Deliberating While Voting: The Antecedents, Dynamics, and Consequences of Talking While Completing Ballots in Two Vote-by-Mail States

An overlooked context for citizen deliberation occurs when voters discuss their ballots with others while completing them at home. Voting by mail (or “absentee voting”) creates an opportunity for informal deliberation in the midst of exercising a basic form of citizen power. We examined this understudied context by blending prior theory with qualitative observations of dyadic and small-group absentee voter discussions to identify common features of such talk, which range from cynical joking and speculation on election outcomes to observing norms of politeness and engaging in heated argument. The hypothesized antecedents and consequences of those behaviors were examined in a survey of 295 Washington and Oregon voters’ recollections of their ballot discussions. Results showed that pro-deliberative features of discussion were reported most often by voters with more formal education and political knowledge. Contrary to hypotheses, the strength of voters’ partisan identities bore no relation to deliberative behavior. Finally, the presence of key discussion features had many of the expected effects on voters’ confidence in ballot choices and their respect for the electoral process, particularly for those voters with less political knowledge.

A growing body of research relevant to deliberative democracy concerns the role of informal political conversation and discussion. Talking to others about public issues can help one gain political information, develop more refined political judgments, participate in civic life, and shape one's public identity and sense of connection to government (Cramer Walsh, 2004;Eveland & Hively, 2009;Gil de Zúñiga & Valenzuela, 2011;Huckfeldt, 2001;Jacobs, Cook, & Delli Carpini, 2009;Nir, 2012). Exposure to alternative arguments during such exchanges can have positive effects on one's political sophistication (Huckfeldt, Johnson, & Sprague, 2004;Price, Capella, & Nir, 2002), though these benefits may be mitigated by one's emotions and cognition during discussion (Kim, 2013). A more skeptical view comes from Mutz (2006), who demonstrates that most political discussion involves like-minded individuals, and when disagreement occurs it can foster cynicism or uncertainty that leads to political disengagement (Mutz, 2006;Torcal & Maldonado, 2014).
The potential benefits and pitfalls of political conversation hold special significance when viewed within the lens of deliberative democracy. Truly deliberative democratic interactions involve unconstrained exchanges among people of equal standing, and well-reasoned analysis of issues based on strong information (Burkhalter, Gastil, & Kelshaw, 2002;Fishkin, 2009). Much of the research on deliberation has focused on formal events and interventions (e.g. Knobloch, Gastil, Reedy, & Cramer Walsh, 2013;Luskin, Fishkin, & Jowell, 2002;Ratner, 2004), though some scholarship has examined how everyday political discussion and conversation can be deliberative (Cramer Walsh, 2004;Jacobs et al., 2009;Moy & Gastil, 2006).
In this essay, we aim to advance the research on informal political discussion and deliberation. We begin by taking a communication-centered approach by looking not only at the aggregate-level features of political conversation (Klofstad, 2007;McLeod, Scheufele, & Moy, 1999;Mondak, 1995;Nir, 2012;Wojcieszak, Baek, & Carpini, 2010) but also at the experience of political conversations (Cramer Walsh, 2004;Eliasoph, 1998). Rather than asking whether those who discuss politics benefit from the practice, we identify some of the common features of political discussions (Kim, 2013;Kwak, Williams, Wang, & Lee, 2005) and the motivations behind citizens' engaging in them (Eveland, Morey, & Hutchens, 2011). This permits us to test the associations of these discussion features with various political outcomes, which we do through the use of voter surveys.
Our other main contribution concerns the focal context of our research. One can use the term "deliberation" to describe diverse forms of civic engagement in which participants have no requirement to reach decisions and no expectations of influence on government (e.g. Nabatchi, Gastil, Leighninger, & Weiksner, 2012), and most research on political conversations treats them as having those features (e.g. Moy & Gastil, 2006;Nir, 2012). However, there is an overlooked real-world setting for consequential political discussion. More jurisdictions than ever permit widespread absentee voting or have adopted vote-by-mail systems. Such reforms have modest effects on voter turnout (Burden, Canon, Mayer, & Moynihan, 2014;Gerber, Huber, & Hill, 2013;Karp & Banducci, 2001;Southwell & Burchett, 2000), but they have the potential for a more substantial impact because those who cast ballots by mail are more likely to discuss politics with others (Richey, 2005). A significant portion of the electorate is talking about politics shortly before or while filling out their mail-in ballot (Reedy, Gastil, & Moy, 2015). These exchanges represent an underappreciated context in which everyday citizens are engaging in consequential discussion prior to exercising their right to vote, which stands as the most fundamental form of public voice and authority in democracy (Dahl, 1989).
To understand better this overlooked phenomenon, we begin by blending a literature review with qualitative observations of actual voter conversations and discussions. We develop hypotheses and test them by collecting survey data pooled from two states that utilize voting by mail in their elections. Our findings show that this context of discussion is fairly common in states that allow voting by mail, and has some features that are both pro-and anti-deliberative. Though political discussion in this context is mainly having positive effects on citizens, the effects of those pro-and anti-deliberative features are not as straightforward as one might expect-and seem to be somewhat unequally distributed across the electorate. We conclude the article by discussing the implications of our findings and suggesting future research on political discussion.

