The Miami Herald Overvote Analysis, Part 1:
by Paul Lukasiak, with Maia Cowan and Nina Mollett
This time, the Herald did far more than simply "spin" its results from its examination of the Florida overvote. It based its conclusions on a wholly unreliable data set, ignoring votes on hundreds of ballots and distorting the nature of votes on thousands of others.
The following analysis details the extent to which the Herald's numerical data is unreliable, and shows how the Herald misrepresented that data. In almost all cases, the inaccuracies and misrepresentations favored the conclusion of net gains for George W. Bush. What the data shows (when it is adjusted for errors made by the Herald where possible), is that under all possible recount scenarios Al Gore received the most votes in Florida.
Among the misrepresentations and inaccuracies, the Herald:
The Miami Herald (a Knight-Ridder newspaper), in its on-line edition, billed its overvote analysis as the "first statewide review of overvotes".(4) Merzer claimed in the story he provided to other members of the Knight-Ridder syndicate that every ballot was hand recounted: "[a]nd now, after an exhaustive hand recount of every Florida ballot..."(5) This statement is false. More than 97% percent of the ballots — all those not counted as overvotes or undervotes by machines — were never examined by hand. Most of the overvoted ballots were not "hand recounted." The implication in the story that circled the globe based on the Herald's review, however, is that every single ballot, certainly every ballot missed by the machines, was examined by hand.
The Herald's analysis of undervotes was conducted by an independent accounting firm, and two people examined each ballot. For the overvote analysis, by contrast, one newspaper employee examined the ballots in each county. As will be demonstrated below, no assumptions should be made that the observations of these reporters were either partial or reliable.
The Herald's methodology for the overvote review contrasts with both the forthcoming review by the NORC (National Opinion Research Center) Consortium of all of Florida's undervoted and overvoted ballots, and the review by BDO Seidman commissioned by the Miami Herald and USA Today. (Although the Herald's analysis of the BDO Seidman undervote data was seriously flawed and biased, there is at this point no reason to question the accuracy of the raw data provided by BDO Seidman.)
Total certified by the Florida Secretary of State + Recount totals ordered included in final tally by FSC Pre-undercount baseline
The overvote analysis uses an entirely different, and extremely confusing, set of baseline numbers.
For its overvote review, the Herald starts with a number that it represents as the results of the mandatory machine recount from November 8th.(6) The Herald claims that this number does not represent any "manually recounted" ballots. In fact, what the number represents is the vote total from the initial (November 15th) statewide certification of votes MINUS the results of the Volusia County full manual recount. (The Volusia County results were included in the November 15th certification.) As will be explained below, this number includes over 10,000 ballots that were recounted by hand in many Florida counties using the same "clear indication of voter intent" standard used in manual recounts in Volusia (as well as Broward, Palm Beach, and Dade Counties.)(7) This number also includes a net loss of 90 Gore votes that resulted from a partial manual recount in Polk County.(8)
In sum, the Herald starts out by excluding some legal, certified votes from its totals because they were hand counted, includes 10,000 other such votes, and further reduces Gore's vote totals based on the results of a partial manual recount in Polk County. This selective inclusion and exclusion of ballots in the Herald's initial "baseline" provided the erroneous totals on which the Herald based its conclusion that Bush might have won under certain standards in Florida.
Then a third adjustment is made to the Herald's numbers. After the machine recount, Orange County reported a total of 966 undervoted ballots. When the Herald initially examined the undervoted ballots, Orange County election officials could only account for 691 of these ballots, a difference of 275 ballots. After the Herald finished reviewing the undervotes, Orange County reported it had found that 433 additional undervotes had been incorrectly identified during the separation process as counted by machine. That means that Orange County is claiming that there were 158 more undervotes than it originally reported. The Herald never inspected these ballots and does not include them in the numbers it reports.
The net result of all of this "fuzzy math" was a "new" baseline that gave George W. Bush a net advantage of 1,133 votes, supposedly excluding undervoted and overvoted ballots. In theory, this number represents all of the ballots that were already counted and therefore not subject to the undervote/overvote review.
