Regression

SCATTERPLOTS AND REGRESSION

1. The table gives weight in pounds and length in inches for 3-month-old
baby girls. Graph the points from the table in a scatterplot and describe
the trend.

Weight (lbs) Length (in)

9.7 21.6

10.2 22.1

12.4 23.6

13.6 25.1

9.8 22.4

11.2 23.9

14.1 25.8

Solution:

Sketch the scatterplot.

1
 3-month-old baby girls
26

25
Length (inches)

24

23

22

21
7 8 9 10 11 12 13 14 15

Weight (pounds)

The points rise from left to right and are fairly linear. We can say that there
is a strong positive linear correlation between the points. There do not
appear to be any outliers in the data.

2. The following values have been computed for a data set of 14 points.
Calculate the line of best fit.

∑
x = 86

∑
y = 89.7

∑
xy = 680.46

x 2 = 654.56
∑

2
Solution:

We’re told that there are 14 items in the data set, so n = 14.

To find the line of best fit, we need its slope and y-intercept. The slope is
given by

n ∑ xy − ∑ x ∑ y
b=
n ∑ x 2 − ( ∑ x)
2

14(680.46) − (86)(89.7)
b=
14(654.56) − (86)2

9,526.44 − 7,714.2
b=
9,163.84 − 7,396

1,812.24
b=
1,767.84

b ≈ 1.0251

The y-intercept is given by

∑ y − b∑ x
a=
n

89.7 − 1.0251(86)
a=
14

89.7 − 88.1599
a=
14

3
 1.5401
a=
14

a ≈ 0.1100

So the line of best fit is

ŷ = bx + a

ŷ = 1.0251x + 0.1100

3. For the data set given in the table, calculate each of the following
values:

x ,(
∑ )
2
2
∑ ∑ ∑ ∑
n, x, y, xy, x

Month 1 2 3 4 5 6 7 8 9 10 11 12
Temperatu
73 73 75 75 77 79 79 81 81 81 77 75
re

Solution:

Expand the original table to calculate the values.

4
 Month, x Temperature, y xy x2

1 73 1(73)=73 12=1

2 73 2(73)=146 22=4

3 75 3(75)=225 32=9

4 75 4(75)=300 42=16

5 77 5(77)=385 52=25

6 79 6(79)=474 62=36

7 79 7(79)=553 72=49

8 81 8(81)=648 82=64

9 81 9(81)=729 92=81

10 81 10(81)=810 102=100

11 77 11(77)=847 112=121

12 75 12(75)=900 122=144

∑
Summing the first column gives x = 78

∑
Summing the second column gives y = 926

∑
Summing the third column gives xy = 6,090

x 2 = 650
∑
Summing the fourth column gives

(∑ )
2
2
Squaring the sum from the first column gives x = 78 = 6,084

5
 4. Use the Average Global Sea Surface Temperatures data shown in the
table to create a line of best fit for the data. Consider 1910 as year 10. Use
the equation to predict the average global sea surface temperature in the
year 2050.

Year Temperature, F

1910 -1.11277

1920 -0.71965

1930 -0.58358

1940 -0.17977

1950 -0.55318

1960 -0.30358

1970 -0.30863

1980 0.077197

1990 0.274842

2000 0.232502

2010 0.612718

Solution:

Start by expanding the table.

6
 Year Temperature, F xy x2

10 -1.11277 -11.1277 100

20 -0.71965 -14.393 400

30 -0.58358 -17.5074 900

40 -0.17977 -7.1908 1,600

50 -0.55318 -27.659 2,500

60 -0.30358 -18.2148 3,600

70 -0.30863 -21.6041 4,900

80 0.077197 6.17576 6,400

90 0.274842 24.73578 8,100

100 0.232502 23.2502 10,000

110 0.612718 67.39898 12,100

∑
Summing the first column gives x = 660

∑
Summing the second column gives y = − 2.5639

∑
Summing the third column gives xy = 3.86392

x 2 = 50,600
∑
Summing the fourth column gives

Squaring the sum from the first column gives

( ∑ x) = 660 = 435,600
2
2

To find the regression line for the data, we need the slope and y-intercept
of the line. The slope is

7
 n ∑ xy − ∑ x ∑ y
b=
n ∑ x 2 − ( ∑ x)
2

11(3.86392) − (660)(−2.5639)
b=
11(50,600) − 435,600

42.50312 + 1,692.174
b=
556,600 − 435,600

1,734.67712
b=
121,600

b ≈ 0.0143

The y-intercept is

∑ y − b∑ x
a=
n

−2.5639 − 0.0143(660)
a=
11

−2.5639 − 9.415188
a=
11

−11.979088
a=
11

a ≈ − 1.0890

Then the equation of the trend line is

ŷ = bx + a

ŷ = 0.0143x − 1.0890

8
To predict average global sea surface temperature in 2050, we’ll need to
plug 150 into this equation.

ŷ = 0.0143(150) − 1.0890

ŷ = 2.145 − 1.0890

ŷ = 1.056

So the predicted sea surface temperature in 2050 is about 1.056∘ F.

