Seite - 428 - in Differential Geometrical Theory of Statistics
Bild der Seite - 428 -
Text der Seite - 428 -
Entropy2016,18, 375
Then,weobtain the followingresult:
sp
s0 0 λζ ζ
P(Imζ−1Z¯n≥ sp)≤ 38α
P(Reζ−1Z¯n≤−s0)≤ α4 P(Imζ−1Z¯n≤−sp)≤ 38α
Figure2.Theconstructionfor the testof thehypothesisEZ=λζwithλ>0.
sp
Z¯n
δH
0 λζ ζ
Figure 3. The critical Z¯n regarding the rejection of ζ. δH bounds the angle between μˆn and any
acceptedζ.
Proposition1. LetZ1, . . . ,Zn be randomvariables takingvalueson theunit circleS1,α∈ (0,1), and letCH
bedefinedas inEquation (8).
(i) CH is a (1−α)-confidence set for the circular populationmean set. In particular, ifEZ = 0, i.e.,
the circular populationmean set equalsS1, then |Z¯n|> √
2s0 with probability atmost α, so indeed
CH=S1withprobabilityat least1−α.
(ii) s0 andsp areof ordern −12 .
(iii) IfEZ =0, then√nδH→0 inprobabilityand theprobabilityof obtaining the trivial confidence set, i.e.,
P(CH=S1)=P(|Z¯n|≤ √
2s0), goes to0exponentially fast.
Proof. (i)holdsbyconstruction.
For (ii), recall Equation (7), from which we obtain the estimates α4 ≤ exp(−ns20/2) resp.
3
8α≤ exp(−ns2p/2), implying that s0 and sp are of order n− 1
2; the same holds stochastically for
δH since Z¯n → EZ a.s. Regarding the second statement of (iii), if μ is unique, consider ζ =−μ;
then, τ = EX¯n < 0 and − √
2s0 is eventually less than τ2 and also α > 2 −n+2 eventually.
Hence, the probability of obtaining the trivial confidence setCH = S1 is eventually bounded by
P(ζ∈CH)≤P(X¯n>−s0)≤P(X¯n> τ2)=P(X¯n−EX¯n>−τ2)≤ exp(−nτ2/8), andthuswillgo to
zeroexponentially fastasn tends to infinity.
3. EstimatingtheVariance
From the central limit theorem for μˆn in case of unique μ, cf. Equation (4), we see that the
aymptoticvarianceof μˆngetssmall if |EZ| isclose to1(thenEZ isclose totheboundaryS1 of theunit
disc,which ispossibleonly if thedistribution isveryconcentrated)or if thevarianceE(Im(μ−1Z))2 in
thedirectionperpendicular toμ is small (if thedistributionwereconcentratedon±μ, thisvariance
wouldbezeroand μˆnwouldequalμwith largeprobability).WhileδH (|Z¯n|beingthedenominator
428
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