UNIT 9 EXTENSION OF REGRESSION

MODELS: DUMMY VARIABLE

CASES
Structure
9.0 Objectives
9.1 Introduction
9.2 The Case of Single Dummy: ANOVA Model
9.3 Analysis of Covariance (ANCOVA) Model
9.4 Comparison between Two Regression Models
9.5 Multiple Dummies and Interactive Dummies
9.6 Let Us Sum Up
9.7 Answers/Hints to Check Your Progress Exercises

9.0 OBJECTIVES
After reading this unit, you will be able to:
 define a qualitative or dummy variable;
 discuss the ANOVA model with a single dummy as exogenous variable;
 specify an ANCOVA model with one quantitative and one dummy
variable;
 interpret the results of dummy variable regression models;
 differentiate between ‘differential intercept coefficient’ and ‘differential
slope coefficient;
 describe the concepts of ‘concurrent, dissimilar and parallel’ regression
models that you encounter while considering ‘differential slope dummies’;
and
 explain how more than two dummies and interactive dummies can be
formulated into a regression model.

9.1 INTRODUCTION
In real life situations, some variables are qualitative. Examples are gender,
choices, nationality, etc. Such variables may be dichotomous or binary, i.e., with
responses limited to two such as in ‘yes’ or ‘no’ situations. Or they may have
more than two categorical responses. We need methods to include such variables
in the regression model. In this unit, we consider some such cases. We limit this
unit to consider regressions in which the dependent variable is quantified. You
may note in passing that when the dependent variable itself is a dummy variable,
we have to deal with them by models such as Probit or Logit. In such models, the

Dr. Pooja Sharma, Assistant Professor, Daulat Ram College, University of Delhi and Prof. B S
Prakash, Indira Gandhi National Open University, New Delhi
OLS method of estimation does not apply. In this unit, we will not consider such Extension of Regression
cases. You will study about them in the course ‘BECE 142: Applied Models: Dummy
Econometrics’. Variable Cases

In this unit, we consider only such cases in which the independent variable is a
dummy variable. Qualitative variables are not straightaway quantified. By
treating them as dummy variables we can make them quantified (or categorical).
For instance, consider variables such as male or female, employed or
unemployed, etc. These are quantifiable in the sense that by treating them as 1 if
‘female, and 0 if ‘male’. Similar examples could be 1 if yes and 0 if no; 1 if
employed and 0 if unemployed, etc. In the above, we have converted a qualitative
response into quantitative form. Thus, the qualitative variable is now quantified.
Such regressions could be a simple regression, i.e., there is only one independent
variable which is qualitative and treated as dummy variable. Or there could be
two independent variables, one of which can be treated as dummy and the other
is its covariant, i.e., there is a close relationship with the variable treated as
dummy. For instance, pre-tax income of persons can be classified above a
threshold level and treated as dummy variable, i.e., above or below the threshold
level income with response taken as 1 or 0. Now, the post-tax income, which is a
co-variant of pre-tax income, can be considered by its actual quantified value.
There could be similar extension of situations where you have to consider
multiple dummies and cases where you have to consider interactive dummies.
The nature of such regressions, particularly for their inference or interpretational
interest, is what we consider in the present unit.

