A Report on Ballot Reconciliation Problems for the 2008 General Election in Montgomery County, Pennsylvania

Research by Stephen J. Strahs and Mavra S. Iano
Written by Stephen J. Strahs
May 13, 2010

Election Reform Network
215-782-8218
www.electionreformnetwork.us

 
Acknowledgments

We wish to thank the Montgomery County Department of Voter Services for their assistance in providing us the necessary access to the voting records at the County Archives.  We are particularly grateful to the staff at the Archives for their help and hospitality over the many days and hours it took for us to complete the work.

Introduction

The Election Reform Network is a non-partisan citizens’ organization based in Montgomery County, Pennsylvania.  Established in 2005, the Network is committed to protecting the integrity of elections and strengthening the democratic process.

Montgomery County in Pennsylvania is the third largest county in the state with voter registration as of the November 08 general election of 584,537 voters.  The county employs the Sequoia AVC Advantage, a direct recording electronic (DRE) full face vote machine for poll place voting.  The machines were the subject of controversy in New Jersey, when in the 2008 primary, it was discovered that they miscounted the number of ballots cast.(i)   Montgomery County is one of 50 of 67 counties in the state using DREs.  The machines, upgraded in 2006 with an audio component for the disabled, have been used in the county since 1996.  Seventy-five additional used machines were purchased for the 2008 election.  

According to the official vote count of the 08 general election, 403,606 voters voted by machine and 23,919 voters used absentee ballots. Total turnout was 73 percent of registered voters, with almost 6 percent of voters using absentee ballots. (ii)

Method

The Election Reform Network performed a study of all election districts (iii) in Montgomery County for the 08 general election to determine if the official count of voters submitted to the state based on the electronic voting machines was consistent with the number of voters appearing at the polls.  This was done by comparing machine counts with the totals for the numbered poll list for each election district (ED).

Because of the very significant discrepancies found in a number of EDs (see below and the excel spreadsheet), we then looked at the actual voter signatures recorded in the alphabetical “district register” - the legal term for poll books - in four EDs.

Key Findings

•    Forty-eight out of the 406 EDs studied (12%)  had significant disparities of nine or more voters between the machine count and the numbered poll lists, while 11 EDs (3%) had disparities of 20 or more voters (see Chart B below).   (iv)
•    Based on the numbered poll lists, Franconia Northeast had a disparity of 47 voters; Whitpain 1 had a disparity of 41 voters; Upper Gwynedd 4 had a disparity of 35 voters; Abington 10-3 had a disparity of 35 voters; and Montgomery 3 had a disparity of 20 voters. (v)
•    304 of the 406 EDs studied (75%) had some disparities between the machine count and the numbered poll lists.
•    Thirty-one of the 304 EDs (10%) with disparities between the machine count and the numbered poll lists had more machine voters than names on the numbered poll list.  The rest, or 90% of the EDs with disparities, had more names on the poll lists than voters recorded in the machine count.  
•    Of the five ED poll books - with actual voter signatures -  reviewed as a check on our poll list data, we found three with poll book signature counts of between 300 and more than 600 fewer signatures than the machine count of voters.  A fourth ED had a disparity of 63 voters in the other direction.
•    The average disparity found between the machine counts and the numbered poll lists was four voters per election district. (vi)

Because of the large discrepancies discovered in some EDs (see above and Summary Chart A below), we decided to investigate an alternative data set, the actual signatures numbered in the alphabetically ordered poll books (PA election law refers to them as “district registers”) for five EDs as a check on our findings.  The EDs for which we reviewed signatures were:  Montgomery 3; Franconia Northeast; Whitpain 1; Upper Gwynedd 4; and Abington 10-3.

For Montgomery 3, with a disparity between the machine count and the numbered poll list of 20 voters, the review of the district register (poll book) found only 1,418 signatures, as compared to the machine count of 2,051. This is 633 voters fewer than the machine count.  

For Franconia Northeast, with a disparity between the machine count and the numbered poll list of 47 voters, the review of the district register (poll book) found only 1,284 signatures, as compared to the machine count of 1,939.    This is 655 voters fewer than the machine count.

