B****r
发帖数: 647 | 1 有三组data，a, b, c
statistically significant: a>c, b>c
如果把a,b合并，还能保证这组新的data>c么？ | P****D
发帖数: 11146 | | s*****r
发帖数: 790 | 3 it is not necessary.
consider this case: a>c is just significant, b>c is just significant,
mean of b is greater than mean of a, they have same sd. the combined data
has mean smaller than b, but a larger sd. when you calibrate the numbers
carefully, you get insignificant results.
【在 B****r 的大作中提到】
: 有三组data，a, b, c
: statistically significant: a>c, b>c
: 如果把a,b合并，还能保证这组新的data>c么？
| c******n
发帖数: 380 | 4 Isn't it an example of Simpson Paradox? | s*****r
发帖数: 790 | 5 it is a little different I guess. Simpson Paradox needs to pool c as well
but here there is only one group of c.
【在 c******n 的大作中提到】
: Isn't it an example of Simpson Paradox?
| B****r
发帖数: 647 | 6 Got it, thanks!
【在 s*****r 的大作中提到】
: it is not necessary.
: consider this case: a>c is just significant, b>c is just significant,
: mean of b is greater than mean of a, they have same sd. the combined data
: has mean smaller than b, but a larger sd. when you calibrate the numbers
: carefully, you get insignificant results.
|
|
