Sunteți pe pagina 1din 14

Dixons Criterion for the

Detection of Outlying
Observation
Outlie
rs

Dirty Data
Wealth of Bill
Gette
Discordant
observations
Extreme points

3/9/17 2
Statistical Definition of
Outlier

According to Barnett & Lewis:

An outlier in a set of data is an


observation that appears to be
inconsistent with the remainder of
the data set.

3/9/17 3
Main Objective
Application of Dixons range-ratio test
to the Uniform Distributions.

3/9/17 4
Dixons Test Statistic

Dixons idea, proposed in 1950.


Detects single extreme point
(outlier).
Data i.i.d.
Arranged sample.
Test Statistic: xn xn1
R10
xn x1
3/9/17 5
Decision Making
Hypothesis,
Calculation,
Critical Values,

Conclusion,

xn is an outlier, if :

xn is not an outlier, if :

3/9/17 6
The Uniform distribution
The general Uniform case;

The standard Uniform case with a =


0&b=1

3/9/17 7
Suppose X be continuous random
variable, with
realizations :
arrangement :

The joint density for ,

Test statistic , the density function


for

and will be :
3/9/17 8
f (x1, xn1, xn )
n!
xn1 n3
f (x1 )dx1 f (t)dt f (xn1 )dxn1 f (xn )dxn
(n 3)! x1

Substitutions:

1
R 1 n2

3/9/17 9
Critical values
n/ 0.005 0.01 0.02 0.05 0.1 0.5

3 0.9950 0.9900 0.9800 0.9500 0.9000 0.5000


4 0.9293 0.9000 0.8586 0.7763 0.6838 0.2929
5 0.8290 0.7845 0.7285 0.6316 0.5358 0.2063
6 0.7340 0.6837 0.6239 0.5271 0.4377 0.1591
7 0.6534 0.6019 0.5427 0.4507 0.3690 0.1295
8 0.5865 0.5358 0.4789 0.3930 0.3187 0.1091
9 0.5309 0.4821 0.4281 0.3481 0.2803 0.0942
10 0.4843 0.4377 0.3867 0.3123 0.2501 0.0829
11 0.4449 0.4005 0.3525 0.2831 0.2257 0.0741
12 0.4113 0.3690 0.3238 0.2589 0.2056 0.0669
13 0.3822 0.3421 0.2993 0.2384 0.1889 0.0611
14 0.3569 0.3187 0.2781 0.2209 0.1745 0.0561
15 0.3347 0.2983 0.2587 0.2058 0.1623 0.0519
3/9/17 10
Example
xi 0.1103, 0.3876, 0.5001, 0.1028, 0.8823, 0.9184,
0.3105, 0.2153, 0.6019, 9.5
H0 ; x10 is an outlier,
H1 ; x10 is not an outlier.
Test Statistic:
xn xn1
r 0.9132
xn x1
0.05
Level of significance:
Decision: The 10th observation is an outlier since
r r10(1 ) 0.3123

3/9/17 11
General Uniform Case

???
The General Uniform Distribution
Parameter Estimation

Maximum Likelihood Estimation Method:

Method of Moments:

3/9/17 13
Refernces

S-ar putea să vă placă și