Sie sind auf Seite 1von 24

NON-PARAMETRIC

METHODS
WHEN TO USE NON-PARAMETRIC
PROCEDURES

 Hypothesis does not involve a population parameter.


 Data obtained on a weaker scale of measurement.
 Assumptions underlying the use of parametric procedure are
not met.
 Results are needed in a hurry and calculations are to be done
by hand.
 Do the data pass the test of normality (Shape of histogram). If
so use parametric, If, not go for non-parametric.
NON-PARAMETRIC TESTS ARE CONCERNED
IN TESTING THE HYPOTHESIS ABOUT
POPULATION FREQUENCY DISTRIBUTIONS
True state of affairs
Non-parametric Parametric
appropriate appropriate
Use non- Right choice Correct level
parametric of significance
Decision but a slight
Of loss of power
statisticia Use Unknown level Right choice
n parametric of significance
possibly too
large
ADVANTAGES AND
DISADVANTAGES
 Less chance for improper application due to
minimum assumptions
 Useful in case of weak measurement scale like
count or rank data.
 Require less calculations .
 Require minimum level of mathematics and
statistics, therefore can be more appreciated by
biologists and social scientists.             
ADVANTAGES AND
DISADVANTAGES
 If the investigator is interested beyond the null
hypothesis then non parametric methods are lacking.
 These does not extract all the information out of the
data. Efforts through transformation are needed to
enable to use parametric procedure.
 Truly non-parametric procedures ( goodness of fit,
test of randomness)
Distribution Free Procedures

 - The Sign test


 - The Wilcoxon’s tests
 - The Mann-Whitney test
 - The Kolmogorov-Smirnov tests
 - The Kruskal-Wallace ANOVA test
 - The Friedman Two Way ANOVA test
SIGN-TEST
ASSUMPTIONS
 Samples are randomly drawn with unknown median
 Measurement scale is at least ordinal
 The variable is continuous.

PROCEDURE
 Record the sign of the difference obtained by
subtracting hypothesized median X- Mo
 Use binomial probability tables to make the
decision on specified level of significance
Sign Test (Example)
Eight judges ranked the flavor of ground
beef patties stored for eight months in two
home freezers. One freezer was set at 0 0F;
the second freezer’s temperature fluctuated
between 00F and 150F. The results are
shown in the table given below:
RANKINGS OF FLAVOR OF GROUND BEEF PATTIES BY 8 JUDGES
(rank 1 is high; rank 2, low)
Judge A B C D E F G H
00F 1 1 2 1 1 1 1 1
00F to 150F 2 2 1 2 2 2 2 2
+ or - + + - + + + + +
In the sample, 7 of 8 judges preferred the 0 0F
patty. If r of n pairs show one sign, then in
testing the hypothesis H0 that half the population
pairs have this sign. The approximate normal
deviate Zc (corrected for continuity) is as
follows:

Zc  ( 2 r  n  1) / n  1.77
At 5% level of significance we are unable to reject
the hypothesis. Sign tables can also be used.
WILCOXON-RANK –SUM TEST
 Hypothesis: Both populations are identical
 Two independent random samples
 Rank the combined sample
 Find rank sum of say first sample
 A x B *
 x*x**x*xxx
 Ranks are all mixed (Indicate Ho)
 Conclusion: Sample A contains more larger ranks
than B indicates negation of Ho
Signed-rank (Wilcoxon) Test
(Example)
One member of a pair of corn seedlings
was treated by a small electric current; the
other was untreated. After a period of
growth, the differences in elongation
(treated-untreated) are shown for each of
10 pairs.
Pair Difference Signed Rank
(mm)
1 6.0 5
2 1.3 1
3 10.2 7
4 23.9 10
5 3.1 3
6 6.8 6
7 -1.5 -2
8 -14.7 -9
9 -3.3 -4
10 11.1 8
Z =(μ-T-0.5)/σ
c

