Reply #35: a simple example [View All]

[breadbox]

Here is a rather simple example which shows that (roughly speaking)
we do need some sort of weighting to be done:
consider the following situation -- there are two precincts,
R and D. In the election R votes 3:1 republican and D votes 3:1
democrat. Suppose that Mitbutsky Polling exit polls 100 people at
each precinct, and gets
23(D), 77(R) at A
78(D), 22(R) at B.
Does this suggest that D scrapes out a win? Perhaps, but perhaps not.
Suppose that after the polls close, we also discover that the number
of people who voted at each location was:
1100 at A
900 at B
Then adjusting the polls would suggest that there were
.23*1100+.78*900= 955 D votes and
.77*1100+.22*900=1045 R votes.
On the other hand, if the turnouts were
900 at A
1100 at B
then we polls would suggest that the results were
.23*900+.78*1100=1065 D votes
.77*900+.22* 900= 935 R votes.

Hence, knowing just the turnout figures at each precinct
helps us to understand the predictions the polls are making.
Of course, the pollsters are *supposed* to poll every 100th person
but that is hard to maintain consistently!