From a posting on AAPORNET, a public opinion listserv:

Date: Mon, 08 Nov 2004 15:41:09 -0600
From: Scott Althaus 
Subject: Re: National Turnout Estimates

In response to [a] query, back in 2000 there was only one
accepted method for calculating turnout. This time, there are now three
alternative ways of calculating voter turnout, which produce three slightly
different turnout trends between 1968 and 2000.

The first divides votes cast for president by the voting-age population
(VAP). This is the "old" way of measuring turnout, and VAP estimates
produce what are to most of us the familiar turnout numbers for past
elections. However, thanks to Michael McDonald and Samuel Popkin's
influential (2001) APSR article, VAP is now widely seen as a problematic
measure that underestimates actual turnout levels.

The other two trends divide votes cast by measures of vote-eligible
population (VEP), which corrects for the errors introduced by VAP alone.
The first VEP estimate (I'll label this McD & P VEP) is the McDonald and
Popkin measure that comes from the following formula: voting-age population
minus non-citizen adults minus ineligible felons plus eligible voters
living overseas. Data for this measure reported below come from the U.S.
Elections Project web page at George Mason University (elections.gmu.edu)
for 2004, and from Table 1 of McDonald and Popkin's APSR article for
1968-2000.

The second VEP estimate (I'll label this Gans VEP, because it is the
measure being used by Curtis Gans for his Committee for the Study of the
American Electorate report, available
http://www.fairvote.org/reports/CSAE2004electionreport.pdf) uses a
different formula: voting-age population minus non-citizen adults. Gans'
estimates reported below are from the first table of that report (p. 12).

Turnout % Est.  1968  1972 1976  1980  1984  1988  1992  1996  2000

          VAP    60.9  55.2  53.5  52.8  53.3  50.3  55.0  48.9  51.2
  McD & P VEP    61.5  56.2  54.8  54.7  57.2  54.2  60.6  52.6  55.6
     Gans VEP    61.9  56.6  55.1  54.7  56.0  53.1  58.1  51.5  54.3

My initial calculations for turnout on Tuesday's election are 52% using VAP
and 56% using McD & P's VEP. However, these are underestimates, since the
official vote totals for 2004 have yet to be finalized with the addition of
absentee and provisional ballots.

If Curtis Gans' estimate of 120 million votes proves correct, then VAP
turnout would be 54%, McD & P VEP would put it at 59% by my calculation,
and Gan's VEP calls it at nearly 60%. These two VEP estimates are both
fairly close; the critical difference for the claim of exceptionally high
turnout is how these three measures produce different turnout estimates for
1992, which had the highest turnout by any measure since 1968. My sense of
the reason for this difference is that the number of ineligible felons was
lower than the number of eligible expatriates in 1992 (see Table 1 of
McDonald and Popkin's APSR article), which was always the case before 1994.
The gap since has been more narrow. In contrast, Gans assumes that these
two numbers will tend to cancel one another out, so they can be ignored.

Note that the only measure that shows projected 2004 turnout levels higher
than 1992 is Gans's (and this presumes that 120 million votes will actually
be cast; if the total ends up being lower, his estimate of turnout will
also go down). Using both of the other two ways of measuring turnout, 1992
had the higher levels of voter participation.

It is also interesting that Gans is now (correctly) using VEP rather than
VAP estimates as comparisons for earlier years, whereas his call last
Wednesday that turnout was higher than any since 1968 was based on an
incorrect comparison of 2004 VEP estimates to VAP estimates from pre-2000
elections. However, in the report released by Gans this past Friday, his
measure of VEP still supports the claim that projected turnout could be
higher than any year since 1968.

Scott Althaus