Using Data

Ertunga Ozelkan, Ph.D.

Professor Systems Engineering
& Engineering Management

Based on Clemen & Reilly, Making Hard Decisions with Decision Tools, Duxbury Press, 2004.

2018 © Dr. E.C. Ozelkan Using Data

Copyright and Usage of this Material

 This material is copyrighted by Dr. Ertunga C. Ozelkan, and

it is shared with you for your education purposes in the class
you are attending. You may not distribute or reproduce this
material for any reason or may not use this material to train
others or may not sell it in any part or in any form without the
written consent of Dr. Ertunga C. Ozelkan.

2018 © Dr. Ozelkan Slide 2 Using Data

2018 © Dr. E.C. Ozelkan

1
 Agenda

 Part 1: Fitting distributions to data

 Part 2: Linear Regression Analysis

2018 © Dr. Ozelkan Slide 3 Using Data

Using Data Part 1:

Fitting Distributions to Data

Ertunga Ozelkan, Ph.D.

Professor Systems Engineering
& Engineering Management

Based on Clemen & Reilly, Making Hard Decisions with Decision Tools, Duxbury Press, 2004.

2018 © Dr. E.C. Ozelkan Using Data

2018 © Dr. E.C. Ozelkan

2
 Objectives

 Construct probability distributions from existing data

 Fit theoretical probability distributions to data

2018 © Dr. Ozelkan Slide 5 Using Data

Constructing Probability Distributions

 Discrete probability
 Create Histograms
 Continuous probability
 Construct CDFs (cumulative density functions)
1. Rank order the data (sort n data points from smallest to largest)
 y1 < y2 < … < yn
2. Calculate centers of adjacent data points xm =(ym + ym+1)/2
3. Assign probability to each center (xm) such that P(X<=xm)=m/n

2018 © Dr. Ozelkan Slide 6 Using Data

2018 © Dr. E.C. Ozelkan

3
 Example: Construct CDFs
y x
rental mid cumulative
rank costs points probability P(X<=64)=1/35
1 52 64.0 0.029
2 76 88.0 0.057
P(X<=88)=2/35
3 100 118.0 0.086 18 325 335.0 0.514
4 136 136.5 0.114 19 345 359.0 0.543
5 137 161.5 0.143 20 373 378.5 0.571
6 186 191.0 0.171 21 384 392.0 0.600
7 196 200.5 0.200 22 400 401.0 0.629
8 205 227.5 0.229 23 402 405.0 0.657
9 250 253.5 0.257 24 408 412.5 0.686
10 257 260.5 0.286
25 417 419.5 0.714
11 264 272.0 0.314
26 422 447.0 0.743
12 280 281.0 0.343
27 472 476.0 0.771
13 282 282.5 0.371
28 480 561.5 0.800
14 283 293.0 0.400
29 643 668.0 0.829
15 303 308.0 0.429
16 313 315.0 0.457 30 693 712.5 0.857
17 317 321.0 0.486 31 732 740.5 0.886
32 749 749.5 0.914
33 750 770.5 0.943
34 791 841.0 0.971
n=35 35 891 1

2018 © Dr. Ozelkan Slide 7 Using Data

Example: Construct CDFs

 Plot the CDF

1.000
Cumulative Probability

0.750

0.500

0.250

0.000
0 200 400 600 800
Yearly Bed Rental Cost

2018 © Dr. Ozelkan Slide 8 Using Data

2018 © Dr. E.C. Ozelkan

4
 Fitting a Theoretical Probability Distribution

 Requires identifying that theoretical distribution that

closely matches the sample data distribution
 Statistically speaking we need to test the following
hypothesis:
 Ho: The data follow a specified distribution (e.g. Binomial,
Normal, etc.)
 Ha: The data do not follow the specified distribution.
 To test the hypothesis a “Test Statistic” needs to be
computed.
Goodness-of-fit Test

2018 © Dr. Ozelkan Slide 9 Using Data

Goodness of Fit

 For Discrete and Continuous Distributions

 Chi-Square Statistic

 For Continuous Distributions

 Kolmogorov-Smirnov Statistic
 Anderson-Darling Statistic

2018 © Dr. Ozelkan Slide 10 Using Data

2018 © Dr. E.C. Ozelkan

5
 Chi-Square Goodness-of-fit Test
 Can be used for Discrete and Continuous Distributions
 It is applied to binned data (i.e., data put into K bins or classes).
 The test is dependent on how the data is binned (e.g. use equiprobable bins,
some software use 0.3s rule where s is the standard deviation)
 Requires sufficient data in order for the chi-square approximation to be valid
(have at least 5 data points per bin, combine bins if data is not sufficient)
 Test Statistic:
K K: number of bins
 2   ( N i  Ei ) 2 / Ei Ni: observed number of data points in bin i
i 1 Ei: expected number of data points in bin i

F: Cumulative Distribution for the distribution being tested

Ei  n[ F (YU )  F (YL )] Yu: is the upper limit for bin i
Yl: is the lower limit for bin i

CS test rejects Ho when n: is the total sample size.

