Download Point Estimation

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Central limit theorem wikipedia , lookup

Transcript
6.1 Point Estimation
What is an estimate?
•
•
•
•
•
•
Want to study a population
Ideally the knowledge of the distribution,
Some parameters of the population may be the
first consideration:
θ
Use the information from a sample to estimate
Estimator θ̂ : defines the procedure (the formula)
Different estimators, each is a rv.
Examples of estimators
•
•
want to estimate the mean µ of a normal dist
Possibilities of estimators:
•
•
•
•
sample mean
X̄ =
sample median
!
Xi
n
average of the smallest and the largest
10% trimmed mean
Which one is best (better)?
•
Error in estimation
θ̂ = θ + error of estimation
•
•
Error is random, depending on the sample
We would like to control the error
•
•
mean 0 (unbiased)
smallest variance
Unbiased Estimator
•
•
estimator: a particular statistic
•
•
•
unbiased: E(θ̂) = θ
•
If possible, we should always choose an unbiased
different estimators: some tend to over (under)
estimate
difference is called the bias
examples: sample mean (yes), sample proportion
(yes) if X is a binomial rv.
Sample Variance
•
•
•
•
•
Need an estimator for the variance
Natural candidate: sample variance
Why divide by n-1, not n?
Make sure the estimator is unbiased
We can not say the same thing for sample standard
deviation
Unbiased Estimators for
Population Mean
• several choices (make things complicated)
• sample mean
• if distribution is continuous and symmetric,
sample median and any trimmed mean
Minimum Variance
•
suppose we have two estimators, both unbiased,
which one do we prefer?
•
•
•
pick the one with smaller variance
minimum variance unbiased estimator (MVUE)
for normal distribution, the sample mean X̄ is the
MVUE for the population mean.
Complications
•
Which estimator is better? depending on the
distribution
•
Cauchy dist: MVUE unknown, sample median is
better
•
•
Uniform dist: X̄e is the best
Trimmed mean quite versatile, quite useful in
practice
Standard Error
•
•
•
•
What quantity shall we use to report the error in
estimation?
!
Standard error:
σθ̂ =
V (θ̂)
Not always available in theory
May involve some guess for other parameters