power and effect size

t/f: power is a function of sample size

true

t/f: to conduct a power analysis, you need an estimate of how large your effect size is

true

factors that affect effect size

-mean differences (large mean difference = larger effect size) -SD (smaller SD = larger effect size)

t/f: a more powerful study is one that has a better chance of rejecting a false null hypothesis

true

t/f: if you have non-significant results or there's no effect then you did NOT have enough power

true

t/f: power analysis gives us an idea of what our sample size should be

true

factors that affect power

-alpha level -sample size -effect size (difference between 2 means, population variance)

effect size

-as difference between the 2 means increases, power increases -as variance increases, power decreases

a priori power analysis

-calculating power before the study is conducted -you should always conduct power analyses a priori

things to consider when estimating effect size

-don't rely on pilot studies for effect size -SE of effect size tends to be overly large (?) -greater probability of overestimating effect size

effect size and power analysis

-if we choose to have a small effect size, our study will have greater power but we'll need a larger n -if we choose to have a large effect size, our study will have less power but we use a smaller n

dangers of low power (and a small n)

-less likely to detect an effect if it exists -effects that are detected and declared to be significant are likely to be *overestimated* in terms of their true effect size (this is b/c only large effects pass the threshold)

estimating effect size for power analyses

-pilot studies are a bad way to determine effect size for a power analysis -replicating published studies based on their given effect size is an uphill battle b/c these studies likely *overestimate* the true effect size (which explains why you don't get the same results when replicating the study based on their effect size)

alpha

-probability of type 1 error -if we increase alpha, our cutoff points move to the left so we're more likely to detect an effect if it exists -also if we increase alpha, beta decreases which increases power (b/c 1-B = power)

how to estimate effect size for power analyses

-we can look at previous studies and estimate effect size -we can use a benchmark effect size

power of .8

80% probability of finding a significant effect if it exists

sample size

as sample size increases, power increases

post hoc power

calculating power after a study has been conducted

the smaller your effect size, the ______ your sample size will need to be

larger

cohen's d

measures effect size between two means in the terms of SD

power

probability of correctly rejecting a false null hypothesis 1-B

when conducting a power analysis: if you know the effect, alpha level and power you want then you can...

solve for the sample size

if power is too low....

you may not detect a treatment effect