In the Eye of the Beholder? Motivated Reasoning in Disputed Elections

The data reported by Jenkins (2004, 117) suggest that there were as many as 471 disputed election cases in the US House where election officials could have influenced the election’s outcome. Similarly, there were 60 disputed elections in the US Senate during this period in which election administrators could have affected the outcome (Jenkins 2005, 59).

Unfortunately, there does not appear to be systematic data available on state legislative and local disputed elections. However, given the frequency of contested congressional, gubernatorial, and other statewide elections, and the substantially larger number of state legislative and local elected offices, disputed elections across all levels of government are not uncommon occurrences.

For example, the state of Oregon allows a great degree of freedom in determining voter intent—“Any vote from which it is impossible to determine the elector’s choice for the office or measure may not be counted” (ORS 254 § 505). In New Hampshire, a majority vote by poll workers is enough to determine voter intent—“If a ballot is marked for any office in a way which does not readily admit of counting … then the ballot shall be counted for that office in accordance with the majority vote of the election officials present and counting votes…” (RSA 63 § 659:64). In Michigan, even if voter intent can be discerned, a ballot could be rejected if a voter fails to make the proper “X” or check mark on the ballot—“Marks other than crosses or check marks used to designate the intention of the voter shall not be counted” (MCL 116 § 803.1(c)). Thus, there are many different standards for determining voter intent and counting ambiguous ballots.

Aside from polls workers, county/state boards of election are important players in disputed elections. If a candidate challenges the initial ballot-counting decision of poll workers, it is possible for a county/state board of election (or a court) to review these decisions. These individuals may or may not be political elites. For example, in Ohio, the county board of election is comprised of four individuals, selected and equally balanced among the major political parties (R.C. 3501.06–3501.07). In Pennsylvania, a county board of elections is comprised of the county commissioners, who are elected officials who represent the two major political parties (25 P.S. §§ 2641–2642). Additionally, membership on the state board of canvassers must be evenly divided between the major political parties in Michigan (MCL § 168.22), and nominees to the state canvassing board are submitted by the central committees of the major political parties (Ibid.). A similar law applies to the state board of elections in North Carolina (N.C.G.S. § 163-19), and the county boards of election in New York (New York Election Law § 3–204). Thus, in several states, members of boards of elections are either selected by or obtained their positions with significant support from the major political parties. As such, these individuals should be likely to support the interests of their political party when reviewing disputed ballots.

The data reported in the 2008 Election Administration and Voting Survey sponsored by the US Election Assistance Commission shows that 10.5% of all poll workers surveyed nationally were under the age of 25 (US Election Assistance Commission 2008). These data a similar to another 2008 national survey of poll workers conducted by the Election Administration Research Center at the University of California, Berkeley and the Verified Voting Foundation. In this survey, 11.1% of poll workers surveyed were under the age of 29 (Election Administration Research Center 2008).

Ballots were obtained from the Minnesota Public Radio website, July 12, 2010. http://​minnesota.​publicradio.​org/​features/​2008/​11/​19_​challenged_​ballots/​round1/​index.​shtml.

Challenges came from either of the two major party candidates on the ballot. Two other minor party candidates also appeared on the ballot. Participants cast only four of 6,816 total votes in favor of the two minor party candidates (we assume this to be a function of respondent error). Participants’ choices were therefore restricted to awarding challenged ballots to one of the two candidates involved in the challenges or discarding the ballot altogether.

In Ohio, college students are permitted to list a school address when registering to vote (http://​www.​sos.​state.​oh.​us/​SOS/​elections/​voterInformation​/​regToVote.​aspx), July 12, 2010.

Kilroy and Stivers engaged in an expensive campaign, spending a combined $5 million, making it the most expensive 2008 House race in Ohio. July 12, 2010. http://​www.​fec.​gov/​DisclosureSearch​/​HSProcessCandLis​t.​do?​ull.

Participants were presented with a female candidate only in the real candidate condition. While presenting these participants with a female candidate might seem problematic because participants in the other candidate conditions were presented only with male candidates, Kilroy’s inclusion was necessary to test Hypotheses 2. We control for the gender of participants in our statistical models. Models estimated without a gender control are not substantively different.

These ballot guides are not assumed to be representative of all possible state ballot guidelines. They are simply examples of two different ways states choose to inform ballot counters and are intended to represent two points on a hypothetical “ambiguity” continuum of recount guides.

The Minnesota 2008 Recount Guide, July 12, 2010: http://​www.​sos.​state.​mn.​us/​Modules/​ShowDocument.​aspx?​documentid=​5236. Specifically, we provided respondents with portions of section 13, entitled “Determining Voter Intent” (8–10). See the “Web Appendix” for more details.

The North Carolina standards, July 12, 2010: http://​www.​sboe.​state.​nc.​us/​getdocument.​aspx?​id=​219. See the “Web Appendix” for more details.

Copies of the ballot counting rules and the ballots themselves are in the online “Web Appendix”.

These nine questions included the five standard political knowledge questions advocated by Delli Carpini and Keeter (1997), as well as four additional and similar questions. We included the four extra questions in order to create more variation among participants.

Pooling the data into a single model and using several interactions would provide the most efficient test of the equality (or lack thereof) of these coefficients. Unfortunately, due to severe multicollinearity between the necessary interaction and lower-order terms we cannot follow this approach. However, Allison (1999, 186) shows that properly specified interactive models and separate regressions are equivalent. Hoetker (2007) demonstrates that although problems of residual variation can occur when comparing logit and probit coefficients, “the researcher can—at a minimum—compare the statistical significance of the coefficients across groups” (338). This is the approach we follow here.

Ballot counters used in the analysis were coded as female with mean levels of knowledge, trust, cognition, and closure.

This figure was taken from the 2008 Minnesota US Senate race. After the recount, Franken held a 225 vote lead, although his lead was expanded to 312 votes after further legal action.