where T is the small rank sum,


and  =n(n+1)/4 and  = (2n+1) /6
In this example, T=15 and n=10
so  =10  (10+1)/4=27.50 and
 = (2  10+1)(27.50)/6  9.81
Thus Z =(27.5-15-0.5)/9.81 =1.22
c
Mann-Whitney Test
It is well established that in an attacked
field more eggs are deposited on tall plants
than on short ones. The records of numbers
of eggs found in 20 plants in a rather
uniform field are given in the following
table. The plants were in two randomly
selected sites, 10 plants each.
Height of Plant Number of Eggs
Less than 23 in. 0 14 18 0 31 0 0 0 11 0
More than 23 in. 37 42 12 32 105 84 15 47 51 65

Z =( μ-T -0.5)/σ
c

where T is the small rank sum,


and  = n 1(n 1+n 2+1)/2 and  = n  /6
2
Count 0 0 0 0 0 0 11 12 14 15
Rank 3.5 3.5 3.5 3.5 3.5 3.5 7 8 9 10
Count 18 31 32 37 42 47 51 65 84 105
Rank 11 12 13 14 15 16 17 18 19 20

T=3.5+3.5+3.5+3.5+3.5+3.5+7+9+11+12=60
 =10  (10+10+1)/2 =105 and  = 10 105/6  13.23
Z c
 ( 105  60  0.5) /13.23  3.36
Non-parametric ANOVAS
When one or several of the assumptions of
analysis of variance are violated one may
use non-parametric methods
Kruskal-Wallis :This is for CR design
Hypothesis:Treatment medians are identical
FRIEDMAN TEST: This is for RCBD
Hypothesis: Population distributions for k treatments
are identical
Kruskal-Wallis “AOV” by
ranks
-  Hypothesis: Several independent samples belong to
identical populations.
-     Rank the observations from lowest to highest ( For ties
give average value).
-      Sum the ranks.
-      Use the following test statistics
 2 2 2

H=
12
 R 1
+R 2
+ + R n
 -3(n+1)
n(n+1) 
 n
1 n 2 n n

R1 = Sum of ranks assigned to n1 observations on the 1st trt


R2 = Sum of ranks assigned to n2 observations on the 2nd trt
Rk = Sum of ranks assigned to nk observations on the kth trt
Example
 In a feeding experiment of some animals the following
results were obtained, the numbers in the table being
the gains in weight in Kg.
Ration 1 Ration 2 Ration 3
7.0 14.0 8.5
16.0 15.5 16.5
10.5 15.0 9.5
13.5

Is there any evidence of difference in rations?


H0: The rations have the same effect.
H1: At least two rations do not have the same effect.

Ranks of Ratios
1 2 3
1 6 2
9 8 10
4 7 3
5
Total 19 21 15

12  192 212 152 


H=  + +  -3(10+1)=1.0636
10(10+1)  4 3 3 

Do not reject H0.


FRIEDMAN TEST
 PROCEDURE
 Rank the treatments in each block from lowest to
highest.
 Obtain the sum of ranks for each treatment.
 Compute the statistic.
12
X 2r = 
bt(t+1) i
ri 2 - 3b(t+1)

 where rij is the rank of the ith treatment; in jth


block.
 ri. is the sum of the ranks of the ith treatment;
2 2
X ~χ
r (t-1)
Example
An experiment was conducted on poultry birds and the weight gain for

.
three blocks and four rations at the end of 10 days was recorded

Rations
A B C D
1 323.4 276.0 375.0 285.6
s
Block

2 344.5 250.5 343.4 224.3


3 313.8 273.9 339.6 248.4

Test whether the effect of rations is same?


H0: The rations have the same effect.
H1: At least two rations do not have the same effect.

Ranks
A B C D
1 3 1 4 2
Block
2 4 2 3 1
3 3 2 4 1
Total 10 5 11 4

12
X 2r = (102  52  112  4 2 )- 3  3(4+1)=7.4
3  4(4+1)

Do not reject H0.

Das könnte Ihnen auch gefallen