The Herald table then claims to add the results of its previous undervote study to its current overvote study, and declares "winners" using four different "standards." What the Herald does not tell you (unless you happen to run across a small, obscure article that accompanies the main stories and tables) is that one final adjustment was made. Over 1,000 of the votes found in Broward County during its undervote study are eliminated from the totals (see below).
Confused? You should be, because the Herald's choices about which ballots to include and exclude in the baseline are based on criteria that were inconsistent. The Herald ignored evidence of legally cast and counted votes, but that included votes that might not even have existed, while at the same time excluding votes that it had already counted.
According to the Herald, the totals from their undervote count "were based on ballots reviewed by the Broward canvassing board, and included more than 1,000 ballots that contained clean punches. Broward officials were uncertain that those ballots hadn't also been included in the machine counts."(9) For its totals, the Herald treated this "uncertainty" as a fact and excluded all the cleanly punched ballots it counted as part of its undervote study. This exclusion decreased Gore's net gain by 531. That 531 is 120 votes more Bush's lead under the criterion by which the Herald said that George W. Bush gained the most votes (the "two-corner chad" criterion). It is more than triple Bush's lead using the "clean punch" standard.
In other words, the Herald presents Bush victories as "certain" even though they were based on numbers the Broward County officials reported as "uncertain." The exclusion of some of these votes may be justified because the data suggest that most of these votes were included in the machine recount and were, in fact, not part of the undervotes At the same time, the data suggest that hundreds of these votes should not have been excluded.
The original undervote count by the Herald included 779 more ballots than Broward reported after its machine recount. During its undervote review, the Herald found an abnormally high number (1,052) of ballots with clear punches for candidates among those "undervotes." Thus it appears likely that some machine counted ballots were included in the "undervote" count.
Rather than adjust the totals to reflect the 779 additional ballots, the Herald chose to completely eliminate all 1,052 cleanly punched ballots from its calculations, leaving 273 ballots completely unaccounted for, and resulting in a net loss of 531 Gore votes.
It has been well established that "hanging" chad become completely detached during the handling and separating process. The review of virtually every punch card county found ballots with fully detached chad among those ballots separated as "undervotes," and there are thousands of unaccounted for "undervoted" ballots in the original Herald review whose "unaccounted" status is readily explained by this phenomenon. In addition to being counted by machine twice, and counted manually, the Broward ballots were once again separated by machine. It is highly likely that these 273 ballots did, in fact, represent legal votes that had been erroneously counted as undervotes in the machine recount — that they should have been included in a review of the undervotes. Nevertheless, the Herald decided to exclude all the cleanly punched ballots it found in its undervote totals from the data on which it based its conclusions in its overvote study.
Gore had a three to one margin in both the total vote from the ballots, and among the "cleanly punched" chad found in the Herald's original undervote review. Thus, it is reasonable to assign these 273 ballots on that basis. Doing so provides Gore with a net 140 votes over the numbers used by the Herald.
In sum, the Herald, faced with the decision of whether to include a category of ballot, makes the choice that benefits Bush but not Gore. There was some justification for excluding some of these ballots from the totals, but in excluding all of them the Herald demonstrated that its methodology not be trusted. Even more important, however, is that in obscuring the use of data that was different from its original undervote review, legitimate questions are raised about the integrity of the entire review.
The reason for hiding this adjustment may be pure pro-Bush bias, or an attempt to hide the flaws of its original undervote review, or a combination of both. But for the purposes of this article, the reason is not nearly as important as the fact that there is clear evidence that the Herald numbers cannot be trusted.
Not all of those votes, however, were from the pool of undervoted and overvoted ballots. 320 ballots were completely missed during the November 8 machine recount. These 320 ballots were (with a few possible exceptions) fully machine-countable, not overvotes or undervotes. They just did not get put into the machines to be counted. The new Herald overvote review included only the machine-counted ballots plus the overvoted ballots. Thus the machine-countable votes that were recovered from those 320 ballots during the manual recount are omitted from the Herald totals.
