(This brief post, which made some skeptical remarks on a study of electronic voting in Snohomish county generated many comments, including some attacking the study, and some in defense of the study by one of the authors, Paul Lehto.  One of the commenters, Eric Earling, wrote me asking if there was some way to post a long comment, since he had much more to say.  I thought the best way to do that was to put his comment in a separate post, as I am doing here.  I have converted the file he sent me to html and fixed a typo or two, but everything past these two introductory paragraphs is Eric's.

Since Paul Lehto was kind enough to reply to some of our criticisms, I want to say that I would be glad to give him space for his own guest post, if he would like to continue the discussion that way. - Jim Miller)

In a previous post Jim Miller raised questions about a Lehto-Hoffman study arguing Dino Rossi benefited from vote irregularities on Election Day in Snohomish County.  In follow-up to comments on the original post, including from study co-author Paul Lehto, I am responding to Paul’s defense of his work:

Paul –

Previously, I noted my strong disagreement to your analysis that there was no Rossi surge toward Election Day in Snohomish County.  But, let me first state where I agree with you.  Investigating new technology, especially touch screen technology with the lack of a paper trail, is important to ensure election security and accuracy. Specifically, I agree with your comments:

I hope you all would agree that our elections should not be based on trust of anyone, rather they are in fact based on openness and the right of the parties and the public to OBSERVE.  When vote counting is done with proprietary software, we are left having no observation of the vote, and no verification of it.  No voter really knows how their vote was counted, or if it was.  No party or candidate should have to trust the other party or ANYONE with the integrity of the vote.

That being said, my biggest complaint with the paper is that I believe you are using a flawed example of what otherwise might be a notable technological problem.  Both Jim Miller and you are more of an expert than I on the intricacies of the specific machines and software in question, and means by which one could in theory engage in election fraud.

It appears you’re starting from the position that there are questions about electronic voting, which appear to have merit and should be pursued.  However, you are using a case study that does not appear accurate at all to help validate this starting position.

Specifically, as noted in the original posts on this topic, I disagree with the notion that Snohomish is a traditionally Democratic county.  It has leaned that way at times, particularly when Republicans have put forward starkly conservative candidates, but based on the information I’ve noted above it is not reflexively Democrat by any means.  As I’ve cited in previous posts on this topic, your use of countywide races is flawed given that non-county wide, partisan races have been comparatively more competitive in the “since 1992” timeframe you utilize.  I would like to know for reference the one exception you found to Democrats doing better on Election Day than Republicans, as cited in your paper during that timeline.

Related to that, I think the sample of elections since 2002 for measuring Republican improvement on Election Day is inadequate given variations in the competitiveness of races, particularly when compared to the weakness of Republican candidates appearing on the ballot countywide in the mid-to-late ‘90’s.   As a fellow Snohomish County resident you should recall that comparison.  Also, as I look at your Chart 1, I see a clear trend of increased Republican advantages on Election Day that began in the mid-90’s, well before the 2002 change in technology in question — I’ll delve into that issue in greater depth shortly.

Before that, however, let’s consider the Rossi race shows a statewide Republican candidate performing better than their Republican counterparts up and down the ticket.  We have recent precedent of that in Slade Gorton outperforming all other Republicans on the ballot countywide by narrowly losing the county to Maria Cantwell in 2000, before the change in technology in 2002.  Related to that, I still find the study’s egregious error on the 2000 results in that race, which overstated Cantwell’s percentage of the vote by 10 percentage points, to be a glaring mistake that causes the reader to reflexively question other points you present.

Specifically, let me respond to your points postulating on whether or not a Rossi surge existed based on the data.  Before doing so, let me disclose for reference I’ve been actively involved and/or observing Snohomish County races since 1992.  I worked actively on Rick White’s campaign in 1998, Slade Gorton’s in 2000, and my father Dave Earling’s run for County Executive in 2003.  I also was in touch at length with individuals affiliated with various campaign efforts in 2004.  As you noted in disclosing your work for a Democrat elected official in the past, we can accept partisan work experience as simply experience, not a lens that automatically colors otherwise rational analysis.

1) “Election Day Was Not All Republicans”: As I have noted with Chart 1 in the study, on Election Day partisan strength, there is an increasing conservative trend starting in the mid-90’s.  I experienced this first hand in 2003 in both the primary and the general.

The primary in particular is of note given in the Executive’s race there were 3 Republican and 3 Democratic candidates; my father was the more moderate of the two leading Republican candidates.   He beat the other credible, and more visibly conservative, Republican roughly 51%-37% in the absentees counted on election day, she almost flipped the trend to roughly 37% -51% for poll votes, and later absentee votes came in at ratios between those two splits.  This left a final tally of 47.7%-42.3%, with 10% for the 3rd candidate, in the Republican primary.  One would assume the technology problems you cite for general elections purposes played no role in what otherwise appears to be a trend of conservative strength on Election Day.

