Seite - 337 - in Differential Geometrical Theory of Statistics
Bild der Seite - 337 -
Text der Seite - 337 -
Entropy2016,18, 421
0.0 0.2 0.4 0.6 0.8 1.0
N = 50, nominal level = 0.1
Ïc 0.00
0.05
0.10
0.15
0.20
0.0 0.2 0.4 0.6 0.8 1.0
N = 50, nominal level = 0.05
Ïc 0.00
0.02
0.04
0.06
0.08
0.10
Figure8.Heatmapof theactual levelof the test forN=50atnominal levels0.1and0.05; thestandard
rule-of-thumb(whereexpectedcountsaregreater than5)applies inside theclosedblackcurvedregion.
5.Discussion
Thispaperhas illustrated thekey importanceofworkingwith theboundaryof theclosureof
exponential familieswhenstudyinggoodness-of-ïŹt testing in thehighdimensional, lowsamplesize
context. Someof thiswork isnew(Section2),whilesomeuses thestructureofextendedexponential
families toaddinsight tostandardresults in the literature (Section3). The last section,Section4,uses
simulationstudies tostart toexploreopenquestions in thisarea.
Oneopenquestionârelatedto theresultsofTheorems1and2âis tosee ifauniïŹedtheory, forall
valuesofα, andover largeclassesofextendedexponential families, canbedeveloped.
Acknowledgments:Theauthorswould like to thanktheEPSRCfor thesupportofgrantnumberEP/L010429/1.
GermainVanBeverwouldalso like to thankFRS-FNRSfor its support throughthegrantFC84444.Wewouldalso
like to thankthereferees forveryhelpfulcomments.
AuthorContributions: All fourauthorsmadecritical contributions to thepaper. R.S.madekeycontribution
to,especially,Section4. P.M.andF.C.providedtheoverall structureandkeycontentdetailsof thepaper.G.V.B.
providedinvaluablesuggestions throughout.
ConïŹictsof Interest:TheauthorsdeclarenoconïŹictsof interest.
AppendixA. ProofofTheorem1
Westartbynotinganimportant recurrencerelationwhichwillbeexploited in thecomputations
below.BydeïŹnition, forany t :=(ti)âRk+1,n=(ni)hasmomentgeneratingfunction
M(t;N) :=E{exp(tTn)}=[m(t)]N
withm(t)=âki=0ai and ai= ai(ti)=Ïie ti. Putting
fN,i(t;r) :=N(r) [m(t)] Nâr ari (0†râ€N),
where
N(r) := NPr= {
1 if r=0
N(Nâ1)...(Nâ(râ1)) if râ{1,...,N} ,
wehave
M(t;N)= fN,i(t;0) (0†i†k) (A1)
andtherecurrencerelation:
âfN,i(t;r)
âti = fN,i(t;r+1)+rfN,i(t;r) (0†i†k; 0†r<N) . (A2)
337
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