The previous review of the undervoted ballots gave Gore a net gain of 33 votes. The net gain of votes for Gore from the overvote count was 12, for a total net gain for Gore of 45 votes. By excluding those 320 votes, the Herald underreported Gore's net gain by at least 53 votes. The actual underreporting of net Gore votes may be 86, because the 33 undervotes found by BDO Seidman may not have been part of the "recovered" votes from the Volusia full manual recount. Also, as noted above, 488 overvotes were originally reported by Volusia County, yet the Herald based its overvote totals on an examination of only 154 ballots. The Herald does not explain the discrepancy between the county's number and the newspaper's.
This result is contradicted by the Orlando Sentinel review of the same ballots, which found 142 ballots (with a net Gore gain of 122 votes) in Gadsden that met the "corrected" criterion for valid overvotes used by the Herald in its overvote analysis ("Bush or Gore oval marked and ovals for other candidates marked and then crossed out"). The Sentinel defined these votes as ballots showing "multiple errors on rejected ballots with 'Bush' or 'Gore' written in"; the example provided with the Sentinel report was completely consistent with the Herald's definition.(10)
In its report, the Herald does not comment on this unlikely and unprecedented result.
The Herald stated that there were no Lee County ballots in either the "write-in/name" or "write in/blank" categories.
Seven votes are not many. The Herald, however, also failed to find a single such vote in any other punch-card county in which it was possible to punch a hole adjacent to the write-in instruction (in most punch-card counties there was no hole next to the write-in instructions; a hole serves no useful purpose for write-in votes and the law specifies it is not required).
Duval County is one of the counties that did provide a hole for the write-in space. According to the Herald, 84% of the nearly 22,000 overvoted ballots in Duval had votes marked on both the first and second page of Duval's two-page presidential ballot. The names of Al Gore and George W. Bush appeared on the first page of that ballot. The "write-in" hole appeared on the second page. There is no way to know how many ballots in Duval County were miscounted as overvotes even though they were legal prescribed as valid votes. Nevertheless, the Herald assertion that not one of the more than 18,000 overvoted ballots were false overvotes — a listed candidate chad detached from the first page and the write-in chad detached on the second — is not credible.
Judging by the Orlando Sentinel report on its own ballot review,(14) the Herald's baseline total and "late absentees" total include votes on more than 10,000 ballots that were rejected by tabulating machines, evaluated by county canvassing boards, and included in the machine-recount totals from November 8.(15) Some of these were votes that were "eaten" by the voting machines, and others were unreadable because of the way the ballot had been folded. Some of these ballots, however, were unreadable because of voter error. Sometimes the voter used the wrong pencil (creating a mark the machine could not detect), and the votes were initially counted as undervotes by the machine. Other ballots were from voters who filled in the ballot wrong, resulting in either overvotes or undervotes.
Most of these "recovered and manually counted" ballots were added to the certified totals in Republican-dominated counties. Most, in fact, were absentee ballots in precinct based optically scanned counties where George W. Bush won the absentee vote by a 2-to-1 (or better) margin.
After deciding what votes the rejected ballots contained, the officials in these counties created duplicates that were fed to the machines. Under Florida law [FS 101.5614 (5)](16) the standard used for counting votes when duplicate ballots are created is the "clear indication of the intent of the voter as determined by the canvassing board" standard. In other words, over 10,000 ballots were recounted by hand under the "voter intent" standard that were included as part of the baseline numbers that the Herald claims represents votes counted by machine. These duplicate ballots were counted by machine, but only after the ballots actually cast had been reviewed by the canvassing boards under the "voter intent" standard.
Unfortunately, the Sentinel article does not indicate what the original 10,000 plus ballots showed in the Presidential race. Doubtless, on some of these ballots the presidential vote was machine readable but rendered unreadable when the machine mangled the ballot. One other ballots, the unreadable vote may have been in categories other than President.
