Today: Sections 13.1 to 13.

3 Finishing what we planned to cover when we started Chapter 9

Five situations we will cover for the rest of this quarter:
ANNOUNCEMENTS: Parameter name and description Population parameter Sample statistic
• We will finish hypothesis testing for the 5 situations today. See For Categorical Variables:
pages 586-587 (end of Chapter 13) for a summary table. One population proportion (or probability) p p̂
• Quiz for week 8 starts Wed, ends Monday at noon Difference in two population proportions p1 – p2 pˆ 1 − pˆ 2
For Quantitative Variables:
One population mean µ x
HOMEWORK (due Monday, Nov 29): Population mean of paired differences
Chapter 13: #15, 24ac, 25 (partial answer in back) µd d
(dependent samples, paired)
Difference in two population means
µ1 − µ2 x1 − x 2
(independent samples)

For each situation will we:

√ Learn about the sampling distribution for the sample statistic
√ Learn how to find a confidence interval for the true value of the parameter
• Test hypotheses about the true value of the parameter
• For independent samples, will see how to do in R Commander only.

Five steps to hypothesis testing – one mean and mean of paired Example for testing one population mean:
difference: Summary Boxes on pages 558-559 and 562.

STEP 1: Determine the null and alternative hypotheses. Is mean human body temperature really 98.6 degrees, or is it
One population mean: Population mean of paired differences lower?
Null hypothesis: H0: µ = µ0 Null hypothesis: H0: µd = 0
Null value is called µ0 Null value = 0 (Note special null value) H0: µ = 98.6 degrees
Ha: µ < 98.6 degrees
Alternative hypothesis is one of these, based on context:
Ha: µ ≠ µ0 Ha: µd ≠ 0
Ha: µ > µ0 Ha: µd > 0 n = 101 blood donors at clinic near Seattle, ages 17 to 84
Ha: µ < µ0 Ha: µd < 0
Sample mean = x = 97.89 degrees,
Sample standard deviation = s = 0.73 degrees
s 0.73
= = 0.073
Standard error = s.e.( x ) = n 101
Example for testing population mean of paired differences: STEP 2:
Verify data conditions. If met, summarize data into test statistic.
Do people gain or lose weight when they quit smoking? American
Journal of Public Health, 1983, pgs 1303-05. Data conditions:
Bell-shaped data (no extreme outliers or skewness) or large sample.
For each person, di = difference in weight (after – before) for Test statistic (remember, use t for means):
people who quit smoking for 1 year. (Positive = weight gain)
sample statistic − null value
µd = population mean weight gain in 1 year for smokers who quit. t=
(null) standard error
H0 : µ d = 0 One population mean: Mean of paired differences:
Ha : µ d ≠ 0 Sample statistic = x Sample statistic = d
Null value = µ0 Null value = 0
n = 322, Sample mean = d = 5.15 pounds, s sd
Sample standard deviation = sd = 11.45 pounds Null standard error = n Null standard error = n
sd 11.45
= = .6381
Standard error of d = n 322 Note that the word “null” is unnecessary in std. error involving means.

Step 2 for the Examples: STEP 3:

Assuming the null hypothesis is true, find the p-value.
Data conditions are met, since both sample sizes are large.
General: p-value = the conditional probability of a test statistic as
extreme as the one observed or more so, in the direction of Ha, if
Example for one mean
the null hypothesis is true.
(Population mean body temperature = 98.6?):
sample statistic − null value 97.89 − 98.6 − .71 Same idea as other situations (see pictures on p. 517), but now we
t= = = = −9.77 need to use the t-distribution with df = n – 1, instead of normal
(null) standard error .73 .0726
101 distribution.

Example for mean of paired differences Alternative hypothesis (similar for µd): p-value is:
(Population mean weight loss after quitting smoking = 0?): Ha: µ > µ0 (a one-sided hypothesis) Area above the test statistic t
Ha: µ < µ0 (a one-sided hypothesis) Area below the test statistic t
sample statistic − null value 5.15 − 0 5.15 Ha: µ ≠ µ0 (a two-sided hypothesis) 2 × the area above |t| = area in
t= = = = 8.07
(null) standard error 11.45 .6381 tails beyond −t and t
322
Use Table A.3 on page 729: Area above 2.20, df = 14
One-Sided p-values for Significance Tests Based on a t-Statistic
Table will provide a p-value range, not an exact p-value.
Can also use Excel or R Commander.
Ex: n = 15, df = 14, t = 2.20
Ha: µ > µ0
p-value = area above 2.20
Since 2.20 is between 2.00
and 2.33, p-value is between
.033 and .018:
.018 < p-value < .033
Double it for two-sided:
Ha: µ ≠ µ0
.036 < p-value < .066
Use with negative values for
Ha: µ < µ0

p-value for our two examples: STEP 4 – using p-values:

