #$ + #& '1 + #9 '2 + #; '1 ∗ '2+)*+,-*./ + 0 (=>?@A B)
Where x2 is a dummy variable
Main effect in Interaction-terms in Hypothesis
Model (1) Model (2)
#& #;
>0 >0 More pronounced

>0 <0 Less pronounced

<0 <0 More pronounced

<0 >0 Less pronounced

32
 Summary
Consider a model:
y = #$ + #& '1 + #) '2 + #+ '1 ∗ '2+-./01.23 + 4
Where x2 is a dummy variable

Interpret the coefficient on #+ :

→ The positive/negative impact of x1 on y is more/less

pronounced for x2 (main group).

→ In other words, compared to x2 (base group), the

positive/negative impact of x1 on y is more/less pronounced
for x2 (main group).

#+ denotes how much unit is lower/higher between main

group and base group.
33
 Example
Board tenure diversity and Investment Efficiency
Dep. Var.: INVESTMENT EFFICIENCY
(1)
BTD 0.001
(0.501)
BTD × SHORT TENURE -0.006*
(-1.907)
SHORT TENURE 0.006***
(2.738)
Control variables INCLUDED
Fixed effects Country, Industry, Year
Observations 81,750
Adjusted R2 0.314
SHORT TENURE is a dummy variable that equals 1 if a firm‘s board tenure is lower than sample
median, 0 ọtherwise.

The coefficient of BTD × LONG TENURE is positive (0.006) and statistically significant at 10% level.
→ The POSITIVE impact of board tenure diversity on firm investment efficiency is less pronounced
for firms that have short tenure.
In terms of economic significance, a one unit increase in the BTD, investment efficiency of short-
tenured firms is 0.006 unit lower than that of long-tenured firms.

34
 Example
Corruption and GDP Growth: The moderating Role of FDI

Corruption GDP Growth

High FDI

Sample: 48 Asian countries

HIGH FDI is a dummy variable that equals 1 if a country‘s FDI is higher than
sample median, 0 ọtherwise.

35
 Example
Corruption and GDP Growth: The moderating Role of FDI
Dep. Var.: GDP GROWTH
(1)
Corruption -0.089***
(-3.501)
Corruption × HIGH FDI 0.050***
(4.102)
HIGH FDI 0.016***
(2.738)
Control variables INCLUDED
Fixed effects Country, Year
Observations 570
Adjusted R2 0.112

HIGH FDI is a dummy variable that equals 1 if a country‘s FDI is higher than sample median, 0
ọtherwise.

The coefficient of Corruption × HIGH FDI is positive (0.050) and statistically significant at 1%.
→ The negative impact of corruption on GDP GROWTH is less pronounced for countries that have
high FDI.
In terms of economic significance, a one unit increase in the Corruption, GDP Growth of high-FDI
countries is 0.050 unit higher than that of low-FDI countries.

36
 A Binary dependent variable:
The linear probability model

37
 The linear probability model

Linear regression when the dependent variable is binary

If the dependent
variable only
takes on the
values 1 and 0

Linear probability mod

(LPM)

In the linear probability model, the coefficients

describe the effect of the explanatory variables on
the probability that y=1
38
The linear probability model
=1 if "in the labor force” of a married
women, =0 otherwise husband‘s income

kidslt6 = number of children less than six years

kidsge6 = number of kids between 6 and 18 years
of age
If the number of kids under six years increases by one,
the probability that the woman works falls by 26.2%
The linear probability model
► Disadvantages of the linear probability model
► Predicted probabilities may be larger than one or smaller than
zero
► Marginal probability effects sometimes logically impossible
► The linear probability model is necessarily heteroskedastic
Variance of Bernoulli
variable
► Heteroskedasticity consistent standard errors need to be
computed

► Disadvantages of the linear probability model

► Easy estimation and interpretation
► Estimated effects and predictions are often reasonably
good in practice