9.2 THE CASE OF SINGLE DUMMY: ANOVA

MODEL
We first consider a simple regression model with only one independent variable.
Further, this independent variable is a dummy variable such as:
𝑌 =𝛽 +𝛽 𝐷 +𝑢 … (9.1)
Here, we take Y as the annual expenditure on food and Di as gender taking the
values 0 if the person is male and 1 if female. The Di’s are thus fixed and hence
non-stochastic. Now, if we assume that 𝑢𝑖 ~𝑁 0, 𝜎2 the OLS method can be
,
applied to estimate the parameters in (9.1). If we do this, the mean food
expenditure for males and females are respectively given by:
E (Yi │ Di = 0) = β1 + β2(0) = β1 … (9.2)
E (Yi │ Di = 1) = β1 + β2 … (9.3)
Here, β1 gives the average or mean food expenditure of males. It is the category
for which the dummy variable is given the value 0. The slope coefficient β2 tells
us by how much the mean food expenditure of females differ from that of the
mean food expenditure of males. Hence, β1 + β2 gives the mean food expenditure
for females. In view of this, it is not correct to call β2 as the slope coefficient
since there is no continuous regression line here. Hence, β2 is the ‘differential
117
Multiple Regression intercept coefficient’. It tells us by how much the value of intercept term differs
Models
between the two categories. A question that arises now is, what would have
happened if we had interchanged the assignment of ‘0’ between the two
categories of males and females ( i.e., if we had assigned the value ‘0’ to
females). You may note that, so long as we have only two categories as in the
present instance, i.e., it is a case of simple regression with only one independent
variable taken as a dummy variable Di with the category of responses
dichotomous or binary, it basically does not matter which category gets the value
of 1 and which gets the value 0. However, some minor difference would be there.
Let us see what this is.
The category to which we assign the value 0 is called as the base category. It is
also called by alternative names such as reference or benchmark or the
comparison category. In such an assignment, the intercept value represents the
mean value of the category that gets the value 0 (which is males in our case
above). What equation (9.3) tells us is, depending on such an assignment, the
mean value of expenditure on food for females is to be obtained by adding the
‘slope coefficient to the intercept value’. If the assignment of dummy is made the
other way, i.e., females 0 and males 1, we see a change in the numerical value of
the intercept term and its t value. Barring this, the R2 value, the absolute value of
the estimated dummy variable coefficient and its standard error, will remain the
same. Let us see this with the help of an example for better understanding.
Consider the data on ‘expenditure on food’ and income for males and females as
in Table 9.1. The data are averages based on the actual number of people (who
are in thousands) in different age groups. We first construct Table 9.2 from the
data in Table 9.1 as below.
Table 9.1: Data on Income and Food Expenditure by Gender
(Figures in $)

Age Food Income Food Income

Expenditure (female) Expenditure (male)
(female) (male)

< 25 1983 11557 2230 11589

25-34 2987 29387 3757 33328

35-44 2993 31463 3821 36151

45-54 3156 29554 3291 35448

55-64 2706 25137 3429 32988

> 65 2217 14952 2533 20437

Source: Table 6-1, Chapter 6, Gujarati.

118
 Table 9.2: Food Expenditure in Relation to Income and Gender Extension of Regression
Models: Dummy
Observation Food Expenditure Income ($) Gender Variable Cases

($)

1 1983 11557 1

2 2987 29387 1

3 2993 31463 1

4 3156 29554 1

5 2706 25137 1

6 2217 14952 1

7 2230 11589 0

8 3757 33328 0

9 3821 36151 0

10 3291 35448 0

11 3429 32988 0

12 2533 20437 0

Source: Table 6-2, Chapter 6, Gujarati.

Results of food expenditure regressed on the gender dummy variable (without
taking into account the income variable at this stage) presents the following
results.
𝑌 = 3176.833 – 503.1667 Di
se = (233.0446) (329.5749)
t = (13.6318) (–1.5267) R2 = 0.1890
The results show that the mean expenditure of males is 3177 $ and that of
females is (3177 – 503 = 2674 $). The estimated Di is not statistically significant
(since its t value is only –1.53). This means that the difference in the food
expenditure between gender is not statistically significant. Recall that we have
assigned the value ‘0’ to males. Hence, the intercept value represents the mean
value for males. In this assignment, to get the mean value of food expenditure of
females, we add the value of the coefficient of the dummy variable to the
intercept value. Now, let us re-assign the value ‘0’ to females and ‘1’ to males.
The regression results that we get are the following:

