490 Chapter 11 Correlation and Regression

Correlation Coefficient
The CORREL function returns the correlation coefficient.

1. Enter the data in columns A and B.

2. Select a blank cell, then click on the fx button.
3. Under Function category, select Statistical. From the Function name list,
select CORREL.
4. Enter the data range (A1:AN, where N is the number of sample data pairs) for the first variable
in Array1. Enter the data range for the second variable in Array2. The correlation coeffi-
cient will be displayed in the selected cell.

11–5 The previous sections stated that if the correlation coefficient is significant, the equation
of the regression line can be determined. Also, for various values of the independent
Coefficient of variable x, the corresponding values of the dependent variable y can be predicted. Sev-
Determination and eral other measures are associated with the correlation and regression techniques. They
Standard Error of include the coefficient of determination, the standard error of estimate, and the predic-
Estimate tion interval. But before these concepts can be explained, the different types of variation
associated with the regression model must be defined.

Types of Variation for Consider the following hypothetical regression model.

the Regression Model x 1 2 3 4 5
y 10 8 12 16 20
The equation of the regression line is y'  4.8  2.8x, and r  0.919. The sample y
values are 10, 8, 12, 16, and 20. The predicted values, designated by y', for each x can
be found by substituting each x value into the regression equation and finding y'. For ex-
ample, when x  1,
y'  4.8  2.8x  4.8  (2.8)(1)  7.6
Now, for each x, there are an observed y value and a predicted y' value. For exam-
ple, when x  1, y  10, and y'  7.6. Recall that the closer the observed values are to
the predicted values, the better the fit is and the closer r is to 1 or 1.
_
The total variation  (y  y)2 is the sum of the squares of the vertical distances
each point is from the mean. The total variation can be divided into two parts: that which
is attributed to the relationship of x and y and that which is due to chance. The variation
_
obtained from the relationship (i.e., from the predicted y' values) is  (y'  y)2 and is
called the explained variation. Most of the variations can be explained by the relation-
ship. The closer the value r is to 1 or 1, the better the points fit the line, and the
_ _
closer  (y'  y)2 is to  (y  y)2. In fact, if all points fall on the regression line,
_ _
 (y'  y)2 will equal  (y  y)2, since y' would be equal to y in each case.
On the other hand, the variation due to chance, found by  (y  y')2, is called the un-
explained variation. This variation cannot be attributed to the relationships. When the un-
explained variation is small, the value of r is close to 1 or 1. If all points fall on the
regression line, the unexplained variation,  (y  y')2, will be 0. Hence, the total varia-
tion is equal to the sum of the explained variation and the unexplained variation. That is,
_ _
 (y  y)2   (y'  y)2   (y  y')2
These values are shown in Figure 11–17. For a single point, the differences are
called deviations. For the hypothetical regression model given earlier, for x  1 and
_
y  10, one gets y'  7.6 and y  13.2.
 Section 11–5 Coefficient of Determination and Standard Error of Estimate 491

Figure 11–17 y
(x, y )
Deviations for the
Unexplained
Regression Equation deviation
(y – y ′)
y –y Total deviation
(x, y ′)
Explained
deviation
y ′ –y
y y
(x,y )

x

The procedure for finding the three types of variation is illustrated next.
STEP 1 Find the predicted y' values.
for x  1 y'  4.8  2.8x  4.8  (2.8)(1)  7.6
for x  2 y'  4.8  (2.8)(2)  10.4
for x  3 y'  4.8  (2.8)(3)  13.2
for x  4 y'  4.8  (2.8)(4)  16
for x  5 y'  4.8  (2.8)(5)  18.8
Hence, the values for this example are as follows:
x y y'
1 10 7.6
2 8 10.4
3 12 13.2
4 16 16
5 20 18.8

STEP 2 Find the mean of the y values.

_ 10  8  12  16  20
y  13.2
5
_
STEP 3 Find the total variation,  (y  y)2.
(10  13.2)2  10.24
(8  13.2)2  27.04
(12  13.2)2  1.44
(16  13.2)2  7.84
(20  13.2)2  46.24
_
 (y  y)2  92.8
_
STEP 4 Find the explained variation,  (y'  y)2.
(7.6  13.2)2  31.36
(10.4  13.2)2  7.84
(13.2  13.2)2  0.00
(16  13.2)2  7.84
(18.8  13.2)2  31.36
_
 (y'  y)2  78.4
492 Chapter 11 Correlation and Regression

STEP 5 Find the unexplained variation,  (y  y')2.

