r******s
发帖数: 2155 | 1 【 以下文字转载自 Business 讨论区,原文如下 】
发信人: pringle (softbutcrisp), 信区: Business
标 题: A statistic question?
发信站: Unknown Space - 未名空间 (Wed Apr 27 18:59:39 2005) WWW-POST
If simple correlation shows that two variables are not significantly
correlated, could the two variables still be statistically associated?
Especially, could the nonsignificant simple correlation be caused by other
confounding factors?
Is it possible that by controlling for the other potential confounding
factors, we could still find th | j****e
发帖数: 245 | 2 Just my 2 cents. Corrections are welcome!
Intuitively it sounds possible to me. However, mathematically speaking, the
Squared Pearson Correlation is the shared proportion of variance between the
two variables. In the given case, it was not significant. It might be helpful
to draw a simple Venn Diagram. Only when significant amount of variance in the
DV is accounted for by other strong predictors that don't correlate with the
original IV, it may be possible for the semipartial/partial coefficient
【在 r******s 的大作中提到】
: 【 以下文字转载自 Business 讨论区,原文如下 】
: 发信人: pringle (softbutcrisp), 信区: Business
: 标 题: A statistic question?
: 发信站: Unknown Space - 未名空间 (Wed Apr 27 18:59:39 2005) WWW-POST
: If simple correlation shows that two variables are not significantly
: correlated, could the two variables still be statistically associated?
: Especially, could the nonsignificant simple correlation be caused by other
: confounding factors?
: Is it possible that by controlling for the other potential confounding
: factors, we could still find th
| A***A
发帖数: 98 | 3 We will probably need a clarification on what is meant by "statistically
associated."
If you mean the two random variables are statistically non-independent, then a
zero correlation does not imply independence. It will only imply independence
if the two variables are jointly normal.
A simple counter-example: y = x^2, then cor(x,y) = 0, but the two are clearly
not independent. It's just that the correlation only captures the linear
association between two variables.
Hmm. If there are more than | d**e
发帖数: 9591 | 4 ###此帖已应当事人要求删除###
【在 r******s 的大作中提到】
: 【 以下文字转载自 Business 讨论区,原文如下 】
: 发信人: pringle (softbutcrisp), 信区: Business
: 标 题: A statistic question?
: 发信站: Unknown Space - 未名空间 (Wed Apr 27 18:59:39 2005) WWW-POST
: If simple correlation shows that two variables are not significantly
: correlated, could the two variables still be statistically associated?
: Especially, could the nonsignificant simple correlation be caused by other
: confounding factors?
: Is it possible that by controlling for the other potential confounding
: factors, we could still find th
|
|
