Vote suppression by machine

[jmknapp]

At yesterday's MLK breakfast in Boston, Kerry directly addressed the issue of vote suppression caused by inadequate numbers of voting machines. He said:

"Voting machines were distributed in uneven ways. In Democratic districts, it took people four, five, 11 hours to vote, while Republicans through in 10 minutes -- same voting machines, same process, our America."

Nowhere was this effect more noticeable perhaps than Franklin County, Ohio, where the largest city is Columbus.

Due to a relatively complex ballot with lots of issues and candidates, voting time in Franklin County averaged about 3.8 minutes per voter. Across the county there were 2,798 voting machines officially in operation (neglecting breakdowns) and 533,564 people who managed to vote.

According to Franklin County BOE figures, there were 640,399 "active voters" in the county. The active voter measure is defined by HAVA and includes a number of factors such as whether someone voted in the last two election cycles. Thus, relative differences in turnout are factored in to the active voter number. Somewhere between the 533,564 and 640,399 numbers is the number of people who would have voted had long lines not prevented many from voting.

Taking the 640,399 number it can be seen that even if there had been a fair allocation of the 2,798 voting machines, there would have been considerable voter suppression, as 640,399 active voters works out to 229 active voters per machine. Since the 13-hour polling day had 780 minutes, that works out to 780/229 = 3.4 minutes per voter available. With the average voting time of 3.8 minutes, this means long lines, particularly during the busy hours. The lines are effectively a poll tax: if you couldn't afford to wait, you didn;t vote.

But even so the allocation of voting machines in Franklin County was not fair, as black and Democratic precincts were shorted voting machines.



Interestingly, the critical range at which the lines would grow exponentially, around 3.4 minutes, neatly splits the population of precincts, with mostly Republican precincts having in excess of 3.4 minutes per voter available.

Funny that.

There are other ways of showing this disparity.

For example:



That one is rather perverse, as those without much education would be expected to require a longer time to vote, but were allocated less time on the average.

[L. Coyote]

1. Too bad we do not have party registration data in Ohio.

It would make a good correlation analysis. Have you done correlations for these variables? It is useful to break this down to a single number.


Is your Franklin County spreadsheet available?

[mgr]

2. Pattern is a bit noisy

jmknapp, always appreciated your efforts.

Could you probably help your analysis by including either error bars, variance, or re-weigh the percentages (possibly by total number of votes in the last two graphs, so you don't have problem with small sample size)to get a tighter fit to your curves.

What type of voting machines are we talking about that created the 3.8 minute margin. I am in a punch card precinct, and if I have my sample ballot completed, it usually takes me a minute to vote (I do absentee now).

You point out that it was not fair to African Americans and registered democrats, but you may also be rigging you results by selecting the higher of the presumed turn outs. What do the patterns look like with the lowest presumed turnout, and do your conclusions change materially?

Mike

[jmknapp]

3. I did some preliminary calcs

One good reason to use the "active voter" number (the "higher of the two turnouts" as you say) is that it is indeed the metric by which the Franklin County BOE allocates machines--so any departures from a flat, fair, adequate allocation seem to go against the stated intent.

A smaller number of people would mean more time on the average to vote, but would not change the results materially I believe--the lines were long all over Franklin County--that much is a given (that there were machine shortages). The question is was there unfair allocation on top of that, and the "active voter" metric gives the best possible measure of absolute demand, if the machines hadn't throttled it.

Personally, at the voting machine I had a computer printout with all my choices on it, yet I'd say I took about three minutes just trying to line them up with the ballot, and double-check. While I was at the front I repeatedly heard poll workers berating voters for exceeding their five minutes, as a two-hour line snaked around the school library and people were anxious.

In the testimony given in several hearings in Columbus, many people told of the five-minute limit and gave data indicating the average throughput. The time to vote of course would be worse in areas of high functional illiteracy and other special needs.

