Site Meter

Friday, October 29, 2004

Bush 47.55% Kerry 45.9 % difference 1.65% standard error of difference due to sampling error alone about 0.5%

is the population weighted average by state of the average state poll since 10/10 (only one observation per tracking poll. Latest poll if none since 10/10).

The very slight shift to Bush and the decline in the standard error are both due to a new poll from New York. It means that, by now, about one third of the sampling variance in the population weighted average is due to Texas a large state with only one poll.

Update: The population weighted average by state of the average state poll since 10/20 (only one observation per tracking poll. Latest poll if none since 10/20).

is Kerry 45.9 % Bush 48% with a standard error due to sampling alone of about 0.77 %.
The very slight shift towards Bush is, naturally, a little bit more marked if the older polls are removed from the sample and not averaged with the newer polls. As mentioned above it is largely due to one new poll each for California and New York.


Under the null of no true change, the standard error of the change in the difference would be root(0.77*0.77 - 0.5*0.5) or 0.58%. The actual change in the change dropping the old polls is Bush gains by 0.45 % which is clearly not significant. I just did a Wu-Hausman test.

No comments: