SOURCE: IS:8900-1978 Method 1 For Single Outlier, Assume A Set of Following Observations

Download as xlsx, pdf, or txt
Download as xlsx, pdf, or txt
You are on page 1of 11

SOURCE: IS:8900-1978

Example. Method 1 For Single Outlier, Assume a set of following observations:


1 By mean and standard deviation
Values x-¯x (x-¯x)2

44 4.4 19.36
42 2.4 5.76
41.5 1.9 3.61
44.5 4.9 24.01
41 1.4 1.96
46.5 6.9 47.61
39 -0.6 0.36
32 -7.6 57.76
38 -1.6 2.56
27.5 -12.1 146.41
mean 39.6
34.37778
sd= 5.86

xn is the largest doughtful observation

T1= 1.177
2.176

Now, check the Table No. 1 for Tn on the basis of level of significance, whether it is greater than calculated
value or not. If calculated Tn is GREATER than Tabulated Value, then it would be considered as outlier and
rejected, otherwise vice versa.

The calculated Tn i.e. T1 is LESS then the Tn given in table on 5% or 1% level of significance, hence the largest
value, 40 willnot be considered as an outlier and will not be rejected.
2 By ratio of differences between relevant order of observations
Values

65.5 Arrange the observations in Ascending order


65.5
66.5
66.5
67
68.5
68.5
70.5
71
72
First, decide whether the outlier to be considered is smallest or largest value
Then, look out the no. of observations
Now, from Table No. 2 select the formula on the basis of smallest or largest value

For the above set of observations, we have to find out the outlier for largest value
From Table No. 2, we are choosing the ratio r11
The formula is
r11= 0.154

The calculated r11 is LESS then the rn given in table on 5% OR 1% level of significance, hence the larg
value, 92.50 will NOT be considered as an outlier and WILL NOT BE rejected.
er than calculated
red as outlier and

ce, hence the largest


ignificance, hence the largest
d.
Example. Method 2 For Two or more Outliers (AT EITHER END), Assume a set of
following observations:
1 One method which can be used is, the previous method can also be taken
repeatedly for detecting the outlier.
2 By mean and standard deviation

N Values x-¯x (x-¯x)2 N Values x-¯x (x-¯x)2


1 89 -1.8 3.24 1 89 -1.800 3.240
2 90 -0.8 0.64 2 90 -0.800 0.640
3 90.5 -0.3 0.09 3 90.5 -0.300 0.090
4 90.5 -0.3 0.09 4 90.5 -0.300 0.090
5 90.5 -0.3 0.09 5 90.5 -0.300 0.090
6 91 0.2 0.04 6 91 0.200 0.040
7 91 0.2 0.04 7 91 0.200 0.040
8 91.5 0.7 0.49
9 91.5 0.7 0.49
10 92.5 1.7 2.89

MEAN= 90.8

8.100

90.500

where k is no. of suspect observations i.e. 2


here "n-k" will be"(13-2=11)"

4.230

0.522

For this method we have compare the value of Lk from Tabulated Value of Lk given
in Table No.3 , If is LESSER than the Tabulated Value then only it will be considered as
an outlier, otherwise not.

In above example the value of Lk= 0.522 is GREATER than the tabulated one, and
hence there is not enough evidence to suspect two largest observations as outlier.
CLAUSE NO. 5.1
Example
Method 3 For Two or more Outliers (atleast one
outlier at each end), Assume a set of following
observations:
1 Test for two outlier, one outlier at each of two ends 2 Test for more than two
N Values x-¯x (x-¯x)2 N Values

1 30.5 -5.250 27.563 1 9.5


2 31 -4.750 22.563 2 19.5
3 32.5 -3.250 10.563 3 30
4 34 -1.750 3.063 4 31
5 34 -1.750 3.063 5 32
6 34 -1.750 3.063 6 38
7 35.5 -0.250 0.063 7 40
8 37 1.250 1.563 8 42.5
9 39 3.250 10.563 9 43.5
10 50 14.250 203.063 10 45

mean 35.75 mean 33.100


31.681
sd= 5.63
= 19.5
= 3.464
from table, R/s 3.68
Now compare the ratio R/s from the value given in Table 4 on the
basis of level of significance, If this value is GREATER then the
value given in the table, then both values will be considered as an
outlier

In above example, Ratio R/s is GREATER than the value given in


table-4, hence both the values (largest and smallest) would be
considered as an outlier.
36.5 37.0 37.5 38.0 39.0
37
37.5
38
39
39.5
39.5
39.5
41.5
50

30.5 31.0 32.5 34.0 34.0


31
32.5
34
34
34
35.5
37
39
50
Test for more than two outlier, atleast one outlier at each end
Ascending order
x-¯x Z-¯Z (Z-¯Z)2 Z-¯Z (Z-¯Z)2
(x-¯x) EXCLUDING SUSPECT
23.600 z1 -0.7 1.04 1.08 z1 -0.7 0.70 0.49
13.600 z2 -0.7 1.04 1.08 z2 -0.7 0.70 0.49
3.100 z3 -0.2 0.54 0.29 z3 -0.2 0.20 0.04
2.100 z4 -0.2 0.54 0.29 z4 -0.2 0.20 0.04
1.100 z5 0.3 0.04 0.00 z5 0.3 -0.30 0.09
-4.900 z6 0.3 0.04 0.00 z6
-6.900 z7 0.3 0.04 0.00 z7
-9.400 z8 0.8 -0.46 0.21
-10.400 z9 1.7 -1.36 1.85
11.900 z10 1.8 1.46 2.13

2
-0.3 U 5= 1.15

2
0.34 U 10= 6.94

E3= 0.166
TABLE VALUE= 0.037 5%
0.013 1%
Now compare the ratio EK from the value given in Table 5 on
the basis of level of significance, If this value is LESSER then the
value given in the table, then all suspect values will be
considered as an outlier

In above example, E3 is LESSER than the value given in table-5,


hence all the suspect values would be considered as an outlier.
39.5 39.5 39.5 41.5 50.0

34.0 35.5 37.0 39.0 50.0

You might also like