Hidden signals in polls
by Winger
Thu Mar 25, 2004 at 01:58:47 AM PDT
This poll was typical last week (ie, prior to the Clarke bomb):
Bush 50
Kerry 47
These numbers are slightly negative for us, right?
But what if I tell you that in 2000, the people sampled above voted like this:
Bush 51
Gore 41
How do you interpret the Bush-Kerry data now?
All of a sudden a slight negative signal (Bush 50, Kerry 47) turns into a positive signal, because of the sampling bias.
How often does this happen in polls?
(Here's the poll from which the above data was taken, a March 16-21 Democracy Corps poll.)
- Winger's diary :: ::

Permalink | 5 comments