Absentee Voter Discussion as a Deliberative Activity
To develop our theoretical model, we begin with a brief analysis of vote-by-mail elections and informal deliberation. We then combine literature on political discussion and informal deliberation with qualitative observations of vote-by-mail discussions. This helps us produce a set of testable hypotheses regarding the antecedents and consequences of the discussion behaviors that occur while voters are filling out their ballots.

Voting by Mail and Political Discussion
During the past two decades in the United States, absentee voting and voting by mail have grown in application and popularity, giving millions of citizens a chance to fill out their ballots at home, or anywhere else, before turning them in through the mail or at a polling place. Vote-by-mail systems can reduce election costs and boost turnout but may also change the act of voting. Citizens voting from home can discuss their vote choices while filling out their mail-in ballots, which is a significant departure from the modern secret ballot system used in the United States for roughly the past century (Schudson, 1999).
Few states vote exclusively by mail, but all permit some degree of "absentee voting," and many of those have relaxed absentee regulations that result in widespread voting by mail ("Voting by Mail," 2012). In some states such voting is often permitted only for those who request a postal ballot owing to a disability or temporary residence outside the state. In other states, such as California, one need only request a mail-in ballot to receive one, leading political parties to encourage absentee registration in the belief that this will boost the turnout of their supporters. Given the familiarity of the term, we refer to "absentee voter discussions" in this paper even when we wish to encompass discussions among voters who live in states that permit all voters to cast their ballots by mail.
Some traditionalists have criticized voting by mail as a solitary act. After all, it eliminates the process through which citizens flock to their local polling places, seeing their neighbors and fellow citizens working as poll volunteers, casting their ballots, and perhaps even standing in long lines to exercise their voting rights (Rose, 2002). Voting by mail can, however, make voting a more social activity. Having two to three weeks to fill out an absentee ballot may give ample opportunity for people to discuss their vote choices with family members, friends, roommates, or coworkers, especially considering that absentee voting encourages general political discussion (Richey, 2005). Recent research has illustrated how political discussion can help citizens learn about issues and candidates (Feldman & Price, 2008;Jacobs et al., 2009;Lee, 2009). For some citizens, these interactions can even occur while they mark their ballots, perhaps giving those political discussions greater importance in their decision making (Reedy et al., 2015). That situation is the focus of the present study. However, it should be noted that other voting alternatives, such as early voting programs that let citizens cast ballots days or weeks before the traditional election day, could have similar effects in spurring political discussion around elections.

Informal versus Highly Structured Deliberation
In studying how vote-by-mail discussions unfold, we hoped to discover how the discussions were structured; how participants interacted with one another; what topics and issues arose; and what sorts of information, arguments and electionrelated media came into play. More specifically, we hoped to learn the degree to which absentee-voting discussions reflected-or contrasted with-deliberative norms (Burkhalter et al., 2002;Chambers, 2003). This theoretical orientation led us to ask whether vote-by-mail discussion participants were able to consider political issues together respectfully, share speaking time, have a free exchange of ideas involving multiple points of view, and reach judgments based on their values and best available information (Burkhalter et al., 2002;Joshua Cohen, 1997).
Prior research has shown that deliberation can have such effects in highlystructured contexts, such as Deliberative Polls (Fishkin, 2009;Luskin et al., 2002), Citizens' Assemblies (Ratner, 2004), or National Issues Forums (Gastil, 2004). Less clear are the benefits of quasi-deliberative conversations that occur spontaneously among citizens. The context of vote-by-mail discussion permits us to study how voters engage in consequential political talk-and to what effect-in their homes, at coffee houses, on the phone, or anywhere a voter chooses to be while completing a mail-in ballot.
By the formal standards of deliberative democracy, such discussions will be imperfect in their design and process (Gastil, Knobloch, & Kelly, 2012). They will normally lack both diversity in participants' viewpoints and completeness in their analysis of voters' ballots. Nonetheless, it will be useful to learn more about the discursive properties of these small scale instances of "enclave deliberation" (Karpowitz, Raphael, & Hammond, 2009). In addition, our analysis is in the spirit of the call for scholars to move beyond asking whether political talk meets deliberative criteria and attend to how discussions arise and why citizens engage in those discussions (Eveland et al., 2011). Our manuscript is grounded in deliberative theory, but we also grapple with these under-analyzed issues of discussion context and citizen motivation.

