MULTIPLE REGRESSION AND

CORRELATION ANALYSIS
• The linear regression equation
using one independent
variable has the form
Y  a  bX
/
• The multiple regression case
extends the equation to
include additional independent
variables. For two independent
variables, the general form of
the multiple regression is: Y  a  b1 X 1  b2 X 2
/

Where :
X1 and X2 are the two
independent variables
• a is the Y intercept
• b1 is the net change in Y for each unit change in
X1, while holding X2 constant. It is called a partial
regression coefficient, a net regression
coefficient , or just a regression coefficient.
• b2 is the net change in Y for each unit change in
X2, while holding X1 holding constant. It is also
called a partial regression coefficient, a net
regression coefficient , or just a regression
coefficient.
• For three independent variables
designated X1, X2, and X3, the general
multiple regression equation is :
• Y/ = a+ B1 X1 +b2 X2 + b3 X3
 Example:

• A Real Estate sells homes in the capital city.

One of the questions most frequently asked by
prospective buyers is: If we purchase this home,
how much can we expect to pay to heat it during
the winter? Three variables are thought to relate
to the heating costs:
• the mean daily outside temperature,
• the number of inches of insulation in the attic,
and
• the age of the furnace
• To investigate, they have selected a
random sample of 20 recently sold homes.
They determined the cost to heat the
home last January, as well as the January
outside temperature, the number of inches
of insulation in the attic, and the age of the
furnace. The sample information is given
in the following table:
Home Heating Cost ($) Mean Outside Attic Insulation Age of furnace
temperature (inches) (years)
(0F)
1 250 35 3 6
2 360 29 4 10
3 165 36 7 3
4 43 60 6 9
5 92 65 5 6
6 200 30 5 5
7 355 10 6 7
8 290 7 10 10
9 230 21 9 11
10 120 55 2 5
11 73 54 12 4
12 205 48 5 1
13 400 20 5 15
14 230 39 4 7
15 72 60 8 6
16 272 20 5 8
17 94 58 7 3
18 190 40 8 11
19 235 27 9 8
20 139 30 7 5
• Determine the multiple regression equation.
Which variables are the independent
variables? Which variable is the dependent
variable? Discuss the regression coefficients.
What does it indicate that some are positive
and some are negative? What is the intercept
value?
• What is the estimated heating cost for a
home if the mean outside temperature is 30
degrees, there are 5 inches of insulation in
the attic, and the furnace is 10 years old?
Solution:
The dependent variable is the January heating cost. There
are three independent variables: the mean outside
temperature, the number of inches of insulation in the attic,
and the age of the furnace.
The general form of a multiple regression equation with
three independent variables is:
Y/ = a+ b1 X1 +b2 X2 +b3 X3
Using Statistical software the multiple regression equation
is:
Y/ = 427 – 4.58X1 – 14.8X2 + 6.10X3
The intercept value is 427. This is the point where the
regression equation crosses the Y – axis
• The regression coefficients for the mean outside
temperature and the amount of attic insulation are
both negative. As the outside temperature increases,
the cost to heat the home will go down. It is an inverse
relationship. For each degree the mean temperature
increases, we expect the heating cost decreases
$4.58 per month.
• The variable “attic insulation” also shows an inverse
relationship: the more insulation in the attic the less
the cost to heat the home. So the negative sign for this
coefficient is logical. For each additional inch of
insulation, we expect the cost to heat the home to
decline $14.80 per month, regardless of the outside
temperature or the age of the furnace.
• The furnace variable shows a direct relationship. For
each additional year old the furnace is, we expect the
cost to increase $6.10per month.
If the mean outside temperature for the month is
30 degrees, there are 5 inches of insulation in
the attic, and the furnace is 10 years old, the
estimated heating cost for the month :
Y/ = a+ b1 X1 +b2 X2 +b3 X3 =427 – 4.58X1 –
14.8X2 + 6.10X3
=427- 4.58(30)- 14.8(5) +6.10(10) =
276.60.

• Self – Review 14 – 1 Page 507.

