wk 4 Flashcards
test statistic =
variance explained by the model = effect
/
variance not explained by the model = error
effect
the model / test we are using
whether we are looking for a difference , relationship of association
NOT inferring cause and effect
what we are looking for
error
things we have not measured / not captured in data analysis
not a mistake but a representation of what we cannot capture in analysis
probability
if probability is small (p
Type 1 error
incorrect reason to reject the null hypothesis
state there is an effect when there is none
false positive
type 2 error
incorrect reason to fail to reject the null
state there is no effect when there was one
failing to find an effect when one exists
false negative
alpha
.05 (5%)
probability of the result occurring by chance if the null hypothesis is true
5% chance of making a type 1 error
beta
0.2 (20%)
probability of making a type 2 error
failing to find an effect
NHST criticisms
focus on Null hypothesis encourages dichotomous thinking bias in literature towards publishing significant results cannot compare the magnitude of effect cannot compare findings across studies
power analysis
attempts to control for type 2 errors
tells the strength of the statistical test to find an effect
ways to undertake power analysis
A priori - before you collect the data and do the analysis, use power to determine sample size is best to provide enough power
Post-hoc- after data collection and inferential statistics
power level
0.8
1 - beta (0.2) = power (0.8)
sufficient power if above 0.8
effect size
attempts to address type 1 errors