5. Compare the scatterplots. The second graph includes extra data

starting in 1880. How does this compare to the plot that only shows 1910 to
2010? Explain trends in the data, and how the regression line changes by
adding in these extra points. Which trend line would be best for predicting
the temperature in 2050?

Average Global Sea Surface Temperatures, 1910-2010

0.9

0.6

0.3
Temperature

-0.3

-0.6

-0.9

-1.2

-1.5
1920 1940 1960 1980 2000 2020

Year

9
 Average Global Sea Surface Temperatures, 1880-2010
0.9

0.6

0.3
Temperature

-0.3

-0.6

-0.9

-1.2

-1.5
1880 1900 1920 1940 1960 1980 2000 2020

Year

Solution:

Adding in these extra three points make the graph from 1880 to 2010
appear more scattered and not as linear as the graph that only includes
the points from 1910 to 2010.

Both data sets have a positive correlation because the general trend of the
scatterplot is to increase as we move from left to right, but we might
consider graphs that are exponential in shape instead of linear. If we use a
line of best fit for the data from 1880 to 2010, it might not be as accurate as
a line predicting only the points from 1910 to 2010.

In other words, the best fit line for 1880 to 2010 would have a weaker
correlation than the line for 1910 to 2010, because the additional points to
the left of the graph are more spread out.

10
But even though cutting off the points makes the line of best fit have a
stronger correlation, it would be good to include them in the data so that
our line of best fit is not misleading.

6. A small coffee shop wants to know how hot chocolate sales are
affected by daily temperature. Find the rate of change of hot chocolate
sales, with respect to temperature.

Daily Temperature, F Hot Chocolate Sales

28 110

29 115

31 108

33 103

45 95

48 93

55 82

57 76

Solution:

Create a scatterplot.

11
 Hot Chocolate Sales by Temperature
120

96
Hot Chocolate Sales

72

48

24

0
30 35 40 45 50 55 60

Daily Temperature, F

From the plot we can see there’s a relatively strong, negative linear
relationship with no outliers. The rate of change is the slope, so we need
to look at the slope of the line of best fit for the data set. The formula for
the slope of the best-fit line is

n ∑ xy − ∑ x ∑ y
b=
n ∑ x 2 − ( ∑ x)
2

Extend the table to find the values we need for the formula.

12
 Daily Hot Chocolate
xy x2
Temperature, F Sales

28 110 3,080 784

29 115 3,335 841

31 108 3,348 961

33 103 3,399 1,089

45 95 4,275 2,025

48 93 4,464 2,304

55 82 4,510 3,025

57 76 4,332 3,249

Sum
326 782 30,743 14,278
:

Plug these values into the slope formula.

8(30,743) − (326)(782)
b=
8(14,278) − (326)2

245,944 − 254,932
b=
114,224 − 106,276

−8,988
b=
7,946

b ≈ − 1.1311

The units of the slope are “hot chocolate sales per degree Fahrenheit.” So
the shop can expect hot chocolate sales to decrease by 1.1311 cups for
every one degree increase in temperature.

13
CORRELATION COEFFICIENT AND THE RESIDUAL

1. What does the shape of this residual plot tell us about the line of best
fit that was created for the data?

Solution:

The shape of this graph tells us that the linear model is probably not the
best choice for our data set, and that we should consider another type of
regression curve, probably one that’s quadratic.

14
 2. What does the shape of this residual plot tell us about the line of best
fit that was created for the data?

Solution:

The points in this residual plot are evenly spaced around the line y = 0. It
has about the same number of points on the left and right, and about the
same number of points above and below 0. It doesn’t appear to have
outliers or interesting features. So the line of best fit for the data is
probably a good one and can be useful for making predictions.

3. Calculate and interpret the correlation coefficient for the data set.

15
 x y

54 0.162

57 0.127

62 0.864

77 0.895

81 0.943

93 1.206

Solution:

To find the correlation coefficient, we use the formula

n − 1 ∑ ( sx ) ( sy )
1 xi − x̄ yi − ȳ
r=

We need to start by finding the means and standard deviations for both x
and y. The means are

54 + 57 + 62 + 77 + 81 + 93
x̄ =
6

x̄ ≈ 70.6667

and

0.162 + 0.127 + 0.864 + 0.895 + 0.943 + 1.206

ȳ =
6

ȳ ≈ 0.6995

16
and the standard deviations are

6
∑i=1 (xi − x̄)2
sx =
n−1

277.7790 + 186.7790 + 75.1117 + 40.1107 + 106.7770 + 498.7760

sx ≈
5

sx ≈ 15.3970

and

6
∑i=1 (yi − ȳ)2
sy =
n−1

0.2889 + 0.3278 + 0.0271 + 0.0382 + 0.0593 + 0.2565

sy ≈
5

sy ≈ 0.4467

Plug these values into the correlation coefficient formula.

