Seite - 336 - in Differential Geometrical Theory of Statistics
Bild der Seite - 336 -
Text der Seite - 336 -
Entropy2016,18, 421
0 50 100 150 200
Index
Figure6.Exponentiallydecreasingvaluesofπi.
4.3.Variation inFiniteSampleAdequacyofAsymptoticDistributionsacross theParameterSpace
Pearson’sχ2 statistic (α=3) iswidelyusedtotest independenceincontingencytables,astandard
rule-of-thumbforitsvaliditybeingthateachexpectedcellfrequencyshouldbeatleast5. Forillustrative
purposes,weconsider 2×2 contingency tables, the relevantN-asymptoticnulldistributionbeing
χ21. We assess the adequacy of this asymptotic approximation by comparing nominal and actual
significance levelsof this test,basedon10,000replications. Particular interest lies inhowtheseactual
levelsvaryacrossdifferentdatagenerationprocesseswithinthesamenullhypothesisof independence.
Figures7and8showtheactual levelof thePearsonχ2 test fornominal levels0.1and0.05 for
samplesizesN=20andN=50,withπr andπcdenotingrowandcolumnprobabilities, respectively.
Theabovegeneral ruleappliesonlyat thecentralblackdot inFigure7,and inside theclosedblack
curvedregion inFigure8. Theactual levelwascomputedforallpairsofvaluesofπr andπc, averaged
using thesymmetryof theparameterspace,andsmoothedusingthekernel smoother for irregular2D
data (implemented in thepackagefields inR). Ineachcase, thewhite tonecontains thenominal level,
while redtonescorrespondto liberalandblue tones toconservativeactual levels.
Thefinitesampleadequacyof this standardasymptotic test clearlyvariesacross theparameter
space. Inparticular, itsnominalandactual levelsagreewellatsomeparametervaluesoutsidethestandard
rule-of-thumbregion;and,conversely,disagreesomewhatatotherparametervalues insideit. Intriguingly,
theagreementbetweennominalandactual levelsdoesnot improveeverywherewithsamplesize.Overall,
theclearpatternsevident inthis lackofuniformityinvitefurthertheoretical investigation.
0.0 0.2 0.4 0.6 0.8 1.0
N = 20, nominal level = 0.1
πc 0.00
0.05
0.10
0.15
0.20
●
0.0 0.2 0.4 0.6 0.8 1.0
N = 20, nominal level = 0.05
πc 0.00
0.02
0.04
0.06
0.08
0.10
●
Figure7.Heatmapof theactual levelof the test forN=20atnominal levels0.1and0.05; thestandard
rule-of-thumb(whereexpectedcountsaregreater than5)appliesonlyat theblackdot.
336
Differential Geometrical Theory of Statistics
- Titel
- Differential Geometrical Theory of Statistics
- Autoren
- Frédéric Barbaresco
- Frank Nielsen
- Herausgeber
- MDPI
- Ort
- Basel
- Datum
- 2017
- Sprache
- englisch
- Lizenz
- CC BY-NC-ND 4.0
- ISBN
- 978-3-03842-425-3
- Abmessungen
- 17.0 x 24.4 cm
- Seiten
- 476
- Schlagwörter
- Entropy, Coding Theory, Maximum entropy, Information geometry, Computational Information Geometry, Hessian Geometry, Divergence Geometry, Information topology, Cohomology, Shape Space, Statistical physics, Thermodynamics
- Kategorien
- Naturwissenschaften Physik