 2  2 ,K c
c: estimated number of parameters
α: significance level for the test
2018 © Dr. Ozelkan Slide 11 Using Data

Kolmogorov-Smirnov Goodness-of-fit Test

 Can be used for Continuous Distributions

 It is proven that the test
 Does not capture spread
 Good in capturing shifts in median values
 Test Statistic:
Dn  max Pn ( xi )  F ( xi )
x

n: number of data points (or sample size)

Pn (xi): cumulative probability based on empirical data and data point xi
F(xi): cumulative probability based on theoretical distribution and data point xi
Dn: maximum vertical distance between the empiric distribution and theoretical distribution

 K-S test rejects Ho when Dn > dn,α where

 0.01 0.05 0.1
d n , 1.63 / n 1.36 / n 1.22 / n

2018 © Dr. Ozelkan Slide 12 Using Data

2018 © Dr. E.C. Ozelkan

6
 Anderson-Darling Goodness-of-fit Test
 Can be used for Continuous Distributions
 It is proven that the test
 Captures deviations in spreads
 Test Statistic:
A2  n  S
 where
2i  1
ln F (Yi )  ln[1  F (Yn1i )]
n
S
i 1 n
n: number of data points (or sample size)
Yi: ordered data
F(Yi): cumulative probability based on theoretical distribution and data point Yi

 A-D test rejects Ho when A2 > an,α,distribution

 The test statistic needs to be computed for each type of distribution to be tested.

2018 © Dr. Ozelkan Slide 13 Using Data

Fitting a Theoretical Probability Distribution

1. Select a few possible distribution functions

 E.g. Normal Distribution N(μ,σ)
2. Estimate the parameters for the distribution functions
 E.g. μ and σ
3. Select a goodness of fit technique and check the goodness
of fit for each distribution function
 Chi-Square Test (Chi-Sq)
 Kolmogorov-Smirnov Test (K-S)
 Anderson-Darling Test (A-D)
4. Select the distribution that yields the best fit

2018 © Dr. Ozelkan Slide 14 Using Data

2018 © Dr. E.C. Ozelkan

7
 Step 1. Select Possible Distributions

 Assume the following possible distributions are selected for

the rental cost data:
 Normal distribution
 Exponential rental mid cumulative 18 325 335.0 0.514
rank costs points probability 19 345 359.0 0.543
1 52 64.0 0.029 20 373 378.5 0.571
2 76 88.0 0.057 21 384 392.0 0.600
3 100 118.0 0.086 22 400 401.0 0.629
4 136 136.5 0.114 23 402 405.0 0.657
5 137 161.5 0.143
24 408 412.5 0.686
6 186 191.0 0.171
25 417 419.5 0.714
7 196 200.5 0.200
26 422 447.0 0.743
8 205 227.5 0.229
9 250 253.5 0.257
27 472 476.0 0.771
10 257 260.5 0.286 28 480 561.5 0.800
11 264 272.0 0.314 29 643 668.0 0.829
12 280 281.0 0.343 30 693 712.5 0.857
13 282 282.5 0.371 31 732 740.5 0.886
14 283 293.0 0.400 32 749 749.5 0.914
15 303 308.0 0.429 33 750 770.5 0.943
16 313 315.0 0.457 34 791 841.0 0.971
17 317 321.0 0.486 35 891 1
2018 © Dr. Ozelkan Slide 15 Using Data

Step 2. Estimate Parameters

If Rental Costs are Normally

Distributed, then we need to
estimate μ and σ
n
ˆ  x   xi / n  380.4
i 1
( y   )2
1 
f N (Y  y |  ,  )  e 2 2 n
 2 ˆ  s   ( xi x )2 /(n  1)  217.6
i 1
E(Y)  
Var(Y)   2