• Exercises: 1, 3. Page 507 – 508.
 MULTIPLE STANDARD ERROR OF
ESTIMATE

• In our previous example, we have estimated the heating

cost to be $276.60. We would expect to find some
random error in this estimate. Sometimes a home with
these statistics would cost more than $276.60 to heat
and other times less. The error in this estimate is
measured by the multiple standard error of estimate .
The standard error is denoted by sy . 123. The
subscripts indicate that three independent
variables are being used to estimate the value of
Y.
• The standard error of estimate describes
the variation around the regression line. A
small standard error indicates the points
are close to the regression line, whereas a
large value indicates the points are
scattered about the regression line. The
standard error is a measure of the error or
variability in the prediction.
Formula for the multiple standard error
of estimate

s y .12...k 
 Y  Y  / 2

n  (k  1)

Y is the observation.
Y/ is the value estimated from the regression nequation.
n is the number of observation in the sample
k is the number of independent variables.
• From our previous example, the Y/ values
are found by using the regression equation
and given in the following table:
Home Mean Attic Insu- Age of Fur- Cost Y/ Y- Y/ (Y-Y/)2
Outside lation(inche Nace(years) Y Residuals
temperature s)
(0F)
1 35 3 6 250 258.90 -8.90 79.21
2 29 4 10 360 295.98 64.02 4098.56
3 36 7 3 165 176.82 -11.82 139.71
4 60 6 9 43 118.30 -75.30 5670.09
5 65 5 6 92 91.90 0.10 0.01
6 30 5 5 200 246.10 -46.10 2125.21
7 10 6 7 355 335.10 19.90 396.01
8 7 10 10 290 307.94 -17.94 321.84
9 21 9 11 230 264.72 -34.72 1205.48
10 55 2 5 120 176.00 -56.00 3136.00
11 54 12 4 73 26.48 46.52 2164.11
12 48 5 1 205 139.26 65.74 4321.75
13 20 5 15 400 352.90 47.10 2218.41
14 39 4 7 320 231.88 88.12 7765.13
15 60 8 6 72 70.40 1.60 2.56
16 20 5 8 272 310.20 -38.20 1459.24
17 58 7 3 94 76.06 17.94 321.84
18 40 8 11 190 192.50 -2.50 6.25
19 27 9 8 235 218.94 16.06 257.92
20 30 7 5 139 216.50 -77.50 6006.25

Total 41695.58
• Here n=20 and k=3 (three independent
variables), so the multiple error of estimate is:

41695.58
 51.05
20  (3  1)
• How do you interpret the 51.05?
• It is the typical error we make when we use
this equation to predict the cost. If the errors
are normally distributed, about 68 percent of
the residuals should be less than 51.05 and
about 95 percent should be less than
2(51.05) or 102.10 .
 THE ANOVA TABLE
• The multiple regression calculations are lengthy. Many
software systems are available to perform the
calculations
• In the analysis of variance table, the variation is
divided into two components: that due to the
treatments and that due to random error. Here the
total is also divided into two components: that
explained by the regression, that is, the independent
variables and the error or unexplained variation.
These two categories are identified in the source
column of the analysis of variance table. In our
example, there are 20 observations, so n=20. The
total number of degrees of freedom is n-1=20-1=19.
The number of degrees of freedom in the regression
row is the number of independent variables. We let k
represent the number of independent variables , so
k=3. The number of degrees of degrees of freedom in
the error row is n-(k+1)=20-(3+1)=16 df.
• The heading “SS” in the middle of the ANOVA
table refers top the sum of squares, or the
variation.
• Total variation = SS total (Y - Y)2 = 212916
• Error Variation = SSE =(Y – Y/) 2 = 41695
• Regression Variation + SSR = SS total – SSE =
212916 – 41695 = 171220.
 The general format of the ANOVA table is:

Source df SS MS F

Regres k SSR MSR=SSR/k MSR/MSE

sion

Error n– SSE MSE=SSE/n-

(k+1) (k+1)

Total n-1 SS
total
• The coefficient of multiple determination,
written as R2 is the percent of the variation
explained by the regression . It is the sum
of squares due to the regression , divided
by the sum of squares total.
Coefficient Of Multiple Determination

SSR 171220
R 
2
  0.804
SStotal 212916
The multiple standard error may be found
directly from the ANOVA table:

SSE 41695
s y.123    51.05
n  (k  1) 20  (3  1)

Self – Review 14 – 2 Page 513.

