Estimation of parameters
description
Transcript of Estimation of parameters
Estimation of parameters
Maximum likelihood principle
What has happened was most likely
Maximize likelihood wrt parameters to obtain estimates
n
ii
n
xfL
xxxf
parameterssample
parameterseofthesamplLikelihood
1
21
)|()(
)|,,,(
]|Pr[
)(
Examples
Binomial distribution
kNkk pp
k
NpL
)1(
)1ln()(lnln pkNpkk
Nl
Observations: k successes in N Bernoulli trials
01
p
kN
p
k
dp
dl
N
kp ˆ
Poisson distribution
N
i i
kN
keL
i
1 )!(
Observations: k1, k2, …, kN
)...(ln 21 NkkkNl
Nkkk
Nd
dl
...21
N
kkk N
...ˆ 21
Normal distributionObservations: X1, X2, …,XN
N
i
iX
NNeL 1
2)(2
1
)2(
11
N
i
iXNl
1
2
2
1ln
N
i
iX
d
dl1
N
iiX
N 1
1̂
N
i
iXN
d
dl1
3
2
N
iiX
N 1
22 ˆ1
ˆ
Exponential distributionObservations: X1, X2, …,XN
N
iiXa
N eaL `1
N
iiXaaNl
`1ln
N
iiX
a
N
da
dl`1
N
iiX
Na
`1
ˆ
Moment estimators
),( axp
N
iiX
NaE
1
1)(
For our examples moment estimators = maximum likelihood
estimators
Are they always the same ?
No
Uniform distributionCauchy distribution
Unbiased and biased estimators
aaxp ˆ ),(
aaE )ˆ(
Variance of estimatorsminimum variance estimators