Effect Soze And Power Analysis Flashcards
Effect size
Size of effect of iv on dv, after finding sig effect, measures prop of explained variance. Diff ways to express: Cohen’s d, eta squared, partial eta squared. R/r squared is %variance explained by line of best fit.
Cohen’s d
Used for t tests: diff between means over variance- gives standard measure over diff units and samples. For anova S looks at max and min mean to get biggest effect, over df, need homo of variance so sd constant.
Eta
Squared: used for one way anova, same as r squared, proportion of variance explained by experiment. Partial: used in factorial anovas, proportion of variance explained by each iv e.g. one per variable .Both scaled between 0 and 1, 0 no effect and 1 is all variance explained by this variable. Eta is sseffect over sstotal. Partial is sseffect over SS residual + sseffect
Reporting effect sizes
Look at table of between subject effects table, in partial eta squared/equivalent column. Report after via value, copy and paste the symbol from google. If eta small, iv effect isn’t big so not worth investing in interventions
Power analysis
Looking at the power of stats tests ability to detect an effect when it’s there and reject null correctly. Depends on sample size, effect size and alpha-run before study. Can work out one from the other 2 e.go if know effect size and power, can calc number of ps
Underpowered studies
Usually too few ps so get false negative but also positive as small sample means small diff looks sig when not. Low power explains failure to replicate as papers favour sig results. Ethics: studies are expensive, inconvenient, painful and dangerous so doing a non powerful study is a waste- now a requirement for ethical approval to show sufficient power
Sample size and power
As sample size increases, so does power. Effect bigger at low sample sizes and tips off. Power of 0.8 considered good. Smaller effect sizes need more ps to achieve high power e.g big effect size like temp and ice cream needs less ps to get high power.
Ways to estimate effect size
Guess if not much precious info using Cohen’s heuristics for general idea or small med or large effect size but bad. Run pilot study W less ps to estimate effect size but not reliable. Find old studies: use results to work out expected effect size. Find or conduct meta anal- calc average effect size
Issues with power analysis
Known as garbage in, garbage out technique, if you put in made up numbersm get out numbers that arent meaningful
Estimates not exact so can’t guarantee results
For complex designs, unlikely to find good estimates of effect size
There is a cheat sheet if don’t understand everything
On docs my love