2018 © Dr. Ozelkan Slide 16 Using Data

2018 © Dr. E.C. Ozelkan

8
 Step 2. Estimate Parameters

If Rental Costs are

Exponentially Distributed, then
we need to estimate m

f E (T  t | m)  me  mt n
mˆ  x   xi / n  380.4
1 i 1
E (T ) 
m
1
Var (T ) 
m2

2018 © Dr. Ozelkan Slide 17 Using Data

Step 3. Compute Goodness of Fit

Normal(380.40, 217.59)
1.0

Normal
Distribution
0.8

0.6

BestFit Student Version

For Academic Use Only
Chi-Sq: 9.2
0.4
K-S: 0.167
A-D:1.027
0.2

0.0
0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

Values in Thousands

< 35.0% 30.0% 35.0% >

0.297 0.464

2018 © Dr. Ozelkan Slide 18 Using Data

2018 © Dr. E.C. Ozelkan

9
 Step 3. Compute Goodness of Fit
Expon(328.40) Shift=+42.617
1.0

0.8
Exponential
Distribution
0.6

BestFit Student Version

For Academic Use Only Chi-Sq: 10.8
0.4 K-S: 0.2396
A-D: 2.195
0.2

0.0
0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

Values in Thousands 1.0

35.0% 30.0% 35.0% >

0.184 0.387

2018 © Dr. Ozelkan Slide 19 Using Data

Step 4. Select Best Distribution

Normal(380.40, 217.59)
Expon(328.40) Shift=+42.617
1.0
1.0

0.8
0.8

0.6

Normal: 0.6
Exponential:
BestFit Student Version BestFit Student Version
For Academic Use Only For Academic Use Only

0.4 Chi-Sq: 9.2 0.4

Chi-Sq: 10.8
K-S: 0.167 K-S: 0.2396
A-D:1.027 A-D: 2.195
0.2 0.2

0.0 0.0
0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

Values in Thousands Values in Thousands

< 35.0% 30.0% 35.0% > 35.0% 30.0% 35.0% >
0.297 0.464 0.184 0.387

2018 © Dr. Ozelkan Slide 20 Using Data

2018 © Dr. E.C. Ozelkan

10
 Lab Exercise

 Do the class example in the computer lab using the @Risk

Software which is part of the Decision Tools Suite.
 Follow instructions on pages 405-411 of your book.
 Use data file posted on Canvas named [Link]

2018 © Dr. Ozelkan Slide 22 Using Data

Using Data Part 2:

Linear Regression Analysis

Ertunga Ozelkan, Ph.D.

Associate Professor
Systems Engineering &
Engineering Management

2018 © Dr. E.C. Ozelkan Using Data

2018 © Dr. E.C. Ozelkan

11
 Objectives

 Get familiar with Linear Regression Models

 Parameter estimation
 Hypothesis testing
 Significance of overall regression
 Significance of each model parameter

2018 © Dr. Ozelkan

Using Data

Why Regression?

 To be able to predict and extrapolate

 Beyond observation points
 To identify a functional relationship to interpolate
 Between observations
 In an Design of Experiments (DOE) setting may be useful
 Again predict and interpolate between the controlled factor levels
 Data might be available from unplanned experiments
 The factor levels might not be fixed
 There might be missing data in planned experiments

2018 © Dr. Ozelkan

Using Data

2018 © Dr. E.C. Ozelkan

12
 Linear Regression

 Most widely used form of regression

 y = a + b·x
y (e.g. Mountain Bike Sales)
o
o
o b
o
o

a
x (e.g. % young
 where
population)
y = predicted (dependent) variable
x = predictor (independent) variable
b = slope of the line
a = value of y when x = 0 (the height of line at the y intercept)

2018 © Dr. Ozelkan Slide 26 Using Data

Computing a and b

y
b
o e4 y=a+bx
e5
o
o e2
e3
o
e1
o
a
x

Find a straight line with a and b such that the

sum of squared errors e12+e22+e32+e42+e52 is
minimum

2018 © Dr. Ozelkan Slide 27 Using Data

2018 © Dr. E.C. Ozelkan

13
 Computing a and b
y
(x4,y4) b
o y=a+bx
(x2,y2)
o
o
(x5,y5)
n=number of data points
o (x3,y3) (xt,yt) : observations
o
a (x1,y1)
x
n n n n n
n xt yt  x t yt  yt x t
t 1 t 1 t 1
b  2
a t 1
b t 1
n
 n  n n
n xt    xt 
2

t 1  t 1 
2018 © Dr. Ozelkan Slide 28 Using Data

Example:
(1) (2) (1)^2 (1) x (2)
No. % Young Population Sales of Mountain Bikes Prediction
1 30 2400 900 72000 2235.135
2 15 1500 225 22500 1408.108
3 35 2200 1225 77000 2510.811
4 25 1800 625 45000 1959.459
5 40 3000 1600 120000 2786.486
total 145 10900 4575 336500 8575.676
3500