Decide whether or not the result is statistically significant based
Example for one mean (normal body temperature): on the p-value.
Ha: µ < 98.6
Examples:
t = –9.77
Mean body temperature:
p-value = area below t = –9.77 for df = 100
p-value = 1.6 × 10-16 < .05, so:
Best we can do from Table A.3 is p-value < .002.
• Reject the null hypothesis.
From Excel, p-value = 1.6 × 10-16
• Accept the alternative hypothesis
• The result is statistically significant
Example for paired differences
(weight gain/loss when quitting smoking): Paired difference, mean weight gain/loss after quitting smoking:
Ha : µ d ≠ 0 p-value = 1.4 × 10-14 < .05, so:
t = 8.07 • Reject the null hypothesis.
p-value = 2 × area above |8.07| for df = 321. • Accept the alternative hypothesis
• The result is statistically significant
Best we can do from Table A.3 is p-value < .004 (take 2 × .002)
From Excel, p-value = 1.4 × 10-14
For tests involving the t-distribution, there is a Substitute Step 3
and 4, called the Rejection Region Approach. Examples (Use α = .05):

Mean body temperature, n = 101, df = 100

Rejection region is the set of test statistic values that will lead us One-sided test Ha: µ < 98.6, Rejection region is t ≤ –1.66
to reject the null hypothesis. Use the bottom row of Table A.2.
Weight gain or loss one year after quitting smoking, df = 321
Alternative hypothesis Column of Table A.2 Rejection region Two-sided test Ha: µd ≠ 0, Rejection region is |t| ≥ 1.98 (use df = 100)
Ha: µ ≠ µ0 Two-tailed α |t| ≥ t*
Ha: µ > µ0 One-tailed α t ≥ t* Rejection region, one-sided, "less than" alternative
T, df=100
Rejection region, two-sided alternative
T, df=100

Ha: µ < µ0 One-tailed α t ≤ –t*

0.05
0.025 0.025
-1.66 0 -1.98 0 1.98
test statistic t test statistic t

Substitute Step 4: Rejection Region Approach

Step 5: Report the conclusion in the context of the situation.
If the test statistic is not in the rejection region:
• Do not reject the null hypothesis. Example 1:
• There is not enough evidence to accept the alternative hypothesis The mean body temperature for healthy human adults is less than 98.6
• The result is not statistically significant degrees.

If the test statistic is in the rejection region: Note: We found a 95% confidence interval for this in an earlier lecture.
• Reject the null hypothesis. It was 97.75 to 98.03 degrees.
• Accept the alternative hypothesis
• The result is statistically significant Example 2:
The mean change in weight for one year after quitting smoking is
For both examples, the test statistic is definitely in the rejection region, significantly different from 0.
so we reject the null hypothesis.
Note: A 95% confidence interval for the mean change is weight is:
5.15 ± 1.97(.638) or 3.89 to 6.41 pounds.
Possible problem: No control group! People gain weight as they age.
 Hypothesis test for difference in two means, independent samples New Example: Work through from start to finish
Called a “two-sample t-test” or “independent samples t-test.”
You already learned how to do this with R Commander. Research question: Can drinking an ice slushie increase
endurance when exercising in hot weather?
Example from Exercise 11.51: Two-sample t-test to compare pulse for
those who do and don’t exercise
Australia study published in Medicine and Science in Sports and
• Data → New data set – give name, enter data Exercise, 2010
• One column for Exercise (Y,N) and one column for pulse
• Statistics → Means →Independent samples t-test • 10 Male volunteers, average age 28
• Choose the alternative (≠,>,<) and conf. level
data: Pulse by Exercises • Two treatments administered to all 10 men:
t = 1.7156, df = 13.014, p-value = 0.05496 o Drink fruit-flavored ice slushie
alternative hypothesis: true difference in means is not
equal to 0
o Drink fruit-flavored cold water
95 percent confidence interval: • Then run on treadmill in 93 degree room until exhausted
-1.727387 15.060720 • Response variable = time until exhaustion
sample estimates:
mean in group N mean in group Y • Order randomized, administered a few weeks apart
72.00000 65.33333 • Did some practice runs to eliminate “learning effect”

Parameter of interest:
µd = mean difference in exhaustion times if everyone in the
population were to run under both conditions.
Hypotheses:
H0: µd = 0 (Slushie and water have same effect on endurance)
Ha: µd > 0 (Slushie improves endurance)
Data and Test Statistic:
3.6
d = 9.5 minutes, sd = 3.6 minutes, so s.e.(d ) = = 1.14
10
9.5 − 0
t= = 8.3, df = 9, p-value ≈ 0.
1.14
Reject H0, conclude ice slushie does increase endurance
compared to drinking cold water.
95% confidence interval is 9.5 ± 2.26(1.14) or 6.9 to 12.1 mins.