2.1 regression

© 2008 Prentice-Hall, Inc.
Regression Models

© 2009 Prentice-Hall, Inc. 4 – 2
Introduction
 Regression analysis is a very valuable
tool for data analyst
 Regression can be used to
 Understand the relationship between
variables
 Predict the value of one variable based on
another variable
 Examples
 Determining best location for a new store
 Studying the effectiveness of advertising
dollars in increasing sales volume

Introduction
 Regression analysis is a form of predictive
modelling technique which investigates
the relationship between a dependent
(target) and independent variable
(s) (predictor). This technique is used for
forecasting, time series modelling and
finding the causal effect
relationship between the variables.

Introduction
 The variable to be predicted is called the
dependent variable
 Sometimes called the response variable
 The value of this variable depends on
the value of the independent variable
 Sometimes called the explanatory or
predictor variable
Independent
variable
Dependent
variable
Independent
variable
= +

Introduction
 Regression models are two types :
 Simple regression model and
 multiple regression model.
 Both are divided into linear and nonlinear
models.

Scatter Diagram
 Graphing is a helpful way to investigate
the relationship between variables
 A scatter diagram or scatter plot is
often used
 The independent variable is normally
plotted on the X axis
 The dependent variable is normally
plotted on the Y axis

Triple A Construction
 Triple A Construction renovates old homes
 They have found that the dollar volume of
renovation work is dependent on the area
payroll
SALES
($100,000’s)
LOCAL PAYROLL
($100,000,000’s)
6 3
8 4
9 6
5 4
4.5 2
9.5 5
Table 4.1

Figure 4.1
12 –
10 –
8 –
6 –
4 –
2 –
0 –
Sales($100,000)
Payroll ($100 million)
| | | | | | | |
0 1 2 3 4 5 6 7 8
ŷ

Figure 4.1
12 –
10 –
8 –
6 –
4 –
2 –
0 –
Sales($100,000)
| | | | | | | |
0 1 2 3 4 5 6 7 8
Ȳ
Ŷ
Ŷ - Ȳ
Y - Ŷ

Simple Linear Regression
where
Y = dependent variable (response)
X = independent variable (predictor or explanatory)
0 = intercept (value of Y when X = 0)
1 = slope of the regression line
e = random error
 Regression models are used to test if there is a
relationship between variables (predict sales
based on payroll)
 There is some random error that cannot be
predicted
e  XY 10

Simple Linear Regression
 True values for the slope and intercept are not
known so they are estimated using sample data
XbbY 10 ˆ
where
Y = dependent variable (response)
X = independent variable (predictor or explanatory)
b0 = intercept (value of Y when X = 0)
b1 = slope of the regression line
^

 Triple A Construction is trying to predict sales
based on area payroll
Y = Sales
X = Area payroll
 The line chosen in Figure 4.1 is the one that
minimizes the errors
Error = (Actual value) – (Predicted value)
YYe ˆ

Least Squares Regression
Errors can be positive or negative so the average error could
be zero even though individual errors could be large.
Least squares regression minimizes the sum of the squared
errors.
Payroll Line Fit Plot
0
2
4
6
8
10
0 2 4 6 8
Payroll ($100.000,000's)
Sales
($100,000)

 For the simple linear regression model, the
values of the intercept and slope can be
calculated using the formulas below
XbbY 10 ˆ
valuesof(mean)average X
n
X
X 

valuesof(mean)average Y
n
Y
Y 





 21
)(
))((
XX
YYXX
b
XbYb 10 

Y X (X – X)2 (X – X)(Y – Y)
6 3 (3 – 4)2 = 1 (3 – 4)(6 – 7) = 1
8 4 (4 – 4)2 = 0 (4 – 4)(8 – 7) = 0
9 6 (6 – 4)2 = 4 (6 – 4)(9 – 7) = 4
5 4 (4 – 4)2 = 0 (4 – 4)(5 – 7) = 0
4.5 2 (2 – 4)2 = 4 (2 – 4)(4.5 – 7) = 5
9.5 5 (5 – 4)2 = 1 (5 – 4)(9.5 – 7) = 2.5
ΣY = 42
Y = 42/6 = 7
ΣX = 24
X = 24/6 = 4
Σ(X – X)2 = 10 Σ(X – X)(Y – Y) = 12.5
Table 4.2
 Regression calculations