3000
b = (5 · 336500 - 145 · 10900) /
Sales of Mountain Bikes

2500
(5 · 4575-1452) = 55
2000

1500 a = (10900-55·145)/5 = 581

1000

500

0
15 20 25 30 35 40 45 50
y= 581+55x
% Young People

2018 © Dr. Ozelkan Slide 29 Using Data

2018 © Dr. E.C. Ozelkan

14
 Example – Multiple Regression
A market researcher for Super
Dollar Super Markets is
studying the yearly amount
families of four or more spend
on food. Three independent
variables are thought to be
related to yearly food
expenditures (Food). Those
variables are: total family
income (Income) in $00, size of
family (Size), and whether the
family has children in college
(College).
2018 © Dr. Ozelkan
Using Data

Example – Multiple Regression

Food
expenditures = a + b1*(Income) + b2(Size) + b3(College)

•The variable college is called a dummy or indicator

variable. It can take only one of two possible outcomes.
That is a child is a college student or not.
•Other examples of dummy variables include gender, the part
is acceptable or unacceptable, the voter will or will not vote
for the incumbent governor.

We usually code one value of the dummy variable as “1”

and the other “0.”
2018 © Dr. Ozelkan
Using Data 31

2018 © Dr. E.C. Ozelkan

15
 Example – Multiple Regression
Fam ily Food Incom e Size Student
1 3900 376 4 0
2 5300 515 5 1
3 4300 516 4 0
4 4900 468 5 0
5 6400 538 6 1
6 7300 626 7 1
7 4900 543 5 0
8 5300 437 4 0
9 6100 608 5 1
10 6400 513 6 1
11 7400 493 6 1
12 5800 563 5 0
2018 © Dr. Ozelkan
Using Data

Example – Multiple Regression

Using MINITAB or Excel:

the regression equation
Y’ = 954 +1.09*X1 + 748*X2 + 565*X3

Food
Expenditure=$954+$1.09*income+$748*size+$565*college

What food expenditure would you estimate for a family

of 4, with no college students, and an income of $50,000
(which is input as 500)?
2018 © Dr. Ozelkan
Using Data

2018 © Dr. E.C. Ozelkan

16
 Example – Multiple Regression

Food
Expenditure=$954+$1.09*income+$748*size+$565*college
Each additional $100 dollars of income per year will
increase the amount spent on food by $109 per year.
An additional family member will increase the amount
spent per year on food by $748.
A family with a college student will spend $565 more per
year on food than those without a college student.
Food Expenditure=$954+$1.09*500+$748*4+$565*0
So a family of 4, with no college students, and an income
of $50,000 will spend an estimated $4,491.
2018 © Dr. Ozelkan
Using Data

Multivariate Linear Regression

 Regression Analysis Model

y   0  1 x1   2 x2   3 x3  
 Given data for n=15 observations, how would we estimate
the parameters?
Output: y Input 1: x1 Input 2: x2 Input 3: x3
Advertising Competition
Promotion Sales ($1000s) ($1000s) Price ($) Price ($)
1 10541 366 90.99 96.95
2 8891 377 90.99 93.99
3 5905 387 94.99 90.99
4 8251 418 96.99 97.95
5 11461 434 92.99 97.95
?
 0 , 1 ,  2 ,  3
6 6924 450 95.95 93.95
7 7347 457 93.95 90.99
8 10972 466 91.95 96.95
9 7811 467 96.95 94.99
10 10559 468 92.95 96.95
11 9825 468 97.99 98.95
12 9130 475 91.95 90.99
13 5116 479 99.95 91.95
14 7830 479 96.99 95.95
15 8388 481 91.95 90.95

