Decision rules and errors (Chapters 5 and 8.1)
Critical value Fa for significance level a: that F value where the cumulative probability distribution reaches 1-a (in other words, the probability that F takes on a value larger than the critical value is a)
Typical
significance levels: 0.05 (=5%, "significant")
0.01 (=1%, "very significant")
Null
Hypothesis H0: no difference exists between groups
Alternative Hypothesis H1:
there is a difference between groups
Let
Fobserved be the observed F value.
If Fa
> critical value, reject H0 and accept H1,
otherwise retain H0
and reject H1.
Type I (a) error: reject H0 even though it is true (wrongly assume treatment effect)
Type II (b) error: accept H0 even though it is false (fail to recognize treatment effect)
Power of an experiment: 1-b: ability to recognize treatment effects
How
to reduce a
errors:
- Increase the significance level (this increases the type 2 error
though)
How
to reduce b
errors:
- Increase the sample size
- Increase the size of the treatment
effect
- Decrease the amount of experimental error
-
Use a more sensitive experimental design