Numerous reports told of people leaving lines because they couldn't wait, had to go to work, whatever. I saw 4 people leave the lines at my polling place, saying that they would come back later. Don't know if they did, but the situation generally got worse, not better, later.

As for scatter the disprecancies are significant by a number of standards tests. The null hypothesis (that there is no correlation) is rejected at the 5% level for the black population in the precinct. Resulting equation is:

minutes = 3.67 - 0.78*B where B is the black ratio 0-1

I.e., voting while black cost 0.78 minutes.

If the ratio of adults in the precinct without a high school diploma is D, the following equation results, also significant at 5%:

minutes = 3.77 - 1.70*D

So the effect of being in a poorly-educated area was even stronger--costing up to 1.7 minutes. Actually no precinct exceeded more than about 60% undiplomaed adults, so about 1 minute poll tax in that case.

Other factors:

K = Kerry fraction in precinct

minutes = 4.27 - 1.34*K

H = median household income in precinct (kilodollars):

minutes = 3.14 - 0.0082*H

Where H ran from a low of about $20K to a high of over $100K at the precinct level. So $100K folks got an extra 0.82 minutes to vote before breaking open the bubbly.

A multiple regression reveals Kerry percent and percent without a diploma as the best two factors to use to predict the allocated voting time.

Here is a bit of that output (formatting will be munged):

Regression Equation Section
Independent Variable/regression coeficient/standard error/T-Value/reject H0 at 5%/power of test
Intercept 4.2268 0.0941 44.898 Yes 1.0000
K -1.0641 0.1945 -5.471 Yes 0.9998
D -0.7790 0.3034 -2.567Yes 0.7282

Estimated Model
4.23 - 1.064*K - 0.78*D

The Kerry percent seems to be the strongest determinant of which precincts got shorted. Kerry percent per precinct ranged from about 20% to 100%, whereas no diploma ranged from 0% to 60%.

The type of voting machine is the Danaher electronic type. You face a large membrane panel of pushbutton switches, with the ballot choices overlayed.

[jmknapp]

4. Correction

The household income formula above should have a + sign, not a - :

minutes = 3.14 + 0.0082*H

That is, the more income in a precinct, the longer people were afforded to vote.

[mgr]

5. There's a lot to chew on here

I appreciate the detailed response. As you may be aware, what you are doing is the nuts and bolts of the vote suppression issue in Ohio. As an Ohio resident, and voter your expertise as an observer, and possibly as a statistician are unparalleled. Carried a little further, your data could bulwark a legal argument that mis-allocation of machines and with arbitrary and capricious time limits may have had three consequences--suppressing the turnout through waiting; suppression of the vote by intimidation in the voting booth; and ballot mis-allocation or spoilage by hurrying the voter with an artificial time limit.

On the Danaher machine, it appears that all choices are on a single screen, about what font would you say? Is it possible that the manner in which the choices were formatted to a single screen aggravated the amount of time making selections? What I am getting at of course, is that eye strain, and weak vision cut across all economic categories.

It would seem that putting a time limit on how long one may be in the booth, and actively informing voters that they were exceeding it would amount to voter intimidation. Hurrying people may lead to improper ballots, such as the spoilage patterns you mapped previously for Cleveland (of course, you are directly familiar with Franklin, not Cuyahoga) or lack of opportunity to address balloting errors. Was this five minute limit applied uniformly throughout Ohio? Why five minutes?

What gives me pause is the use of percentage on the x axis rather than numerical values. This can disguise small or large populations that may reweigh the curve when extrapolating to the numbers disenfranchised because they were forced by the long lines to make economic decisions regarding whether they could afford to stay and vote. Since these are precinct level data, are the precincts equivalent to the appropriate magnitude, that percentage remains a determinant of the pattern, or is there such variability in turnout and total number of voters in a precinct that this could be a problem?

