12. Power

Type I error = H₀ is true, but researcher rejects it

Type II error = H₀ is false, but researcher does not reject it

Chance of type II error: β

• if α increases → β decreases + Power increases

• if α decreases → β increases (but not by same amount) + Power decreases

• less spread → more power

• bigger sample → more power

Type II error: researcher concludes there is no difference bet. 2 groups, when the difference does exist in pop.

Inverse of Type II error: researcher conclude there is dif. bet. groups and the difference does exist → what we want

Change of inverse of type II error: 1 - β → power = chance of correctly rejecting H₀

Large samples → higher power

(+) descriptive stats are more accurate

(+) standard error is smaller

(+) ability to differentiate bet. groups increases

(-) more expensive

(-) sometimes practically unfeasible

• higher α → greater chance of Type I error

• researcher need to find balance bet. small α and high power

One-sided test - if dif turns out to be in opposite dir. you can’t reject H₀ even if p < α

Two-sided test - less power, not theory driven