6 − 1 [( 15.3970 ) ( ) ( 15.3970 ) ( )
1 54 − 70.6667 0.162 − 0.6995 57 − 70.6667 0.127 − 0.6995
r= +
0.4467 0.4467

( 15.3970 ) ( ) ( 15.3970 ) ( )
62 − 70.6667 0.864 − 0.6995 77 − 70.6667 0.895 − 0.6995
+ +
0.4467 0.4467

( 15.3970 ) ( ) ( 15.3970 ) ( )]
81 − 70.6667 0.943 − 0.6995 93 − 70.6667 1.206 − 0.6995
+ +
0.4467 0.4467

17
 1
r=
5 [ (−1.0825)(−1.2033) + (−0.8876)(−1.2816) + (−0.5629)(0.3683)

+(0.4113)(0.4189) + (0.6711)(0.5217) + (1.4505)(1.1339)]

1
r= (1.3026 + 1.1375 − 0.2073 + 0.1723 + 0.3501 + 1.6447)
5

1
r= (4.3999)
5

r ≈ 0.88

The positive correlation coefficient tells us that the regression line has a
positive slope. The fact that the positive value is closer to 1 than it is to 0
tells us the data is strongly correlated, or that it most likely has a strong
linear relationship. If we looked at a scatterplot of the data and sketched in
the regression line, we’d see that this was true.

4. Calculate the residuals, draw the residual plot, and interpret the
results. Compare the results to the r-value in the previous problem. The
equation of the line of best fit for the data is

ŷ = 0.0257x − 1.1142

18
 x y

54 0.162

57 0.127

62 0.864

77 0.895

81 0.943

93 1.206

Solution:

Create a table to find the residual of each value.

x Actual y Predicted y Residual

54 0.162 0.2736 -0.1116

57 0.127 0.3507 -0.2237

62 0.864 0.4792 0.3848

77 0.895 0.8647 0.0303

81 0.943 0.9675 -0.0245

93 1.206 1.2759 -0.0699

A plot of the residuals is

19
 0.40

0.30

0.20

0.10

-0.10

-0.20

-0.30

-0.40
60 70 80 90 100

From the residual plot, it looks like the data had an outlier at x = 62. We
already have a somewhat strong positive linear correlation from the
correlation coefficient from the previous problem of r ≈ 0.88, so it’s likely
the relationship would be even stronger without the outlier.

5. The table shows average global sea surface temperature by year.

Calculate and interpret the correlation coefficient for the data set. Leave
the years as they are.

20
 Year Temperature, F

1880 -0.47001

1890 -0.88758

1900 -0.48331

1910 -1.11277

1920 -0.71965

1930 -0.58358

1940 -0.17977

1950 -0.55318

1960 -0.30358

1970 -0.30863

1980 0.077197

1990 0.274842

2000 0.232502

2010 0.612718

Solution:

Since this is a larger set of data, it can be nice to use a program like Excel
to expand the table.

21
 Year Temperature, F (xi-x)2 (yi-y)2

1880 -0.47001 4,225 0.02414

1890 -0.88758 3,025 0.32827

1900 -0.48331 2,025 0.02845

1910 -1.11277 1,225 0.63703

1920 -0.71965 625 0.16404

1930 -0.58358 225 0.07233

1940 -0.17977 25 0.01819

1950 -0.55318 25 0.05691

1960 -0.30358 225 0.00012

1970 -0.30863 625 0.00004

1980 0.077197 1,225 0.15353

1990 0.274842 2,025 0.34748

2000 0.232502 3,025 0.29935

2010 0.612718 4,225 0.85997

Sum: 27,230 -4.40480 22,750 2.98985

Mean: 1,945 -0.31463

The standard deviation for x and y are

14
∑i=1 (xi − x̄)2 22,750
sx = = ≈ 41.8330
n−1 13

14
∑i=1 (yi − ȳ)2 2.98986
sy = = ≈ 0.4796
n−1 13

22
Now that we have the means and standard deviations, we can find the
correlation coefficient. If we expand the table, then we’ll be able to pull
just the one sum out of the table to plug into the correlation coefficient
formula.