Are there differences in percentage turnout from the precincts with greater time available to cast one's vote to the ones that were hurried? If they are higher, when correcting for the difference, what might be the total number of voters affected? Is this number sufficient to allow one to re-evaluate the outcome of the election in Ohio? Or do other factors need to be considered from elsewhere in Ohio?

Mike

[jmknapp]

9. Precinct size doesn't matter

The precincts tend to be around the same size in population, so the distribution of dots gives a pretty good indication of the absolute numbers of people affected. Here's a histogram of votes cast per precinct in Franklin County's 788 precincts:



Such variation in vote size that does exist is only slightly correlated with Kerry %:



This chart gives some idea of the effect on turnout. I'd have to rework it to use minutes:



As for being directly familiar with Franklin County, of course I only observed a small part of it, but my polling place seemed to be a microcosm of all the problems reported elsewhere. There is a lot of informtion on the record thanks to voter suppression hearings held in Columbus and Cleveland as well, and on voteprotect.org.

I was a volunteer for ACT on election day, busy with GOTV efforts, but I wish I could have cloned myself, alas, to observe as many polling places as possible.

As for my personal observations, I can relate the following.

On election morning, I went to ACT headquarters downtown at 5:30am to start the long day. First on the agenda was “lit drops”—putting get-out-the-vote messages on doorknobs in the wee hours as people were waking up. My friend and I pulled a precinct on the near east side, a minority area where I normally would not venture, certainly not at dark. But it was very cool. Mostly black, with a few whites, mostly single-family dwellings that were not half bad. One black fellow was going through the neighborhood at 6am bellowing like a town crier “Get out and vote! Get out and vote!” I never saw such enthusiasm to get out and vote from the few people I did talk to, out getting their morning paper or whatever. One black bus driver from the nearby Children’s Hospital drove by and I asked through his open door “Are you going to vote today?” He replied with an enthusiastic yes, and when I asked who for he shook his head at my stupid question and said “Kerry of course.” He went on to say that he had made an excuse with his boss that he had a doctor’s appointment that afternoon, so he could get out to vote. I kills me that this is exactly the type of working man who would have been stymied by the long lines at the polls that day. I doubt that he was able to vote.

One old white lady buttonholed me on her doorstep and told me that she had eight children and “the war in Iraq is not worth a single one of them.” I could see one of her daughters and grandchildren inside. She started crying about the prospects of Bush getting re-elected and I said “Don’t worry too much, I have a good feeling about today.”

I did talk to one Bush supporter. He was out getting his paper and I asked who he was voting for today. He said forthrightly “President Bush.” When I asked why he said “He makes me feel safe.” War is peace I guess.

By the time we had finished the lit-drop, it was past 6:30 and the polls were open. The predicted rain had not started yet, but it was coming soon and would continue throughout the day. We passed by a polling place off Parsons Ave., a library, and the line was out the door snaking around the block. I thought it was a remarkable sight, a little exciting, but at the time I didn’t realize the voter suppression that these lines represented.

According to the official results, in the precinct we worked Kerry got 454 votes to Bush’s 39. Turnout was 507 out of 1332 registered, or a paltry 38%. According to board of election figures, there are 953 “active voters” in this precinct—people who have voted in the last two elections, along with newly-registered voters. Only 4 voting machines were allocated, meaning there were about 236 active voters per machine. There was a problem at this polling place reported to 1-866-OUR-VOTE: “Polling place was not open when it was supposed to be at 6:30 a.m. Voting machines have not been set up and presiding Judge with book of registered voters had not shown up.”

After finishing this lit-drop activity we went to a “visibility” operation—holding signs on Main St. also in a minority area at the Kelton intersection. There was a lot of enthusiastic horn-honking and smiles from the passers-by, some of whom were probably amused at all the white X'ers and yuppies in the 'hood.

