Election 2004 Questions

Robert Koehler writes:

Robert C. Koehler | Common Wonders: Was the election of 2004 stolen? Thus is the question framed by those who don't want to know the answer. Anyone who says yes is immediately a conspiracy nut.... So let's not ask that question. Let's simply ask why the lines were so long and the voting machines so few in Columbus and Cleveland and inner-city and college precincts across the country, especially in the swing states, causing an estimated one-third of the voters in these precincts to drop out of line without casting a ballot; why so many otherwise Democratic ballots, thousands and thousands in Ohio alone, but by no means only in Ohio, recorded no vote for president (as though people with no opinion on the presidential race waited in line for three or six or eight hours out of a fervor to have their say in the race for county commissioner); and why virtually every voter complaint about electronic voting machine malfunction indicated an unauthorized vote switch from Kerry to Bush....

We might also ask why so many Ph.D.-level mathematicians and computer programmers and other numbers-savvy scientists are saying that the numbers don't make sense.... And we might, no, we must, ask... about those exit polls, which... last November went haywire, predicting Kerry by roughly the margin by which he ultimately lost to Bush. This swing is out of the realm of random chance, forcing chagrined pollsters to hypothesize a "shy Republican" factor as the explanation.... And the numbers are still haywire. A few days ago, Terry Neal wrote in the Washington Post about Bush's inexplicably low approval rating in the latest Gallup poll, 45 percent, vs. a 49 percent disapproval rating. This is, by a huge margin, the worst rating at this point in a president's second term ever recorded by Gallup, dating back to Truman.... Bush mustered low approval ratings immediately before the election, surged on Election Day, then saw his ratings plunge immediately afterward.

Comments

[alan]

I have wanted for some time to get hold of one of those voting machines from Ohio in my lab to examine it. Anyone have any idea of how to bring this about?

I can think of several ways that the voting machines can be rigged to A) bias the election in favor of one side or the other, and B) do it in such a way that it fools the tests the officials use to detect software error or perfidy.

It is quite telling that the manufacturers don't allow their "customers" to audit the source code of the systems. As if driving a Microsoft Access data base is a big trade secret. However these things can be reverse engineered and I would be willing to do it if someone would do the work of getting me a sample.

[P O'Neill]

He's right that there are smart people who think there's a serious issue, which is disturbing. But there two issues that fall short of outright election stealing: (1) state-level institutional inertia and control that benefits Republicans, such as better funded systems in areas more inclined to vote Republican, and (2) well-funded get-out-the-vote operations.

All that Republican campaign money got them a very slick GOTV operation, and the election campaign was focused around polarising but non-policy issues like Swift Boat, coded gay-bashing, Kerry = next 9/11 etc. So quite a few people got worked up to keep that gay Khmer Rouge-loving French-haired Commie out of office, but fundamentals reasserted themselves once the election was over.

[RonK, Seattle]

Those who are impressed by the "PhD vote" re exit poll interpretation would be wise to bring themselves up to date on the analytical state of play.

http://www.dailykos.com/story/2005/4/29/161938/921

An upcoming conference/confrontation should shed a great deal more light on what can and cannot be inferred from these data.

[Ottnott]

The get-out-the-vote explanation is very pausible, but you then have to explain the large variance between the official vote tallies and the exit polling.

There are plausible explanations for that variance, but they have always struck me as after-the-fact attempts to create a plausible explanation.

Personally, I think the election was almost "stolen" by GOP control of key positions in the agencies that operate elections--the long polling lines, disappearing or late absentee-ballot requests, etc.

I say "almost", because I'm not confident that the effect was enough to change the outcome.

But it certainly was enough to change the outcome of an even tighter race. Voter rights are a joke in the U.S., but what can you expect in a country where one of the highest interpreters of the law says that U.S. citizens don't have a constitutional right to vote.

Strangely enough, that same interpreter of the law ruled that the same citizens apparently have the right not to be subject to county-to-county variations in method in a vote recount even though there were wide county-to-county variations in method in the original count.

Go figure. Or weep.

[s9]

The problem I have found is that you can ask all those other questions instead of the question about whether the 2004 election was "stolen" and you will still be branded a conspiracy nut and laughed out of the room.

At best.

At worst, you'll be ignored, as if you were speaking about the weather— in Welsh. The very idea that America's national elections are now a fraud is Unthinkable.

[Prior Probability]

There is only one reason to have a voting machine that doesn't leave a paper trail. That reason is that you want to steal an election.
That is also the only reason to buy a voting machine from a vendor who will not allow an independent audit of the software used. And the only reason for a vendor not to allow such an audit.

[alan]

So how do I get my hands on one? With the 2004 software set.

[robert west]

Alan - you likely can't.

