l******o 发帖数: 3764 | 1 本人不懂统计,能找到的Literature里的数据和俺仅会的那几个标准的power analysis
公式都对不上,晕了,请牛人帮忙
1.reference paper: n=200, 其中nA=26, nb=174, outcome mu_A/mu_B=12.6 (95%CI 3
.39-46.7), p=0.0002
H0: mu_A/mu_B=1
我的sample n=60, 假设一样的frequency, nA=8, nB=52, 一样的effect size和SD,
alpha=0.05, power是多少呢?
2.(和1不相关)reference paper只有个box plot和p(见插图), 另外说明了这个p是
用nonparametric two-sample t test算的。
同样,假设我的sample size =60, same frequency (then TT=54, TG+GG=6), effect
size and SD, then power=?
图例如下:The (+) sign and line inside each box indicate mean and median,
and the upper and lower limits of the box are 75th and 25th percentiles,
respectively. The vertical bars above and below designate the maximum and
minimum values, respectively. The
solid circles outside the box are outlier values. P values indicate the
statistical significance of rank sum tests (nonparametric two-sample t test)
. | l******o 发帖数: 3764 | | c***z 发帖数: 6348 | 3 Is it binary response? Take a look at this:
www.unc.edu/~rls/s151-2010/class23.pdf
Three rules to take home:
1. when n > 30, the t-score approaches the z-score and we can use the theory
of normal distributions, on which all the sample size stuff is based;
2. since we are guessing standard error anyway, there is no precise minimal
sample size to talk about;
3. when there isn't enough sample, we can resample with replacement, i.e.
bootstrapping. |
|