On Mon, 27 Feb 2012 08:31:02 -0800, Marie Helweg-Larsen wroteL >I have a simple statistical question. > >I have a sample of 307 people. 111 are in the red group and 196 are the >blue group. > >The correlation between variables x and y in the red group is r= .226 (n=111), >p <.05 and in the blue group r=.164 (n=196), p<.05. However, when I run the >correlation between x and y in the entire sample (red and blue combined, no >missing data) I get a negative correlation, r=-.142 (n=307), p < .05. >Now what doesn't make sense to me that two groups individually have positive >and significant correlations but the two groups combined can have a negative >and significant correlation. >So you stats tipsters. Is that statistically possible?
I believe you have an instance of "Simpson's Paradox". Wikipedia has an entry on it (yadda-yadda); see: http://en.wikipedia.org/wiki/Simpson%27s_paradox The little figure on the top right seems to describe your situation. If memory serves, Guildford & Fructher's textbook on statistics in psychology and education had examples of this type of situation (I remember the nice looking graphs) under the heading of "non-homogeneous groups", that is, the correlations are one way within the groups and overall do not match up (such as an overall positive correlation but negative correlations within the groups). I lost my copy of Guilford & Fruchter some time ago but your library might have access to a copy which you can use to verify what I said. -Mike Palij New York University [email protected] --- You are currently subscribed to tips as: [email protected]. To unsubscribe click here: http://fsulist.frostburg.edu/u?id=13090.68da6e6e5325aa33287ff385b70df5d5&n=T&l=tips&o=16309 or send a blank email to leave-16309-13090.68da6e6e5325aa33287ff385b70df...@fsulist.frostburg.edu