2018 © Dr. Ozelkan Slide 35 Using Data

2018 © Dr. E.C. Ozelkan

17
 Multivariate Linear Regression

 Regression Analysis Model

y   0  1 x1   2 x2  ...   k xk  
 Given data for n observations, how would we estimate the
parameters? (out:output, in: input)
data out in1 in2 ... ink
1 y1 x11 x21 ... xk1
?
2 y2 x12 x22 ... xk2  0 , 1 ,  2 ,...,  k
... ...    
n yn x1n x2n ... xkn
2018 © Dr. Ozelkan Slide 36 Using Data

Multivariate Regression

 Note that for each data you can write an equation in which
the parameters are unknowns

data out const . in1 in2 ... ink error

1 y1   0  x111  x21  2  ...  xk1  k   1
2 y2   0  x12 1  x22  2  ...  xk2  k   2
... ... ...     ...
n yn  0  x1n 1  x2n  2  ...  xkn  k   n

2018 © Dr. Ozelkan Slide 37 Using Data

2018 © Dr. E.C. Ozelkan

18
 Multivariate Regression

 In practice to represent this set of equations in a compact

form and for programming purposes matrix convention is
used and preferred. Therefore, let’s define the following
matrices:
 y1  1 x11 x21 ...xk1   1  1 
 2 1 2     
y x1 x22 ... xk2 
y   , X   ,    ,    2 ,
2

 ...        ...   ... 

 n      
1 x1  k   n 
n
y  x2n ... xkn 

 Which results in the following simplified representation:

y  X  
2018 © Dr. Ozelkan Slide 38 Using Data

Computing the Parameters

 It can be shown that the least square estimators for the

parameters that minimize the error of predictions can be
obtained using:

  ( X T X ) 1 X T y

 where XT represents the transpose of the data matrix X

 where (XTX)-1 represents the inverse of the expression XTX

2018 © Dr. Ozelkan Slide 39 Using Data

2018 © Dr. E.C. Ozelkan

19
 Remarks

 Note that the simple linear regression analysis is a specific case of the
multivariate linear regression case
 thus the parameter estimation for a simple linear regression using the
parameter estimation formula for the multivariate linear regression would
result in the same estimates as the simple linear regression expressions
described earlier.
 Main assumption of a regression is that the errors are random
 In fact a check on this assumption is necessary for a valid regression
analysis.
 To make statistical testing we also assume that errors are normally
distributed with a mean μ=0 and standard deviation of σ
 You also assume that the input variables have significant impact on the
output variable
 Again this assumption needs to be checked as well.

2018 © Dr. Ozelkan Slide 40 Using Data

Regression Outputs

 A typical regression software would yield

 Estimated parameters

 0 , 1 ,  2 ,...,  k
 Correlation between the actual outputs and predicted outputs (as
given by regression)
 Statistical significance of the parameters
 Statistical significance of the regression
 Would provide graphs for the
 Regression prediction errors (also called residuals)
 Line fit between the actual outputs and predicted outputs
 Normal probability plot which indicates if the errors are random and
normally distributed.

2018 © Dr. Ozelkan Slide 41 Using Data

2018 © Dr. E.C. Ozelkan

20
 Statistical Tests for Regression

 Typically the following two types of tests are conducted:

1. Test for Significance of Overall Regression
2. Test for Significance of Individual Regression Parameters
 This test is repeated for each parameter!

2018 © Dr. Ozelkan Slide 42 Using Data

Test for Significance of Overall Regression

 The following is the regression hypothesis to be tested

H 0 :  0  1   2  ...   k  0
H 1 :  j  0, for at least one j
 Rejection of H0 (Null hypothesis) implies at least one input
variable is significant to define this regression model.
 To test this regression an F-test is conducted
MS R SSR / k where SSR indicates
F0   the sum of squares
MS E SSE /(n  k  1) regression and SSE
the sum of squares
 H0 is rejected if F0 > Fα,k,n-k-1 error. SEE NEXT
PAGE FOR MORE
2018 © Dr. Ozelkan Slide 43 DETAILS
Using Data

2018 © Dr. E.C. Ozelkan

21
 Remark on the Test for Significance of
Overall Regression
 The test presented on the previous page assumes that the total variation
in the data behavior will be partially explained by the regression model
and partially by the random error. In other words:

SSTotal  SS Regression  SS Error

 Since the Sum of Squares indicates variability, this analysis is also called
the Analysis of Variance (ANOVA) for Regression and represented
conveniently in a table format as follows:
ANOVA Table for Regression
Source of Sum of Degrees of Mean Square F0 Significance of F0
Variation Squares Freedom
Regression SSR k MSR=SSR/k MSR/MSE α level at which
MSR/MSE is significant
Error (or SSE n-k-1 MSE=SSE/(n-k-1)
Residual)
Total SST n-1 Number of inputs

