My analysis of precinct-level data from the 2004 election in Franklin
County, Ohio, uses different statistical methods than Ben Highton used
in his PS paper, but as far as the relationship between voting
machines and voter turnout is concerned we reach qualitatively similar
conclusions. My analysis also looks at a measure of the long lines in
the county and includes measures of precinct racial composition. If
one uses the November 2004 electorate as the standard, the allocation
of voting machines clearly and disproportionately reduced turnout
among African American voters. For the paper, see
http://macht.arts.cornell.edu/wrm1/franklin2.pdf
That's an update of the paper I originally wrote in July, 2005, in
response to one of the infamous DOJ letters. I've updated it to take
into account new data I just received from Franklin County regarding
the information they used to allocate voting machines to precincts.
Apparently they used a measure of the active voter electorate computed
as of mid-June, 2004. While the allocation of machines discriminates
heavily against African American voters if it is compared to the
November electorate, if it is compared to the electorate measured as
of June then precincts that have a high proportion of African
Americans and precincts that have a low proportion of African
Americans have on average virtually equal numbers of voters per
machine.
Average Number of Voters per Voting Machine
Proportion
African American November June
Low 213 178
Medium 226 172
High 242 176
Details are in the paper.
> registration appears in the numerator. The amount of deadwood on the
> registration rolls is probably correlated with turnout (e.g., precincts with
> higher levels of deadwood could be poorer neighborhoods with lower turnout).
Using active voter counts, as I do, should eliminate the deadwood
concern. The "turnout" effects are only slightly smaller if one looks
at active voters instead of registered voters.
> voting machines in their precincts. But when one looks carefully at the
> numbers in the report, the number of votes per voting machine was lower in
> predominantly African-American precincts than in other precincts. For the
The point about "the number of votes per voting machine" is one I flag
in my paper as being particularly specious. That has nothing to do
with a lack of bias in the allocation of voting machines. Instead it
traces to two other effects. My paper mentions one of them: voters in
precincts with a high proportion of African Americans took longer to
vote than did voters in precincts with a low proportion. Some reports
I've read trace that to there being longer ballots in the places where
African Americans were tending to vote; I have not managed to lay
eyes on the ballots. The other factor is that African American voters
seem to have been more determined to vote than white voters were, on
average. Survey data (collected by the DNC for the DNC study of Ohio
2004) shows that African Americans were more likely than whites were
to return to the polls after having to leave due to the long lines
(see page 3 in Section III of the DNC report). That's probably the
primary reason the significant disparities in voting machine provision
did not produce even larger disparities in voter turnout.
Walter Mebane
* - * - * - * - * - * - * - * - * - * - * - * - * - * - * - * - * - *
Walter R. Mebane, Jr. email: wrm1@cornell.edu
Professor office voice: 607/255-3868
Department of Government cell: 607/592-0546
Cornell University fax: 607/255-4530
217 White Hall WWW: http://macht.arts.cornell.edu/wrm1/
Ithaca, NY 14853-7901
* - * - * - * - * - * - * - * - * - * - * - * - * - * - * - * - * - *
Michael McDonald writes:
> My problem with Highton's analysis comes from a different angle. Highton
> calculates his dependent variable (turnout rates) as a percent of registered
> voters and for one of his independent variables (registrants per machine)
> registration appears in the numerator. The amount of deadwood on the
> registration rolls is probably correlated with turnout (e.g., precincts with
> higher levels of deadwood could be poorer neighborhoods with lower turnout).
> The findings could therefore be an artifact of measurement issues, as
> precincts with higher number of registrants per machine are the same as
> those with lower turnout rates due to the confounding presence of deadwood,
> not long lines.
>
> I investigated constructing citizen voting-age population for the Franklin
> County precincts after I had a similar discussion with Walter Mebane through
> the election law list-serve a few months back (Walter is one of the
> co-authors of the Social Science Research Council report cited by Highton,
> who is also a co-author on the same report). Unfortunately, I concluded
> there was no reliable way to construct a CVAP measure for the 2004
> precincts. We are left, then, with an analysis that is known to be flawed -
> I don't think anyone would seriously argue that deadwood is not present on
> voter registration files - though there is no mention of the limitations of
> the measures in Highton's article. Perhaps in the future I can get a handle
> on the magnitude of the potential error induced by calculating turnout rates
> based on voter registration rather than the eligible population.
>
> One other point about the SSRC report: a claim that came out of Franklin
> County was that African-Americans were discriminated against by a lack of
> voting machines in their precincts. But when one looks carefully at the
> numbers in the report, the number of votes per voting machine was lower in
> predominantly African-American precincts than in other precincts. For the
> claim to be true that machines were inefficiently distributed into
> African-American communities creating a situation for long lines, it must
> have been that African-American voters all showed up to vote at the same
> time and long lines ensued, while in White precincts, voters came in at a
> steady pace throughout the day. I don't find this credible, and I don't
> think that election officials have the capacity to predict timing of voting
> patterns. While this is not Highton's argument, it arises indirectly when
> he calculates predicted effects, finding inefficient distribution of
> machines effectively gave Bush a net of 6,000 votes in Franklin County, Ohio
> (race is, of course, related to presidential vote).
>
> All this said, I have personally witnessed voters stepping out of even
> relatively short lines, so I do not doubt long lines induced people not to
> vote. I just don't buy that it would have increased turnout by 7.7% to have
> 250 registered voters per machine in every precinct. The predicted effects
> are likely attenuated by the presence of deadwood.
>
> ==================================
> Dr. Michael P. McDonald
> Visiting Fellow, Brookings Institution
> Assistant Professor, Dept of Public and International Affairs
> George Mason University
> 4400 University Drive - 3F4
> Fairfax, VA 22030-4444
>
> Office: 703-993-4191
> Fax: 703-993-1399
>
> mmcdon@gmu.edu
> http://elections.gmu.edu/
>
> -----Original Message-----
> From: owner-election-law_gl@majordomo.lls.edu
> [mailto:owner-election-law_gl@majordomo.lls.edu] On Behalf Of Jeffrey MA
> Hauser
> Sent: Tuesday, February 07, 2006 1:17 PM
> To: election-law
> Subject: Re: Electionlawblog news and commentary 2/7/06
>
> << Message: 06 (5.75 KB) >> FORGIVE ME if this reflects a failure to
> appreciate the statistical
> analysis nomenclature, but doesn't Benjamin Highton's article on voter
> turnout during the 2004 presidential election in Franklin County
> understate the effect because his model, unlike his essay, ignores that
> which made 2004 unique within generally low-turnout precincts?
>
> That is, his model seems to fail to incorporate a way to address these
> questions:
>
> 1. Isn't voter turnout in low-turnout districts more elastic than in
> high-turnout with respect to Presidential years?
>
> 2. Aren't new registrants, especially those being hounded by field
> organizers, much more likely to vote than otherwise demographically
> indistinguishable people?
>
> Re 1 & 2, wouldn't the incredible increases in turn out in Cuyahoga,
> despite declining population, need to be accounted for in this analysis?
>
> Finally -- I should know this, but do not -- who appoints the county
> Democratic election board members?
>