Exercise: 5,6.
 EVALUATING THE REGRESSION
EQUATION
• Using a scatter Diagram.
• Correlation Matrix
Correlation Matrix: A matrix showing the
coefficients of correlation between all
pairs of variables.
A correlation matrix is also useful in
analyzing the factors involved in the cost
to heat a home.
• Cost is the dependent variable, Y. We are
particularly interested in independent
variables that have a strong correlation
with the dependent variable. We may wish
to develop a simpler multiple regression
equation using fewer independent
variables and the correlation matrix helps
us identify which may be relatively more
important.
 A correlation matrix

Cost Temp. Insul. Age

Cost 1

Temp. - 0.81151 1

Insul. - 0.2571 -0.10302 1

Age 0.536728 - 0.48599 0.063617 1

• As we can see, temperature has the
strongest correlation with cost,
• 0.81151. The negative sign indicates the
inverse relationship. Age has a stronger
correlation with cost than insulation and
the correlation between cost and the age
of the furnace is direct. It is 0.536728.
• A second use of the correlation matrix is to
check for multicollinearity.
• MULTICOLLINEARITY: Correlation among the
independent variables.
• Multicollinearity can distort the standard error of
estimate and may, therefore, lead to incorrect
conclusions as to which independent variables
are statistically significant. In this case, the
correlation between the age of the furnace and
the temperature is the strongest, but it is not
large enough to cause a problem.
• A common rule of thumb is that correlation
among the independent variables between – .
70 and .70 do not cause difficulties.
• The usual remedy for multicollinearity is to drop
one of the independent variables that are
strongly correlated and recompute the
regression equation.
 GLOBAL TEST:

• Testing Whether the Multiple Regression Model Is

Valid
The ability of the independent variables X1, X2, …,
Xk to explain the behavior of the dependent
variable Y can be tested. The test used is
referred to as the global test. Basically, it
investigates whether it is possible all the
independent variables have zero net regression
coefficients. To put it another way, could the
amount of explained variation, R2 , occur by
chance?
• To relate this question to the heating cost example, we
will test whether the independent variables ( amount of
insulation in the attic, mean daily outside temperature,
and age of furnace) are capable of effectively
estimating home heating costs.
• In our example, there are three independent variables.
b1, b2, and b3 are sample net regression coefficients.
The corresponding coefficients in the population are
given the symbols 1, 2, 3. We now test whether
the net regression coefficients in the population are
zero.
• The null hypothesis is:
• H0: 1 = 2 = 3 = 0
The alternate hypothesis is:
• H1 : Not all the s are 0.
• [If the null hypothesis is true, it implies the regression
coefficients are all zero and logically, are of no use in
estimating the dependent variable ( heating cost).]
Should that be the case, we would have to search for
some other independent variables – or take a different
approach – to predict home heating costs.
• To test the null hypothesis that the multiple regression
coefficients are all zero, we employ the F distribution.
Let  = .05 level of significance.
• The df can be found in the software summary in
the analysis of variance table:
ANOVA TABLE

Source DF SS MS F p
Regression 3 171220 57073 21.90 0.000
Residual Error 16 41695 2606

Total 19 212915

[ No.of independent variables, k = 3, the df in the numerator, n – (k+1) = 20-(3+1) =16 df in

denominator.]
• The value of F is found from the following
equation:
Global Test

SSR / k MSR
F 
SSE /[n  (k  1)] MSE

Inserting these values in the formula

SSR / k 171220 / 3
F   21.90
SSE /[n  (k  1)] 41695 /[20  (3  1)]
The critical value is found from in Appendix G. It is 3.24.
• The decision rule is: if the computed F is greater
than 3.24, reject H0 and accept the alternate
hypothesis, H1. The computed value of F is 21.90,
which is in the rejection region. The null hypothesis
that all the multiple regression coefficients are zero is
rejected. The p-value is 0.000 from the above table, so
it is quite unlikely that H0 is true. The alternate
hypothesis is accepted, indicating that not all the
regression variables (amount of insulation, etc.) do
have the ability to explain the variation in the
dependent variable (heating cost).
• The outside temperature, the amount of insulation and
age of the furnace have a great bearing on heating
costs.
 Evaluating Individual Regression
Coefficients
• The next step is to test the variables individually
to determine which regression coefficients may
be 0 and which are not.
• [Why is it important to find whether it is possible
that any of the s equal 0? If a  could equal 0, it
implies that this particular independent variable
is of no value in explaining any variation in the
dependent value. If there are coefficients for
which H0 cannot be rejected, we may want to
eliminate them from the regression equation.]
• We will now conduct three separate tests of
hypothesis – for temperature, for insulation, and
for the age of the furnace.

For temperature: For insulation: For furnace age:

H0: 1 = 0 H0: 2 = 0 H0: 3 = 0
H1: 1  0 H1: 2  0 H1: 3  0