An Inductive Approach to Theorizing Vote-by-Mail Discussion
Though there is a great deal of prior research on political deliberation and conversation in general, the absence of prior work in the specific context of voteby-mail discussion led us to take a more inductive approach to identifying the key discursive features (and likely consequences) of this unique context. Direct observation of such talk has intrinsic value as a form of open-ended research, but its primary purpose was to crystalize hypotheses that could then be tested quantitatively.
With this aim in mind, the first author observed six absentee voter discussions during the 2007 general election 1 in King County, Washington, which ranged in 1 In 2007, Washington state was in the process of transitioning from a system of relaxed absentee voting regulations to an entirely vote-by-mail system. King County, the site of this study, had substantial absentee voting at the time, but was not yet completely a vote-by-mail county. The size from a dyad to a group of twenty people. These gatherings consisted of friends, acquaintances, colleagues, or family members who had planned to fill out their absentee ballots together and discuss their vote choices. One observation involved a pair of roommates, two looked at married couples, two were groups of neighbors/friends who gathered regularly for each election, and one was a large group consisting of both friends and acquaintances invited to discuss that year's ballot over a potluck dinner. The first author found out about these discussions through snowball social sampling in the weeks leading up to the 2007 general election; all of these discussions were planned by the participants, and the researcher obtained their permission to observe these naturally occurring discussions and take field notes with minimal disruption to the group.
The point of such a variety of groups was to begin our analysis with a broad array of discussion settings, and the half-dozen cases varied not only in size but also duration, focus, and process. The large voting party, for instance, had a holistic approach to the absentee ballot, urging participants to shout out a particular office, candidate, or policy measure on the ballot that they wanted to discuss. Other groups, such as the younger married couple, started at the top of the absentee ballot and worked through it systematically. In order to avoid unduly affecting the group discussions, participants were not asked about their political ideology or party identification beforehand, and the nature of the larger informal voter gatherings made post-discussion questions quite difficult. However, many participants expressed moderate to liberal views during the discussions, as one might expect in a heavily Democratic city such as Seattle.
Sensitized by prior theory on political discussion and deliberation, observational field notes taken during these sessions focused on the tone and substance of the talk. Special areas of emphasis included how respectful or argumentative people were in these conversations; the topics covered in discussions, like ballot initiatives or local candidate elections; the power dynamics in the conversations, such as the pattern of interruption and monopolization of speaking turns; and the effect of topic expertise and political knowledge on the discussions, such as people deferring to an expert's opinion on a ballot measure.
The key conservational elements we extracted from these notes for subsequent testing were either ones that occurred more than once across the discussions (Owen, 1984) or constituted "critical events" that played a vital role in the dynamic or decision making of a particular group (Poole & Baldwin, 1996). discussions all happened within two to three weeks of election day, which is when mail-in ballots were sent to citizens.

Expertise, Heuristics, and Information
Of the themes extracted, one was prominent enough to be nearly invariant across the groups and, thus, not ideal for study as a variable across discussions. Each discussion was organized for the purpose of sharing expertise and information, and this constituted the principal theme of talk. Voters had gathered during an offyear election campaign to make decisions on mostly unfamiliar candidate slates and challenging public policy questions, and they turned repeatedly to their discussion partners for assistance. This finding fits with prior research showing the informative potential of voter discussion (e.g. Huckfeldt, 2001;Katz, 1957), particularly when voters are well engaged cognitively with the discussion (Kwak et al., 2005). Though political discussion is often seen as an activity for only the most knowledgeable citizens, recent work has shown that political knowledge is not always strongly connected with talking about politics and public issues (Jacobs et al., 2009;Reedy et al., 2015), meaning that these discussions could be an important context for less-informed citizens to learn about issues and candidates.
Much of the expertise shared in discussions involved indirect information. Rather than focusing on the background of a candidate or the details of a ballot measure, discussants pointed to key endorsements as "selling points" or as proxy decision makers (see Lupia & McCubbins, 1998;Popkin, 1994). During each discussion, at least one participant (and often many participants) also referenced physical or digital materials-most commonly the official state Voter's Guide.
Though appeals to expertise often helped voters make more informed decisions, sometimes another dynamic occurred. In two of the discussions, the person who appeared to have the most general political expertise wound up being the focal point of their discussions. Participants started to anticipate the experts' analysis and even ask for their opinions before anyone else got a chance to speak. In those cases the expert gained some measure of control over the direction and terms of the discussion, though not going so far as to effectively "take it over." Disparities in political knowledge and social status can cause problems in the public sphere, whether it involves those with greater status dominating discussions (Newman, 2009), or political discourse focusing on the interests of the more knowledgeable and engaged (Delli Carpini & Keeter, 1996). It remains to be seen if these issues are cropping up regularly in vote-by-mail discussions, but it certainly warrants attention in our analysis of this deliberative context.

