Soit et quatre variables aléatoires telles que , où sont des paramètres inconnus. Supposons également que ,Alors lequel est vrai?Y 1 , Y 2 , Y 3
Y1,Y2,Y3 Y 4Y4 E ( Y 1 ) = θ 1 - θ 3 ; E ( Y 2 ) = θ 1 + θ 2 - θ 3 ; E ( Y 3 ) = θ 1 - θ 3 ; E ( Y 4 ) = θ 1 - θ 2 - θ 3E(Y1)=θ1−θ3; E(Y2)=θ1+θ2−θ3; E(Y3)=θ1−θ3; E(Y4)=θ1−θ2−θ3 θ 1 , θ 2 , θ 3θ1,θ2,θ3 V a r ( Y i ) = σ 2Var(Yi)=σ2 i = 1 , 2 , 3 , 4.i=1,2,3,4. A. θ 1 , θ 2 , θ 3
θ1,θ2,θ3 sont estimables.B. θ 1 + θ 3
θ1+θ3 est estimable.C. est estimable et \ dfrac {1} {2} (Y_1 + Y_3) est la meilleure estimation linéaire sans biais de \ theta_1- \ theta_3 .θ 1 - θ 3
θ1−θ3 12 (Y1+Y3)12(Y1+Y3) θ1-θ3θ1−θ3 D. θ 2
θ2 est estimable.
La réponse est donnée est C qui me semble étrange (parce que j'ai eu D).
Pourquoi j'ai eu D? Depuis, E(Y2−Y4)=2θ2
Pourquoi je ne comprends pas que C pourrait être une réponse? D'accord, je vois que Y1+Y2+Y3+Y44
Veuillez me dire où je me trompe.
Également publié ici: /math/2568894/a-problem-on-estimability-of-parameters
la source
self-study
tag ou quelqu'un viendra et fermera votre question.Réponses:
Cette réponse met l'accent sur la vérification de l'estimabilité. La propriété de variance minimale est de ma considération secondaire.
Pour commencer, résumez les informations en termes de forme matricielle d'un modèle linéaire comme suit: Y : = [ Y 1 Y 2 Y 3 Y 4 ] = [ 1 0 - 1 1 1 - 1 1 0 - 1 1 - 1 - 1 ] [ θ 1 θ 2 θ 3 ] + [ ε 1 ε 2 ε 3 ε 4 ] : =X β + ε ,
Si la matrice du modèle XX est de plein rang, le paramètre original ββ admet une des moindres carrés uniques estimer β = ( X ' X ) - 1 X ' Y . En conséquence, tout paramètre φ , défini comme une fonction linéaire φ ( β ) de β est estimable dans le sens où elle peut être de manière non ambiguë estimée par les données via les moindres carrés estimer β en tant que φ = p ' β .β^=(X′X)−1X′Y ϕ ϕ(β) β β^ ϕ^=p′β^
La subtilité apparaît lorsque XX n'est pas de plein rang. Pour avoir une discussion approfondie, nous fixons d'abord quelques notations et termes ci-dessous (je respecte la convention de l'approche sans coordonnées des modèles linéaires , section 4.8. Certains des termes semblent inutilement techniques). De plus, la discussion s'applique au modèle linéaire général Y = X β + εY=Xβ+ε avec X ∈ R n × kX∈Rn×k et β ∈ R kβ∈Rk .
Comme mentionné ci-dessus, lorsque le rang ( X ) < k , toutes les fonctions paramétriques ϕ ( β ) ne sont pas estimables. Mais, attendez, quelle est la définition du terme estimable techniquement? Il semble difficile de donner une définition claire sans déranger un peu l'algèbre linéaire. Une définition, qui je pense est la plus intuitive, est la suivante (à partir de la même référence susmentionnée):rank(X)<k ϕ(β)
Interprétation. La définition ci-dessus stipule que le mappage de la variété de régression M à l'espace de paramètres de ϕ doit être un à un, ce qui est garanti lorsque rang ( X ) = k (c'est -à- dire lorsque X lui-même est un à un). Lorsque rang ( X ) < k , on sait qu'il existe β 1 ≠ β 2 tel que X β 1 = X β 2M ϕ rank(X)=k X rank(X)<k β1≠β2 Xβ1=Xβ2 . La définition estimable ci-dessus exclut en effet les fonctionnelles paramétriques structurellement déficientes qui entraînent elles-mêmes des valeurs différentes même avec la même valeur sur M , ce qui n'a pas de sens naturellement. En revanche, une fonction paramétrique estimable ϕ ( ⋅ ) permet le cas ϕ ( β 1 ) = ϕ ( β 2 ) avec β 1 ≠ β 2 , tant que la condition X β 1 = X β 2 est remplie.M ϕ(⋅) ϕ(β1)=ϕ(β2) β1≠β2 Xβ1=Xβ2
There are other equivalent conditions to check the estimability of a parametric functional given in the same reference, Proposition 8.4.