In the 2003 general, my father was behind roughly 45%-55% in absentee ballots counted on Election Day.   Yet narrowly won the poll voting, and lost the later absentees by a lesser margin than the first batch leading to a final 48%-52% tally.  While there is some merit to the individual campaign dynamics contributing to this phenomena in the primary and general, the two elections were different enough I think it is safe to anecdotally conclude there was a more conservative poll voting trend in the county that year; which again fits your Chart 1.

Part of the reason for the growing strength of Republicans on Election Day, and increasing strength of Democrats in absentees, is likely in part due to increased voter registration efforts by Democrats and their allies (especially unions) in Snohomish County, and across the state and country for that matter.  I, and other observers, began to see this take effect in the late 90’s and through the 2000 campaign in particular.  Those Democrat-related voter registration efforts have done a better job of encouraging new registrations to vote via absentee, while Republicans have lately focused more on the last several days of the campaign as the peak in their Get Out The Vote efforts.

I think this does much to validate the 2004 trend of early absentees favoring Democrats in Snohomish County with comparative Republican strength on Election Day, and a trend somewhere in between in late absentees.

2) “There Was No Late Rossi Surge”:  Again, I disagree on your use of CNN exit polls (derived from the media consortium that had notable problems not only in 2004 but also in 2002, let alone 2000 when it was in different form).  Much like the study’s error in the Gorton/Cantwell race, the lack of credible sourcing on that issue casts doubt on the validity of your argument.

I again refer you to Survey USA polling for more reliable data.  Their website has a glitch which is currently omitting their October polling, but does have an interesting spreadsheet showing their accuracy in 2004 compared to other polling firms.  Rasmussen and Mason-Dixon were also quite accurate compared to their peers but I don’t recall them doing as much regular public polling on the Governor’s race in WA.  I assume you could track down the public polling done by Survey USA from that company or from KING in Seattle or KHQ in Spokane that paid for and released the statewide polling.  I vividly recall not only Survey USA showing the gap between Rossi and Gregoire closing, even flipping slightly in the closing weeks, but Mason-Dixon polling done for KOMO also showed a similar trend — as I found in old KOMO news stories available online.  The following site may be of value in examining polling showing Rossi gaining the last several weeks.

Also, in looking at other counties showing a Rossi surge, I would suggest sticking with larger counties like Snohomish that have voted both ways in statewide races since 1992.  This would include Pierce, Spokane, and Kitsap.  Jefferson is a solidly Democratic county and thus not easily comparable.  Meanwhile, neither Whatcom nor Skagit are similar enough in their size or countywide demographics to make them comparable either.  Whatcom in particular has a unique demographic group in students at Western Washington University, where Democrats have done extensive registration work and where, as I experienced in 2000, those students (leaning strongly Democratic) voted notably close to or on Election Day.

Meanwhile, I recall seeing voting trends in Pierce, Spokane, and Pierce, perhaps not as notable as Snohomish, but similar nonetheless, between the absentees counted on Election Day and poll voting results — with later absentees again falling in between the two ratios.  Obviously these counties used a variety of poll voting technology.  I recall the surge between Rossi’s vote totals in the absentees counted on Election Day in King versus poll results was also highly statistically significant.  All of this at least indirectly supports and is consistent with my theory described above as to why Democrats have been improving in absentee voting while Republicans are showing increased strength in votes cast at later dates, including Election Day.

Lastly on this point, I should note it is unreliable to compare what the Snohomish County Auditor claims as the number of ballots returned on a certain date and what the state is estimating as the number of ballots returned.  First, there are too many variations in absentee voting (both as a percentage of overall votes and in rates of ballot return) in various counties to make this comparison valid.  Moreover, in my personal experience with the Snohomish County Auditor’s office in 2000, 2003, and 2004, they never had an accurate tracking system to say how many ballots they had received, or remained to be counted after the election.  The disparities between their claims and the actual number of ballots they ended up counting were always disturbingly high and inconsistent.   This lack of precise tracking has been raised by others in faulting King and other counties during the current post-election mess and defiantly deserves further attention as part of broader election reform.

3) “Democrats Did Not Avoid Touch Screens in Large Enough Numbers”: I think your point here asks the wrong question, or at least approaches the issue from the wrong angle.  As I’ve noted, the data indicates, Democratic strength in the absentees has steadily grown, notably in conjunction with Republican strength in poll voting.  I would argue that Democrats are not necessarily avoiding touch screens, rather in the aggregate they transitioned increasingly to the use of absentee ballots as a positive choice beginning in the mid-90’s, not as a negative reaction to recent touch screen technology.