I can't speak to Ohio, but I have a good deal of knowledge about how these things are done in California, and it turns out that the software set is often different in each precinct (because each precinct can, theoretically, have a different set of contests and/or a different ordering of candidates within contests).

In some counties all of the processing is done centrally, in which case there is a unified software set. However, in places like San Mateo county, the vote processing is done on a per-precinct basis, and thus the software on each machine isn't the same.

[Dubblblind]

Hey, nothing much at stake here. Just our democracy. And anyone who questions what is going on is a crackpot? Sounds to me like the real (crack)pots are calling the kettle black.

[alan]

Robert, I live in California too, but it is Ohio I want to look at. That's one of several places that there is such variance between polls and official tally.

If the software is different from precinct to precinct then I would only need to see one to start with. If it could be shown that one test case was corrupted somehow, that would be enough to trigger an official investigation of all of them. Not just Ohio but everwhere similar machines were used.

It seems kind of stupid that the actualy software would be different between precincts though. You would think they would have come up with a re-usable program that would simply have a different configuration fileset loaded.

I suspect Ohio law would provide a way for citizen review. If not, perhaps a court test would be in order. Unfortunately, I have neither the time or resources but if I did, that is what I would do.

[Depressed]

I just clicked through on the link provided. I quickly found out that the ratio of voters to machines varied from precinct to precinct by over 3 to 1. The low ratios were all (relatively) favorable to Bush, the high ratios (and hence the longest waits) to Kerry.
Just a coincidence, no doubt.

[Steve in Seattle]

I looked at the election results nationally, in Florida, and in Ohio by precinct and I can't find anything which is ridiculous in terms of statistical variations for Bush in electronic voting machines. Many of studies that I've seen on the Internet exclude some aspect of the problem in some significant way. They don't examine swings in previous elections, they ignore differences in absentee ballots or other voting process changes, they don't track what happened with other candidates, etc.

Florida, in particular, requires a massive conspiracy to have true fraud. There were different e-vote machine manufacturers, several different models, and lots of different software versions. But more importantly, there was a new voting process that allowed people to vote for more than a week, including weekends! This is in addition to absentee ballots.

In Florida, the Bush vote tracked closely with the Martinez vote, except among women. The easiest explanation for that was that the other Senate candidate in Florida was a woman.

It is quite possible that the election was stolen, but at some point we need to acknowledge that really strong electioneering is not election theft in our country. Unless you are willing to classify a lot of elections as "stolen".

[Richard Cownie]

From the Edison-Mitofsky report, it seemed that
there was a very high WPE (within-precinct error)
for all the automated counting methods (punchcard,
optical, and touchscreen), but a much lower WPE
for paper/hand-count precincts. That would suggest
that strange things happened in the automated
tabulation process.

We should all be jumping up and down demanding
access to the complete unweighted precinct-level
exit poll data - that would give a much better
chance of figuring out what might have happened.

Some people say "move on, stop worrying about the
past" - but obviously if the election of 2004
was fixed, then it's very likely that future
elections would also be fixed in the same way.

[Steve in Seattle]

re: WPE in Edison-Mitofsky report

My read of the report says that no systemic explanation existed for the high WPEs, including no plausible linkage to type of voting machine. (the report explicitly says this at least 3 times) Rather than a systemic explanation in voting fraud, the report blames the errors on management failures in the human process of polling.

[Richard Cownie]

>My read of the report says that no systemic
>explanation existed for the high WPEs, including
>no plausible linkage to type of voting

They say that because the high WPEs are not
specific to touchscreens as some people had feared
or suspected. But I haven't heard any innocent
explanation for the high WPEs for all the automated
methods compared to paper/hand-count.

The "explanation" of differential response rates
seems very inadequate - there doesn't seem to
be any good evidence for it, or any explanation
for why this would be a bigger problem in 2004
than in earlier elections.

Anyway, it's shocking to me that no-one has been
allowed to analyze the complete precinct-level
data. Something strange happened either in the
exit polling or the vote-counting, and without
that precinct-level data it's hard to test
hypotheses about what it might have been.

[Arbogast]

2004 Bush 59,834,866 Kerry 56,373,514
2000 Bush 50,456,169 Gore 50,996,116
That's all you need to know. Exit polls showed overwheming numbers of new voters voting for Kerry, yet Bush had 9.4 million new voters and Kerry had less than 5.5 million. In other words, if you were a new voter, you were approximately twice as likely, nation-wide, to vote for Bush than Kerry. That datum is not supported by any other datum in the 2004 dataset: exit polls, etc. etc.
This was theft on a massive basis, democracy destroying theft.

[Richard Cownie]

>That's all you need to know. Exit polls showed
>overwheming numbers of new voters voting for Kerry,
>yet Bush had 9.4 million new voters and Kerry had
>less than 5.5 million.

