Cleveland Study on Registration Errors

A new study projects that 10,000 votes may be compromised in Cuyahoga County and some 35,000 votes statewide because of clerical and voter errors, unless we take immediate action.

Early in 2004, volunteers for the Greater Cleveland Voter Registration Coalition began hearing from some citizens that they had never received confirmation of registration from the Cuyahoga County Board of Elections. Fortunately, the Coalition had copied every single registration form before they handed them in to the Board of Elections. Two preliminary follow-up studies on 275 registrations confirmed that there was indeed a problem. The next step was to compare nearly 2500 submitted registrations to those listed as “active” voters on the Board of Election rolls. The initial result was that about 13% were not found on the rolls, 6 weeks or more after submission. The Board next kindly gave us access to the detailed records on these registrations, which revealed that about half of the 13% had been legitimately deleted because of death, voter moving out of county, imprisonment for felony after registration, lack of citizenship, duplication, etc. Other supposedly missing registrations arrived later on the rolls after the Board caught up with its backlog, even though they had been submitted months before.

We next traced the fate of the remaining problem submissions and found that that over 5% of the nearly 2500 registrations were not on the rolls at all, incorrectly entered, not updated for address change, or suffered from voter errors (See detail below.). In most of these cases, the voter would have no way of knowing they were not registered or were registered with errors until election day, unless they happened to check with the Board of Elections. Also, after subtraction for duplicate registration forms and registrations legitimately cancelled, the final number analyzed was closer to 2200 (see below).

By knowing that at least 5% of voters might have their votes compromised, and that the Cuyahoga County Board of Elections had received about 200,000 applications, we computed that about 10,000 votes in Cuyahoga County would be compromised. To determine the statewide numbers of received registrations, in early September, we called larger county Boards of Election to ask for their numbers of new 2004 registrations/change-of-addresses, until we had numbers from Boards that collectively handle 48% of Ohio's 2002 registered voters. From those total numbers, we projected that about 760,000 new cards had been submitted statewide since January 1, 2004. Since some of these would be later legitimately disqualified (deceased, moved out of county, imprisoned, duplications, etc), we estimated that 5% of 700,000 or 35,000 votes statewide would be compromised if the results found in Cuyahoga County apply to registrations statewide.

Details of Study:

Our team spent at least 60 man-hours working on this study. We actually found a total error rate of 7%, but reduced it to "at least 5%" for public consumption in case we had made any errors. The Cuyahoga County Board of Elections has received our information but as yet has not checked it. Although there may be some changes in the following numbers after they are reviewed by the Board of Elections, and although some of the errors could arguably be moved from one category to another, we believe the conclusion is firm that about 10,000 Cuyahoga citizens are at risk for having their votes compromised.

Of this 7% of errors, the majority (n=104) were due to clerical error in the following categories:
No record found although the card was definitely submitted (13)
Entry errors of name spelling (16), address (41), or date of birth (9): total (66)
Address change on registration card not updated, so old address still on rolls (25)

The remainder (n=60) were voter errors:
Form inadequately filled out (32)
Street address given did not exist (13)
Card undeliverable, presumably because voter had moved (15)

The total number of registrations is around 2183, not including duplicates and registrations legitimately cancelled.

For details of assumptions made, see below.

Which voters are most likely to be affected by registration errors?

One group of voters most subject to the errors described here are those who change address and therefore must repeatedly re-register, thus increasing the overall chance of error. This includes youth, who are four times more likely to move within a year than the general population , low income people who are 90% more likely to move than those with high incomes, and African Americans who are 45% more likely to move in one year than non-Hispanic whites (computed from Table 8, US Census data, ). We suspect but don’t have evidence that populations with lower educational level are also more prone to error in filling out registration forms.

What can be done to “save” the registrations that are not or are erroneously recorded?

Thanks to the alarm raised by CASE, there was a successful media campaign in several major urban centers to alert voters to call their Board of Elections if they hadn’t received positive confirmation that their registrations had been correctly entered. Now, after the Registration deadline of Oct. 4, we need different actions and messages, since many voters may still lose their votes for the reasons laid out in this study.

Send us your ideas ASAP. Here are some possible actions to be discussed in the next few days:
· Public campaign and if necessary, legal action, to allow correction of clerical and voter errors between Oct. 4 and Nov. 2, so that these errors do not lead to disenfranchisement.
· Development of a major campaign to educate voters and absentee voters who live in the 66 counties using punch cards how to avoid errors (a huge source of lost votes). This would include signage and advice at the polling places.
· A major public outreach campaign to warn voters who are subject to Provisional Balloting to be sure they vote in the correct precinct (unless Sec. of State Blackwell’s decision is reversed)s), and in any event, how to fill out a correct provisional ballot.