119
Multiple Regression 𝑌= 2673.667 + 503.1667 Di
Models
se = (233.0446) (329.5749)
t= (11.4227) (–1.5267) R2 = 0.1890
Thus, we notice that the mean food consumption expenditures of the two genders
have remained the same. The R2 value is also the same. The absolute value of the
dummy variable coefficient and their standard errors are also the same. The only
change is in the numerical value of the intercept term and its t value.
Another question that we may get is: since we have two categories, male and
female, can we assign two dummies to them? This means we consider the model
as:
Yi = β1 + β2 D2i + β3Di + ui … (9.4)
where Y is expenditure on food, D2 = 1 for female and 0 for male and D3 = 1 for
male and 0 for female. Essentially, we are trying to see whether we can assign
two dummies for male and female separately? The answer is ‘no’. To know the
reason for this, consider the data for a sample of two females and three males, for
which the data matrix is as in Table 9.3. We see that D2 = 1 – D3 or D3 = 1 – D2.
This is a situation of perfect collinearity. Hence, we must always use only one
dummy variable if a qualitative variable has two categories, such as the gender
here.
Table 9.3: Data Matrix for the Equation

Gender Intercept D2 D3

Male Y1 1 0 1

Male Y2 1 0 1

Female Y3 1 1 0

Male Y4 1 0 1

Female Y5 1 1 0

A more general rule is: if a model has the common intercept β1, and the
qualitative variable has m categories, then we must introduce only (m – 1)
dummy variables. If we do not do this, we get into a problem of estimation called
as the ‘dummy variable trap’. Finally, note that when we have a simple
regression model with only one dummy variable as considered here, the model
considered is also called as the ANOVA model. This is because there is no
second variable from which we are seeking to know the impact or variability on
the dependent variable. When we have this, we get what we call as an ANCOVA
model. We take up such a case in the next section.

120
9.3 ANALYSIS OF COVARIANCE (ANCOVA) Extension of Regression
Models: Dummy
MODEL Variable Cases

In economic analysis, it is common to have among explanatory variables some of

which are qualitative and some others quantitative. Such models are called as
Analysis-of-Covariance (ANCOVA) models. Here, we shall consider a model
that has both a quantitative and a dummy variable among the regressors. In
general, regression models containing a combination of quantitative and
qualitative variables are called ANCOVA models. Here, the quantitative
variables are called covariates or control variables. ANCOVA models are an
extension of the ANOVA models. They provide a method of statistically
controlling the effects of covariates (i.e., a quantitative explanatory variable) in a
model that includes both the type of variables with the qualitative variable treated
as a dummy variable. The quantitative variable considered is usually a covariate
in the sense that it bears close association with the main variable. Because of this,
exclusion of covariates from a model results in model specification error. In the
example considered above, we regressed ‘food expenditure’ on only gender
dummy [𝑌 = 𝛽 + 𝛽 𝐷 + 𝑢 ]. Now, let us consider another variable, ‘income
after taxes’, i.e., disposable income (a covariate of food expenditure) as an
explanatory variable (Xi). The model now is
𝑌 =𝛽 +𝛽 𝐷+𝛽 𝑋 +𝑢 … (9.5)
where Y = expenditure on food ($), X = after tax income ($), D = 1 for female and
= 0 for male. Let us now consider, for better appreciation, the result for the
regression in equation (9.5) obtained from the data in Table 9.2 as follows:
^ = 1506.244 − 228.9868𝐷 + 0.0589𝑋
𝑌
t = (8.0115) (–2.1388) (9.6417)
2
R = 0.9284
The dummy variable coefficient is statistically significant. Therefore, we reject
the null hypothesis that there is no difference in the average value of expenditure
on food for male and female. In other words, we conclude that gender has a
significant impact on consumption or food expenditure. Note that this difference
in consumption expenditure is inferred holding the effect of after-tax income
constant. Likewise, holding the gender differences constant, the after tax income
coefficient is significant. The slope coefficient for ‘after tax income’ indicates
that the mean food expenditure [i.e., the marginal propensity to consume (MPC)]
increases by 6 cents for every additional dollar of increase in the disposable
income. Note that since we have taken ‘0’ for males, the intercept term relates to
the MPC for males. For female MPC, we have to add the intercept value to the
coefficient of gender dummy (i.e., 1506.2 – 228.9 = 1277.3). Thus, the equations
for the MPC of females and males can be respectively written as:
Mean food expenditure for females: Ŷi  1277.2574  0.0589X i

Mean food expenditure for males: Ŷi  1506.2440  0.0589X i

121
Multiple Regression Since the MPC or the slope is same for both the gender, the two regressions are
Models
parallel as in Fig. 9.1 below.

