On Wednesday, April 2, 2014 10:59:39 AM UTC-4, micky wrote:
On Wed, 02 Apr 2014 05:08:26 -0400, micky
wrote:
OT Unless the percentages match, no such thing as a statistical dead
heat
I've wondered about this off and on, and I once asked on another
newsgroup, but until today, I've never remembered to google it.
http://iase-web.org/documents/papers/isi53/292.pdf
http://boards.straightdope.com/sdmb/...ad.php?t=30816
The more I think about this, the more things become clear. When the say
the difference is within the margin of error, is there a 95% chance the
guy who is ahead, is ahead? No. But I just realized that the chance
never goes below 50+%, because if it were lower than 50, the odds would
be that he is behind. And if his poll numbers exceed the other guy, the
odds are he's ahead, not behind.
Consider a poll on candidates Smith versus Jones. We poll one person who picks Smith. Our poll indicates Smith with 100% of the vote, but we wouldn't put much confidence in the result of that poll.
So we poll another person who picks Jones. Now, with a sample size of 2, we have 50% Smith and 50% Jones but how reliable is a poll of two people when there are thousands of people who will be voting?
We go out ringing doorbells again and end up with a hundred responses, 52 for Smith and 48 for Jones, a 4% difference between the two. Based on the sample size and other parameters*, we might find that the results of this poll give us a 99% confidence that the results of the sample are within 5% of the results had we polled the entire electorate.
You could think of it as meaning that Smith should end up with somewhere between 57% (52% + 5%) and 47% (52% - 5%)of the vote and Jones should end up with somewhere between 53% (48% + 5%) and 43% (48% - 5%)of the vote. Since we're 99% confident that either candidate could end up with over 50% and win, it's a "statistical dead heat."
* Here's a good reference on all that:
http://en.wikipedia.org/wiki/Sample_size_determination
Paul