For Whitpain 1, with a disparity between the machine count and the numbered poll list of 41 voters, the review of the district register (poll book) found 2,384 signatures, showing more voters signing in than recorded in the machine count.  The machine count was 2,321 voters, or 63 voters less than the number of poll book signatures.  (vii) Both the district register and poll list showed substantially more voters than the machine count.

For Upper Gwynedd 4, with a disparity between the machine count and the numbered poll list of 35 votes, we found the number of voters signed in to be 2,021 (viii).  This number, when compared to the machine count of 1,705, is 316 votes more.  Both the district register and poll list showed substantially more voters than the machine count.

For Abington 10-3, with a disparity between the machine count and the numbered poll list of 35 voters, the review of the district register found the number of voters signed in to be 746, the same number as in the numbered poll list.  (ix)

We have no ready explanation for the large disparities between the machine counts and the signature numbers in the five poll books.  After considering the possibility that absentee ballots and provisional ballot voters were wrongly assigned numbers in the poll books, we determined that there was little chance of this being so, except for the case of Abington 10-3.  First, in three of the five EDs, the poll book signature numbers are much smaller than the machine counts, so inflated numbers could not account for the problem.  Second, if all absentee ballots were incorrectly included in the Upper Gwynedd 4, they would still only account for 110 of the 316 total discrepancy. For Abington 10-3, however, we did see evidence of 20 absentee voters being mistakenly assigned a poll book signature number.  There was a total of 34 absentee ballots in that ED. With regard to provisional ballots as a source of the disparities, there were only 612 provisional applications in the county spread across 400-plus EDs, making it extremely unlikely that they were a factor.  


Chart A: Machine Count-Poll List-Poll Book Comparisons

Chart B: Election Districts With Machine Count – Poll List Disparities of 9 Or More

47 Election Districts with Machine Count-Poll List Disparities of 9 Or More  


Significance of the Findings: A Wake Up Call

There is no single explanation for the disturbing discrepancies between the numbered poll lists and the machine counts of voters or the machine counts and the district register signature counts. However, a full investigation is long overdue.  When the study findings were brought to the attention of the Department of Voter Services, no explanation was offered.

Montgomery County has employed the Sequoia AVC Advantage vote machines for almost 14 years and as far as we are aware, there has never been a systematic attempt to reconcile the number of voters appearing at the polls with the recorded machine counts. The Election Reform Network has called for such a study more than once, including in our General Election 08 Report.

In fact, the Pennsylvania election code requires that counties compare the records of the number of voters with the number of actual ballots cast prior to the certification of election results.  If in any election district the number of votes for an office exceeds the total number of voters, the county is required to investigate.  And until the matter is resolved, the votes from that election district are to be excluded from the official count. Allegheny County is believed to comply with this requirement. (x)

Even if the machines are operating according to standard, our findings indicate, at the least, clear problems with the most basic election day procedures at the polls.  The documentation of who votes is no mere academic question; it speaks to the integrity of the vote and helps determine, in accordance with Pennsylvania election law, who is eligible to vote in future elections. (xi) 

One plausible explanation for fewer votes recorded on the machines than voters present at the polls is the inadequate Sequoia machine interface with the voter.  Often referred to as the “fleeing voter problem,” this requires that after making her  selections, the voter must then  remember to push a “cast vote” button down to the right of the ballot.  There is considerable anecdotal evidence over the years of Sequoia’s tenure in the county that voters do forget to press the “cast” button and “flee” the booth without their choices being recorded.  There has also been much controversy about the proper response of local election officials in those circumstances.  The dilemma is whether, after gaining an agreement between the party representatives, to stick a  hand into the booth to press the “cast vote” button while not seeing the voter’s selections, or void the vote.  Unfortunately, despite their awareness of the problem, the county election board has apparently chosen not to investigate it or make a concerted effort to eliminate it.

Another possible explanation for the loss of votes also relates to the machine interface.  The problem, documented by a Princeton research team, is that voters may believe they are legitimately voting when, in fact, the machine has not been activated and will not record any votes.  This may happen, according to the researchers, because when voters press their selections they light up for about two seconds and when they press “cast vote,” the button also lights up, which may lull voters into believing their votes will be counted.   (xii)

Together, our findings should constitute a “wake-up call” for election authorities to tighten procedures to strengthen trust in the election system by putting into practice rigorous ballot reconciliation techniques that, at the least, can scrupulously compare voter attendance through sign-ins and numbered poll lists with machine counts.  This is an essential but still limited step toward making a more complete reassessment of the existing vote machine technology.  Machines whose results can not be independently verified or independently recounted require that step.