I'm somewhat inclined to agree with your conclusion,
but your argument is not valid. You can't assume
that all the old voters voted exactly the same as
they did in 2000 - there's no evidence for that.
As an alternative (and equally unprovable) theory,
suppose that a large bloc of voters favor the
status quo - in 2000 they would have voted for
Gore, in 2004 they would have voted Bush.

[Dave S]

Diebold:
Manufactuer of these paper-less voting machines also manufactures ATM's, the bank-card machines for instant cash. Except- the bank knows, and you know there has to be a paper receipt to clear up any post-transaction discrepancies. Because Money is important, dammit.
Somehow, Diebold decided or was twisted into making paperless voting machines for just that reason, so any discprepancies cannot be cleared up.

Remember: money important
elections not important

Find out who made these decisions.

[Steve in Seattle]

I should have responded more subtantively in the last post.

WPE as an unbiased measure of anything useful have been called into question by Liddle (search for Febble Fancy Function). After reviewing the USCV research which questioned WPEs, there was no strong evidence indicating that we should reverse what we already know from experience -- exit polls are too blunt of a tool to easily estimate whether fraud occurred in elections.

The undertone of the Edison-Mitofsky report is that they admitted management incompetence at staffing and training their people. Based on the available data, I would tend to agree with their own harsh criticism of their effort.

WRT to error rates of automated methods compared to paper ballots with hand-counts: 3% of rural precincts in the U.S. use paper ballots with hand-counts. The rest of the country doesn't use the method because it is too slow to count votes and error rates increase as the number of counters involved increases. I have not seen a scientific study of error rates with election equipment, especially with equipment over time (some election equipment does degrade). But I would not be surprised that any technology more complicated than an experienced counter counting a small number of ballots by hand is going to have a higher error rate.

Per the Liddle research, the WPE measure behaves as a function of partisan voting. As such, while the true error rate for a given voting technology may remain constant, the WPE will fluctuate with the partisanship of the precinct. Especially when the vote is split deeply partisan.

I am also not surprised that Mitofsky doesn't want to release the raw data. The methodology behind its collection is a trade-secret for the practioners. And I disagree that the precint level exit poll data will tell you anything of substance about election fraud in the 2004 election. The election polling process is too fraught with errors to give any information of substance in a tight election unless the fraud is obviously massive, systemic, and poorly implemented.

[Richard Cownie]

>The election polling process is too fraught with
>errors to give any information of substance in a
>tight election unless the fraud is obviously
>massive, systemic, and poorly implemented.

So are you claiming that it was just luck that
exit polls were much closer to the official
result in all elections before 2004 ? That
seems to be the sticking point: something
happened in 2004 that had not happened to the
same degree in earlier elections/exit-polls.
What was it ?

I think it's very possible that there was indeed
massive and systemic fraud in this election: we
haven't seen enough evidence yet to eliminate
that possibility. From the reports I read, both
the Bush campaign and the Kerry campaign believed
that Kerry had won on the evening of polling day -
which makes me think that two sets of private
polls were confirming what the NEP poll said.

I'll note also that when there was a discrepancy
between exit polls and official results in
Ukraine, the US government took that seriously
enough to push to get the result overturned.
I'm getting sick of American exceptionalism: if
exit polls work in the rest of the world, why
wouldn't they work here ?

>WRT to error rates of automated methods compared
>to paper ballots with hand-counts: 3% of rural
>precincts in the U.S. use paper ballots with
>hand-counts. The rest of the country doesn't use
>the method because it is too slow to count votes
> and error rates increase as the number of
>counters involved increases.

In the UK, it's all paper ballots counted by hand.
Most of the constituencies report their results
by about 3am, and the new prime minister takes
power the next day. What's "too slow" about
that ?

[anonymous]

Last week this was reported online:

OHIO 2004: 6.15% Kerry-Bush vote-switch found in probability study

Defining the vote outcome probabilities of wrong-precinct voting has revealed, in a sample of 166,953 votes (1/34th of the Ohio vote), the Kerry-Bush margin changes 6.15% when the population is sorted by probable outcomes of wrong-precinct voting.

The Kerry to Bush 6.15% vote-switch differential is seen when the large sample is sorted by probability a Kerry wrong-precinct vote counts for Bush. When the same large voter sample is sorted by the probability Kerry votes count for third-party candidates, Kerry votes are instead equal in both subsets.

Read the revised article with graphs of new findings:

The 2004 Ohio Presidential Election: Cuyahoga County Analysis
How Kerry Votes Were Switched to Bush Votes

http://jqjacobs.net/politics/ohio.html

A small spreadsheet too:

http://jqjacobs.net/politics/xls/cuyahoga_t_tests.xls