[an error occurred while processing this directive]

REVISITING THE U.S. VOTING SYSTEM: A RESEARCH INVENTORY

November 27-28, 2006

Convened by the American Association for the Advancement of Science

Frederick Conrad, Michael J. Hanmer, Michael W. Traugott

VOTER CONFIDENCE IN THE NEW GENERATION OF ELECTION TECHNOLOGY

The introduction of new voting technology, especially the recent spate of adoptions, was triggered by a series of administrative and legal disasters associated with the 2000 election in Florida. The impetus behind the influx of new technology was a concern about a loss of public trust and confidence in the American electoral system. The main issues highlighted in the 2000 elections were concerns about whether voters’ intentions were recorded accurately, as most notably in the case of the “butterfly” ballot used in Palm Beach County, as well as whether repeated handling of some kinds of ballots, specifically punch cards that had hanging, dimpled, and pregnant chads, could produce reliable totals of how many votes each candidate received.

Since the 2000 election, several public pollsters have collected data about the level of confidence that citizens have about whether their votes will be counted accurately and whether the results can manipulated to produce a desired outcome. When the 2004 presidential election rolled around, for example, only 35 percent of a national sample of 920 “likely voters” interviewed in a CBS News poll a week before the election indicated they had “a lot” of confidence that the votes would be counted properly. In a mid-October poll, 60 percent of the respondents said they were “very” or “somewhat worried” that officials in other parts of the country would try to “manipulate the vote counts to favor their candidate” while one-quarter expressed the same level of concern that officials “in their area” would do the same. Just before the 2006 election, the Gallup Poll reported that only 28 percent of a sample of 1,002 adults felt “very confident” that the votes would be accurately cast and counted in that election, not any difference from 2004. Almost half (47%) felt “somewhat confident” about such accuracy.

Public attitudes about various aspects of acceptance of new voting technology are best viewed in the context of a broader body of research that deals with general attitudes about science and technology as well as studies of how the public responds to the adoption of specific new technologies. In brief summary, this research suggests that the acceptance of technology is a function of a number of factors that include personal characteristics such as education, a general set of attitudes about science and technology as well as their role in society, personal experiences with the technology, and confidence and trust in the specific technology. This research provides a framework or context within which we can better understand reactions to the new voting technology, employing the following model.

Figure 1. A General Model of Acceptance of New Voting Technology

In order to pursue these relationships, we collected data under the auspices of the Time Sharing Experiments in the Social Sciences (TESS) project. In a sample of 1,214 respondents interviewed late in 2005 and early 2006, we used a series of randomly administered vignettes designed to capture the effect of three different aspects of the new voting technology - access cards, paper receipts and methods for transmitting vote totals - that might relate to voters’ confidence in the electoral process as a function of the use of access cards, the use of a paper receipt, and different transmission technologies for sending vote totals.

In a preliminary analysis, we observe that overall, respondents were reasonably confident that their votes would be accurately recorded. This general confidence in accuracy was moderated by the presence of a paper receipt in the voting vignette. When the voting description included a paper receipt, respondents were more confident their vote would recorded accurately. The mean confidence scores (adjusted for the covariates) are 3.19 (SE = .034) when a paper receipt is mentioned and 3.03 (SE = .034) when it is not (where 1 = “not at all confident” and 4 = “very confident”). Although the effect is only two tenths of a scale point, the sample size is large enough so that we can detect even this subtle effect: respondents are sensitive to the potential benefits of a paper receipt for insuring accurate recording of votes. At the time the data were collected, the media had given some attention to the issue of voter verification of electronic voting though the coverage was not as widespread as was when the 2006 election approach. The effect may well be larger if the data were collected today.

Voters’ confidence in the accuracy with which their vote was recorded was also sensitive to the method of transmission. Respondents did not begin our experiment already concerned about transmission of vote tallies, but if the voting vignette mentioned transmission – particularly over the Internet – this increased respondents’ concern. More specifically, respondents’ confidence was highest when there was no mention of transmission and lowest under either of the internet transmission descriptions. If the transmission process was described as copying data to a CD that is taken by a courier to a central location, this seemed both to make respondents aware of the transmission issue and to reassure them that their votes would be accurately recorded; CD transmission led to confidence that is indistinguishable from confidence when no transmission was mentioned. However, Internet transmission led to reliably lower confidence than no mention, whether the transmission is encrypted or unencrypted.

When respondents were asked to evaluate the likelihood that someone could figure out how they voted, they reported greater likelihood when an access card was present than when it was not. While overall respondents were not very concerned that this will take place, they do seem to have some concern that access cards (as we described them) could compromise anonymity.

The good news is that the observed effects are small in the initial stages of the deployment of the new technology. However, the data analyzed here were collected outside of a general election campaign. But it remains to be seen whether media coverage of the 2006 election, the first in which there is relatively complete change in voting systems, will change this. This research report is being completed at a time when the last details of the 2006 election are not yet known, and the canvassing and recounts are not yet complete nor all of the results certified.

While the 2006 election generally looks pretty clean, partly due to the use of more computerized equipment to produce accurate tabulations that diminish the prospect of uncovering serious arithmetic errors, there have been a few incidents suggesting important human error. In Sarasota County, Florida, in Katherine Harris’s House district, there were 18,000 “under votes” in the House race, out of a total of about 237,000 cast and with a 373 vote margin separating the two candidates. This high rate of undervoting appeared on one kind of touch screen DRE machine and did not appear on absentee ballots or early votes recorded on paper. This suggests that human error was the probable cause when the ES&S machines were programmed. Several widely distributed news stories like this could have a noticeable impact on public confidence in the electoral system.

Our original model suggests an important role for the media as most citizens will not experience problems themselves but will read about it happening to others. This indicates the likely benefit of pursuing two additional avenues of research. One is a systematic content analysis of the news to confirm our suspicions about the different media frames used in the new voting technology. Furthermore, there is likely to be some benefit from pursuing further research based upon the third person effect: “I don’t expect to experience problems but I know many others will.” We will incorporate these conceptual elements in new data collections to support such analyses.