Year Temp, F (xi-x) (xi-x)/sx (yi-y) (yi-y)/sy ((xi-x)/sx)((yi-y)/sy)

1880 -0.47001 -65 -1.55380 -0.15538 -0.32398 0.50340

1890 -0.88758 -55 -1.31475 -0.57925 -1.20778 1.58793

1900 -0.48331 -45 -1.07571 -0.16868 -0.35171 0.37834

1910 -1.11277 -35 -0.83666 -0.79814 -1.66418 1.39235

1920 -0.71965 -25 -0.59761 -0.40502 -0.84450 0.50468

1930 -0.58358 -15 -0.35857 -0.26895 -0.56078 0.20108

1940 -0.17977 -5 -0.11952 0.13486 0.28119 -0.03361

1950 -0.55318 5 0.11952 -0.23855 -0.49739 -0.05945

1960 -0.30358 15 0.35857 0.01105 0.02304 0.00826

1970 -0.30863 25 0.59761 0.00600 0.01251 0.00748

1980 0.077197 35 0.83666 0.39183 0.81699 0.68354

1990 0.274842 45 1.07571 0.58947 1.22909 1.32214

2000 0.232502 55 1.31475 0.54713 1.14080 1.49987

2010 0.612718 65 1.55380 0.92735 1.93359 3.00441

Sum
11.00042
:

The correlation coefficient is then

23
 n − 1 ∑ ( sx ) ( sy )
1 xi − x̄ yi − ȳ
r=

1
r= (11.00042)
14 − 1

1
r= (11.00042)
13

r ≈ 0.8462

There’s a strong positive linear relationship between the year and the
temperature of the ocean’s surface.

6. Calculate the residuals and create the residual plot for the data in the
table. Compare this with the r-value we calculated in the last question and
interpret the results. Use the equation for the regression line
ŷ = 0.0143x − 28.332.

24
 Year Temperature, F

1880 -0.47001

1890 -0.88758

1900 -0.48331

1910 -1.11277

1920 -0.71965

1930 -0.58358

1940 -0.17977

1950 -0.55318

1960 -0.30358

1970 -0.30863

1980 0.077197

1990 0.274842

2000 0.232502

2010 0.612718

Solution:

Use the equation of the regression line to find predicted values of

temperature, and add those values to the table.

25
 Year Actual y Predicted y Residual

1880 -0.47001 -1.448 0.97799

1890 -0.88758 -1.305 0.41742

1900 -0.48331 -1.162 0.67869

1910 -1.11277 -1.019 -0.09337

1920 -0.71965 -0.876 0.15635

1930 -0.58358 -0.733 0.14942

1940 -0.17977 -0.59 0.41023

1950 -0.55318 -0.447 -0.10618

1960 -0.30358 -0.304 0.00042

1970 -0.30863 -0.161 -0.14763

1980 0.077197 -0.018 0.15917

1990 0.274842 0.125 0.149842

2000 0.232502 0.268 -0.035498

2010 0.612718 0.411 0.201718

Make a plot of the residuals.

26
 1.00

0.75

0.50

0.25

-0.25
1890 1910 1930 1950 1970 1990 2010

The residual plot is a good example of why finding the correlation

coefficient is not enough.

This plot makes it look like using an exponential regression would be a

better fit for the data. The residuals are not above and below the line y = 0
in a random pattern. Which means that even though the correlation
coefficient of r ≈ 0.8449 says there is a strong positive linear relationship
between year and temperature, it probably won’t do as good of a job as
we think at making predictions for the future because another type of
model would be better.

27
COEFFICIENT OF DETERMINATION AND RMSE

1. Linda read an article about the predictions of high school students and
their GPA. The article studied three factors, the number of volunteer
organizations each student participated in, the number of hours spent on
homework, and the student’s individual scores on standardized tests.

The article concluded that the number of hours spent on homework are
the best predictor of GPA, because they found 24 % of the variance in GPA
to be from hours spent on homework, 15 % from the number of volunteer
organizations, and 11.5 % from individual scores on standardized tests.

What is the coefficient of determination for the line-of-best-fit that has y

-values of high school GPA and x-values of hours spent on homework? Is
the line of best fit a good predictor of the data? Why or why not?

Solution:

Remember the percent of the variation in y that can be explained by the x

-values is the coefficient of determination or the r 2 value.

In this context, the percent of the variance in GPA due to hours spent on
homework is 24 % . So, we’re talking about a least squares line where
r 2 = 0.24. This is a very weak positive relationship, so the line of best fit is
probably not a good predictor of the connection between hours spent on
homework and GPA.

28
 2. For the data in the table, calculate the sum of the squared residuals
based on the mean of the y-values.

x y

1 3.1

2 3.4

3 3.7

4 3.9

5 4.1

Solution:

First calculate ȳ.

3.1 + 3.4 + 3.7 + 3.9 + 4.1

ȳ =
5

18.2
ȳ =
5

ȳ = 3.64

The formula for a residual is

residual = actual − predicted

In this case, the predicted value is the mean of the y-values, ȳ = 3.64. Let’s
expand the table and calculate the residuals.