As a final point, I should note you discussed the positive bump up President Bush received with poll votes, consistent with what I’ve argued is a growing conservative trend on Election Day.  In the aggregate, President Bush’s improved performance in 2004 (45.47% in Snohomish) v. 2000 (43.61%) is consistent with his trend of increasing his vote share in most states across the country.  Thus, not indicating improved Republican strength on Election Day in Snohomish County led to results inconsistent with other results across the country or state that obviously utilized a multiplicity of voting technologies.

4) “Democratic Absentees”:  I’ve noted above at length why I think it is reasonable to postulate Democrats have been increasing their relative strength in absentee voting in Snohomish County.  As that party and its allies have improved registration efforts based on encouraging absentee voting, Republicans since the 2000 cycle have focused heavily on the “72 Hour effort” before Election Day as well as utilizing improved databases and voter contact efforts in those closing days.   Resulting Republican strength in recent years during the later days of the election is thus rational.  Moreover, your concern over the two peaks in the electronic voting also raises a question that deserves further consideration.

That database work I’ve mentioned by the GOP has focused at length on specific demographic groups that tend to cluster in certain communities, especially “ex-urbs” and other new, relatively affordable, single-family housing.  At one point you said the twin peaks might be explained by unequal precinct shifts on Election Day.  Based on the database targeting I’ve noted above, combined with a late-in-the-election focus by Republicans on voter turnout, such variances could explain the twin peak that concerns you.  Moreover, consider that in 2000 74.94% of 338,000 registered voters in the county cast ballots.  In 2004, 84.31% of the 352,000 registered voters cast ballots.   Considering natural turnover in registered voters, the overall increase in the number of registered voters, and the increase in overall voter turnout, we’re talking about a notably different county electorate in 2004 than 2000.  Furthermore, remembering much of the growth in the county has occurred in the aforementioned “ex-urbs” targeted by Republicans, this is further evidence in support of a plausible and explainable Rossi surge on Election Day.  Based on the variety of post-election analysis of voter turnout patterns, demographic targeting, etc. across the country in the 2004 cycle, I would argue that theory deserves further examination, especially since it contradicts your argument against a Rossi surge.

Also, the copy of your paper Jim Miller linked to appears to jump from footnote 9 to 11, thus omitting footnote 10 which thus leaves me unable to view your proof that Gregoire had a 5-10 point lead going into the election.  The link I gave you above indicates that was not the case.  Related to such polls, I obviously can’t provide the Rossi campaign’s internal tracking polls, but can vouch for the fact they showed steady gains over the last three weeks, including an actual lead the last few days of the election.  I simply offer that as anecdotal support to the public polls that show the gap narrowing statewide as the election approached.  I note the next point understanding that one can’t use it to reliably prove one point either way: yet, using the very rationale in campaign polling you noted in the study, one would assume the Gregoire campaign would have been trumpeting their own polling that wasn’t showing a Rossi surge — if such polling existed, though I doubt it based on the other evidence available.  Whether you accept that idea or not, it’s an interesting point to consider.

Clearly, I disagree with your hypothesis that there is no explanation for Rossi’s Election Day success in Snohomish County.  You have raised some intriguing points regarding potential flaws in touch screen voting.  I’m not expert enough to say their valid, and I don’t think even if they are they valid that they swung the poll voting they way you argue given the other points I’ve raised.

Specifically, you’ve cited the supposed one in a trillion chance that the Snohomish absentees vary as they did with the poll votes.  I fundamentally disagree with insisting that voting patterns have to fit mathematical models.  In fact, I differ mightily with some contributors to Sound Politics who have argued that deviations in mathematical vote tracking models are evidence implying fraud.  While the human errors in this election (and the registration process beforehand) have obviously been many — and in my opinion turned the election from Rossi to Gregoire — voting trends are not just numerical patterns, they include emotion of the voters in the aggregate as well.  Political observers of all partisan backgrounds have noted how Gregoire was fading at the end, while Rossi’s momentum continued to build — all for well-publicized reasons more lengthy than I should articulate here.

Whether you agree with my analysis or not, I believe I’ve put forward a significant number of plausible and consistent reasons as to why the vote totals generally broke the way they did; and those reasons include voter decisions based on emotional response to the closing days of the campaign.   To be fair, I haven’t done the time-consuming number crunching to back up each of my points in detail, but in fairness, your study is the one that should offer clear and convincing proof of your case study.  It was not difficult to find logical reasons to explain the data in question, reasons which you have not addressed, and in many ways contradict your assumptions.

I wish you the best of luck in your further investigations into the true accuracy of this voting technology.  It is a task that must be done to ensure valid, reliable elections.  But in this case, your chosen case study is too deeply flawed in its starting points to support what may in the end be a valid thesis regarding the technology’s flaws.