Announcements:

Final quiz begins after class today and ends on

Monday at noon.
For Monday discussion, teams will discuss results
Chi-Square Test
from articles in medical journals. Article summaries
are posted on course website (in daily calendar).
Section 15.3 for
Homework (due Mon, Nov 29):
Goodness-of-Fit
Chapter 15: #6, #26bd, #35

Research Question 15.3 Testing Hypotheses about One

For a categorical variable with k categories, Categorical Variable
are the population proportions (or Situation: Similar to binomial, but there can be more
than two possible outcomes. Called multinomial.
probabilities) falling into each of the k
categories as specified? Measure a single categorical variable on each person
or trial.
Examples:
p
Each person or trial falls into one of k mutually
Are digits 0 to 9 equally likely to be drawn in lottery? exclusive categories.
In genetics, is offspring ratio 9:3:3:1, as expected by Null hypothesis specifies the probabilities of falling
Mendels laws? into each of the k categories.
Is death from sudden infant death syndrome equally Alternative hypothesis is that those are not all
likely in all 4 seasons? correct.
3 4

Example 15.8 Pennsylvania Daily Number Use same 5 steps of hypothesis testing
Called chi-square goodness-of-fit test
State lottery game: Three-digit number made by Step 1: Determine the null and alternative hypotheses.
drawing a digit between 0 and 9 from each of three
different containers. H0: The probabilities for k categories are p1, p2, . . . , pk.
Ha: Not all probabilities specified in H0 are correct.
Lets examine draws from the first container.
If numbers
b randomly
d l selected,
l t d eachh value
l would ld be
b
equally likely to occur. So, k = 10 and on each draw Note: Probabilities in the null hypothesis must sum to 1.
there is probability 1/10 of getting each digit (0, ..., 9)
Pennsylvania Lottery Example:
H0: p1 = p2 = ...= p10 = 1/10
H0: p = 1/10 for each of the 10 possible digits
Ha: Not all probabilities are 1/10. Ha: The 10 digits are not all equally likely.
5 6

1
Goodness of Fit (GOF) Test (continued) Example 15.8 Pennsylvania Daily Number
Step 2: Verify necessary data conditions, and if met, Data: n = 500 days between 7/19/99 and 11/29/00
summarize the data into an appropriate test statistic.
Data condition needed: At least 80% of the expected
counts are greater than 5 and none are less than 1. Test
statistic:
2 (Observed Expected) 2

Expected

where the expected count for the ith category

is computed as npi.

7 8

Step 3: p-value of Chi-square Test

Example, continued Large test statistic => evidence that values in null
are not correct (observed counts dont match
Expected count = 500 ( 1/10) = 50 for each digit
expected counts).
p-value = probability the chi-square test statistic could
g or larger
have been as large g if the null hypothesis
yp were
true.
Chi-square probability distribution used to find p-value.
Degrees of freedom: df = k 1
This is because we are free to specify k 1 totals, then
the last one is determined.
10

Chi-square Distributions Finding the p-value from Table A.5, p. 732:

Skewed to the right distributions. Look in the corresponding df row of Table A.5.
Minimum value is 0. Scan across until you find where the statistic falls.
Indexed by the degrees of freedom.
If value of statistic falls between two table entries,
Note: p-value is between values of p (column headings)
Mean of the distribution for these two entries.
entries
= degrees of freedom. If value of statistic is larger than entry in
rightmost column (labeled p = 0.001),
p-value is less than 0.001 (written as p < 0.001).
If value of statistic is smaller than entry in
leftmost column (labeled p = 0.50),
p-value is greater than 0.50 (written as p > 0.50).
11 12

2
 Step 4: Making a Decision Example 15.8 Daily Number (cont)
Chi-square goodness of fit statistic:
Large test statistic => small p-value
=> evidence that the proportions are not as specified. From Table A.5 (page 732) gives areas to the right of the
chi-square value, because thats the p-value in this
Two equivalent rules: situation.: Example: Chi-square value = 6.04.
j H0 when
Reject df = k 1 = 10 1 = 9
p-value > 0.50 (note it is 0.736)
p-value 0.05 Rejection region: Above 16.92.
Chi-square statistic is greater than the entry in Result is not statistically significant;
the 0.05 column of Table A.5 (the critical value). the null hypothesis is not rejected.
That defines the rejection region.

