MUST READ: End notes from the Baiman-Simon paper.

I have posted my election probability analysis at DU time and time again. And I will continue to do so.

This is from Baiman and Simon.
I'm in good company.

My only criticism: they focus on the battleground states (4 out of 11 beyond the MOE). I have analyzed the probability of Bush exit poll deviations beyond the MOE occurring in SIXTEEN (16) states, while NONE so deviated for Kerry.

And while 41 states deviating by any amount for Bush, only 10 did for Kerry. The state poll deviations, along with the National Exit Poll deviation, are the TWO MAJOR SMOKING GUNS which confirm one another.

As far as the National Election Exit poll (13,047 sample-size) is concerned, their analysis pretty much coincides with mine.

They reference the Excel NORMDIST function in calculation of the probability. Sound familiar?

NOTE: I HAVE INSERTED CAPITALIZED HEADINGS FOR CONVENIENCE
.......................................................

Scoll down two screens for the paper:
http://spiral-stairs.dailykos.com/story/2004/12/30/223723/91


SNIP

SAMPLE-SIZE MOE
22 Ibid, p. 2, Table. Calculation of the margin of error may be checked as follows: Calculate the standard error of a random sample using the formula , where p = Kerry percentage of the vote (0.481) and N = the sample size (13,047). The fact that an exit poll is a cluster sample, grouping respondents by precinct, rather than a fully homogenized random sample of the target venue, increases the
standard error by 30% to 0.00568 (see Merkle, D. and Edelman, M. "A Review of the 1996 Voter News Service Exit Polls from a Total Survey Error Perspective," in Election Polls, the News Media and
Democracy, ed. P.J. Lavrakas, M.W. Traugott, New York: Chatham House, pp. 68 - 72). Ninety-five percent of the time, a result predicted on the basis of a random sample will be within 1.96 standard errors, or ±0.011 (1.1%) for a sample of this size.

STANDARD DEVIATION
23 It is dramatic because a 2.7% "miss" at these levels of precision is extremely unlikely to occur. The statistician's measure of such likelihood is known as a "standard deviation." A result which is off, as in this case, by 4.7 standard deviations is without question "dramatic:" the odds against its occurrence are
enormous.

DISCRIMINATORY VOTER SUPPRESSION
24"Discriminatory voter suppression" refers to methods that disproportionately reduce voter turnout in precincts that favor one candidate, for example through disproportionate allocation of voting machines.

Because state level exit polls are weighted sums of precinct voting shares, disproportionate changes in turnout can contribute to a discrepancy in state exit polls relative to the actual vote.


THE NORMAL DISTRIBUTION
25 Probability of a 48.1% vote share assuming an exit poll vote share of 50.8%: P(0.481) = 1 - NORMDIST(0.481, 0.508, 0.005686, True) = 0.0000010424 (where NORMDIST is an Excel spreadsheet
function that gives the probability of obtaining 0.481 for a normal distribution with a mean of 0.508 and a standard deviation of 0.005686). 1/0.0000010424 = 959,336.

NUMBERS FROM THE WASHINGTON POST
26 Based on final election numbers from the Washington Post, Nov. 24, 2004.

27 For reference, even a clearly "male-skewed" 50% male, 50% female sample would have resulted in a Kerry victory 93.5% of the time.

AN MIT GRADUATE STUDENT'S ANALYSIS
28 There is some intriguing evidence to the contrary, drawn from an analysis performed by William Kaminsky, a graduate student at MIT. Kaminsky finds that in 22 of the 23 states which break down their
voter registrations by party ID the ratio of registered Republicans to registered Democrats in the final, adjusted exit poll was larger than the ratio of registered Republicans to registered Democrats on the official registration rolls. In other words, the adjustments performed on the exit polls in order to get them to agree
with the official tallies would, if valid, require Republicans to have won the get-out-the-vote battle in essentially every state. We find this requirement implausible, and indeed observational evidence pointed to just the opposite: massive new voter turnout, which virtually always cuts in favor of the challenger; huge lines in Democratic precincts; unadjusted exit poll data showing apparently greater Democratic turnout; etc. Exit polls appropriately stratified to official party ID percentages, which would effectively neutralize
any suspected "reluctant Bush responder" phenomenon by including the expected proportions of Republican and Democratic voters, would on the basis of Kaminsky's analysis have yielded results at least
as favorable to Kerry as those upon which we have relied in our calculations.

VOTE COUNT DISCREPANCIES IN THE BATTLEGRTOUND STATES
29 A complete analysis of all 45 states and the District of Columbia for which comparable exit poll data is available shows that four out of the 11 battleground states had exit poll/vote count discrepancies that were outside of a standard 5% (one-tail) margin of error, whereas this was the case for only one of the 35 nonbattleground
states. Moreover, all of these statistically significant discrepancies were in favor of Bush. This data is at odds with claims of "systemic" pro-Kerry exit poll skew. See Baiman, R. Dec. 19, 2004 at:

http://www.freepress.org/departments/display/19/2004/997
(some figures have been updated by the author to reflect more recent data).

WHO WAS UNWILLING TO TALK TO THE EXIT POLLSTERS?
30 It is by no means self-evident that either candidate's supporters were systemically more likely to be intimidating or more easily intimidated. While it might be more reasonably argued that voters finding themselves in a dwarfed minority in their communities might have been less willing to be exit poll respondents, in light of the even division of the national electorate, any such tendencies would have resulted in a wash, with no net effect on the validity of the national exit poll. We would of course welcome the release by Edison/Mitofsky and/or the National Election Pool of the data which would facilitate further analysis of these and other factors.

THE CATCH-22
31 It should be clear that more is at stake than the presidency itself. Use of computerized vote counting will only increase, as mandated by law. Vote counting is the bedrock protocol of a democracy and meaningful reform of a broken counting system is dependent on an expression of public will ultimately
exercised at the ballot box and fairly, accurately, and honestly tabulated. If the system has broken down and is no longer counting accurately and honestly, there is no effective democratic mechanism to bring pressure upon a governing majority to reform a vote counting status quo which is seen to work in its favor.

This is, as may be seen, a potentially crippling catch-22 for a democracy.

http://www.democraticunderground.com/discuss//duboard.php?az=view_all&address=203x268654

TIA, what do you do with all your brilliant analysis? Does it ever go to Conyers, Arnebeck, etc? I hope so because it is too good not to be used by those working on investigations.

with the other stat folks and perhaps Will -

This would be a book I'd buy copies to give away!