After such a verbose background introduction, let's come back to your question.
A. ββ itself is non-estimable for the reason that rank(X)<3rank(X)<3 , which entails Xβ1=Xβ2Xβ1=Xβ2 with β1≠β2β1≠β2 . Although the above definition is given for scalar functionals, it is easily generalized to vector-valued functionals.
B. ϕ1(β)=θ1+θ3=(1,0,1)′βϕ1(β)=θ1+θ3=(1,0,1)′β is non-estimable. To wit, consider β1=(0,1,0)′β1=(0,1,0)′ and β2=(1,1,1)′β2=(1,1,1)′ , which gives Xβ1=Xβ2Xβ1=Xβ2 but ϕ1(β1)=0+0=0≠ϕ1(β2)=1+1=2ϕ1(β1)=0+0=0≠ϕ1(β2)=1+1=2 .
C. ϕ2(β)=θ1−θ3=(1,0,−1)′βϕ2(β)=θ1−θ3=(1,0,−1)′β is estimable. Because Xβ1=Xβ2Xβ1=Xβ2 trivially implies θ(1)1−θ(1)3=θ(2)1−θ(2)3θ(1)1−θ(1)3=θ(2)1−θ(2)3 , i.e., ϕ2(β1)=ϕ2(β2)ϕ2(β1)=ϕ2(β2) .
D. ϕ3(β)=θ2=(0,1,0)′βϕ3(β)=θ2=(0,1,0)′β is also estimable. The derivation from Xβ1=Xβ2Xβ1=Xβ2 to ϕ3(β1)=ϕ3(β2)ϕ3(β1)=ϕ3(β2) is also trivial.
After the estimability is verified, there is a theorem (Proposition 8.16, same reference) claims the Gauss-Markov property of ϕ(β)ϕ(β) . Based on that theorem, the second part of option C is incorrect. The best linear unbiased estimate is ˉY=(Y1+Y2+Y3+Y4)/4Y¯=(Y1+Y2+Y3+Y4)/4 , by the theorem below.
The proof goes as follows:
Therefore, option D is the only correct answer.
Addendum: The connection of estimability and identifiability
When I was at school, a professor briefly mentioned that the estimability of the parametric functional ϕϕ corresponds to the model identifiability. I took this claim for granted then. However, the equivalance needs to be spelled out more explicitly.
According to A.C. Davison's monograph Statistical Models p.144,
For linear model (1)(1) , regardless the spherity condition Var(ε)=σ2IVar(ε)=σ2I , it can be reformulated as
E[Y]=Xβ,β∈Rk.
It is such a simple model that we only specified the first moment form of the response vector YY . When rank(X)=krank(X)=k , model (2)(2) is identifiable since β1≠β2β1≠β2 implies Xβ1≠Xβ2Xβ1≠Xβ2 (the word "distribution" in the original definition, naturally reduces to
"mean" under model (2)(2) .).
Now suppose that rank(X)<krank(X)<k and a given parametric functional ϕ(β)=p′βϕ(β)=p′β , how do we reconcile Definition 1 and Definition 2?