Male

Food Expenditure
Female

After-tax expenditure

Fig. 9.1 Mean Food Expenditure for Male and Female

The model signifies the role and the impact of both the type of variables
(quantitative and qualitative) in explaining a dependent variable. Specifically, in
the example considered, the after tax expenditure is seen to affect the food
expenditure of both males and females.
Check Your Progress 1 [answer questions in about 50-100 words]
1) Define a qualitative variable.
..............................................................................................................................
..............................................................................................................................
..............................................................................................................................
..............................................................................................................................
..............................................................................................................................

2) Specify a regression model with a single dummy variable. Mention its

features from the point of view of interpretation of estimated coefficients.
..............................................................................................................................
..............................................................................................................................
..............................................................................................................................
..............................................................................................................................
..............................................................................................................................

122
3) What happens if the base value is reassigned for the dummy variable, say Extension of Regression
gender, in a simple regression model as in equation (9.1)? Models: Dummy
Variable Cases
.............................................................................................................................
.............................................................................................................................
.............................................................................................................................
.............................................................................................................................
.............................................................................................................................

4) What is meant by ‘dummy variable trap’? How do we avoid it?

.............................................................................................................................
.............................................................................................................................
.............................................................................................................................
.............................................................................................................................
.............................................................................................................................

5) Distinguish between an ANOVA model and an ANCOVA.

.............................................................................................................................
.............................................................................................................................
.............................................................................................................................
.............................................................................................................................
.............................................................................................................................

6) What is an advantage of ANCOVA model? What is a consequence of

omitting the inclusion of a covariant in an ANOVA model?
.............................................................................................................................
.............................................................................................................................
.............................................................................................................................
.............................................................................................................................
............................................................................................................................

7) Specify the general form of an ANCOVA model with one qualitative and one
quantitative variable. What does the slope oefficient for the quantitative
variable considerd indicate in general?
.............................................................................................................................
.............................................................................................................................
.............................................................................................................................
.............................................................................................................................
.............................................................................................................................

123
Multiple Regression
Models
9.4 COMPARISON BETWEEN TWO REGRESSION
MODELS
In the example considered above, i.e., for both the ANOVA and the ANCOVA
models, we saw that the slope coefficients were same but the intercepts were
different. This raises the question on whether the slopes too could be different?
How do we formulate the model if our interest is to test for the difference in the
slope coefficients too? In order to capture this, we introduce a ‘slope drifter’. For
the example of consumption expenditure for male or female considered above, let
us now proceed to compare the difference in the consumption expenditure by
gender by specifying the model with dummies as follows:

𝑌 = 𝛽 + 𝛽 𝐷 + 𝛽 𝑋 + 𝛽 (𝐷 𝑋 ) + 𝑢 … (9.6)

Note that the additional variable added is DiXi which is in multiplicative or

interactive form. In (9.6), we have taken Di = 0 for males and Di = 1 for females.
Now, the ‘mean food expenditure’ for males is given by:

E (Yi │ Di = 0, Xi) = β1 + β3Xi … (9.7)

{since Di = 0}

The ‘mean food expenditure’ for females is given by:

E (Yi │ Di = 1, Xi) = β1 + β2Di + (β3 + β4Di) Xi

= (β1 + β2) + (β3 + β4) Xi … (9.8)

