Environmental Engineering Reference
In-Depth Information
Grubbs' test
This test is recommended by the U.S. EPA (1992). First arrange the data in an
increasing order (x
1
;
x
n
), the lowest (x
1
) and the highest (x
n
) data points can be
tested for outlier by the following t-statistic:
x
2
; ...
T ¼
x
x
1
s
x
n
x
or
T ¼
ð2
:
25Þ
s
The T value is then compared with a critical T value (Table 2.3) at the specified
sample size and selected a (normally 5%). The data point is deemed to be an outlier
if T
<
T
critical
. The test can be continued to test for further possible outlier(s)
(i.e., x
2
;
). EPA also recommends the use of Grubbs' test for log-normal data
after the log transformation.
x
n1
; ...
Table 2.3 Critical values used for the Grubbs' test
Critical T
Critical T
Degree of
Degree of
freedom (n1)
T
0.05
T
0.01
freedom (n1)
T
0.05
T
0.01
3
1.153
1.155
13
2.331
2.607
4
1.463
1.492
14
2.371
2.659
5
1.672
1.749
15
2.409
2.705
6
1.822
1.944
16
2.443
2.747
7
1.938
2.097
17
2.475
2.785
8
2.032
2.221
18
2.504
2.821
9
2.110
2.323
19
2.532
2.854
10
2.176
2.410
20
2.557
2.884
11
2.234
2.485
30
3.103
2.745
12
2.285
2.550
50
2.956
3.336
Dixon's test
This test is used for a small number of outliers with a sample size between 3 and 25
(Gibbons, 1994). First arrange data in an increasing order, x
1
<
x
n
,
then calculate the D value based on the number of observations (Table 2.4). An
x
2
<
x
3
< ...<
Table 2.4 Formula to calculate D values for the Dixon's test
n
If x
n
is suspected
If x
1
is suspected
x
n
x
n1
x
n
x
1
x
2
x
1
x
n
x
1
37
D
10
¼
D
10
¼
x
n
x
n1
x
n
x
2
x
2
x
1
x
n1
x
1
810
D
11
¼
D
11
¼
x
3
x
1
x
n1
x
1
x
n
x
n
2
x
n
x
2
1113
D
21
¼
D
21
¼
Search WWH ::
Custom Search