29
 x y e

1 3.1 -0.54

2 3.4 -0.24

3 3.7 0.06

4 3.9 0.26

5 4.1 0.46

Now we just need to find the squares of these residuals and add them
together.

x y e e2

1 3.1 -0.54 0.2916

2 3.4 -0.24 0.0576

3 3.7 0.06 0.0036

4 3.9 0.26 0.0676

5 4.1 0.46 0.2116

Sum: 0.632

So the sum of the squared residuals is about 0.632.

3. Use the same data as the previous question to calculate the sum of
the squared residuals based on the least squares regression line,
ŷ = 0.25x + 2.89.

30
Solution:

The formula for a residual is

residual = actual − predicted

In this case the predicted value is based on the regression line,

ŷ = 0.25x + 2.89. Let’s expand the table and calculate the residuals.

x Actual y Predicted y e

1 3.1 3.14 -0.04

2 3.4 3.39 0.01

3 3.7 3.64 0.06

4 3.9 3.89 0.01

5 4.1 4.14 -0.04

Now we just need to find the squares of these residuals and add them
together.

Predicte
x Actual y e e2
dy

1 3.1 3.14 -0.04 0.0016

2 3.4 3.39 0.01 0.0001

3 3.7 3.64 0.06 0.0036

4 3.9 3.89 0.01 0.0001

5 4.1 4.14 -0.04 0.0016

Sum: 0.007

31
So the sum of the squared residuals is 0.007.

4. Based on the previous two questions, in which we found the sum of

the squared residuals based on the mean of the y-values and then the line
of best fit, what percentage of error did we eliminate by using the least
squares line? What is the term for this error?

Solution:

The sum of the squared residuals for the mean of the y-values was

residuals2 = 0.632
∑

The sum of the squared residuals for the line of best fit was

residuals2 = 0.007
∑

This means using the line of best fit reduces the error by

0.632 − 0.007

0.625

This is

0.625
= 0.9889 = 98.89 %
0.632

32
of a reduction in error by using the least squares regression line. This is
another way to calculate the coefficient of determination, so r 2 = 0.9889.

5. What is the RMSE of the data set and what does it mean?

x y

1 3.1

2 3.4

3 3.7

4 3.9

5 4.1

Solution:

To find RMSE, we’ll use the formula

∑ residuals2
RMSE =
n

We already calculated the residual sum of squares.

33
 Predicte
x Actual y e e2
dy

1 3.1 3.14 -0.04 0.0016

2 3.4 3.39 0.01 0.0001

3 3.7 3.64 0.06 0.0036

4 3.9 3.89 0.01 0.0001

5 4.1 4.14 -0.04 0.0016

Sum: 0.007

The sum of the squared residuals was 0.007, so RMSE will be

0.007
RMSE = ≈ 0.0374
5

RMSE is the standard deviation of the residuals, which means that

• 68 % of the data points will be within ± 0.0418 of the regression line,

• 95 % of the data points will be within ± 2(0.0418) of the regression

line, and

• 99.7 % of the data points will be within ± 3(0.0418) of the regression

line.

Since the RMSE we found is a small standard deviation, the data points are
going to be more tightly clustered around the line-of-best-fit and the
correlation in the data will be stronger.

34
 6. Calculate the RMSE for the data set, given that the least squares line is
ŷ = 0.0028x + 1.2208.

x y

5 1.25

10 1.29

12 1.17

15 1.24

17 1.32

Solution:

To find RMSE, we’ll use the formula

∑ residuals2
RMSE =
n

We can calculate the residual sum of squares.

35
 Predicte
x Actual y e e2
dy

5 1.25 1.2348 0.0152 0.00023104

10 1.29 1.2488 0.0412 0.00169744

12 1.17 1.2544 -0.0844 0.00712336

15 1.24 1.2628 -0.0228 0.00051984

17 1.32 1.2684 0.0516 0.00266256

Sum: 0.01223424

The sum of the squared residuals was 0.01223424, so RMSE will be

0.01223424
RMSE = ≈ 0.0495
5

36
CHI-SQUARE TESTS

1. We want to know whether a person’s geographic region of the United

State affects their preference of cell phone brand. We randomly sample
people across the country and ask them about their brand preference.
What can we conclude using a chi-square test at 95 % confidence?

iPhone Android Other Totals

Northeast 72 33 8 113

Southeast 48 26 7 81

Midwest 107 50 10 167

Northwest 59 33 10 102

Southwest 61 27 9 97

Totals 347 169 44 560

Solution:

Start by computing expected values.

