Professional Documents
Culture Documents
When making a decision to reject or not reject H0, there are two types of error we wish to
avoid:
Balancing the risks of these two types of error is the basis for calculating the dividing line
between accepting or rejecting H0. It is analogous to a court decision, where a defendant is
on trial. The null hypothesis is innocent (presumption of innocence). The two errors are:
If the jury is convinced ‘beyond all reasonable doubt’ that the defendant is guilty then he is
convicted. The ‘beyond all reasonable doubt’ criterion is there to avoid type I errors if
possible, even though it might lead to the occasional type II error. Twelve jurors decide on
the case and have to reach a unanimous decision. If they disagree, the judge will sometimes
accept a majority (10-2) verdict. This presumably raises the risk of a type I error while
lowering the risk of a type II error.
The punishment can influence the balance of error probabilities. In civil cases (e.g. a libel
action) where no criminal activity is involved, the jury decides ‘on the balance of
probabilities’. The defendant, if convicted, would not have a criminal record, nor go to jail,
so the court is more willing to make a type I error and reduce the probability of a type II
error.
Similarly, in a criminal case, if the punishment is the death penalty the jury may be less
willing to convict since the consequences of a type I error are catastrophic...
True situation
Decision H0 true H0 false
Accept H0 Correct decision Type II error
Reject H0 Type I error Correct decision
and let us suppose that x ~ N(, 2/n). The distributions of x under H0 and H1 are shown
below:
H1 H0
I II
_
xD 100 x
If the decision line is set at x D then areas I and II show the probabilities of making type I and
type II errors respectively. (Area I gives the area under H0 to the left of x D, which would
imply falsely rejecting H0.) Moving the line left (right) means reducing (increasing) the
probability of making a type I error and reducing (increasing) the probability of a type II
error. The optimal position of x D comes from balancing the two error probabilities (and the
costs of the two types of error).
The optimal position of x D is difficult (if not impossible) to establish, since the precise
position of the distribution under H1 is unknown. Hence we compromise by setting the
probability of a type I error equal to 5% (usually). Hence any value of the test statistic falling
into the 5% tail of the Normal distribution rejects the null hypothesis.