(10  7.6)2  5.76
(8  10.4)2  5.76
(12  13.2)2  1.44
(16  16)2  0.00
(20  18.8)2  1.44
 (y  y')2  14.4
Notice that
total variation  explained variation  unexplained variation
92.8  78.4  14.4

Coefficient of The coefficient of determination is the ratio of the explained variation to the total varia-
Determination tion and is denoted by r 2. That is,
Objective 5. Compute the
explained variation
coefficient of determination. r2 
total variation

Historical Note For the example, r 2  78.4/92.8  0.845. The term r 2 is usually expressed as a per-
centage. So in this case, 84.5% of the total variation is explained by the regression line
Karl Pearson recom- using the independent variable.
mended in 1897 that the
French government close
Another way to arrive at the value for r 2 is to square the correlation coefficient. In
all its casinos and turn the this case, r  0.919, and r 2  0.845, which is the same value found by using the varia-
gambling devices over to tion ratio.
the academic community
to use in the study of
probability. The coefficient of determination is a measure of the variation of the dependent variable that is explained
by the regression line and the independent variable. The symbol for the coefficient of determination is r 2.

Of course, it is usually easier to find the coefficient of determination by squaring r

and converting it to a percentage. Therefore, if r  0.90, then r 2  0.81, which is equiv-
alent to 81%. This result means that 81% of the variation in the dependent variable is
accounted for by the variations in the independent variable. The rest of the variation,
0.19 or 19%, is unexplained. This value is called the coefficient of nondetermination and
is found by subtracting the coefficient of determination from 1. As the value of r ap-
proaches 0, r 2 decreases more rapidly. For example, if r  0.6, then r 2  0.36, which
means that only 36% of the variation in the dependent variable can be attributed to the
variation in the independent variable.

Formula for the Coefficient of Nondetermination

1.00  r 2

Standard Error of When a y' value is predicted for a specific x value, the prediction is a point prediction.
Estimate However, a prediction interval about the y' value can be constructed, just as a confidence
Objective 6. Compute the interval was constructed for an estimate of the population mean. The prediction interval
standard error of estimate. uses a statistic called the standard error of estimate.
 Section 11–5 Coefficient of Determination and Standard Error of Estimate 493

The standard error of estimate, denoted by s est is the standard deviation of the observed y values about
the predicted y' values. The formula for the standard error of estimate is


 y  y ' 2
sest 
n2

The standard error of estimate is similar to the standard deviation, but the mean is
not used. As can be seen from the formula, the standard error of estimate is the square
root of the unexplained variation—i.e., the variation due to the difference of the
observed values and the expected values—divided by n  2. So the closer the observed
values are to the predicted values, the smaller the standard error of estimate will be.
The next example shows how to compute the standard error of estimate.

Example 11–12 A researcher collects the following data and determines that there is a significant rela-
tionship between the age of a copy machine and its monthly maintenance cost. The re-
gression equation is y'  55.57  8.13x Find the standard error of estimate.
Machine Age, x (years) Monthly cost, y
A 1 $ 62
B 2 78
C 3 70
D 4 90
E 4 93
F 6 103

Solution
STEP 1 Make a table, as shown.
x y y' y  y' (y  y')2
1 62
2 78
3 70
4 90
4 93
6 103
STEP 2 Using the regression line equation, y'  55.57  8.13x, compute the
predicted values y' for each x and place the results in the column labeled y'.
x1 y'  55.57  (8.13)(1)  63.70
x2 y'  55.57  (8.13)(2)  71.83
x3 y'  55.57  (8.13)(3)  79.96
x4 y'  55.57  (8.13)(4)  88.09
x6 y'  55.57  (8.13)(6)  104.35
STEP 3 For each y, subtract y' and place the answer in the column labeled y  y'.
62  63.70  1.70 90  (88.09)  1.91
78  71.83  6.17 93  (88.09)  4.91
70  79.96  9.96 103  (104.35)  1.35
STEP 4 Square the numbers found in step 3 and place the squares in the column
labeled (y  y')2.
494 Chapter 11 Correlation and Regression

STEP 5 Find the sum of the numbers in the last column. The completed table
follows.
x y y' y  y' (y  y')2
1 62 63.70 1.70 2.89
2 78 71.83 6.17 38.0689
3 70 79.96 9.96 99.2016
4 90 88.09 1.91 3.6481
4 93 88.09 4.91 24.1081
6 103 104.35 1.35 1.8225
169.7392

STEP 6 Substitute in the formula and find sest.

 
 y  y' 2 169.7392
sest    6.51
n2 62
In this case, the standard deviation of observed values about the predicted
values is 6.51.

The standard error of estimate can also be found by using the formula


 y2  a  y  b  xy
sest 
n2

Example 11–13 Find the standard error of estimate for the data for Example 11–12 by using the preced-
ing formula. The equation of the regression line is y'  55.57  8.13x.