In sum, this report uncovered flagrant incongruities between machine counts and the voter sign-in and record-keeping processes as related to the maintenance of the numbered poll lists and poll books.  Lists were at times non-sequential and difficult to decipher.  Close to 50 election districts had list totals that differed with the machine counts by at least nine voters.  In terms of the district registers, three of the five poll books reviewed for their five election districts had numerical discrepancies with the ED machine counts of more than 300 voters.  The evidence demands further inquiry and a more rigorous approach to ballot reconciliation.

Moreover, when put in the context of some recent elections in Montgomery County with razor-thin margins between winners and losers, indifference to these issues is difficult to accept.  In 2008 a race for state legislature was won by 419 votes.  In the 2009 primary, the difference in the Democratic race for Towamencin tax collector was all of 35 votes.  And in the 2009 general election, the Conshohocken mayoralty was won by a single vote.  

Recommendations

There is no evidence that the County has tried to reconcile these very significant discrepancies and either fix the machines or fix the sign-in and voter record-keeping processes. Until these different counts can be reconciled, the full accuracy of elections in Montgomery County can only be suspect. We therefore call upon the County to:

1)    Undertake a full investigation of the disparities uncovered and draw clear conclusions, to the extent possible about their causes and what measures to be taken to eliminate them.

2)    Institute a regularized procedure for reconciling voter attendance with machine counts for every election done prior to certification of results, in compliance with state election law.
 
3)    Take every step to ensure that the machine counts and the district election registers are reconciled for next week’s primary election before results are official.  

These steps are essential to providing to all citizens and voters in Montgomery County the fundamental assurance that the total number of voters recorded on the machines is consistent with the total number of voters who sign in at each precinct.  Such an assurance is a basic requirement for a fair and accurate elections process.

i iSee New York Times editorial, March 22, 2008, “Unreliable Voting in New Jersey”

ii 2008 General Election November 4, 2008 Summary Report Montgomery County, Pennsylvania Official Report

iii Documentation was only available for 406 of the 423 precincts. No records of any kind were available for eleven EDs in Cheltenham and no numbered poll list was found for North Whales 2 as well as for five subdivided EDs.

iv The total disparity found between the machine count and the numbered poll list as a proportion of the total machine count for the 406 precincts amounts to .4%.

v For Montgomery 3, where we found an initial disparity between the numbered poll list and the machine count of 260 votes, we determined that 240 names on two pages of names on the poll list were not in the district register and may belong to another precinct, thus reducing the discrepancy between the machine count and the numbered poll list to 20.

vi See associated spreadsheet analysis

vii However, even the number of signatures in the poll book requires interpretation. This is so because there were two sets of numbers in the poll book, a clear procedural error. One set had the highest reaching 1,800, while a second set, which had an “A” preceding every number, went up to 584. We determined that the sum of the highest number from both sets was the total number of voters, which amounted to 2,384.

viii Unfortunately, even this number, found on p. 190 of Volume H-R, merits further scrutiny since we saw few numbers in the 1800s, 1900s and 2000s while searching for the highest number (in non-sequential order). It was not possible to count the precise number of signatures in the book and we therefore tentatively accepted, for purposes of this study, the largest number found.

ix There is some question about the signature count in the district register since it was discovered that 20 numbered signature spaces in the ED’s poll book had been stamped “Applied for Absentee Ballot.” If the district register amount is reduced by 20 signatures, the discrepancy between it and the machine count is down to 15 voters.

Moreover, voters using absentee ballots, rather than voting at the polls by machine, should not be assigned a number in the poll book since those numbers are used to account for only voters who voted by machine on that day.

x See Section CS 3154(b) of the Pennsylvania election code

xi See 25PACS 1510 of the Pennsylvania election code

xii For a clear description of the machine interface problem and voter uncertainty with regard to when a machine is or is not activated to process their vote, see Andrew Appel, et al, 2008, Insecurities and Inaccuracies of the Sequoia AVC Advantage 9.00H DRE Voting Machine, Part III: User-Interface Inaccuracies and Insecurities Can Disenfranchise Voters