Chapter 14 Nonparametric Inferences
Nonparametric Inferences based on a samples
1. Comparing a Location Parameters: a Independent Samples
(Completely Randomized Design: One-way Fixed Effects Design)
2. Comparing a Location Parameters: a Paired Samples
(Completely Randomized Block Design)
Nonparametric Test for Correlation
1
Nonparametric Inferences
Nonparametric Inferences based on a samples
1. Comparing a Location Parameters: a Independent Samples
(Completely Randomized Design: One-way Fixed Effects Design)
In comparing a independent populations, the one-way analysis of variance (ANOVA) procedure
was employed. To apply the ANOVA procedure using the F distribution, we assume that the
populations from which the samples were drawn were normally distributed with equal variance,
σ 2 . However, if the populations being sampled are not normally distributed, then we cannot
apply the ANOVA procedure. In such cases, nonparametric techniques are used.
The most widely used nonparametric technique for comparing several populations is the
Kruskal-Wallis one-way analysis of variance by ranks. When only two samples are being
considered, the Kruskal-Wallis test is equivalent to the Mann-Whitney test.
a. Assumptions
1. Y11 , Y12 ,..., Y1n1 is a random sample of size n1 from population 1.
Y21 , Y22 ,..., Y2n2 is a random sample of size n 2 from population 2.
Ya1 , Ya 2 ,..., Yana is a random sample of size n a from population a.
2. a samples are independent.
3. Each Yi is a continuous random variable.
4. The variable Yi is measured on at least ordinal scale.
b. Hypotheses of Interests
H0 : The a populations have identical distribution
vs H1 : At least one population differ with respect to location
or H0 : μ1 ... μ a
vs H1 : At least one μ i μ i'
c. Construction of a Test Statistic
1. Combine all a samples.
2. Rank all n1 n 2 ... n a observations. Assign tied observations the mean of the rank
positions.
3. Sum the ranks of the observations from each sample:
ni
R i rank Yij , i 1,..., a
j1
2
4. Test Statistic:
n i N 1
2
12 a
1
T R i
N N 1 i 1 n i
2
12 a
R i2 a
or T 3 N 1,
N N 1 i 1 n i
N n i n1 ... n a
i 1
T ~ Kruskal-Wallis (n1, n2,…, na) Table 25 of the Cambridge Table
Note that T approaches the chi-square distribution with a 1 degrees of freedom as N
approaches infinity.
A Hypothesis Test for a Location Parameters: a Independent Samples
Hypothesis: H0 : μ1 ... μ a
vs H1 : At least one μ i μ i'
Test Statistic:
12 a
R i2
T
N N 1 i 1 n i
3 N 1
ni
R i rank Yij , i 1,..., a
j1
T ~ Kruskal-Wallis (n1, n2, … , na)
Rejection Region:
Reject H0 if T Wα, n1 , n2 , ..., na
α P T Wα, n1 , n 2 , ..., na
P Kruskal-Wallis n , n , ..., n
1 2 a Wα, n1 , n2 , ..., na
Example:
The data shows cortisol levels in three groups of patients who delivered between 38 and 42
weeks of gestation.
Group 1: Patients before the onset of labor at elective Caesarean section.
Group 2: Patients at emergency Caesarean section during induced labor.
3
Group 3: Patients in whom spontaneous labor occurred and who delivered either normally
or by Caesarean section.
Group 1: 262 307 211 323 454 339 304 154 287 356
Group 2: 465 501 455 355 468 362
Group 3: 343 772 207 1048 838 687
By using the Kruskal-Wallis test, conclude whether these data indicate a difference in mean
cortisol levels among the three groups of patients? Use 0.05.
Solution:
Factor: Group of patients
Samples: 3 groups
Sampling units: 22 patients
Cortisol levels are recorded from each group of patients – 3 independent samples.
Group 1 Group 2 Group 3
10 6 6
Independent samples
Data Type:
Response variable: Cortisol Levels
Parameter: μ Estimator: Y
Computations:
Group 1 Rank Group 2 Rank Group 3 Rank
154 1
211 3 207 2
262 4
287 5
304 6 355 11 343 10
307 7 362 13
323 8
339 9
356 12
4
454 14 455 15 687 19
465 16 772 20
468 17 838 21
501 18 1048 22
n1 = 10 R1 = 69 n2 = 6 R2 = 90 n3 = 6 R3 = 94
Study Objective:
To compare mean cortisol levels among three groups of patients.
Assumptions
1. Y11 , Y12 ,..., Y1,10 is a random sample of size 10 from population 1.
Y21 , Y22 ,..., Y26 is a random sample of size 6 from population 2.
Y31 , Y32 ,..., Y36 is a random sample of size 6 from population 3.