4
6
24
6

 X
X
7
6
42
6

Y
Y
251
10
512
21 .
.
)(
))((






XX
YYXX
b
24251710  ))(.(XbYb
XY 2512 .ˆ Therefore

4
6
24
6

 X
X
7
6
42
6

Y
Y
251
10
512
21 .
.
)(
))((






XX
YYXX
b
24251710  ))(.(XbYb
XY 2512 .ˆ Therefore
sales = 2 + 1.25(payroll)
If the payroll next
year is $600 million
000950$or5962512 ,.)(.ˆ Y

Measuring the Fit
of the Regression Model
 Regression models can be developed
for any variables X and Y
 How do we know the model is actually
helpful in predicting Y based on X?
 We could just take the average error, but
the positive and negative errors would
cancel each other out
 Three measures of variability are
 SST – Total variability about the mean
 SSE – Variability about the regression line
 SSR – Total variability that is explained by
the model

Measuring the Fit
 Sum of the squares total
2
)(  YYSST
 Sum of the squared error
   22
)ˆ( YYeSSE
 Sum of squares due to regression
  2
)ˆ( YYSSR
 An important relationship
SSESSRSST 

Measuring the Fit
Y X (Y – Y)2 Y (Y – Y)2 (Y – Y)2
6 3 (6 – 7)2 = 1 2 + 1.25(3) = 5.75 0.0625 1.563
8 4 (8 – 7)2 = 1 2 + 1.25(4) = 7.00 1 0
9 6 (9 – 7)2 = 4 2 + 1.25(6) = 9.50 0.25 6.25
5 4 (5 – 7)2 = 4 2 + 1.25(4) = 7.00 4 0
4.5 2 (4.5 – 7)2 = 6.25 2 + 1.25(2) = 4.50 0 6.25
9.5 5 (9.5 – 7)2 = 6.25 2 + 1.25(5) = 8.25 1.5625 1.563
∑(Y – Y)2 = 22.5 ∑(Y – Y)2 = 6.875 ∑(Y – Y)2 = 15.625
Y = 7 SST = 22.5 SSE = 6.875 SSR = 15.625
^
^^
^^
Table 4.3

 Sum of the squares total
2
)(  YYSST
 Sum of the squared error
   22
)ˆ( YYeSSE
 Sum of squares due to regression
  2
)ˆ( YYSSR
 An important relationship
 SSR – explained variability
 SSE – unexplained variability
SSESSRSST 
Measuring the Fit
For Triple A Construction
SST = 22.5
SSE = 6.875
SSR = 15.625

Measuring the Fit
Figure 4.2
12 –
10 –
8 –
6 –
4 –
2 –
0 –
Sales($100,000)
| | | | | | | |
0 1 2 3 4 5 6 7 8
Y = 2 + 1.25X
^
Y – Y
Y – Y
^
YY – Y
^

Coefficient of Determination
 The proportion of the variability in Y explained by
regression equation is called the coefficient of
determination
 The coefficient of determination is r2
SST
SSE
SST
SSR
r  12
 For Triple A Construction
69440
522
625152
.
.
.
r
 About 69% of the variability in Y is explained by
the equation based on payroll (X)

Correlation Coefficient
 The correlation coefficient is an expression of the
strength of the linear relationship
 It will always be between +1 and –1
 The correlation coefficient is r
2
rr 
8333069440 .. r

Correlation Coefficient
*
*
*
*
(a) Perfect Positive
Correlation:
r = +1
X
Y
*
* *
*
(c) No Correlation:
r = 0
X
Y
* *
*
*
* *
* **
*
(d) Perfect Negative
Correlation:
r = –1
X
Y
*
*
*
*
* *
*
*
*
(b) Positive
Correlation:
0 < r < 1
X
Y
*
*
*
*
*
*
*
Figure 4.3

Using Computer Software
for Regression
Program 4.1A

for Regression
Program 4.1B

for Regression
Program 4.1C

for Regression
Program 4.1D

for Regression
Program 4.1D
Correlation coefficient is
called Multiple R in Excel

Assumptions of the Regression Model
1. Errors are independent
2. Errors are normally distributed
3. Errors have a mean of zero
4. Errors have a constant variance
 If we make certain assumptions about the errors
in a regression model, we can perform statistical
tests to determine if the model is useful
 A plot of the residuals (errors) will often highlight
any glaring violations of the assumption

