Power and Sample Size Flashcards
(12 cards)
How large a sample do I need
not too small bc if so it will be waste of money resources since you will not get anything out of that study
larger than necessary, study wastes resources
therefore must estimate power to determine sample size
precision analysis
estimate how large a sample is needed to produce a given level of precision at a fixed confidence level
Two types of statistical error
Type 1 error (alpha)
Type 2 error (beta)
Statistical power
1-beta; probability that the null hypothesis will be rejected when it is false
Under Neyman-Pearson, alpha is
set by investigator
Under Neyman-Pearson, beta
is estimated and will be determined by such characteristics as sample size, trait variance and effect size
Power analysis applies to
all major types of epi data:
- dichotomous (binary)
- multiple categories
- continuous
- time-to-event (in this case, power will be a function of the number of events)
as alpha decreases
so will the power. It will decrease
But beta increased
given in a fixed sample size:
if alpha increases, beta decreases
if alpha decreases, beta increases
without changing the null hypothesis, the only way to increase power while using a smaller alpha level is to increase sample size
increases precision
allows a smaller effect to be detected
Factors affecting statistical power
Z = a standard normal random variable μ1 = mean under HA μ0 = mean under H0 σ = population standard deviation Zα = critical value at significance level of α n = sample size
which will have more power? one sided or two sided?
with all else being equal, one sided test will have more power