37
 iPhone Android Other Totals

Northeast 72 (70.02) 33 (34.10) 8 (8.88) 113

Southeast 48 (50.19) 26 (24.44) 7 (6.36) 81

107
Midwest 50 (50.40) 10 (13.12) 167
(103.48)
Northwest 59 (63.20) 33 (30.78) 10 (8.01) 102

Southwest 61 (60.11) 27 (29.27) 9 (7.62) 97

Totals 347 169 44 560

Now we’ll check our sampling conditions. The problem told us that we
took a random sample, and all of our expected values are at least 5, so
we’ve met the random sampling and large counts conditions. And even
though we’re sampling without replacement, 560 is far less than 10 % of the
US population, so we’ve met the independence condition as well.

We’ll state the null and alternative hypotheses.

H0: Cell phone brand preference isn’t affected by geographic region.

Ha: Cell phone brand preference is affected by geographic region.

Calculate χ 2.

2 (72 − 70.02)2 (33 − 34.10)2 (8 − 8.88)2

χ = + +
70.02 34.10 8.88

(48 − 50.19)2 (26 − 24.44)2 (7 − 6.36)2

+ + +
50.19 24.44 6.36

(107 − 103.48)2 (50 − 50.40)2 (10 − 13.12)2

+ + +
103.48 50.40 13.12

38
 (59 − 63.20)2 (33 − 30.78)2 (10 − 8.01)2
+ + +
63.20 30.78 8.01

(61 − 60.11)2 (27 − 29.27)2 (9 − 7.62)2

+ + +
60.11 29.27 7.62

χ 2 ≈ 0.0560 + 0.0355 + 0.0872

+0.0956 + 0.0996 + 0.0644

+0.1197 + 0.0032 + 0.7420

+0.2791 + 0.1601 + 0.4944

+0.0132 + 0.1760 + 0.2499

χ 2 ≈ 2.6759

The degrees of freedom are

df = (number of rows − 1)(number of columns − 1)

df = (5 − 1)(3 − 1)

df = (4)(2)

df = 8

With df = 8 and χ 2 ≈ 2.6759, the χ 2-table gives

39
 Upper-tail probability p

df 0.25 0.20 0.15 0.10 0.05 0.025 0.02 0.01 0.005 0.003 0.001 5E-04

7 9.04 9.80 10.75 12.02 14.07 16.01 16.62 18.48 20.28 22.04 24.32 26.02

8 10.22 11.03 12.03 13.36 15.51 17.53 18.17 20.09 21.95 23.77 26.12 27.87

9 11.39 12.24 13.29 14.68 16.92 19.02 19.68 21.67 23.59 25.46 27.88 29.67

We’re off the chart on the left, which means we will definitely not exceed
the alpha level α = 0.05. Therefore, we’ll fail to reject the null hypothesis,
and conclude that geographic region of the country does not affect cell
phone brand preference.

2. A beverage company wants to know if gender affects which of their

products people prefer. They take a random sample of fewer than 10 % of
their customers, and ask them in a blind taste test which beverage they
prefer. What can the company conclude using a chi-square test at α = 0.1?

Beverage

A B C Totals

Men 35 34 31 100

Women 31 33 36 100

Totals 66 67 67 200

Solution:

Start by computing expected values.

40
 Beverage

A B C Totals

Men 35 (33.0) 34 (33.5) 31 (33.5) 100

Women 31 (33.0) 33 (33.5) 36 (33.5) 100

Totals 66 67 67 200

Now we’ll check our sampling conditions. The problem told us that we
took a random sample and that we sampled less than 10 % of the
population, so we’ve met the random sampling and independence
conditions. And all of our expected values are at least 5, so we’ve met the
large counts condition as well.

We’ll state the null and alternative hypotheses.

H0: Gender does not affect beverage preference.

Ha: Gender affects beverage preference.

Calculate χ 2.
2 2 2
(35 − 33) (34 − 33.5) (31 − 33.5)
χ2 = + +
33 33.5 33.5

(31 − 33)2 (33 − 33.5)2 (36 − 33.5)2

+ + +
33 33.5 33.5

4 0.25 6.25 4 0.25 6.25

χ2 = + + + + +
33 33.5 33.5 33 33.5 33.5

χ 2 ≈ 0.1212 + 0.0075 + 0.1866 + 0.1212 + 0.0075 + 0.1866

χ 2 = 0.6306

41
The degrees of freedom are

df = (number of rows − 1)(number of columns − 1)

df = (2 − 1)(3 − 1)

df = (1)(2)

df = 2

With df = 2 and χ 2 = 0.6306, the χ 2-table gives

Upper-tail probability p

df 0.25 0.20 0.15 0.10 0.05 0.025 0.02 0.01 0.005 0.003 0.001 5E-04

1 1.32 1.64 2.07 2.71 3.81 5.02 5.41 6.63 7.88 9.14 10.83 12.12

2 2.77 3.22 3.79 4.61 5.99 7.38 7.82 9.21 10.60 11.98 13.82 15.20

3 4.11 4.64 5.32 6.25 7.81 9.35 9.84 11.34 12.84 14.32 16.27 17.73

We’re off the chart on the left, which means we will definitely not exceed
the alpha level α = 0.1. Therefore, we’ll fail to reject the null hypothesis,
and conclude that gender does not affect beverage preference.