Residual Plots
 A random plot of residuals
Figure 4.4A
Error
X

Residual Plots
 Nonconstant error variance
Errors increase as X increases, violating the
constant variance assumption
Figure 4.4B
Error
X

Residual Plots
 Nonlinear relationship
Errors consistently increasing and then consistently
decreasing indicate that the model is not linear
Figure 4.4C
Error
X

Estimating the Variance
 Errors are assumed to have a constant
variance ( 2), but we usually don’t know
this
 It can be estimated using the mean
squared error (MSE), s2
1
2


kn
SSE
MSEs
where
n = number of observations in the sample
k = number of independent variables

Estimating the Variance
71881
4
87506
116
87506
1
2
.
..





kn
SSE
MSEs
 We can estimate the standard deviation, s
 This is also called the standard error of the
estimate or the standard deviation of the
regression
31171881 ..  MSEs

Testing the Model for Significance
 When the sample size is too small, you
can get good values for MSE and r2 even if
there is no relationship between the
variables
 Testing the model for significance helps
determine if the values are meaningful
 We do this by performing a statistical
hypothesis test

 We start with the general linear model
e  XY 10
 If 1 = 0, the null hypothesis is that there is
no relationship between X and Y
 The alternate hypothesis is that there is a
linear relationship (1 ≠ 0)
 If the null hypothesis can be rejected, we
have proven there is a relationship
 We use the F statistic for this test

 The F statistic is based on the MSE and MSR
k
SSR
MSR 
where
k = number of independent variables in the model
 The F statistic is
MSE
MSR
F 
 This describes an F distribution with
degrees of freedom for the numerator = df1 = k
degrees of freedom for the denominator = df2 = n – k – 1

 If there is very little error, the MSE would
be small and the F-statistic would be large
indicating the model is useful
 If the F-statistic is large, the significance
level (p-value) will be low, indicating it is
unlikely this would have occurred by
chance
 So when the F-value is large, we can reject
the null hypothesis and accept that there is
a linear relationship between X and Y and
the values of the MSE and r2 are
meaningful

Steps in a Hypothesis Test
1. Specify null and alternative hypotheses
010 :H
011 :H
2. Select the level of significance (). Common
values are 0.01 and 0.05
3. Calculate the value of the test statistic using the
formula
MSE
MSR
F 

Steps in a Hypothesis Test
4. Make a decision using one of the following
methods
a) Reject the null hypothesis if the test statistic is
greater than the F-value from the table in Appendix D.
Otherwise, do not reject the null hypothesis:
21
ifReject dfdfcalculated FF ,,
kdf 1
12  kndf
b) Reject the null hypothesis if the observed significance
level, or p-value, is less than the level of significance
(). Otherwise, do not reject the null hypothesis:
)( statistictestcalculatedvalue-  FPp
value-ifReject p

Step 1.
H0: 1 = 0 (no linear relationship between
X and Y)
H1: 1 ≠ 0 (linear relationship exists
between X and Y)
Step 2.
Select  = 0.05
625015
1
625015
.
.

k
SSR
MSR
099
71881
625015
.
.
.

MSE
MSR
F
Step 3.
Calculate the value of the test statistic

Step 4.
Reject the null hypothesis if the test statistic
is greater than the F-value in Appendix D
df1 = k = 1
df2 = n – k – 1 = 6 – 1 – 1 = 4
The value of F associated with a 5% level of
significance and with degrees of freedom 1
and 4 is found in Appendix D
F0.05,1,4 = 7.71
Fcalculated = 9.09
Reject H0 because 9.09 > 7.71

F = 7.71
0.05
9.09
Figure 4.5
 We can conclude there is a
statistically significant
relationship between X and Y
 The r2 value of 0.69 means
about 69% of the variability
in sales (Y) is explained by
local payroll (X)

r2 coefficient of determination
 The F-test determines whether or not there
is a relationship between the variables.
 r2 (coefficient of determination) is the best
measure of the strength of the prediction
relationship between the X and Y variables.
• Values closer to 1 indicate a strong prediction
relationship.
• Good regression models have a low
significance level for the F-test and high r2
value.

Coefficient Hypotheses
 Statistical tests of significance can be performed
on the coefficients.
 The null hypothesis is that the coefficient of X (i.e.,
the slope of the line) is 0 i.e., X is not useful in
predicting Y
 P values are the observed significance level and
can be used to test the null hypothesis.
 Values less than 5% negate the null hypothesis and
indicate that X is useful in predicting Y
 For a simple linear regression, the test of the
regression coefficients gives the same information
as the F-test.

