Qualitative Predictors

Credit Data Set

• Balance: Average Credit Card Debt
• Age
• Cards: Number of credit cards
• Education: Years of education
• Income (in thousands of dollars)
• Limit: Credit Limit
• Rating: Credit Rating
• Gender
• Student
• Married
• Ethnicity: Asian, African American, Caucasian
Qualitative Predictors with Two Levels
• Consider only “Balance” (response variable) and Gender (qualitative
predictor).
• For a qualitative predictor, we simply create an indicator/ dummy
variable that takes on two possible values.
• Based on the “Gender” variable, we can have the following dummy
variable:
1, if 𝑖th person is female
𝑥𝑖 = ൜ .
0, if 𝑖th person is male
Qualitative Predictors with Two Levels
• Fit the following model:
𝛽0 + 𝛽1 + 𝜀𝑖 , if 𝑖th person is female
𝑦𝑖 = 𝛽0 + 𝛽1 𝑥𝑖 + 𝜀𝑖 = ቊ .
𝛽0 + 𝜀𝑖 , if 𝑖th person is male
• So 𝛽0 is the average credit card balance among males.
• And 𝛽0 + 𝛽1 as the average credit card balance among females.
• Therefore 𝛽1 is the average difference in credit card balance between
females and males.
Regression Table
Coefficients Std. Error t-statistic P-value
Intercept 509.80 33.13 15.39 <0.0001
Gender[Female] 19.73 46.05 0.43 0.67
Interpretation
• From the previous Table, we observe that the average credit card
debt for males is $509.80.
• Females are expected to carry $19.73 in additional debt.
• However, we note that the p-value for the dummy variable is very
high. This indicates that there is no statistical evidence of a difference
in average credit card balance between the genders.
Qualitative Predictors with More than Two
Levels
• Consider the variable “ethnicity” as a predictor.
• Ethnicity has three levels. So we need to create two dummy variables.
• The first dummy variable may be
1, if 𝑖th person is Asian
𝑥𝑖1 = ൜ .
0, if 𝑖th person is not Asian
• Similarly, the second dummy variable could be
1, if 𝑖th person is Caucasian
𝑥𝑖2 = ൜ .
0, if 𝑖th person is 𝑛𝑜𝑡 Caucasian
Regression Table
Coefficients Std. Error t-statistic P-value
Intercept 531.00 46.32 11.46 <0.0001
Ethnicity[Asian] −18.69 65.02 −0.29 0.7740
Ethnicity[Caucasian] −12.50 56.68 −0.22 0.8260
Interpretation
• From the previous Table, we see that the estimated average balance
for the African American is $531.00.
• The Asian category are expected to have $18.69 less debt than the
African American category.
• The Caucasian category are expected to have $12.50 less debt than
the African American category.
• However, we observe that the p-values associated with the coefficient
estimates for the two dummy variables are very large.
• This suggests that there is no statistical evidence of a real difference
in credit card balance between the ethnicities.
Extension to Quantitative and Qualitative
Variable
• Consider the predictors “income” (quantitative predictor) and
“student” (qualitative predictor) along with balance as response
variable.
• Suppose the model takes the following form:
balance𝑖
𝛽2 , if 𝑖th person is student
≈ 𝛽0 + 𝛽1 × income𝑖 + ቊ .
0, if 𝑖th person is not a student
𝛽0 + 𝛽2 , if 𝑖th person is a student
= 𝛽1 × income𝑖 + ቊ .
𝛽0 , if 𝑖th person is not a student
Extension to Quantitative and Qualitative
Variable
Coefficients Std. Error t-statistic P-value
Intercept 211.14 32.46 6.51 <0.0001
Income 5.98 0.56 10.75 <0.0001
Student[Yes] 382.67 65.31 5.86 <0.0001
 Extension to
Quantitative and
Qualitative Variable
• This suggests that the average
effect on balance of a one-unit
increase in income does not
depend on whether or not the
individual is a student.
• This represents a potentially
serious limitation of the model,
since in fact a change in income
may have a very different effect
on the credit card balance of a
student versus a non-student.
Inclusion of Interaction Term
• This limitation can be resolved if we add an interaction variable,
created by multiplying income with the dummy variable for student.
• So the new model is
balance𝑖
𝛽2 + 𝛽3 × income𝑖 , if 𝑖th person is student
≈ 𝛽0 + 𝛽1 × income𝑖 + ቊ .
0, if 𝑖th person is not a student
(𝛽0 + 𝛽2 ) + (𝛽1 + 𝛽3 ) × income𝑖 , if 𝑖th person is student
=ቊ .
𝛽0 + 𝛽1 × income𝑖 , if 𝑖th person is not a student
Inclusion of Interaction Term
Coefficients Std. Error t-statistic P-value
Intercept 200.62 33.70 5.95 <0.0001
Income 6.22 0.59 10.50 <0.0001
Student[Yes] 476.68 104.35 4.57 <0.0001
Income × Student[Yes] -2.00 1.73 -1.16 0.25
 Inclusion of
Interaction
Term
• We observe that the slope for
students is lower than the slope
for non-students.
• This indicates that increases in
income are associated with
smaller increases in credit card
balance among students as
compared to non-students.