2. The 3 samples are independent.
3. Each Yi is a continuous random variable.
4. The variable Yi is measured on at least ordinal scale.
A Hypothesis Test for a Location Parameters: a Independent Samples
Hypothesis: H0 : μ1 ... μ a
(Cortisal levels do not differ)
vs H1 : At least one μ i μ i'
Test Statistic:
12 a
R i2
T 3 N 1
N N 1 i 1 n i
12 69 2 90 2 94 2
3 22 1
22 22 1 10 6 6
9.232
T ~ Kruskal-Wallis (n1 = 10, n2 = 6, n3 = 6)
Rejection Region:
Reject H0 if T Wα,n1 ,n2 ,...,na
W0.05, 10, 6, 6 5.991 (from chi-square table with 2 degrees of freedom)
5
(Remark: T approaches the chi-square distribution with a 1 = 3 – 1 = 2 degrees of freedom as
N = 10 + 6 + 6 = 22 is large)
Conclusion:
T = 9.232 W0.05, 10, 6, 6 5.991 . H0 is rejected at α 0.05 . There is a difference in mean
cortisol levels among the three groups of patients.
2. Comparing a Location Parameters: a Paired Samples
A nonparametric technique commonly used for comparing “a paired populations” is the
Friedman test.
a. Assumptions:
1. There are “a samples”:
Sample 1: Y11 , Y12 ,..., Y1n is a random sample from population 1.
Sample 2: Y21 , Y22 ,..., Y2n is a random sample from population 2.
Sample a: Ya1 , Ya 2 ,..., Yan is a random sample from population a.
2. a samples are dependent. Each observation is measured using “a” different methods, e.g.,
( Y11 , Y21 ,...,Ya1 ) and ( Y12 , Y22 ,..., Ya2 ) refer to the first and second observations, respectively,
that are measured using “a” different methods.
3. Each Yi is a continuous variable.
b. Hypotheses of Interests
H0 : The a populations have identical distribution
vs H1 : At least one population differ with respect to location
or H0 : μ1 ... μ a
vs H1 : At least one μ i μ i'
c. Construction of a Test Statistic
1. Rank observations within each block.
2. Sum the ranks of the observations from each sample:
ni
R i rank Yij , i 1,..., a
j1
6
4. Test Statistic:
a
12
T
ab a 1 i 1
R i2 3b a 1 ,
a = number of treatments, b = number of blocks
T ~ Friedman (a, n) Table 24 of the Cambridge Table
Note that T approaches the chi-square distribution with a 1 degrees of freedom as n
approaches infinity.
A Hypothesis Test for “a” Location Parameters: “a” Paired Samples
Hypothesis: H0 : μ1 ... μ a
vs H1 : At least one μ i μ i'
Test Statistic:
a
12
T
ab a 1 i 1
R i2 3b a 1 ,
R i rank Yij ,
n
i 1,...,a
j1
T ~ Friedman (a, n)
Rejection Region:
Reject H0 if T Wα, a, n
α P T Wα, a, n
P Friedman a, n Wα, a, n
Example:
A researcher compared three methods of determining serum amylase values in patients with
pancreatitis. Each specimen taken from each patient was measured using each of three methods.
Method of Determination
7
Specimen A B C
1 4000 3210 6120
2 1600 1040 2410
3 1600 647 2210
4 1200 570 2060
5 840 445 1400
6 352 156 249
7 224 155 224
8 200 99 208
9 184 70 227
By using the Friedman test, conclude whether these data indicate a difference among the three
methods? Use 0.05.
Solution:
Factor: Method of determination
Treatment/samples: 3 groups
Experimental Units: 9 patients
Block: 9 patients
Each specimen taken from each patient was measured using each of three methods.
Data Type:
Response variable: Serum amylase values
Parameter: μ Estimator: Y
Computations:
Method of Determination (Treatment/sample)
Specimen (Block) A B C
1 2 1 3
2 2 1 3
3 2 1 3
4 2 1 3
5 2 1 3
6 3 1 2
7 2.5 1 2.5
8 2 1 3
9 2 1 3
R1 = 19.5 R2 = 9 R3 = 25.5
Study Objective:
To compare three methods of determining serum amylase values.
8
A Hypothesis Test for a Location Parameters: a Paired Samples
Hypothesis: H0 : μ1 ... μ a vs H1 : At least one μ i μ i'
(Methods do not differ)
Test Statistic:
a
12
T
ab a 1 i 1
R i2 3b a 1
12
19.5 92 25.5 3 9 3 1 15.5
2 2
3 9 4
T ~ Friedman (a = 3, n = 9)
Rejection Region:
Reject H0 if T Wα, a, n W0.05, 3, 9 6.222
Conclusion:
T = 15.5 > W0.05, 3, 9 6.222 . H0 is rejected at α 0.05 . There is a difference in mean serum
amylase values among the three methods.