{since Di = 1}

In equation (9.8), (β1 + β2) gives the mean value of Y for the category that
receives the dummy value of 1 when X is zero. And, (β3 + β4) gives the slope co-
efficient of the income variable for the category that receives the dummy value of
1. Note that the introduction of the dummy variable in the ‘additive form’ enables
us to distinguish between the intercept terms of the two groups. Likewise, the
introduction of the dummy variable in the interactive (or multiplicative) form
(i.e., 𝐷 𝑋 ) enables us to differentiate between the slope coefficients (or terms) of
the two groups. Depending on the statistical significance of the differential
intercept coefficient, β2, and the differential slope coefficient, β4, we can infer
whether the female and male food expenditure functions differ in their intercept
values, or their slope values, or both. There can be four possibilities as shown in
Fig. 9.2. Fig. 9.2 (a) shows that there is no difference in intercept or the slope
coefficient of the two food expenditure regressions. Such regression equations
are called ‘Coincident Regressions’.

124
 Extension of Regression
Models: Dummy
Y Y Variable Cases

X X
0 0
(a) Coincident Regressions (b) Parallel Regressions

X
X
0
0
(c) Concurrent Regressions (d) Dissimilar Regressions

Fig 9.2 Comparison of Regression Equations

Fig. 9.2 (b) shows that the two slope coefficients are the same but intercepts are
different. Such regressions are referred to as ‘Parallel Regressions’. Fig. 9.2 (c)
shows that the two regressions have the same intercepts but
different slopes. Such regressions are referred as ‘Concurrent Regressions’. Fig.
9.2 (d) shows that the two intercepts and the two slope coefficients are both
different. Such regressions are called ‘Dissimilar Regressions’.

9.5 MULTIPLE DUMMIES AND INTERRACTIVE

DUMMIES
We often might require to consider more than one dummy variables. Besides,
there could be cases where we might be interested in seeing for the impact of
dummy variable interactions. Let us consider a case as given below.
Yi = β1 + β2D2i + β3D3i + β4Xi + ui … (9.7)

125
Multiple Regression where Y is income, X is education measured in number of years of schooling, D2
Models
is gender (0 if male, 1 if female), D3 is if in reserved segment or group (e.g.
SC/ST/OBC) taking the value 0 if ‘not in reserved segment’, i.e., in general
segment and 1 if ‘in reserved segment’. Here, gender (D2) and reservation (D3)
are qualitative variables and X is quantitative variable. In this formulation (for
example, equation 9.7) we have made an implicit assumption that the differential
effect of gender is constant across the two segments of reservation. We have
likewise assumed that the differential effect of reservation is constant across the
two genders. This means if the average income is higher for males than for
females, it is so whether the person is in the general segment or in the reservation
segment. Likewise, it is assumed here that if the average income is different
between the two reservation segments, it is so irrespective of gender. However, in
many cases, such assumptions may not be tenable. This means, there could be
interaction between gender and reservation dummies. In other words, their effect
on average income may not be simply additive as in (9.7) but could be
multiplicative. If we wish to consider for this interactive effect, we must specify
the model as follows:
Yi = β1 + β2D2i + β3D3i + β4(D2i D3i) +β5Xi + ui … (9.8)
In equation (9.8), the dummy variable D2iD3i is called as ‘interactive or
interaction dummy’. It represents the joint or simultaneous effect of two
qualitative variables. Taking expectation on both sides of equation (9.8), i.e., by
considering the average effect on income across gender and reservation, we get:
E (Yi │ D2i =1, D3i = 1, Xi) = β1 + β2 + β3 + β4 + β5Xi … (9.9)
Equation (9.9) is the average income function for female reserved category
workers where β2 is the differential effect of being female, β3 is the differential
effect of being in the reserved segment and β4 is the interactive effect of being
both a female and in reserved segment. Depending on the statistical significance
of various dummies, we need to make relevant inferences. The specification can
easily be generalized for more than one quantitative variable and more than two
qualitative variables.
Check Your Progress 2 [answer questions within the given space in about 90-
100 words]
1) What is meant by a ‘slope drifter’? When is it introduced and for what use?
Specify a general model with such a ‘slope drifter’ and comment on the
additional variable introduced.
.............................................................................................................................
.............................................................................................................................
.............................................................................................................................
.............................................................................................................................
.............................................................................................................................

