Lecture 14 Non-parametric hypothesis testing H igh M edium Low P ristine 33 51 6 25 34 43 28 27 32 75 22 38 47 19 29 60 21 49 46 64 31 30 25 31 25 34 93 24 28 57 T-TE S T Medium Low P ristine High 0.145265 0.172254 0.931288 Medium 1 0.081749 Low 0.211812 H igh Medium Low P ristine 14 6 30 23 12 10 20 22 15 2 27 11 8 29 19 4 28 7 9 3 16 18 23 16 23 12 1 26 20 5 T-TES T Medium Low P ristine High 0.121074 0.414821 0.09406 Medium 0.5811 0.020992 Low 0.103828 The ranking of data The ranking of data eliminates outliers and non- linearities. In most cases it reduces within group variances. All parametric tests can be applied to ranked data! Spiders on Mazurian lake islands Disturbance
Lecture 14 Non-parametric hypothesis testing. Spiders on Mazurian lake islands. The ranking of data. Disturbance. The ranking of data eliminates outliers and non-linearities . In most cases it reduces within group variances . All parametric tests can be applied to ranked data!. - PowerPoint PPT Presentation
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Lecture 14Non-parametric hypothesis testing
High Medium Low Pristine33 51 6 2534 43 28 2732 75 2238 47 1929 60 21
49 4664 31
302531253493242857
T-TESTMedium Low Pristine
High 0.145265 0.172254 0.931288Medium 1 0.081749Low 0.211812
High Medium Low Pristine14 6 30 2312 10 20 2215 2 2711 8 2919 4 28
7 93 16
1823162312126205
T-TESTMedium Low Pristine
High 0.121074 0.414821 0.09406Medium 0.5811 0.020992Low 0.103828
The ranking of dataThe ranking of data eliminates outliers and non-linearities.In most cases it reduces within group variances.
All parametric tests can be applied to ranked data!
Spiders on Mazurian lake islands
Disturbance
Effects of ranking
High Medium Low Pristine High Medium Low Pristine33 51 6 25 14 6 30 2334 43 28 27 12 10 20 2232 75 22 15 2 2738 47 19 11 8 2929 60 21 19 4 28
Z -1.67038 =(EB21-EB22)/EB22P(Z) 0.047422 =ROZKŁAD.NORMALNY(EB22;EB21;EB23;PRAWDA)Double sided 0.094844
As in the case of the t-test does the ranked ANOVA result in lower significance levels. Ranking levels off the within group heterogeneity (lower within group variance). The test
is less conservative.
Raw data Ranked data
Kruskal-Wallis test or Kruskal-Wallis one way ANOVA by ranks
2
1
12 3( 1)( 1)
ri
i i
RKW N
N N n
KW is approximately χ2 distributed. Values can be taken from a c2 table with r-1
degrees of freedom
Raw data ANOVA
High Medium Low Pristine High Medium Low Pristine33 51 6 25 14 6 30 2334 43 28 27 12 10 20 2232 75 22 15 2 2738 47 19 11 8 2929 60 21 19 4 28
49 46 7 964 31 3 16
30 1825 2331 1625 2334 1293 124 2628 2057 5
Raw data Ranked data
ni 5 2 7 16Ri 71 16 74 298
Ri2 / ni 1008.2 128 782.2857 5550.25KW 3.370783
Chi2(KW;3) 0.337912
The ANOVA gave the more conservative result
Random skewers
Diversity of ground beetles along an elevational gradient