To: rusin@math.niu.edu
Subject: The Statistics of the Florida Recount
Date: Tue, 28 Nov 2000 12:09:56 -0500

Thank you, Professor Rusin, for the interesting statistical analysis of the
Florida voting situation you posted on Yahoo.  It has been forwarded,
probably rather widely, from there.  I would just like to comment on one of
the assumptions in your analysis, the 1% error rate.

There are two types of punched card voting machines, as well as several
other totally different types of machines used by various Florida counties.
One punched card type, the Datamatic, is used in several predominantly
suburban counties and another type, the Votamatic, is used in several more
urban counties, including Dade and Palm Beach, according to the Washington
Post.  The design of the Datamatic machine virtually eliminates the
possibility of hanging or dimpled chads, which are not prevalent but are
very possible on a Votamatic machine.

According to one of the inventors of the Votamatic machine, speaking in a
press conference organized by one of Mr. Gore's lawyers, the machines used
in Palm Beach, Miami, and elsewhere are quite error prone by now.  It seems
that the original design called for a high quality soft plastic backing to
the card at the locations where the chads are to be punched.  That would
allow the stylus to depress the chad sufficiently that it would be likely to
come all the way out.  With less than optimal pressure, of course, a chad
would be likely to "hang."  It would be quite unlikely for a chad to become
merely "dimpled" or "pregnant."  But that is when the machines are made with
high quality soft plastic and are fairly new.

Over a period of time, the patent on the Votamatic machine expired, and a
number of companies other than the original manufacturer start to produce
and sell them.  The quality of the plastic varied among manufacturers.  Now,
any soft plastic contains a volatile "plasticizer" which evaporates slowly
over time and with use.  As the plasticizer degrades, the plastic becomes
harder.  Thus, in some older and well-used machines, the plastic may be so
resistant to the stylus that a clean punch is quite hard to achieve.  In
such a case, "dimpled" or "pregnant" chads, as well as hanging chads become
not uncommon.  But the hardness of the plastic also varies over the area of
the card.

According to a former Election Commissioner for Palm Beach County, speaking
at the same press conference, the usual practice is to design the ballot
such that the votes for the most important office or decision are aligned
with column one on the punch card.  Successive votes for elections of
declining importance are placed in successive columns.  Thus, the
deterioration of the plastic would differentially affect the more important
races more so than the less important ones.  By 1996-1998, it was noticed
that the number of ballots which had registered votes for the more minor
elections, but not for the most important election (say, national Senate or
House or Presidential races) were increasing.  That is anomalous.
Apparently, the rate at which such votes are normally cast on machines of
unquestioned accuracy is about three percent.  That is, about three percent
of voters intentionally do not vote for the major offices, but do vote for
the minor offices.

[See below for amendation --djr]

Over recent elections in Palm Beach County, these rates increased to between
5% and 8%, by 1998.  I do not know what the rate was in the 2000 election.
Thus, it is plausible, although by no means certain, that the difference
between the raw rate of ballots with no votes for President, but votes for
lesser offices and the 3% background rate for such ballots are due to
problematic Votamatic machines.  It is important to note that only a portion
of the machines would have the defect; it would not be consistent over all
precincts or even within a precinct necessarily.  It would depend on the
manufacturer and the history of the individual machines.

Still, if these claims by the Votamatic inventor and the former election
supervisor are accurate, then the number of expected valid votes for
President which were improperly rejected due to machine defects, would be
somewhere between 2% and 5% or possibly slightly more.  As you noted, such
machine errors would be expected to affect the votes counts for each
candidate in proportion to the number of non-problematic votes each received
on similar machines.  It would also be true that some valid votes might be
reflected in merely dimpled, rather than hanging, chads.  It is equally true
that some dimpled chads could reflect the possibility that some people had
started to vote for President, changed their minds, and then deliberately
left the chad in place.
Thus, there would always remain a degree of subjectivity in the assessment
of such ballots.  I do not know of any objective estimates of the percentage
of people who start to poke a chad and then change their minds.

Nonetheless, even after allowing for uncertainties in a hand counting
process, it seems much more likely that the number of improperly uncounted
ballots in counties using Votamatic machines is at least double the one
percent you initially assumed.

[personal reflections deleted --djr]


[Note: this respondent has claimed a 3% intentionally-blank rate as the
norm. This is inconsistent with our findings. This respondent has
acknowledged not having a specific reference for this figure, so 
we must continue to treat the true numerical value of this rate as
speculative. --djr]