Comments on: Further Analysis of Electronic Voting Patterns

By: Mike Hout

Mike Hout — Thu, 25 Nov 2004 18:17:21 +0000

Colleagues, We do not claim that our results show cheating. I cannot think of a way that stats could be definitive on that question. We have tried to stick to words like “discrepancy” and “statistical anomaly” to communicate how open to interpretation these results are. Personally I suspect hardware or software errors. In interviews and public comments since seeing Andy Gelman’s plots, I have stressed that Broward and PB are key to the results. This brief note does not address all your concerns, I know. We are working up an FAQ and blog site to help us manage the flood of correspondence we are getting. It is too soon to predict when it will be working.

By: Alex

Alex — Wed, 24 Nov 2004 00:06:17 +0000

esp, ***As for their analysis, I’d give it high marks if it were an undergraduate paper, lower marks if it were a graduate paper, and I must consider it borderline delusional as a research consortium effort.*** Exactly! If this was an undergrad paper, I’d say, great job. :))

By: Jim Miller

Jim Miller — Tue, 23 Nov 2004 01:17:24 +0000

Curious what different patterns different people find. I saw the surge in votes for Bush this year in Palm Beach county as evidence for fraud in 2000. You can find a post on my site explaining why, if you are not familiar with that argument. By the way, strong Republican votes in Palm Beach are not unprecedented. I believe George H. W. Bush carried the county in 1988. That’s one of the many, many reasons Republicans (and open minded Democrats such as Pat Caddell) suspect fraud there in 2000.

By: Craig Newmark

Craig Newmark — Sun, 21 Nov 2004 19:03:28 +0000

I think commenter tomf hits the nail on the head. The model doesn’t account for a possible switch in the Jewish vote, a vote which is exceptionally large in those three counties. If variables to control for the number of Jewish residents are included in the model, the electronic voting variables become insignficant. For details, see the post on my blog, http://newmarksdoor.typepad.com/mainblog/2004/11/more_on_the_stu.html

By: esp

esp — Sun, 21 Nov 2004 18:40:10 +0000

I’m reminded of a time in the late 60’s during my graduate studies in statistics when our campus hosted a visit by a noted non-parametric statistician. He talked about his analysis of a particular data set and told us that his findings had hardly been noted by the academic community, but that his data had been quoted repeatedly. I am now retired from teaching, but if I were in the classroom you can bet my students would all be taking a crack at the Berkeley data. I very much appreciate the UC group’s making the data available to us. From the looks of this blog we are about to wear it out. As for their analysis, I’d give it high marks if it were an undergraduate paper, lower marks if it were a graduate paper, and I must consider it borderline delusional as a research consortium effort.

By: Alex

Alex — Sun, 21 Nov 2004 14:24:46 +0000

Unfortunately, I think the Berkley study is flawed. It does not account for the increase of Republican voters between 2000 and 2004. See http://alex.strashny.com/2004/11/electronic_voting_machines_florida_and_bush_votes_in_2004.html

By: Anthony

Anthony — Sun, 21 Nov 2004 05:54:52 +0000

Dan Weintraub, the “California Insider“, quotes an email which points out that there is an alternate explanation for the results: Democratic party cheating with punch-card ballots depressed the Bush vote in 2000 in the two or three counties with the biggest variations, and such cheating was impossible using electronic voting machines.

By: Kieran Healy

Kieran Healy — Sat, 20 Nov 2004 22:16:25 +0000

What exactly is a “working paper”? It’s a paper that is paying its debt to society rather than leeching off of the teat of the state. [Try the veal, I’m here till Thursday.] It’s just a paper circulated for comment and subject to revision. It’s not your first draft, but it has yet to go through the peer-review process and get published in a journal.

By: msk08

msk08 — Sat, 20 Nov 2004 22:09:02 +0000

What exactly is a “working paper”?

By: TomF

TomF — Sat, 20 Nov 2004 19:17:16 +0000

I am almost certain that, immediately in the wake of the election, someone said/wrote that Bush had way outperformed his 2000 vote in Broward and Palm Beach, and that some of this was the result of a swing in support from Jewish voters because of Isreal and the mid-east. Given what’s said above—that the entire shift is explained in Broward and Palm Beach—I’d look at that possible explanation, first.

By: Zaoem

Zaoem — Sat, 20 Nov 2004 16:24:20 +0000

You probably can’t do the analysis on the precinct level because precincts are usually not constant across elections.

By: tib

tib — Sat, 20 Nov 2004 14:38:01 +0000

yami, FL makes precinct data available by county. I am not sure all counties have posted results (FL Dept. of State has links http://enight.dos.state.fl.us/). Also, ncec.org collects precinct level results for most races.

By: yami

yami — Sat, 20 Nov 2004 05:13:44 +0000

I know it would be a huge pain to collect the data, but other than that, why isn’t this being done on a district or precinct level? Seems like you could get better demographic controls that way.

By: ogmb

ogmb — Sat, 20 Nov 2004 03:54:06 +0000

Still didn’t read the study (and increasing thinking it might not be worth it), but anybody who wants to analyze Optical vs. eVote results needs to restrict themselves to counties with 40,000-600,000 registered voters. The smallest county with eVote is Sumter (40,532 reg. voters), the largest with OpScan is Orange (531,774). I don’t know how anyone would want to draw inferences from counties outside this range, as they have no approximate match in the respective control group.

By: A Simple Citizen

A Simple Citizen — Sat, 20 Nov 2004 03:51:07 +0000

This was what I posted on another Blog when we were talking about Broward on Nov. 10, 2004. I don’t know when these numbers were posted on Broward’s Election site, but this is what it was when I looked at it 11/10/04. 2000 (66.23% turnout) 588,107 total ballots cast—563,846 for B & G B: 177,328 (30.93%) G: 386,518 (67.42%) 2004 (still unofficial) (68.75% turnout) 709,630 total ballots cast—698,455 for B & K B:244,638 (36.61%) K:453,817 (64.21%) B: +67310 K over G: +67299 134,609 ballot difference from 2000-2004. Each received 50% of the total. I’m just sayin’ it’s a little too coincidental. Thanks for letting me post.