power Flashcards
what does effect size estimate
the magnitude of an effect
d statistics are use dwhen
the response variable is continuous while the predictor is categorical
- ANOVA
r statistics are used when
two variables are continious
- correlation or regression
odds ratio is used when
the response variable is binary and the predictor variable is binary or continuous
- logiostic regression
what test used to do d statistics
cohens d or hedges d
what test used to do r statistics
correlation coefficients including Pearsons, sermons, and pointbiserial
what tests used to do odds ratio
comparative risk measurements
interpretation if you get cohens d of 0.2
small effect size
interpretation if you get cohens d of 0.5
medium effect size
interpretation if you get cohens d of 0.8
large effect size
interpretation of you get r2 value f 0.1
small effect size
interpretation of you get r2 value f 0.3
medium effect size
interpretation of you get r2 value of 0.5
large effect size
interpretation of you get a 2x2 table value of 1.5
small effect size
interpretation of you get a 2x2 table value of 3.5
medium effect size
interpretation of you get a 2x2 table value of 9
large effect size
what is power in statistics
the probably of correctly finding a real pattern
type I error
false positive
type 2 error
false negative
effect size=
difference in means/SD
power increases as effect size ______
increases
power increases as sample size ____
increases
what influences power
- effect size, alpha (type I error), and sample size
two types of power analysis
priori
post hoc
when is priori power analysis useful
for setting up a large experiment with some pilot data (can calculate the sample size needed)
when is post hoc p[ower analysis useful
useful for deciding how powerful your conclusion is (usually when results were not significant)
five steps in conducting a power analysis
- select the type of power analysis desired (prior post hoc)
- select the expected study design that reflects your hypothesis of interest (t test, anova, etc)
- select a power analysis tool that supports your design
- provide 3 of the 4 parameters (alpha, power, expected effect size)
- solve for the remaining parameter, usually sample size
