• Nenhum resultado encontrado

.u sp .b r/ ˜ti ag omm

N/A
N/A
Protected

Academic year: 2022

Share ".u sp .b r/ ˜ti ag omm"

Copied!
4
0
0

Texto

(1)

w ww

.ime

.u sp .b r/ ˜ti ag omm

Maximum likelihood estimation in uniform distributions

Tiago M. Magalh˜ aes

Department of Statistics, University of S˜ao Paulo, Brazil

Abstract

We have some difficulties to obtain the maximum likelihood estimador (MLE) in non regularity cases, for example, when the support of an distribution depends on an unknown parameter. We exemplify possible situations for obtaining the MLE through the uniform distribution.

Keywords: Maximum likelihood estimation, non regularity cases, uniform distri- bution

1 Preliminary considerations

IfX has an uniform distribution in the interval (θ1, θ2) (notation: X ∼U(θ1, θ2), your probability density function is given by

f(x) = 1

θ2−θ1 I12)(x), (1)

whereI12)(x) is an indicator function, i.e., I12)(x) =

1, x∈(θ1, θ2) ; 0, otherwise.

2 Examples

2.1 Uniparametric cases

Example 1. Let X1, . . . , Xn be independent random variables, each X`, ` = 1, . . . , n having a density given by (1), where (θ1, θ2) = (−θ, θ). The likelihood function for θ, denoted byL(θ), is given by

L(θ) = 1

n n

Y

`=1

I(−θ,θ)(x`). (2)

Email: tiagomm@ime.usp.br; last modification: July 17, 2014

1

(2)

w ww

.ime

.u sp .b r/ ˜ti ag omm

2 EXAMPLES 2

Note that, theL(θ), in (2), can be write as L(θ) =

1 2θ

n n

Y

`=1

I(0,θ)(|x`|). (3)

Now, note that, the expression (2) has maximum when Qn

`=1I(0,θ)(|x`|) = 1, i.e., if 0<|x1|< θ, . . ., 0<|xn|< θ, if 0<max{|x1|, . . . ,|xn|}< θ. But

I(0,θ)(max{|x1|, . . . ,|xn|}) = I(max{|x1|,...,|xn|},+∞)(θ). Then, the expression (3) can be write as

L(θ) = 1

n

I(max{|x1|,...,|xn|},+∞)(θ). Finally, the θ that maximizes L(θ) is given by

θˆ= max{|x1|, . . . ,|xn|}.

Example 2. Let X1, . . . , Xn ∼ U(δ − θ, δ +θ), where δ is a known parameter, i.e., eachX`, ` = 1, . . . , n have an uniform distribution centered in δ. In this case, note that X`−δ∼ U(−θ, θ), then

θˆ= max{|x1 −δ|, . . . ,|xn−δ|}

is the MLE ofθ.

Example 3. LetX1, . . . , Xnbe independent random variables, eachX` ∼U(θ−δ, θ+δ), whereδ is a known parameter. The likelihood function forθ is given by

L(θ) = 1

n n

Y

`=1

I(θ−δ,θ+δ)(x`). (4)

The expression (4) has maximum whenQn

`=1I(θ−δ,θ+δ)(x`) = 1, i.e., ifθ−δ < x1 < θ+δ, . . ., θ−δ < xn< θ+δ. Then, the expression (4) has maximum when

θ−δ < x(1)< x(n)< θ+δ, (5)

where x(1) = min{x1, . . . , xn} and x(n) = max{x1, . . . , xn}. In Figure 1, we present the area represented by expression (5) .

By (5), we can write (4) as L(θ) =

1 2δ

n

I(θ−δ,x(n)) x(1)

I(θ−δ,θ+δ) x(n)

(6)

= 1

n

I(x(1),θ+δ) x(n)

I(θ−δ,θ+δ) x(1) .

By (5), we also have thatθ < x(1)+δ and θ > x(n)−δ. Finally, the expression (4) can be write as

L(θ) = 1

n

I(x(n)−δ,x(1)) (θ). (7) The θ that maximizes L(θ), in (7), is given by

θˆ∈ x(n)−δ, x(1)+δ , i.e., in this case the MLE ofθ is not unique.

(3)

w ww

.ime

.u sp .b r/ ˜ti ag omm

2 EXAMPLES 3

x(1) x(n)

θ − δ θ + δ

θδθ+δ x(1)=x(n)

Figure 1: Area where Qn

`=1I(θ−δ,θ+δ)(x`) = 1.

2.2 Biparametric case

Example 1. LetX1, . . . , Xn ∼U(θ−ϑ, θ+ϑ), whereθand ϑare unknown parameters.

The likelihood function for (θ, ϑ) is given by L(θ, ϑ) =

1 2ϑ

n n

Y

`=1

I(θ−ϑ,θ+ϑ)(x`). (8)

The expression (8) has maximum when Qn

`=1I(θ−ϑ,θ+ϑ)(x`) = 1, i.e., if θ−ϑ < x1 <

θ+ϑ, . . .,θ−ϑ < xn< θ+ϑ. Then, the expression (8) has maximum when

θ−ϑ < x(1) < x(n) < θ+ϑ. (9)

By inequalities (9), we have the following relations x(n)−ϑ < θ < x(1)

ϑ > x(n)−θ and ϑ > θ−x(1) (10)

From second row of (10),

ϑ >max

θ−x(1), x(n)−θ . (11)

If, in (11), max

θ−x(1), x(n)−θ =x(n)−θ, we have that x(n)−θ > θ−x(1), solving this inequation, we have θ ≤ x(1)+x2 (n). If, in (11), max

θ−x(1), x(n)−θ = θ−x(1),

(4)

w ww

.ime

.u sp .b r/ ˜ti ag omm

2 EXAMPLES 4

θ−x(1) > x(n)−θ, in this case, we haveθ ≥ x(1)+x2 (n). Then, the MLE of θ is θˆ= x(1)+x(n)

2 . (12)

Replacing (12) in (11), we have ϑ >max

x(1)+x(n)

2 −x(1), x(n)− x(1)+x(n) 2

= x(n)−x(1)

2 . (13)

The MLE of ϑ is obtained by the inequation (13). But, note that, in (8), ∀θ fixed, L(θ, ϑ) is maximized by the lower ϑ (see the ratio 1/2ϑ), such that L(θ, ϑ)>0. Then,

ϑˆ= x(n)−x(1) 2 is the MLE ofϑ.

Referências

Documentos relacionados

Assim, educação não é educação da alma, como queria Platão, nem do pensamento que pensa a si mesmo, como propôs Hegel, mas é o processo mesmo de desenvolvimento

Riddley - Notes on the botany of Fernando Noronha.. Desmidiaeeae &#34;in&#34; Symbolae

We have begun to develop an assessment tool based on the stepwise inventory, with evaluation metrics built in for each step. We have maintained a prospective database of RARP

Sub-banda permanece sem coloração e o restante da fita apresenta coloração rosa vivamente colorida, adquirindo tonalidade mais escura na banda.. Não apresenta fluorescência

Finally, in paragraph 5 we give some illustrative results and a sufficient condition (5.4) for the Luikov number Lu in order to obtain when the temperature distribution has a

In the previous part, we have made use of the classical homogenization procedure to obtain separately an elastic effective potential and a viscoplastic effective potential..

Contudo, nessa transgressão sentimos estranhamento pela ameaça do impossível que não sabemos explicar porque não é familiar, em consequência, aparece o medo como outro

Diante desse contexto, o presente trabalho apresenta a proposta de um modelo de reaproveitamento do óleo de cozinha, para a geração de biodiesel que