Nonparametric Test for Correlation
In correlation, we want to measure the degree or strength of the relationship between variables.
The nonparametric correlation is used to measure relationship, if any, exists between data from
populations with unknown distributions.
The Spearman rank correlation coefficient was introduced by Spearman in 1904.
a. Assumptions
1. The data consist of a random sample of n pairs:
X1, Y1 , X 2 , Y2 , ..., X n ,Yn is a random sample of n pairs.
b. Hypothesis of Interest
H0 : There is no correlation between X and Y
vs (A) H1 : There is an inverse relationship (negative correlation) between X and Y
(B) H1 : There is a direct relationship (positive correlation) between X and Y
9
(C) H1 : X and Y are either directly or inversely related
Construction of a Test Statistic
1. Each X is ranked relative to all other observed values of X from smallest to largest.
The rank of the ith value of X is denoted by: R Xi
2. Each Y is ranked relative to all other observed values of Y from smallest to largest.
The rank of the ith value of Y is denoted by: R Yi
3. Test Statistic:
n n
6 R X i R Yi 6 d i2
2
r 1 i 1
1 i 1
, r ~ Spearman(n)
n n 1
2
n n 1
2
Note:
1. r is the measure of association
2. 1 r 1
r = 1 indicates a perfect direct relationship and r = 1 indicates a perfect inverse relationship.
Hypothesis Tests for Correlation
(A) Hypothesis: H0 : There is no correlation between X and Y
vs H1 : There is an inverse relationship between X and Y
Test Statistic:
n n
6 R X i R Yi 6 d i2
2
r 1 i 1
1 i 1
, r ~ Spearman n
n n 1
2
n n 1
2
Rejection Region:
Reject H0 if r Wα, n
α P r Wα, n P Spearman n Wα, n
(B) Hypothesis: H0 : There is no correlation between X and Y
vs H1 : There is a direct relationship between X and Y
10
Test Statistic:
n n
6 R X i R Yi 6 d i2
2
r 1 i 1
1 i 1
, r ~ Spearman n
n n 2 1 n n 2 1
Rejection Region:
Reject H0 if r Wα, n
α P r Wα, n P Spearman n Wα, n
(C) Hypothesis: H0 : There is no correlation between X and Y
vs H1 : X and Y are either directly or inversely related
Test Statistic:
n n
6 R X i R Yi 6 d i2
2
r 1 i 1
1 i 1
, r ~ Spearman n
n n 2 1 n n 2 1
Rejection Region:
Reject H0 if r Wα or r Wα
,n ,n
2 2
Example:
A researcher noted a marked inter-observer variation in blood pressure readings. They found that
doctors who read high on systolic tended to read high on diastolic. The data shows mean systolic
and diastolic blood pressure readings by 14 doctors.
Doctor 1 2 3 4 5 6 7
Systolic 141.8 140.2 131.8 132.5 135.7 141.2 143.9
Diastolic 89.7 74.4 83.5 77.8 85.8 86.5 89.4
Doctor 8 9 10 11 12 13 14
Systolic 140.2 140.8 131.7 130.8 135.6 143.6 133.2
Diastolic 89.3 88 82.2 84.6 84.4 86.3 85.9
Is there a direct relationship between systolic and diastolic readings by doctors? Use = 0.05.
Computations:
11
Systolic (Xi) R (Xi) Diastolic (Yi) R (Yi) di = R (Xi) ̶ R (Yi) d i2
141.8 12 89.7 14 -2 4
140.2 8.5 74.4 1 7.5 56.25
131.8 3 83.5 4 -1 1
132.5 4 77.8 2 2 4
135.7 7 85.8 7 0 0
141.2 11 86.5 10 1 1
143.9 14 89.4 13 1 1
140.2 8.5 89.3 12 -3.5 12.25
140.8 10 88 11 -1 1
131.7 2 82.2 3 -1 1
130.8 1 84.6 6 -5 25
135.6 6 84.4 5 1 1
143.6 13 86.3 9 4 16
133.2 5 85.9 8 -3 9
d 2
i 132.5
A Hypothesis Test for Correlation
Hypothesis:
H0 : Systolic and diastolic blood pressure readings by doctors are uncorrelated
vs H1 : There is a direct relationship between systolic and diastolic blood pressure readings
Test Statistic:
n n
6 R X i R Yi 6 d i2
2
r 1 i 1
1 i 1
n n 1
2
n n 1
2
6 132.5
1 1 0.29 0.71
14 14 1
2
Rejection Region:
Reject H0 if r Wα, n W0.05, 14 0.4575 (Refer to Table 13 of the Cambridge Table)
Conclusion:
r 0.71 W0.05, 14 0.4575. H0 is rejected at α 0.05. Doctors who read high on systolic tend to
read high on diastolic blood pressures.
12