3. A coffee company wants to know whether or not drink and pastry

choice are related among their customers. The company randomly
sampled fewer than 10 % of their customers, and recorded their drink and
pastry orders. What can the restaurant conclude using a chi-square test at
99 % confidence?

42
 Bagel Muffin Totals

Coffee 38 34 72

Tea 25 29 54

Totals 63 63 126

Solution:

Start by computing expected values.

Bagel Muffin Totals

Coffee 38 (36) 34 (36) 72

Tea 25 (27) 29 (27) 54

Totals 63 63 126

Now we’ll check our sampling conditions. The problem told us that we
took a random sample and that we sampled less than 10 % of the
population, so we’ve met the random sampling and independence
conditions. And all of our expected values are at least 5, so we’ve met the
large counts condition as well.

We’ll state the null and alternative hypotheses.

H0: Pastry preference isn’t affected by beverage preference.

Ha: Pastry preference is affected by beverage preference.

Calculate χ 2.

43
 (38 − 36)2 (34 − 36)2 (25 − 27)2 (29 − 27)2
2
χ = + + +
36 36 27 27

42 4 4 4
χ = + + +
36 36 27 27

2 8
2
χ = +
9 27

χ 2 ≈ 0.52

The degrees of freedom are

df = (number of rows − 1)(number of columns − 1)

df = (2 − 1)(2 − 1)

df = (1)(1)

df = 1

With df = 1 and χ 2 = 0.52, the χ 2-table gives

Upper-tail probability p

df 0.25 0.20 0.15 0.10 0.05 0.025 0.02 0.01 0.005 0.003 0.001 5E-04

1 1.32 1.64 2.07 2.71 3.81 5.02 5.41 6.63 7.88 9.14 10.83 12.12

2 2.77 3.22 3.79 4.61 5.99 7.38 7.82 9.21 10.60 11.98 13.82 15.20

3 4.11 4.64 5.32 6.25 7.81 9.35 9.84 11.34 12.84 14.32 16.27 17.73

We’re off the chart on the left, which means we will definitely not exceed
the alpha level α = 0.01. Therefore, the coffee company will fail to reject
the null hypothesis, and conclude that beverage preference does not
affect pastry preference.

44
 4. A school district wants to know whether or not GPA is affected by
elective preference. They randomly sampled fewer than 10 % of their
students, and recorded their elective preference an GPA. What can the
school district conclude using a chi-square test at α = 0.1?

GPA range

<2 2 3 4+ Totals

Music 12 26 31 34 103

Theater 21 22 23 21 87

Art 36 29 29 32 126

Totals 69 77 83 87 316

Solution:

Start by computing expected values.

GPA range

<2 2 3 4+ Totals

12 26 31 34
Music 103
(22.49) (25.10) (27.05) (28.36)
21 22 23 21
Theater 87
(19.00) (21.20) (22.85) (23.95)
36 29 29 32
Art 126
(27.51) (30.70) (33.09) (34.69)

Totals 69 77 83 87 316

45
Now we’ll check our sampling conditions. The problem told us that we
took a random sample and that we sampled less than 10 % of the
population, so we’ve met the random sampling and independence
conditions. And all of our expected values are at least 5, so we’ve met the
large counts condition as well.

We’ll state the null and alternative hypotheses.

H0: Elective choice doesn’t affect GPA.

Ha: Elective choice affects GPA.

Calculate χ 2.
2 2 2 2
(12 − 22.49) (26 − 25.10) (31 − 27.05) (34 − 28.36)
χ2 = + + +
22.49 25.10 27.05 28.36

(21 − 19.00)2 (22 − 21.20)2 (23 − 22.85)2 (21 − 23.95)2

+ + + +
19.00 21.20 22.85 23.95

(36 − 27.51)2 (29 − 30.70)2 (29 − 33.09)2 (32 − 34.69)2

+ + + +
27.51 30.70 33.09 34.69

χ 2 ≈ 4.89 + 0.03 + 0.58 + 1.12 + 0.21 + 0.03

+0.00 + 0.36 + 2.62 + 0.09 + 0.51 + 0.21

χ 2 = 10.65

The degrees of freedom are

df = (number of rows − 1)(number of columns − 1)

df = (3 − 1)(4 − 1)

