s********9
发帖数: 74 | 1 It was mentioned that SAS gets shortcoming in survey data analysis. Is
anyone familiar with this area?
When sample size is large enough, everything is significant. Even when
correlation coefficient is 0.3, the correlation is significant.
Usually, which software do you use and what do you do to get reliable
results? |
c******n
发帖数: 380 | 2 proc power给定其它条件可以得出sample size，从原来的data中随机挑一些再做test
，不知道说的对不对 |
d******o
发帖数: 59 | 3 其实这个是统计学的通病。只要size够大所有coefficient都是significant.因为
standard errors会小。不过我觉得significant和correlation coefficient关系不大
要紧。0.3已经不小啦。
很多bayesian现在argue这个问题。
sas不行，是因为它variance没有用准确的方法estimate. 你可以用SUDAAN，或者R里一
个叫survey的package. |
s********9
发帖数: 74 | 4 Thank you for the suggestion and the information!
As far as I know a lot of CRO still use SAS, do you know if they have any
special method or requirement when they use SAS to process the survey data? |
d******o
发帖数: 59 | 5 SUDAAN can be used in SAS. It is such a typical software for survey data
analysis. |