Solution
STEP 1 Make a table.
STEP 2 Find the product of x and y values and place the results in the third column.
STEP 3 Square the y values and place the results in the fourth column.
STEP 4 Find the sums of the second, third, and fourth columns. The completed table
is shown here.
x y xy y2
1 62 62 3,844
2 78 156 6,084
3 70 210 4,900
4 90 360 8,100
4 93 372 8,649
6 103 618 10,609
 y  496  xy  1778  y2  42,186

STEP 5 From the regression equation, y'  55.57  8.13x, a  55.57, and b  8.13.
STEP 6 Substitute in the formula and solve for sest.
 Section 11–5 Coefficient of Determination and Standard Error of Estimate 495


 y 2  a  y  b  xy
sest 
n2


42,186  55.57 496  8.13 1778 
  6.48
62
This value is close to the value found in Example 11–12. The difference is
due to rounding.

Prediction Interval The standard error of estimate can be used for constructing a prediction interval (sim-
ilar to a confidence interval) about a y' value.
Objective 7. Find a When a specific value x is substituted into the regression equation, one gets y' which
prediction interval. is a point estimate for y. For example, if the regression line equation for the age of a ma-
chine and the monthly maintenance cost is y'  55.57  8.13x (Example 11–12), then the
predicted maintenance cost for a 3-year-old machine would be y'  5.57  8.13(3) or
$79.96. Since this is a point estimate, one has no ideas how accurate it is. But one can
construct a prediction interval about the estimate. By selecting an  value, one can
achieve a 1   confidence that the interval contains the actual mean of the y values that
correspond to the given value of x.
The reason is that there are possible sources of prediction errors in finding the
regression line equation. One source occurs when finding the standard error of estimate,
sest. Two others are errors made in estimating the slope and the y intercept, since the
equation of the regression line will change somewhat if different random samples are
used when calculating the equation.

Formula for the Prediction Interval about a Value y '


1 n x  X  2
y'  t  2 sest 1  y y'
n n  x2   x 2


1 n x  X  2
 t  2 sest 1 
n n  x2    x 2

with d.f.  n  2

Example 11–14 For the data in Example 11–12, find the 95% prediction interval for the monthly main-
tenance cost of a machine that is 3 years old.

Solution
STEP 1 Find  x,  x2, and X .
20
 x  20  x2  82 X  3.3
6
STEP 2 Find y' for x  3
y'  55.57  8.13x
y'  55.57  8.13(3)  79.96
STEP 3 Find sest
sest  6.48, as shown in Example 11–13.
496 Chapter 11 Correlation and Regression

STEP 4 Substitute in the formula and solve : t /2  2.776, d.f.  6  2  4 for 95%


1 nx  X  2
y'  t 2 sest 1  y y'
n n  x 2   x 2


1 nx  X 2
 t 2 sest 1 
n n  x2   x 2


1 63  3.3  2
79.96  2.776 6.48  1  y 79.96
6 682   20  2


1 63  3.3  2
 2.7766.48  1 
6 682   20  2
79.96  2.776 6.48 1.08  y 79.96  2.776 6.48 1.08 
79.96  19.43 y 79.96  19.43
60.53 y 99.39

Hence, one can be 95% confident that the interval 60.53 y 99.39
contains the actual value of y.

Exercises
11–67. What is meant by the explained variation? How is 11–81. Compute the standard error of estimate for
it computed? Exercise 11–13. The regression line equation was found in
Exercise 11–41.
11–68. What is meant by the unexplained variation? How
is it computed? 11–82. Compute the standard error of estimate for
Exercise 11–14. The regression line equation was found in
11–69. What is meant by the total variation? How is it
Exercise 11–42.
computed?
11–83. Compute the standard error of estimate for
11–70. Define coefficient of determination.
Exercise 11–15. The regression line equation was found in
11–71. How is the coefficient of determination found? Exercise 11–43.
11–72. Define coefficient of nondetermination. 11–84. Compute the standard error of estimate for
11–73. How is the coefficient of nondetermination found? Exercise 11–16. The regression line equation was found in
Exercise 11–44.
For Exercises 11–74 through 11–79, find the coefficients
11–85. For the data in Exercises 11–13, 11–41, and 11–81,
of determination and nondetermination and explain the
find the 90% prediction interval when x  20 years.
meaning of each.
11–86. For the data in Exercises 11–14, 11–42, and 11–82,
11–74. r  0.81
find the 95% prediction interval when x  $1100.
11–75. r  0.70
11–87. For the data in Exercises 11–15, 11–43, and 11–83,
11–76. r  0.45 find the 90% prediction interval when x  4 years.
11–77. r  0.37 11–88. For the data in Exercises 11–16, 11–44, and 11–84,
find the 98% prediction interval when x  47 years.
11–78. r  0.15
11–79. r  0.05
11–80. Define standard error of estimate for regression.
When can the standard error of estimate be used to
construct a prediction interval about a value y'?