Well, by manipulating notations and words, we can show that (the "proof" is rather trivial) the estimability of ϕ(β)ϕ(β) is equivalent to that the model (2)(2) is identifiable when it is parametrized with parameter ϕ=ϕ(β)=p′βϕ=ϕ(β)=p′β (the design matrix XX is likely to change accordingly). To prove, suppose ϕ(β)ϕ(β) is estimable so that Xβ1=Xβ2Xβ1=Xβ2 implies p′β1=p′β2p′β1=p′β2 , by definition, this is ϕ1=ϕ2ϕ1=ϕ2 , hence model (3)(3) is identifiable when indexing with ϕϕ . Conversely, suppose model (3)(3) is identifiable so that Xβ1=Xβ2Xβ1=Xβ2 implies ϕ1=ϕ2ϕ1=ϕ2 , which is trivially ϕ1(β)=ϕ2(β)ϕ1(β)=ϕ2(β) .
Intuitively, when XX is reduced-ranked, the model with ββ is parameter redundant (too many parameters) hence a non-redundant lower-dimensional reparametrization (which could consist of a collection of linear functionals) is possible. When is such new representation possible? The key is estimability.
To illustrate the above statements, let's reconsider your example. We have verified parametric functionals ϕ2(β)=θ1−θ3ϕ2(β)=θ1−θ3 and ϕ3(β)=θ2ϕ3(β)=θ2 are estimable. Therefore, we can rewrite the model (1)(1) in terms of the reparametrized parameter (ϕ2,ϕ3)′(ϕ2,ϕ3)′ as follows
E[Y]=[1011101−1][ϕ2ϕ3]=˜Xγ.
Clearly, since ˜XX~ is full-ranked, the model with the new parameter γγ is identifiable.
la source
Apply the definitions.
I will provide details to demonstrate how you can use elementary techniques: you don't need to know any special theorems about estimation, nor will it be necessary to assume anything about the (marginal) distributions of the Yi. We will need to supply one missing assumption about the moments of their joint distribution.
Definitions
All linear estimates are of the form tλ(Y)=4∑i=1λiYi
An estimator of θ1−θ3 is unbiased if and only if its expectation is θ1−θ3. By linearity of expectation,
θ1−θ3=E[tλ(Y)]=4∑i=1λiE[Yi]=λ1(θ1−θ3)+λ2(θ1+θ2−θ3)+λ3(θ1−θ3)+λ4(θ1−θ2−θ3)=(λ1+λ2+λ3+λ4)(θ1−θ3)+(λ2−λ4)θ2.
Comparing coefficients of the unknown quantities θi reveals λ2−λ4=0 and λ1+λ2+λ3+λ4=1.
In the context of linear unbiased estimation, "best" always means with least variance. The variance of tλ is
Var(tλ)=4∑i=1λ2iVar(Yi)+4∑i≠jλiλjCov(Yi,Yj).
The only way to make progress is to add an assumption about the covariances: most likely, the question intended to stipulate they are all zero. (This does not imply the Yi are independent. Furthermore, the problem can be solved by making any assumption that stipulates those covariances up to a common multiplicative constant. The solution depends on the covariance structure.)
Since Var(Yi)=σ2, we obtain
Var(tλ)=σ2(λ21+λ22+λ23+λ24).
The problem therefore is to minimize (2) subject to constraints (1).
Solution
The constraints (1) permit us to express all the λi in terms of just two linear combinations of them. Let u=λ1−λ3 and v=λ1+λ3 (which are linearly independent). These determine λ1 and λ3 while the constraints determine λ2 and λ4. All we have to do is minimize (2), which can be written
σ2(λ21+λ22+λ23+λ24)=σ24(2u2+(2v−1)2+1).
No constraints apply to (u,v). Assume σ2≠0 (so that the variables aren't just constants). Since u2 and (2v−1)2 are smallest only when u=2v−1=0, it is now obvious that the unique solution is
λ=(λ1,λ2,λ3,λ4)=(1/4,1/4,1/4,1/4).
Option (C) is false because it does not give the best unbiased linear estimator. Option (D), although it doesn't give full information, nevertheless is correct, because
θ2=E[t(0,1/2,0,−1/2)(Y)]
is the expectation of a linear estimator.
It is easy to see that neither (A) nor (B) can be correct, because the space of expectations of linear estimators is generated by {θ2,θ1−θ3} and none of θ1,θ3, or θ1+θ3 are in that space.
Consequently (D) is the unique correct answer.
la source