46
 df = (2)(3)

df = 6

With df = 6 and χ 2 = 10.65, the χ 2-table gives

Upper-tail probability p

df 0.25 0.20 0.15 0.10 0.05 0.025 0.02 0.01 0.005 0.003 0.001 5E-04

5 6.63 7.29 8.12 9.24 11.07 12.83 13.39 15.09 16.75 18.39 20.52 22.11

6 7.84 8.56 9.45 10.64 12.59 14.45 15.03 16.81 18.55 20.25 22.46 24.10

7 9.04 9.80 10.75 12.02 14.07 16.01 16.62 18.48 20.28 22.04 24.32 26.02

The χ 2 value just clears α = 0.1, which means that the school district can
reject the null hypothesis and conclude that elective choice affects GPA. If
they had set a higher confidence level of 95 % (with α = 0.05), they would
not have been able to reject the null.

5. An airline wants to know if people travel constantly throughout the

year, or if travel is more concentrated at specific times. They recorded
flights taken each quarter, and recorded them in a table (in hundreds of
thousands). What can the airline conclude using a chi-square test at 95 %
confidence?

Quarter Jan-Mar Apr-Jun Jul-Sep Oct-Dec Total

Flights 3.97 4.58 4.73 5.14 18.42

Solution:

47
With 18.42 (or 18,420,000) total flights, the expected number of flights in
each quarter would be 18.42/4 = 4.605.

Quarter Jan-Mar Apr-Jun Jul-Sep Oct-Dec Total

Flights 3.97 4.58 4.73 5.14 18.42

Expecte
4.605 4.605 4.605 4.605 18.42
d

We’ll state the null and alternative hypotheses.

H0: Number of flights taken is not affected by quarter.

Ha: Number of flights taken is affected by quarter.

Calculate χ 2.

2 (3.97 − 4.605)2 (4.58 − 4.605)2 (4.73 − 4.605)2 (5.14 − 4.605)2

χ = + + +
4.605 4.605 4.605 4.605

χ 2 ≈ 0.0876 + 0.0001 + 0.0034 + 0.0622

χ 2 = 0.1533

The degrees of freedom are n − 1 = 4 − 1 = 3. With df = 3 and χ 2 = 0.1533,

the χ 2-table gives

Upper-tail probability p

df 0.25 0.20 0.15 0.10 0.05 0.025 0.02 0.01 0.005 0.003 0.001 5E-04

2 2.77 3.22 3.79 4.61 5.99 7.38 7.82 9.21 10.60 11.98 13.82 15.20

3 4.11 4.64 5.32 6.25 7.81 9.35 9.84 11.34 12.84 14.32 16.27 17.73

4 5.39 5.99 6.74 7.78 9.49 11.14 11.67 13.28 14.86 16.42 18.47 20.00

48
We’re off the chart on the left, which means we will definitely not exceed
the alpha level α = 0.05. Therefore, the airline will fail to reject the null
hypothesis, and conclude that number of flights taken is not affected by
quarter.

6. A sandwich company wants to know how their sales are affected by

time of day. They recorded sandwiches sold during each part of the day.
What can the sandwich company conclude using a chi-square test at
α = 0.1?

Time of day Midday Afternoon Evening Total

Sales 213 208 221 642

Solution:

With 642 total sandwiches sold, the expected number of sandwiches in

each period would be 642/3 = 214.

Time of day Midday Afternoon Evening Total

Sales 213 208 221 642

Expected 214 214 214 642

We’ll state the null and alternative hypotheses.

H0: Number of sandwiches sold is not affected by time of day.

49
 Ha: Number of sandwiches sold is affected by time of day.

Calculate χ 2.

(213 − 214)2 (208 − 214)2 (221 − 214)2

2
χ = + +
214 214 214

1 36 49
χ2 = + +
214 214 214

86
2
χ =
214

χ 2 ≈ 0.4019

The degrees of freedom are n − 1 = 3 − 1 = 2. With df = 2 and χ 2 = 0.4019,

the χ 2-table gives

Upper-tail probability p

df 0.25 0.20 0.15 0.10 0.05 0.025 0.02 0.01 0.005 0.003 0.001 5E-04

1 1.32 1.64 2.07 2.71 3.81 5.02 5.41 6.63 7.88 9.14 10.83 12.12

2 2.77 3.22 3.79 4.61 5.99 7.38 7.82 9.21 10.60 11.98 13.82 15.20

3 4.11 4.64 5.32 6.25 7.81 9.35 9.84 11.34 12.84 14.32 16.27 17.73

We’re off the chart on the left, which means we will definitely not exceed
the alpha level α = 0.1. Therefore, the sandwich company will fail to reject
the null hypothesis, and conclude that number of sandwiches sold is not
affected by time of day.

50
51