13 14

Step 5: Report the Conclusion in Context New Example: Is Sudden Infant Death
Syndrome (SIDS) Seasonal?

Conclusion: Pennsylvania lottery digits drawn are not Data from King County, Washington
statistically different from whats expected by chance. Define p1, p2, p3, p4 to be the proportion of deaths from
SIDS that happen in the winter, spring, summer and fall.
They are defined so that the seasons have about equal days.
Step 1: Determine the null and alternative hypotheses.
H0: p1 = , p2 = , p3 = , p4 =
Ha: Not all probabilities specified in H0 are correct.

Note: Probabilities in the null hypothesis must sum to 1.

15 16

Step 2: Verify necessary data conditions, and if met, Example, continued

summarize the data into an appropriate test statistic.
Data condition needed: At least 80% of the expected
Total n = 78 + 71 + 87 + 86 = 322
counts are greater than 5 and none are less than 1. Test Expected count = 322 ( 1/4) = 80.5 for each
statistic: season.
2 (Observed Expected) 2

Expected (78 80.5)2 (71 80.5) 2 (87 80.5)2 (86 80.5)2

2
80.5 80.5 80.5 80.5
where the expected count for the ith category
is computed as npi. = 2.10
Example: Counts for the 4 seasons were 78, 71, 87, 86
Are these different enough to conclude a difference exists in
the population?
17

3
 Step 4: Making a Decision
Step 3: Finding the p-value
Large test statistic => small p-value
Degrees of freedom = 4 1 = 3. => evidence that the proportions are not as specified.
Two equivalent rules:
From Table A.5, smallest entry is 2.37, the value Reject H0 when
with
ith .50
50 belo
below it.
it So
So, for our
o r test statistic of p-value 0.05; in our example it is.
2.10 all we can say is p-value > .50. Chi-square statistic is greater than the entry in
the 0.05 column of Table A.5 (the critical value).
Rejection region approach: That defines the rejection region. In our example,
For df = 3, reject the null hypothesis if the test the test statistic is not in the rejection region.
statistic is greater than 7.81. (Ours is not.) So we do not reject the null hypothesis.
20

Step 5: Report the Conclusion in Context Use of chi-square test in genetics

Based on Mendels laws, expect certain ratios of
Conclusion: Sudden infant death syndrome proportions phenotypes. Can be tested using chi-square
across seasons are not statistically different from goodness-of-fit tests.
whats expected by chance (i.e. all seasons being Example: In a dihybrid cross (AaBb x AaBb), the
equal).
q ) expected
p proportions
p p of 4 pphenotypes
yp are 9:3:3:1.
Data from classic experiment with Starchy/sugary and
Green/white seedlings, progeny of 3839 self-
fertilized heterozygotes (Starchy/green,
Starchy/white, Sugary/green, Sugary/white):
1997, 906, 904, 32.
Null hypothesis probabilities are 9/16, 3/16, 3/16, 1/16
21

Results from Minitab for this example

Chi-Square Goodness-of-Fit Test
Test Contribution
Category Observed Proportion Expected to Chi-Sq
1 1997 0.5625 2159.44 12.219
2 906 0.1875 719.81 48.159
3 904 0
0.1875
1875 719
719.81
81 47
47.130
130
4 32 0.0625 239.94 180.205
N DF Chi-Sq P-Value
3839 3 287.714 0.000

Because the p-value is 0.000, reject the null hypothesis.

Conclude that in this case, the genetics did not work out to
be the 9:3:3:1 ratio expected.
Note that the largest contribution to the large test statistic
is (4)Sugary/white. Observed = 32, expected = 239.94.