Reply #23: no, as Febble says, at most the sheet "refutes" a straw man [View All]

[OnTheOtherHand]

As you will recall, E/M in January (echoing earlier statements) wrote inter alia: "Our investigation of the differences between the exit poll estimates and the actual vote count point to one primary reason: in a number of precincts a higher than average Within Precinct Error most likely due to Kerry voters participating in the exit polls at a higher rate than Bush voters."

The spreadsheet does not address that issue. As far as I can tell, it has no relevance to the actually-existing exit poll debate whatsoever (unless people find that it pulls numbers together in a handy way). If anything, it is likely further to confuse the debate.

To give some idea of how the spreadsheet promotes confusion, consider this explanation of the pre-election analysis (on the "instructions" tab of the spreadsheet) :

"The MoE defines the range in which we can expect the final results to occur (with 95% probability). For example, if a poll with a 3% MoE yields 51% for a candidate, then the probability that his/her actual result will fall between 48%-54% is 19 out of 20 (95%)."

That would be true if sampling error were the only source of error. Good luck with that. The electoral-vote.com database has seven different results from Ohio completed on October 31 or November 1 (I'm only counting the Zogby tracking poll once). Bush was ahead in six of these polls, by up to 6 points; Kerry led the seventh poll by 7 points. It would be brave to assume that these differences were due to sampling error alone.

So, if six out of seven polls showed Bush ahead in Ohio, why does the spreadsheet have Kerry ahead by exactly three points there? Heck if I know, and heck if the spreadsheet cites any sources. However, I suspect that its designer has chosen to ignore the likely voter results -- literally ignore them, as if they never existed -- and rely on registered voter results. That controversial decision certainly doesn't warrant the 100% confidence it is given in this exercise.

Similarly, the explanation of the Monte Carlo simulation (which applies to both the pre-election and the exit poll analysis) says, "This model consists of 200 trial 'elections' using state win probabilities, to determine the expected Electoral Vote and win probability." Well, unless I'm much confused, these "win probabilities" assume that there is no bias whatsoever in the exit polls -- even though every exit poll we've looked at has overstated the Democratic candidate's performance.

The spreadsheet seems to be designed to give one interpretation of the pre-election polls, and one interpretation of the exit polls, the aura of simple fact. Kerry led in the pre-election polls; the exit polls were unbiased. Just enter numbers, press buttons, and see the statistical proof!