Analysis of Variance (ANOVA) Table
 When software is used to develop a regression
model, an ANOVA table is typically created that
shows the observed significance level (p-value)
for the calculated F value
 This can be compared to the level of significance
() to make a decision
DF SS MS F SIGNIFICANCE
Regression k SSR MSR = SSR/k MSR/MSE P(F > MSR/MSE)
Residual n - k - 1 SSE MSE =
SSE/(n - k - 1)
Total n - 1 SST
Table 4.4

ANOVA for Triple A Construction
 Because this probability is less than 0.05, we
reject the null hypothesis of no linear relationship
and conclude there is a linear relationship
between X and Y
Program 4.1D
(partial)
P(F > 9.0909) = 0.0394

Multiple Regression Analysis
 Multiple regression models are
extensions to the simple linear model
and allow the creation of models with
several independent variables
Y = 0 + 1X1 + 2X2 + … + kXk + e
where
Y = dependent variable (response variable)
Xi = ith independent variable (predictor or explanatory
variable)
0 = intercept (value of Y when all Xi = 0)
I = coefficient of the ith independent variable
k = number of independent variables
e = random error

Multiple Regression Analysis
 To estimate these values, a sample is taken
the following equation developed
kk XbXbXbbY  ...ˆ 22110
where
= predicted value of Y
b0 = sample intercept (and is an estimate of 0)
bi = sample coefficient of the ith variable (and is
an estimate of i)
Yˆ

Jenny Wilson Realty
 Jenny Wilson wants to develop a model to
determine the suggested listing price for houses
based on the size and age of the house
kk XbXbXbbY  ...ˆ 22110
where
= predicted value of dependent variable (selling
price)
b0 = Y intercept
X1 and X2 = value of the two independent variables (square
footage and age) respectively
b1 and b2 = slopes for X1 and X2 respectively
Yˆ
 She selects a sample of houses that have sold
recently and records the data shown in Table 4.5

Jenny Wilson Realty
SELLING
PRICE ($)
SQUARE
FOOTAGE
AGE CONDITION
95,000 1,926 30 Good
119,000 2,069 40 Excellent
124,800 1,720 30 Excellent
135,000 1,396 15 Good
142,000 1,706 32 Mint
145,000 1,847 38 Mint
159,000 1,950 27 Mint
165,000 2,323 30 Excellent
182,000 2,285 26 Mint
183,000 3,752 35 Good
200,000 2,300 18 Good
211,000 2,525 17 Good
215,000 3,800 40 Excellent
219,000 1,740 12 Mint
Table 4.5

Jenny Wilson Realty
Program 4.2
21 289944146631 XXY ˆ

Evaluating Multiple Regression Models
 Evaluation is similar to simple linear
regression models
 The p-value for the F-test and r2 are
interpreted the same
 The hypothesis is different because there
is more than one independent variable
 The F-test is investigating whether all
the coefficients are equal to 0

Evaluating Multiple Regression Models
 To determine which independent
variables are significant, tests are
performed for each variable
010 :H
011 :H
 The test statistic is calculated and if the
p-value is lower than the level of
significance (), the null hypothesis is
rejected

Jenny Wilson Realty
 The model is statistically significant
 The p-value for the F-test is 0.002
 r2 = 0.6719 so the model explains about 67% of
the variation in selling price (Y)
 But the F-test is for the entire model and we can’t
tell if one or both of the independent variables are
significant
 By calculating the p-value of each variable, we can
assess the significance of the individual variables
 Since the p-value for X1 (square footage) and X2
(age) are both less than the significance level of
0.05, both null hypotheses can be rejected

Binary or Dummy Variables
 Binary (or dummy or indicator) variables
are special variables created for
qualitative data
 A dummy variable is assigned a value of
1 if a particular condition is met and a
value of 0 otherwise
 The number of dummy variables must
equal one less than the number of
categories of the qualitative variable

Jenny Wilson Realty
 Jenny believes a better model can be developed if
she includes information about the condition of
the property
X3 = 1 if house is in excellent condition
= 0 otherwise
X4 = 1 if house is in mint condition
= 0 otherwise
 Two dummy variables are used to describe the
three categories of condition
 No variable is needed for “good” condition since
if both X3 and X4 = 0, the house must be in good
condition

Jenny Wilson Realty
Program 4.3

Jenny Wilson Realty
Program 4.3
Model explains about
90% of the variation
in selling price
F-value
indicates
significance
Low p-values
indicate each
variable is
significant
4321 369471623396234356658121 XXXXY ,,,.,ˆ 

Model Building
 The best model is a statistically significant
model with a high r2 and few variables
As more variables are added to the model, the
r2-value usually increases
For this reason, the adjusted r2 value is often
used to determine the usefulness of an
additional variable
The adjusted r2 takes into account the number
of independent variables in the model
When variables are added to the model, the
value of r2 can never decrease; however, the
adjusted r2 may decrease.

Model Building
SST
SSE
SST
SSR
 12
r
 The formula for r2
 The formula for adjusted r2
)/(SST
)/(SSE
1
1
1Adjusted 2



n
kn
r
 As the number of variables increases, the
adjusted r2 gets smaller unless the increase due
to the new variable is large enough to offset the
change in k

Model Building
 It is tempting to keep adding variables to a model
to try to increase r2
 The adjusted r2 will decrease if additional
independent variables are not beneficial.
 As the number of variables (k) increases, n-k-1
decreases.
 This causes SSE/(n-k-1) to increase which in turn
decreases the adjusted r2 unless the extra variable
causes a significant decrease in SSE
 The reduction in error (and SSE) must be sufficient
to offset the change in k

Model Building
 In general, if a new variable increases the adjusted
r2, it should probably be included in the model
 In some cases, variables contain duplicate
information
 When two independent variables are correlated,
they are said to be collinear (e.g., monthly salary
expenses and annual salary expenses)
 When more than two independent variables are
correlated, multicollinearity exists
 When multicollinearity is present, hypothesis
tests for the individual coefficients are not valid
but the model may still be useful

Nonlinear Regression
 In some situations, variables are not linear
 Transformations may be used to turn a
nonlinear model into a linear model
*
* **
** *
* *
Linear relationship Nonlinear relationship
* *
** **
*
*
**
*

Colonel Motors
 The engineers want to use regression analysis to
improve fuel efficiency
 They have been asked to study the impact of
weight on miles per gallon (MPG)
MPG
WEIGHT
(1,000 LBS.) MPG
WEIGHT
(1,000 LBS.)
12 4.58 20 3.18
13 4.66 23 2.68
15 4.02 24 2.65
18 2.53 33 1.70
19 3.09 36 1.95
19 3.11 42 1.92
Table 4.6

Colonel Motors
Figure 4.6A
45 –
40 –
35 –
30 –
25 –
20 –
15 –
10 –
5 –
0 – | | | | |
1.00 2.00 3.00 4.00 5.00
MPG
Weight (1,000 lb.)





 




Linear model
110 XbbY ˆ

Colonel Motors
Program 4.4
 A useful model with a small F-test for
significance and a good r2 value

Colonel Motors
Figure 4.6B
45 –
40 –
35 –
30 –
25 –
20 –
15 –
10 –
5 –
0 – | | | | |
1.00 2.00 3.00 4.00 5.00
MPG
Weight (1,000 lb.)





 




Nonlinear model
2
210 weightweight )()(MPG bbb 

Colonel Motors
 The nonlinear model is a quadratic model
 The easiest way to work with this model is to
develop a new variable
2
2 weight)(X
 This gives us a model that can be solved with
linear regression software
22110 XbXbbY ˆ

Colonel Motors
Program 4.5
 A better model with a smaller F-test for
significance and a larger adjusted r2 value
21 43230879 XXY ...ˆ 

Cautions and Pitfalls
 If the assumptions are not met, the statistical
test may not be valid
 Correlation does not necessarily mean
causation
 Your annual salary and the price of cars may be
correlated but one does not cause the other
 Multicollinearity makes interpreting
coefficients problematic, but the model may
still be good
 Using a regression model beyond the range of
X is questionable, the relationship may not
hold outside the sample data

Cautions and Pitfalls
 t-tests for the intercept (b0) may be ignored
as this point is often outside the range of
the model
 A linear relationship may not be the best
relationship, even if the F-test returns an
acceptable value
 A nonlinear relationship can exist even if a
linear relationship does not
 Just because a relationship is statistically
significant doesn't mean it has any
practical value

2.1 regression

More Related Content

What's hot (20)

Similar to 2.1 regression (20)

Recently uploaded (20)

2.1 regression