APPENDIX
Estimating Lost Votes
Votes are lost for two different reasons. First, some votes are not counted
because of voter mistakes or confusion casting ballots or because of equipment
unreliability or errors. Second, some voters want to vote but are denied a
ballot because of a failure in the system.
For example, the voter’s registration is out of date or the voter cannot
obtain an absentee ballot.
From election returns we can estimate how much loss is due to equipment
problems. This is discussed at length in the equipment section of the report.
Depending on the assumptions made about the fraction of uncounted votes for
president, we estimate that approximately 1.5 million votes for president were
intended to be cast but not counted in November 2000 (see Part I, “Lost Votes”).
In addition, approximately 2.5 million votes were lost in races for Senate or
governor over the last complete election cycle.
Vote loss due to accessibility is much harder to estimate because we must
rely on survey data. Researchers do not always write survey questions right, and
people are notoriously bad at answering surveys. The best available data come
from the 2000 Current Population Survey’s (CPS) Voter Supplement File.
Respondents who said they did not vote in the November 2000 elections were
asked: “What was the main reason [you/name] did not vote?” Respondents were
given
eleven different reasons, and the responses are provided for both 1996 and 2000
in the accompanying table “Why Registered Voters Say They Don’t Vote.”
There were approximately 40 million registered voters who did not participate in the November 2000 elections. Over 7 percent of the CPS survey respondents said they did not vote because of “Registration problems (i.e., didn’t receive absentee ballot, not registered in current location).” Just under 3 percent report “long lines and short hours” as an obstacle. In addition, 11 percent cite “other reasons,” some of which might be related to registration.
There are three scenarios for these data.
- First, we could take them at face value, which we do in the text. That is, registered voters who could not vote because of registration problems went to the polls but could not vote because of a mix up. Also, registered voters who said the lines were too long would have voted if the lines were shorter. That yields an estimate of 3 million lost votes due to registration problems and 1 million lost votes due to lines.
- Second, the registration numbers could be higher. The registration
question lists a couple of reasons for registration
problems. It is well known in survey research that such “prompting” affects the way that people answer questions. Because the text of the question is not exactly right, the respondent may have cited “other reasons” even though the problem really lies in the registration system. Nearly 11 percent citing other reasons might include registration problems. A more liberal reading of these data, then, would put the numbers with registration problems higher than 3 million. - Third, the real numbers encountering registration problems could be lower. We do not know exactly how many of the 7.4 percent who stated that they had registration problems actually had problems and how many are blaming the system. Our subjective belief is that a conservative estimate of the number of lost votes due to registration problems is 1.5 million voters.
This guess is based on the assumption that half of the 7.4 percent did not
participate because they were not registered in their current location. Similar
inflation of the responses might occur with the category “lines” because
some people might just be complaining about the system.
Rather than relying only on the 2000 CPS responses, we have also looked at
responses to the same question in other recent years (1996, 1998). Estimates
from those years imply somewhat smaller numbers of lost votes due to
registration and polling place problems, about 50% less than the 2000 estimates:
1.5 million due to registration, and 480,000 due to lines and hours (in addition
to the 1.5 million lost due to ballot problems).
Nevertheless, adding up these three numbers still yields a very large number of
lost votes: 3.5 million.
Of these, we expect that half could be recovered with changes in voter
registration and polling place operations, and equipment improvements.
These calculations suggest that the number of lost presidential votes in 2000 is between 3.5 and 5.5 million, but possibly higher, depending on the reliability of the CPS data. We round these figures up to 4 to 6 million, because we are being conservative about the baseline number of registered non-voters and because these figures do not include other problems, for which we do not have an accounting. In addition, we estimate that equipment problems account for another 2.5 million votes in Senate and gubernatorial elections over the last cycle for those offices.