For further information, contact: Norman Robbins, MD, PhD,
Citizens Alliance for Secure Elections, [email protected]

Assumptions and Errors in the Cleveland Study

Submission dates. It is important to establish that there was sufficient time for registration/change of address cards analyzed in this study to be entered on the official rolls of the Cuyahoga County Board of Elections (BOE). 98% of the study cards with clerical or voter errors were filled out by registrants between Aug 2003 and July 11, 2004. In general, beginning the spring of 2004, cards were submitted to the BOE by The Greater Cleveland Voter Registration Coalition on average about every 2 weeks after the registration date. Since the detailed analysis of the fate of these cards was done in the first 2 weeks of September, at a time when the Board had caught up with its major backlog, and the detailed analysis was done on computers at the BOE, 98% of the cards would have been submitted approximately 7 weeks before that time (about 2.5% were submitted 5 weeks ahead of the analysis). The vast majority were submitted at least 2 months earlier, and most far earlier. Nonetheless, some cards, not found on the BOE analyzed in the first pass in late August, did appear in early September after a huge backlog was cleared. Therefore, it is possible that some of the cards reported in this study as “not found” or with address changes not updated, are now on the rolls.

The “STX” category of cards. When an address entered on the rolls does not correspond to an existing city address or the address of some former voter, the application is dubbed “STX” i.e. the street address does not exist. Until that address can be further followed up with actual street and address mapping, or if it is not eventually found, the registration is not considered legitimate. However, we found cases where the STX category may have been inappropriate because the entered name and especially address differed from that on the original card. We grouped these under clerical errors (not STX or voter errors), since the application in these cases might be rejected through no fault of the registrant.

Sample size, deletions and checking for duplicates. The original sample of about 2451 cards included 59 cards which at the first pass in late August were at first listed as “not found” but in the early September (when we had access to the BOE computers) were determined to have been legitimately deleted by the BOE. Reasons included moving out of county, death, and imprisonment. These were deleted from both the sample size and the numbers reported as clerical or voter errors. Next, we determined that in 1280 of the cards in the “found” category, there were 123 duplicates, i.e. about 10% of the sample size. Since finding duplicates was a slow and tedious process, we estimated that in the remaining 921 “found” cards, there was a 10% duplication rate also, so we subtracted both this estimated 10% of duplicates and the actually determined number of duplicates (in the sample of 1280) to arrive at the final sample size. This sample also includes 29 non-duplicate “founds” and the 164 non-duplicate cards with clerical or voter errors. In this way, we computed the final sample size to be 2183, and it was on this basis that we concluded that in fact, there were about 7% (164/2183) errors.

Estimating that 10,000 votes were possibly compromised in Cuyahoga County, and 35,000 statewide.
(a) Cuyahoga County. The BOE told us that about 200,000 registrations and change of name or address forms had been submitted between Jan. 1 and end of July, 2004. As noted in the main description, we chose to use “at least 5% error” rather than the actual 7% error in order to allow for errors or assumptions and to be very conservative. Taking 5% of 200,000, we arrived at 10,000 Cuyahoga votes subject to the errors we described (Were we to have used our 7% error rate, we would have said 14,000 votes were subject to errors). There are factors which both work to increase or decrease this 10,000 estimate, but the final conclusion is still that this estimate is conservative and likely to be an underestimate. First, since 10-20% of the 200,000 submissions may be duplicates, the actual number of non-duplicate applications could be as low as 175,000. On the other hand, we don’t know the level of care taken by all registrars to reduce voter errors. This might increase the percentage errors found if we could analyze the whole 200,000 received. Another major factor affecting our estimate is that we did not include cards received by the BOE in 2003, when registration began to pick up, increasing as the year progressed. If we imagine that say, 50,000 new applications were received in 2003, then we would estimate that 5% of 250,000 total new registrations/updates would be affected, i.e. 12,500 votes.

(b) Statewide. In early September, from 13 county BOEs, not including Cuyahoga, we obtained the estimated number of registrations/change of address/name from Jan. 2004 to end of July. These plus those of Cuyahoga comprised about 48% of the registered voters in Ohio. In these counties, we also computed the percent of new registrations as a percent of registered voters in 2002, which averaged about 9%. From this data, we projected the registration of the entire state (assuming 9% had been registered in the rest of the state’s counties except Cuyahoga), adding on Cuyahoga County at the end (The registrations in Cuyahoga County were higher than elsewhere, so we couldn’t use them as typical). This led to an estimate of 760,000 new registrations/change of address statewide. However, we dropped about 60,000, estimating they were legitimate deletions (see above), leaving about 700,000. 5% of these is 35,000 votes that may be compromised because of clerical or voter errors.