2018 © Dr. Ozelkan Slide 44 Using Data

Computing Expressions

n
SS Total ( SST )   ( yi  y ) 2  SS Regression  SS Error
i 1
Average of all
n outputs
SS Regression ( SS R )   ( yˆ i  y ) 2 Individual
i 1 observations
n n
SS Error ( SS E )   ( yi  yˆ i ) 2   ei
2
Regression
prediction for
i 1 i 1 each individual
SS T  SS R  SS E observation

SS R  SST  SS E

2018 © Dr. Ozelkan Slide 45 Using Data

2018 © Dr. E.C. Ozelkan

22
 Computing Expressions (cont.)

SS R SS
R2   1 E
SST SST
2 2 SS E (n  k  1)
Radjusted  Ra  1 
SST (n  1)
2
Ra  R 2

Takes into account degrees of freedom!

In plain English:
Comparing variance explained by
regression to general variation of outputs

2018 © Dr. Ozelkan Slide 46 Using Data

Tests on Individual Regression Parameters

 The following is the parameter hypothesis to be tested

H0 :  j  0
H1 :  j  0
 If H0 (Null hypothesis) is not rejected, input variable xj can be
removed from the regression model.
 This test is repeated for each parameter!
 To test this hypothesis a t-test is conducted
ˆ j ˆ j
t0   where Cjj is the jth

standard error of ˆ j
diagonal element of
ˆ C jj
2
(XTX)-1
 H0 is rejected if |t0| > tα/2, n-k-1

2018 © Dr. Ozelkan Slide 47 Using Data

2018 © Dr. E.C. Ozelkan

23
 Reading the Regression Output
SUMMARY OUTPUT
Adj. R-Square shows Confidence Level for
Regression Statistics correlation that is Significance of
Multiple R 0.896661 between 0 and 1. High is Regression = 1-
R Square 0.804002 Significance F, thus
good! Here it is
Adjusted R Square
0.730502
Standard Error
572.6871 0.730~73% very good! low “Significance F”
Observations 12 shows good
regression relation!
ANOVA Here =1-
df SS MS F Significance F
Regression 3 10762903 3587634 10.93889 0.003334
0.003334=0.9967~99.
Residual 8 2623764 327970.5 67% indicates very
Total 11 13386667 good regression
relation!
Coefficients
Standard Error t Stat P-value Lower 95%Upper 95%
Intercept 954.0238 1580.839 0.603492 0.562897 -2691.4 4599.445
Income 1.0918 3.153489 0.34622 0.738107 -6.18016 8.36376
Size 748.3837 302.9535 2.470293 0.038693 49.77177 1446.996
Student 564.5213 495.1349 1.140136 0.287212 -577.262 1706.304

Confidence Level for Significance of

parameters= 1- (P-value), thus low p-value
means significant input! Here, for Size=1- Confidence interval for the parameters at
0.038693~0.961 ~96.1% indicates that Size is a selected confidence level give an
significant,
2018 © Dr. Ozelkan but Income and Student are not! Slide 48 indication of the parameter uncertainty!
Using Data

Using Minitab
Regression Analysis: Sales versus Population

The regression equation is

Sales = 581 + 55.1 Population

Running
Predictor Coef SE Coef T P
Constant 581.1 409.3 1.42 0.251
Population 55.14 13.53 4.07 0.027
Regression in
S = 260.284 R-Sq = 84.7% R-Sq(adj) = 79.6%
Minitab:
Stat=>regression
PRESS = 602182 R-Sq(pred) = 54.65%

Analysis of Variance
=> regression
Source DF SS MS F P
Regression 1 1124757 1124757 16.60 0.027
Residual Error 3 203243 67748
Total 4 1328000

2018 © Dr. Ozelkan

Using Data

2018 © Dr. E.C. Ozelkan

24
 Lab Exercise

 Do the regression analysis in the computer lab using the MS

Excel Software
 Activate “regression” tools by selecting Tools=>add-ins=>Analysis
ToolPack (put a check in the box on the left side) and then pressing
OK.
 Do the simple regression shown on slides.
 Do the multivariate regression by using the data given in the
file named [Link] that corresponds to Table 10.5

2018 © Dr. Ozelkan Slide 50 Using Data

2018 © Dr. E.C. Ozelkan

25