Parametric and Non-parametric Methods (Chapter 18)

Parametric methods Non-Parametric methods

H0: Median = 450

Ha: Median ≠ 450

1
 Chapter 18 - Nonparametric Tests

Frank Wilcoxon (1882 -1965)

“The basic idea of replacing actual
sample data by their ranks, which
seems so utterly simple in retrospect,
proved to be inspirational to the further
development of the entire field of
nonparametric statistics.
The elegant simplicity of these tests
led to their widespread adoption and
the fact that Wilcoxon in 1945 was not
even aware of all the advantages of his
new methods does not dim the luster
of his contribution.”

2
 Confidence Intervals: Class Exercise
The school psychologist is currently busy studying
stress during exam time. She has developed a
questionnaire that she believes measures stress.
A score above 80 indicates stress at a dangerous
level.
A random sample of 16 senior learners revealed an
average stress level of 84.25.
You may assume that stress levels are normally distributed with
standard deviation of 8.

µ = 80
3
Nonparametric Tests (Chapter 18) linked to Parametric Tests (Ch 9 & 10)

Previously: Now:
• Normal distribution or big • Small samples
samples • No assumptions about the specific
• Quantitative data form of the population’s probability
distribution
• Categorical or quantitative data

4
 STK110: Chapter 9: Hypothesis Testing
Median: 50% of the values are less than the median and 50% of the
values are more than the median.

450

5
 18.1 The Sign Test

In conducting the sign test, we compare each observation to the

hypothesized value of the population median.

✔Observation > than the hypothesized value, record a “+” sign

✔Observation < than the hypothesized value, record a “-” sign
✔Observation = the hypothesized value, eliminate the observation
and proceed with the smaller sample.

6
 18.1 The Sign Test Example: The sign of the differences
The following data is weekly sales of a new product and the
manufacturer’s estimate that the median sales should be $450 per
week on a per store basis
Store Weekly Sign Store Weekly Sign
Number sales($) Number sales($)
56 485 + 63 474 +
19 562 + 39 662 +
36 415 - 84 380 -
128 860 + 102 515 +
12 426 - 44 721 +
H0: Median = 450 H0: p = 0.5
Ha: Median ≠ 450 Ha: p ≠ 0.5
Test statistic: The number of plus signs = 7

Decision rule: Go to the Binomial table with n=10 and p = 0.5 (or use
the BINOM.DIST function) to get the p-value = probability to get a
value of 7 or more. 7
 18.1 The Sign Test Example: The sign of the differences

H0: Median = 450 H0: p = 0.5

Ha: Median ≠ 450 Ha: p ≠ 0.5
Test statistic: The number of plus signs = 7
Decision rule: BINOM.DIST(6,10,0.5,TRUE) = 0.8281
And p-value = 1- 0.8281 = 0.1719, but, because we are testing 2-
sided, the p-value = 2(0.1719) = 0.3438 > 0.05 and the null hypothesis
can not be rejected
Conclusion:
The
hypothesis
that the
median sales
per week per
shop is $450
cannot be
rejected
8
 Normal approximation of the Binomial Distribution

The normal approximation to binomial probabilities works well even for moderately
large n as long as p is not close to zero or one

1.5 2.5

Continuity
correction:

9
Normal approximation of the Binomial Distribution: Example Home sales

One year ago the median price of a new home was $236000. Current downturn in
economy has real estates using sample data on recent home sales to determine if the
population median price of a new home is less today than a year ago. A random
sample of 61 recent new home sales was investigated resulting in one price the same
as the hypothesized median (ignore this one) and 22 plus signs. (part of the Excel file
HomeSales is shown.