126
2) Differentiate between the four type of regressions that we might get when Extension of Regression
considering a model of the type in equation (9.6) with two slope drifters 𝛽 Models: Dummy
Variable Cases
and 𝛽 as therein.
.............................................................................................................................
.............................................................................................................................
.............................................................................................................................
.............................................................................................................................
.............................................................................................................................

3) List the four types of regression models, with dummy variables to

accommodate different cases or situations, as we have considered in this unit.
Specify their difference by their name and features.
.............................................................................................................................
.............................................................................................................................
.............................................................................................................................
.............................................................................................................................
.............................................................................................................................

9.6 LET US SUM UP

This unit makes a distinction between qualitative and quantitative variables. It
has considered three types of models in which the focus is kept on inclusion of
qualitative variables in the regression models. The first of such models is
considered is a simple regression model. In this, we have considered only one
dummy variable, as an independent variable, on the RHS of the regression
equation. This equation is of the form: 𝑌 = 𝛽 + 𝛽 𝐷 + 𝑢 . Analysis in this
form is called as ANOVA. Quite often, we would be committing a specification
bias if we consider the regression model in this form. This happens because the
variable Yi will be clearly related to a variable Xi which is a quantitative variable.
To accommodate this, we considered the second type of model in which we
included a co-variant (Xi) into the regression equation: 𝑌 = 𝛽 + 𝛽 𝐷 + 𝛽 𝑋 +
𝑢 . Analysis in this form is called as ANCOVA. In both these type of models, our
focus was only on observing the significance of difference in the intercepts. But
in practice, we do encounter a number of situations in which not only the
intercept, but the slope too could vary between categories. To allow for this kind
of situation, we considered a third type of model in which we accommodated for
the interactive effect of the ‘dummy variable with the quantitative variable’, i.e.,
DiXi. The regression model considered for this kind of an analysis is of the
form: 𝑌 = 𝛽 + 𝛽 𝐷 + 𝛽 𝑋 + 𝛽 (𝐷 𝑋 ) + 𝑢 . In this situation, we noted that we
could come across four possibilities viz. coincidental, parallel, concurrent and
dissimilar regressions. We have finally considered the case where a regression
model may have to be formulated to accommodate more than one qualitative

127
Multiple Regression variable and a case where we might be interested in examining for the interactive
Models
effect of the two qualitative variables. For this, we considered models such as Yi
= β1 + β2D2i + β3D3i + β4(D2i D3i) +β5Xi + ui.

9.7 ANSWERS/ HINTS TO CHECK YOUR PROGRESS

EXERCISES
Check Your Progress 1
1) A qualitative variable is one which has a categorical response such as yes/no
or employed/unemployed or male/female. If the response is limited to two,
as in these cases, it is called as a dichotomous variable. The responses can be
more than two. But they may be classified as 1, 2, 3, …….. Such responses
are unambiguous or categorical. Hence, a qualitative variable is also called
as dummy variable or categorical variable.

