2

What are some examples of a statistical result being counterintuitive? Especially with the regression analysis.

amoeba
  • 93,463
  • 28
  • 275
  • 317
sbmm
  • 73
  • 1
  • 10
  • 2
    Any result can be counter intuitive depending on what is your intuition. If you don't know math, everything will be strange to you. – Aksakal Jan 16 '15 at 13:13
  • 1
    See [Romano & Siegel (1986), *Counterexamples in Probability & Statistics*](http://www.crcpress.com/product/isbn/9780412989018). – Scortchi - Reinstate Monica Jan 16 '15 at 13:24
  • 2
    This might end up being closed as opinion-based, but actually could be a very interesting community wiki list. Maybe moderators could convert it into a community wiki, hoping to save the question. The answers could cover e.g. Simpson's paradox, Lindsley's paradox, Stein's phenomenon, etc. See also [Category:Statistical paradoxes](http://en.wikipedia.org/wiki/Category:Statistical_paradoxes) on Wikipedia. I upvote the question. – amoeba Jan 16 '15 at 13:25
  • Thanks a lot, the book and the wiki, both are very helpful. – sbmm Jan 16 '15 at 13:33
  • 2
    Damit, closed. I was halfway writing a response and hoping that someone would point me in the right direction in understanding the example. I do believe that there is a meaningful distinction between statistical paradoxes and counter-intuitive everyday results. – LauriK Jan 16 '15 at 13:51
  • 2
    @amoeba I agree it's an interesting question, and I don't think that a question about "counterintuitive results, especially in regression" is much of a duplicate of "paradoxes". – Glen_b Jan 16 '15 at 13:53
  • But I found it many times in the texts that Simpson's paradox is a counter-intuitive results in statistics. I'm giving Simpson's paradox as an example since it might happen as a result of a regression analysis. I'm searching for another possible counter-intuitive results which might happen as a result of regression analysis. – sbmm Jan 16 '15 at 14:48
  • To be more precise, I have different sets of data, which I was told to consider them separately in the regression analysis since it might result in counter-intuitive results. I don't know what are the possible problems if I combine these data in my regression model (I should mention that a single data set is high dimensional p>>n, so I'm using LASSO). One of the problems I found is the Simpson's paradox, which makes sense if I combine data sets. But I'm not aware of any other problems by combining all data sets in the regression analysis. – sbmm Jan 16 '15 at 15:01

0 Answers0