10
Normal approximation of the Binomial Distribution: Example Home sales

H0: Median ≥ 236 000 H0: p = 0.5 Note that sample size was
reduced by one value that was
Ha: Median < 236 000 Ha: p ≠ 0.5 the same as the median value

p-value = P(x ≤ 22.5) …note the

correction factor of .5

= P(z ≤ -1.94) …Using Excel’s NORM.S.DIST function

= 0.0262

Conclusion: Because this p-value <0.05, we reject the null hypothesis

and conclude that the median price of a new home is less than the $236
000 median price a year ago.
11
 18.1 The Sign Test: Class exercise

A high school student decided to keep track of the number

of hours spent studying for 30 days. He reported that he
spent more than 4 hours studying on 21 days and less
than 4 hours on 9 days. He is interested to know if the
median number of hours spent studying is more than 4.

Do a complete hypothesis test (on a 5% level) and use the

normal distribution to approximate the binomial probability
of 21 plus signs

12
 18.1 The Sign Test: Class exercise

H0: Median ≤ 4
Ha: Median >4

p-value = P(x > 21 - 0.05 ) …note the correction factor of .5

Rejection rule: Reject null hypotheses if p < 0.05

Decision: Reject the null hypotheses because p < 0.05

Conclusion: We do have enough evidence to

conclude that the median number of hours studied is
more than 4

13
14
 STK 110 Statistical Inference Concerning (µ1-µ2)

µ = 125 µ = 135
…mmmm, is there a
significant difference
between the 2 population
means?

15
 STK 110 Statistical Inference Concerning (µ1-µ2)

Two independent populations – population variances known

16
 Nonparametric Tests: two independent samples

Using Ranks for Comparing Two Populations

Distributions: Independent Random Samples

H0: The two populations are identical

Ha: The two populations are not identical

17
 Nonparametric Tests: two independent samples

H0: µY = µX
Ha: µY > µX

?
The nonparametric analogue tests for equality of medians in the two
distributions.
18
 18.3 The Mann-Whitney-Wilcoxon Test: Example
Two fuel additives are being tested to determine their effect on petrol mileage.
Seven cars were tested with additive 1 and nine cars with additive 2. Data shows
miles per gallon obtained with the two additives. Test whether there is a significant
difference between petrol mileage for the two additives.

H0: The two population of additives are identical

Ha: The two population of additives are not identical

19
 18.3 The Mann-Whitney-Wilcoxon Test, Normal approximation:
Example
Pooled Ranks
Mileage per gallon petrol
16.7 1
Additive 1 Additive 2
17.3 2
17.3 18.7
17.5 3
18.4 17.8
17.8 4
19.1 21.3
18.2 5
16.7 21.0
18.4 6
18.2 22.1
18.6 7
18.6 18.7
18.7 8.5
17.5 19.8
18.7 8.5
20.7
19.1 10
20.2
19.8 11
Step 1: Pool the data from the two groups 20.2 12
and order the values from lowest to highest 20.7 13
Step 2: Assign ranks to the values in 21 14
increasing order. When there are ties, the 21.3 15 20
mean ranks are assigned. 22.1 16
 18.3 The Mann-Whitney-Wilcoxon Test, Normal approximation:
Example
Pooled Ranks
Mileage per gallon petrol
16.7 1
Additive 1 Additive 2
17.3 2
17.3 2 18.7 8.5
17.5 3
18.4 6 17.8 4
17.8 4
19.1 10 21.3 15
18.2 5
16.7 1 21.0 14
18.4 6
18.2 5 22.1 16
18.6 7
18.6 7 18.7 8.5
18.7 8.5
17.5 3 19.8 11
18.7 8.5
20.7 13
19.1 10
20.2 12
19.8 11
34 20.2 12
Step 3: Return to the two separate samples
20.7 13
and show ranks.
21 14
Step 4: Use the sum of the ranks for 21.3 15 21
sample 1 as the test statistic W
22.1 16
 18.3 The Mann-Whitney-Wilcoxon Test, Normal approximation:
Example

P(W ≤ 34) …note the correction factor of .5

p-value = 2(.0040) <0.05

Reject H0 and conclude that additives are not identical

22