2) The model in this case can be 𝑌 = 𝛽 + 𝛽 𝐷 + 𝑢 . We are considering the

dependent variable Yi as quantitative variable. The Di’s are thus fixed and
hence non-stochastic. Di is taken a dichotomous, i.e., it takes the values 0
and 1. In such cases, the factor or entity which is assigned the value 0, is
called as the base category. The estimated value of the mean of Yi, given Di
= 0, is given by 𝛽 . Here, 𝛽 is not strictly the slope coefficient but is the
‘differential intercept coefficient’. The estimated value of the mean of Yi,
given Di = 1, is given by 𝛽 + 𝛽 .
3) The mean value of Yi for the two gender classes, the R2 value, the absolute
value of the estimated dummy variable coefficient and the standard errors
will be the same. The numerical value of the intercept term and its t value
will change.
4) The number of responses to the dummy variable is called as ‘categories’ of
response. If the dummy variable refers to gender of the respondent, there are
two categories of response viz. male and female. If we assign two separate
dummies in such cases, we encounter a situation of perfect collinearity.
Hence, we will not get unique estimates or one of the two parameters is not
estimable. This situation is called as ‘dummy variable trap’. To avoid this
situation, the general rule is if we have m categories, we limit the number of
dummies to ‘m – 1’. The models should also have a common intercept β1.
5) If the regression model considered has only one independent variable in
general, and that variable is a dummy variable as considered here in
particular, then the variation or the sources of variability that is sought to be
identified for the dependent variable is limited to that one variable. In such
cases, the regression model considered is called as an ANOVA model. If the
independent variables considered are two, with one considered as dummy
variable, and the other variable considered is related to the dummy variable,
then such models are called as ANCOVA model.

128
 In other words, regression models in which some independent variables are Extension of Regression
qualitative and some others are quantitative, are called as ANCOVA models. Models: Dummy
Variable Cases
6) The advantage is that ANCOVA models provide a method of statistically
controlling the effects of covariates. The consequence of excluding a
covariant from being included in the model is that the model suffers from
‘specification error’. The consequence of committing specification errors are
that the ideal assumptions required for the OLS estimators to be efficient are
violated. Consequently, they lose out on their efficiency properties.

7) The general form of the model is like: 𝑌 = 𝛽 + 𝛽 𝐷 + 𝛽 𝑋 + 𝑢 . The slope

coefficient indicates the rate of increase (or decrease) in the ‘marginal
propensity to consume (MPC)’. This is when the dependent variable Y relates
to a consumption variable like expenditure on food and the quantitative
independent variable is like disposable income as considered here.
Check Your Progress 2
1) In regression models with one intercept and one slope coefficients, our
interest might be to test to know whether: (i) the intercept terms are
statistically different and (ii) the slope coefficients are statistically
different? For investigating the second question, we need to introduce what
is called as a ‘slope drifter’. The model specified with such a drifter would
be like: 𝑌 = 𝛽 + 𝛽 𝐷 + 𝛽 𝑋 + 𝛽 (𝐷 𝑋 ) + 𝑢 .The additional variable
introduced here is DiXi. It is a multiplicative variable in the interactive form.
Here 𝛽 and 𝛽 are the two slope drifters which helps us infer for the
statistical difference in the intercept values and the slope values respectively.
2) We get a ‘coincident regression’ when there is no difference both in intercept
as well as the slope. We get a ‘parallel regression’ when the two intercept
terms are different but the two slope coefficients are the same. We get a
‘concurrent regression’ when the two regressions have the same intercept but
different slopes. We get two ‘dissimilar regressions’ when both the intercept
terms and the slope coefficients are different.

3) (i) 𝑌 = 𝛽 + 𝛽 𝐷 + 𝑢 . (ii) 𝑌 = 𝛽 + 𝛽 𝐷 + 𝛽 𝑋 + 𝑢 . (iii) 𝑌 = 𝛽 +

𝛽 𝐷 + 𝛽 𝑋 + 𝛽 (𝐷 𝑋 ) + 𝑢 . (iv) Yi = β1 + β2D2i + β3D3i + β4(D2i D3i)
+ β5Xi + ui. The first is the ANOVA model in which we have considered only
one single dummy variable as the independent variable. The second is the
ANCOVA model in which we have considered one qualitative dummy
variable and another quantitative exogenous variable related to the dummy
variable, the omission of which would lead to a ‘specification bias’. The third
involves an interactive variable (𝐷 𝑋 ) in which we try to see whether both
the slopes and the intercept coefficients differ. In this, there is a possibility of
getting four different type of regressions viz. coincident, parallel, concurrent
and dissimilar regressions. The fourth situation considered involves a
interactive dummy variable like: Yi = β1 + β2D2i + β3D3i + β4(D2i D3i) +β5Xi +
ui .
129