Under the statute, these duplicate ballots should not include any invalid votes. Thus, there are certainly many cases of "undervoted" and "overvoted" ballots that show up in the machine totals as either completely unambiguous votes, or as "no vote" for President.
It should also be noted that in duplicating these optically scanned counties, the election officials in these counties were in clear and unambiguous violation of the statutes. Under the statutes, ballot cards (punch card ballots) must be duplicated if "damaged or defective" and the names of the candidate does not appear on those cards. Where punch card are used and the name appears on those cards, the county officials can either duplicate those cards, or count them by hand. But when "any paper ballot [e.g., optically scanned ballot] is damaged or defective so that it cannot be counted properly by the automatic tabulating equipment, the ballot shall be counted manually at the counting center by the canvassing board."
According to the Orlando Sentinel, at least one county (Bay) has refused access to the original ballots on which their illegally duplicated ballots are based.
As noted above, an indeterminate number of these original ballots probably contained unambiguous votes for President that would have been counted by machine were it not for other circumstances. But because the Herald never bothered to review the original, hand counted ballots (and because the Sentinel has not provided a breakdown of what is on the original ballots) at this point we have no idea what appeared on the ballots that were actually cast by the voters.
The significant point is that the Herald misrepresents its data when it states that its baseline numbers do not include any ballots that were hand-recounted. In fact, the baseline number from the November 8 machine recount definitely includes the ballots that were hand-counted in Seminole, Polk, Taylor, Hamilton, and other counties. Nearly all the manual counts resulted in net increases in votes for Bush.
The Sentinel article focuses on duplicated ballots, and does not discuss ballots that were, as prescribed by law, manually counted but not duplicated. There is a very strong possibility that other counties followed the prescribed procedure and included non-duplicated hand counted ballots with overvotes in their certified totals, and that these ballots were among those separated out and recounted by the Herald. Certainly, the Herald makes no mention of any procedures used in its overvote review to guard against such an eventuality.
The Herald's failure to acknowledge that hand-recounted ballots are included in the baseline is significant because it undermines the Herald's own criteria for determining who might have received more votes.
These manual recounts have a significance beyond the media ballot reviews: If not for the selective hand recounts and duplication of unreadable ballots by a few majority-Republican counties, George W. Bush would have lost the first machine recount and the final certified tally.
Although the Sentinel does not provide a full breakdown of these ballots, it gives intriguing hints as to how election officials in Republican-dominated counties manipulated the vote counting to maximize the Bush advantage. For example, in Escambia County, which went for Bush by a margin of 64% to 35% (the absentee margin was even greater, 72% to 27%) ballots are scanned at the precinct, not at a central location. Escambia County officials chose to disable the machine feature that identifies ballot errors and allows the voters to submit a corrected (countable) ballot. Escambia's overvote rate was ten times that of counties that provided voters with overvote protection.
Although Escambia County did not protect citizens from overvoting at the polls, officials went to great lengths to make sure that the intent of every absentee voter was reflected in the county totals. More than 2,400 absentee ballots were duplicated because the machine could not read the original. Given the almost 3-to-1 margin for Bush in Escambia's absentee ballots, it is easy to see how the decision to protect absentee voters from rejected ballots resulted in an advantage for Bush of hundreds of votes that would not have been counted had they been treated the same way as votes cast in the polls.
Conversely, the Herald recount totals show that, had voters at the polls been given the same consideration as absentee voters, Al Gore would have received a net gain of 157 votes — a 2-to-1 overvote margin for Gore. These lost Gore votes comprise less than 12% of the overvoted ballots in Escambia County. By extrapolating the results of the 12% of overvoted ballots to the pool of overvoted ballots in Escambia, it is easy to see how the decision not to provide overvote protection at the polls could have, by itself, cost Al Gore the election.
None of these factors, however, are mentioned in the analysis of the overvote. The Herald blames voter error for Gore's "defeat," ignoring the decisions made by county and state officials.
