This preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full Document
Unformatted text preview: ∙ Other than Type I and Type II errors, there are two other possibilities: we fail to reject H when we should because H is true or we reject H against H 1 when we should because H 1 is true. ∙ Some authors use the language “reject H ” and “accept H .” The latter is less desirable than “fail to reject H ” for reasons we will see. 15 4 . Size and Power ∙ Because we want to make type I errors relatively rare, we need to formalize the notion of what we mean by “rare.” ∙ Suppose first that the null hypothesis is simple: H : . Then the size of a test, usually denoted , is the probability of a Type I error: P Reject H  H is true P Reject H  ∙ I will also use significance level to refer to the size of a test. 16 ∙ If the null hypothesis is composite, the probability of a Type I error generally depends on the value of ∈ Θ . In that case, we define the size as the maximum of the probability of Type I error: max ∈ Θ P Reject H  (where, in our applications, the max will always exist). ∙ By definite, for any ∈ Θ , P Reject H  ≤ . ∙ The probability of a Type II error is a function over Θ 1 : P Fail to Reject H  , ∈ Θ 1 17 ∙ It is very useful to define the power function as P Reject H  , ∈ Θ P Reject H where the second definition is just shorthand notation. ∙ Notice that is the maximum of over ∈ Θ . ∙ For ∈ Θ 1 , 1 − P Type II error 18 ∙ Ideally, we could have 0, all ∈ Θ 1, all ∈ Θ 1 , which effectively means we would never make a mistake. But this is impossible in practice. ∙ If we want to make the size of a testing procedure “small,” the cost is that power against alternatives is reduced. 19 5 . Test Statistics and Rejection Rules DEFINITION : Given a set of observations X 1 , X 2 ,..., X n , a test statistic T h X 1 ,..., X n is a (measurable) realvalued function. ∙ A test statistic, like an estimator, must be computable for any outcome of data. It cannot depend on unknown parameters. It is viewed as a random variable because it is a function of the random vectors X 1 , X 2 ,..., X n . ∙ For a given sample of data, it is sometimes useful to let t h x 1 ,..., x n denote the particular outcome of T (that is, a number). 20 ∙ Given specification of the null and alternative hypotheses, we will use test statistics and rejection rules to obtain tests with a given size. We can then study power properties to choose among tests....
View
Full Document
 Fall '12
 Jeff
 Normal Distribution, Null hypothesis, Statistical hypothesis testing, alternative hypotheses

Click to edit the document details