ECON 251 Other modeling topics: Introduction
Research Methods Regression analysis is one of the most commonly used
techniques in statistics.
It is considered powerful for several reasons:
• It can cover variety of mathematical models
―linear relationships.
―non - linear relationships.
―qualitative variables.
• It provides efficient methods for model building, to select the
best fitting set of variables.
6. Multiple Regression III
1 2
Polynomial Models Polynomial models with one predictor variable
The independent variables may appear as functions of a First order model (p = 1)
number of predictor variables. y = b 0 + b 1x + e
• Polynomial models of order p with one predictor variable:
y = b 0 + b 1x1 + b 2x2 + …+ b pxp + e
Second order model (p = 2)
• Polynomial models with two predictor variables:
y = b 0 + b 1x + eb 2x2 + e
y = b 0 + b 1x1+ b 2x2 + e
y = b 0 + b 1x1+ b 2x2 + b 3x1x2 + e
b2 < 0 b2 > 0
3 4
1
Polynomial models with one predictor variable Example – Revenue vs. Profit
Third order model (p=3) Build a model that can predict profits based on revenue
y = b 0 + b 1x + b 2 + eb 3
x2 x3 + e
b3 < 0 b3 > 0
5 6
Revenue vs. Profit: Linear model Revenue vs. Profit: Linear model
How good is this model?
Why is the independent variable
insignificant?
Could we do better?
Profit = 4.9944 + 0.0134 * Revenue 7 8
2
Revenue vs. Profit: Quadratic model Revenue vs. Profit: Quadratic model
How good is this model?
Is it better than a linear model?
Could we do better?
Profit = 4.014 + 0.524 * Revenue – 0.062 * Revenue2
9 10
Revenue vs. Profit: Cubic model Revenue vs. Profit: Cubic model
Why are none of the variables
significant?
Do you see any contradictions?
Profit = 4.335 + 0.266 * Revenue + 0.004 * Revenue2
11 12
– 0.005 * Revenue3
3
Polynomial models with two predictor variables Example – Advertising and Sales
How is advertising expenditure related to sales?
• typically, higher advertising expenditures yield higher sales
• First order model • First order model with But, does each dollar of advertising expenditure have the
y = b 0 + b 1x1 + b 2x2 + e interaction same effect on the amount of sales?
y = b 0 + b 1x1 + b 2x2
The effect of one predictor variable on y
+b 3x1x2 + e Example: Tyler Personal Care
is independent of the effect of the other
predictor variable on y. • dependent variable: sales (in 1000s)
The two variables interact
to affect the value of y. • independent variables: price, advertising expenditure (in
X2 = 3 X2 = 3 $1000s)
X2 = 2
X2 = 1 X2 = 2 Plot sales against price
X2 =1
x1 x
13 1 14
Example – Advertising and Sales Example – Advertising and Sales
Clearly, higher advertising expenditure yields higher sales,
but the effect of higher advertising expenditure depends on
the price of a product
• at higher selling price the effect of advertising expenditure
347 ______________
282
43 • there is some ________ between price and advertising
expenditure
Model:
Sales = β0 + β1 Price + β2 Advertising Expenditure + β3 INT + ε
where
What do you notice? INT = Price * Advertising Expenditure
What is the effect of higher advertising expenditure?
15 16
4
Example – Advertising and Sales Qualitative Independent Variables
Model is In many real-life situations one or more independent
________ variables are qualitative.
Including qualitative variables in a regression analysis model
is done via indicator variables.
An indicator variable (I) can assume one out of two values,
All three
“zero” or “one.”
independent For example:
variables are
____________ 1 if a degree earned is in Economics
I=
There is evidence that the effect advertising expenditure on sales 0 if a degree earned is not in Economics
depends on the price of the product
We cannot study the effect of one variable on y independently of the
17 18
other variable
Example – Car Sale Price Example – Car Sale Price
The dealer believes that color is a variable that affects a Solution (Car Sale Price.xls)
car’s price. • the proposed model is
Three color categories are considered: y = b 0 + b 1(Odometer) + b 2I1 + b 3I2 + e
• White I1 = 1 if the color is white To represent a qualitative variable that has
• Silver 0 if the color is not white
m possible categories (levels), we must create
• Other colors I2 = 1 if the color is silver m-1 indicator variables.
0 if the color is not silver
Note: Color is a qualitative variable. White car
Other color
And what about “Other colors?” Set I1 = 0 and I2 = 0 Silver color
19 20
5
Example – Car Sale Price Example – Car Sale Price
There is insufficient There is sufficient
Create two variables: White and Silver and fill in the values: evidence to infer that a
either 0 or 1 evidence to infer that a
white color car and a car silver color car sells for
of “Other color” sell for a a larger price than a car
In Excel: different auction price. of the “Other color”
• use the IF function category.
Price Color Odometer white silver
5318 1 37388 1 0
white car
=IF(B2=2,1,0)
=IF(B2=1,1,0) 21 22
Example – Car Sale Price Example
If we disregard the fact that variable White is insignificant:
Create and identify indicator variables to represent the
From Excel we get the regression equation following qualitative variables.
PRICE = 6350 - 0.0278(ODOMETER) + 45.2 I1 + 148 I2 • Religious affiliation (Christian, Hindu, Jew, Muslim, Other)
For one additional mile the auction price
decreases by 2.78 cents. A white car sells, on the average,
for $45.2 more than a car of the “Other color” category
• Working shift (8:00am to 4:00pm, 4:00pm to 12:00 midnight,
Price 12:00 midnight to 8:00am)
A silver color car sells, on the average, for $148
The equation for a
car of silver color more than a car of the “Other color” category
Price = 6350 - .0278(Odometer) + 45.2(0) + 148(1)
The equation for a
• Supervisor (Ringo Star, Ronaldo, Derek Jeter, and Quentin
car of white color Tarantino)
The equation for a Price = 6350 - .0278(Odometer) + 45.2(1) + 148(0) ―Assume there are no other supervisors
car of the “Other color”
Price = 6350 - .0278(Odometer) + 45.2(0) + 148(0) ―Assume there are other supervisors
category. 23 24
Odometer