One can only assume from past articles that the Miami Herald editorial board believes, for whatever reasons, that it cannot run the headline that Al Gore would have won any fair recount in all counties — indeed, by any possible consistent standard, he should have won the mandated second machine count on November 8. The most charitable explanation of this reluctance is that the Herald might regard such a result as potentially destabilizing to the country, given that George W. Bush has been installed in power. It may be, too, that political or economic pressure has been brought to bear on the newspaper. For whatever reasons, the Herald has abdicated its journalistic duty to present the unbiased truth. Their "inconclusive" results on the recount have been achieved by using false data. Hundreds of votes were omitted. Others may have been double counted. Thousands are mislabeled. Unable to sort out the confusing spin of numbers upon numbers presented in the Herald report, the average citizen will not question their accuracy or the validity of the conclusion. But if you follow the ball, bouncing all over Florida, the truth emerges.
Part II will examine how the Herald's narrative was biased toward creating the impression of a Bush victory when, in fact, all the evidence shows that Gore got the most legal votes in Florida.
Note:In the interest of fairness, I wrote to Martin Merzer, the primary author of the Herald analysis, asking for comment. Below (17) is my letter to him, his response, and my response to that.
Attached is a (not quite final edit version) critique of the methodology used by the Herald and its partners in its overvote analysis.
A number of on-line journalism sites have expressed an interest in the piece, based upon its first draft. I wanted to give you an opportunity to respond to it, and I will be happy to see to it that your response is included on sites where it is published. Please note, however, that it may well be published on-line prior to your responses, inasmuch as you have, to date, refused to respond to two previous inquiries regarding your articles on this subject.
First of all, we're having trouble sending email back to you - it comes back as undeliverable. Second of all, you're already posted this. Third of all, if this gets through, here is our response, which I hope you include in its entirety, including the websites:
We stand by our methodology, our ballot reviews and our reporting. Anyone interested in examining our calculations in depth can find supplemental material on these websites — material that was printed in The Miami Herald on Friday, May 11.
Dear Mr. Merzer:
First off, please allow me to apologize if mail was bounced back to you. Apparently, at some point on Sunday evening (until about 8:00 a.m Monday) the e-mail server was having problems.
Re: the substance of your response, in fact it was the web sites that you cited that I used as the basis for my critique.
Inasmuch as your overvote analysis acts as an indictment of the methodology in your undervote analysis (forcing you, in the undervote analysis, to add 400+ undervotes from Orange County and eliminating over 1,000 undervotes in Broward County) I would suggest that your overwhelming confidence in the overvote survey methodology may be misplaced somewhat.
It is clear that your approach missed the votes recovered in Volusia County in the manual count as a result of 320 ballots not counted by machine in the "machine recount".
It is clear that your approach missed ballots that other people found in Gadsden County.
It is clear that your approach did not find seven ballots with fully legal votes that the supervisor of elections there personally examined and that were in the pool of overvotes you supposedly "examined" (And don't take my word for them being legal votes. Read the Orlando Sentinel accounts on these votes. Or you can take the word of the lawyer for the Florida Association of County Supervisors of Elections. Or do what I did, and call up a Florida law school (I used the University of Florida), ask for the leading expert there on Florida election law, then ask him or her about the status of these votes. Or you can just read my piece cited in the previously attached article that lays out the relevant statutes and codes, and read them for yourself, and draw your own conclusions.)
It is clear that your approach mislabeled as "machine counted" votes that were first counted by hand, then duplicated based on a "voter intent" standard.
In other words, it is clear that there are serious problems with your numbers — or at the very least, serious problems with the way in which you explained how those numbers were arrived at.
In any case, by addressing the issues I raise, you can clear up ambiguities in your report, or acknowledge errors made, thereby enhancing the Herald's (and your own) reputation for integrity. Conversely, your unwillingness to address the specific issues raised in my critique will merely serve to raise further questions about the integrity and impartiality of the Herald's overvote analysis.
|Copyright © 2003 by Paul Lukasiak|
Permission is granted to link to this page or to reproduce the contents. In fact, we think the information is so important we encourage you to spread it far and wide as fast as you can.