Then it was mid-day and time for the volunteers to go their respective ways to vote before regrouping in the afternoon. At my own polling place I observed most of the problems described across the county that day. There was a long line inside the building and no pollworkers to direct anyone. I asked a voter if this was the only line and was told that it was. That was incorrect, as there were two lines, one for each of two precincts, but the line for the other precinct (mine!) was very short. So I waited in this wrong line, barely moving, for about half an hour. I saw several people leaving the line because they couldn’t wait. Old people were made to stand, although one lady had a chair that she was dragging along with her. Finally a harried poll worker came by and I saw she had some kind of list of voters. I ran her down and was able to determine that I was in the wrong line, and was able to vote after only about another 20 minute wait. From there I was able to see the two Republican lawyers the likes of which infested the polls in Ohio on election day. They were hovering over the backs of the poll workers, talking with them (I couldn’t overhear). Each had a yellow legal pad on which they were jotting notes.

One black man with two kids in tow was being denied a ballot. He said that he had voted there before. The octogenarian poll worker kept repeating insipidly, “We want your vote to count, sir.” He was made to fill out a provisional ballot.

When I got into the booth, which used Danaher electronic voting machines, I had my voting choices on a computer printout, but even then the ballot was so complicated that it took me at least 3 minutes to vote. Under “Vote for President” there was the name Michael Badnarik with a blank space beneath. That was momentarily confusing, until I realized that they had put a sticker over Nader’s name, and the rest of the candidates were below.

While in the booth I heard poll workers berating other people for taking more than 5 minutes. In my precinct, Kerry received 274 votes to Bush’s 121. Turnout was 51%.

[skids]

6. Franklin study link.

This subject has been fairly well covered, culminating in the following study.

http://uscountvotes.org/index.php?option=com_content&task=view&id=65&Itemid=43

(EDIT: while Thomas at uscountvotes is probably frantically trying to fix their webserver, here are some other links:

http://www.indybay.org/uploads/franklinvotemachines.pdf

http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=203x121922

)

(SECOND EDIT: to be clear, jmknapp's material is new, as noone as far as I can recall worked with the "minutes allocated" figures. So it is an interesting additional perpective.)

[mgr]

7. Not the same analysis

What the link does is set the table, it is not related to actual election day outcomes, but characterizes a possible pre-election outcome of suppression of democratic votes. Knapp's work is post election, and quantifies that effect. His has greater utility as it demonstrates what that potential effect identified in Lohrentz's paper is.

The other weakness is that one cannot review Lohrentz's methodology and calculations, but infer them. Together, the two analyses are damning. If there are any discrepancies in the poll book, there is the BOE (and SOS?)that is likely to go down.

Mike

[skids]

8. The broken link goes further.

It is to Liddle's analysis. Hopefully will be fixed soon.

[jmknapp]

10. Measure of suppression

Here's a pretty interesting measure of the suppression caused by lack of voting machines. Since Democratic precincts traditionally tend to have lower turnout than Republican precincts, it's hard to disentangle that effect from that caused by shorting machines.

However, if a two-variable regression analysis is done using the Kerry% (K) and the number of active voters per machine (V) as the independent variables, and turnout as the dependent variable, the following statistically significant relationship is obtained (P < 0.05 for both variables):

turnout percent = 83.0 - 0.039*V - 0.32*K

So the suppression component caused by V (number of active voters per machine) alone is 0.039*V. I.e., a precinct with V=250 would have 250*0.039 = 9.8% suppressed vote on that basis alone.

Countywide, Kerry got 56% and V averaged 229. Plugging those into the equation yields a turnout of 56%. Actual turnout at the polls (i.e., excluding absentee and provisionals) was 56%, so that's a good check on the model.

The really interesting thing is that it allows the total suppression countywide to be estimated. Since the average V was 229, that gives 8.9% of suppression. Out of 845,720 total registered voters, that represents 845720*0.089 = 75,269 suppressed votes in this one county.

[mgr]

11. Bravo n/t

[Hugin]

12. Kick

This is important!