Avoiding Controversy
Even though most of the observed discussions involved lively talk, none of them featured outright arguing or even "reasonable" hostility (Tracy, 2010). In most cases, voters treaded carefully around controversial topics or changed the subject when discussions turned too contentious. After all, these were gatherings of friends, family members, and acquaintances, not opponents on a political stage. Even though this made for respectful discourse, which is a key aspect of democratic deliberation (Burkhalter et al., 2002;Mansbridge, Hartz-Karp, Amengual, & Gastil, 2006), it sometimes meant that voters failed to examine a potential point of important disagreement.
For example, when one of the voting parties discussed the King County Prosecutor race, a participant said, "This is something I feel strongly about. Satterberg is a Republican, but I'm voting for him. He was Maleng's right-hand man and people in that office feel he's the heir apparent." Another participant replied in a slightly raised voice, "But this guy is a Republican! He raises money for them." He then added that the Democratic opponent supported gun control-a position presumed to be popular among fellow discussants.
No one followed up on this exchange. After a lull in the conversation, the first speaker's husband made a joke about unseasoned candidates, and the group briefly talked about the duties of the county prosecutor. When the Satterberg supporter reiterated her view and added that the Republican had been "in office for a long time" as deputy prosecutor, another participant simply said, "Wow. Let's move to something else." The group never returned to the topic.

Joking and Cynicism
Whether releasing tension from intense discussions (Bales, 1970) or for its intrinsic pleasure, discussants frequently joked with each other about the election, making light of some candidates or their campaign messages. Some participants joked about seemingly ridiculous candidates, such as those who subscribed to bizarre conspiracy theories. Others quipped about the most blatantly partisan campaign advertisements.
At one of the voter gatherings, a participant noted that a candidate's picture in the Voter's Guide looked more like a police mug shot than a publicity photo. The participant noted that this candidate made few public appearances. As the group passed-around the Voter's Guide and laughed at the menacing photo, one quipped that the candidate "isn't talking to anyone, he doesn't want them to see him!" In some discussions the participants grew weary of political posturing and partisan messages in campaign materials, and they responded with sardonic remarks about those tactics. Their joking and mocking seemed to serve a dual purpose: It helped reinforce their cynicism about politics (Capella & Jamieson, 1997;Eliasoph, 1998), and it helped them express their decisions about candidates and issues. Most participants lashed out at those campaigns utilizing attack ads or inflammatory language in their communication, citing the ads as examples of "politics as usual" and using them as reasons to vote against the aggressive campaign. In effect, it sharpened the sense of separation between the discussants' "ingroup" of reasonable citizens against the "outgroup" of politicians (Tajfel & Turner, 1986), boosting the discussants' cohesion at the cost of stereotyping public officials and candidates.

Deliberative Moments
Reflecting on these discussions directly in relation to conceptions of deliberative democratic discussion (Burkhalter et al., 2002) is useful. As some of the preceding examples suggest, there were moments of more rigorous analysis and times where groups declined to deliberate thoroughly. Some discussions maintained a high standard of mutual respect, while others slipped below that standard. Given that deliberative democracy stands as an ideal (Chambers, 2003), one could not expect any voter discussion to always show the analytic depth and democratic social relationships required for a fully deliberative process. We did, however, look for instances of critical moments-high points of deliberative practice where groups came closer to that ideal, since even the most democratic groups can only maintain those states briefly (Gastil, 1993).
There was a handful of such moments in these discussions. One example that stands out came from the gathering of older adults, when they considered Proposition 1, a regional measure that aimed to channel billions of dollars in tax revenue into a massive project expanding both roads and public transit. Even as Ben 2 helped lead off the discussion with a rather strong rejection of the policy proposal, he did so by giving concrete reasons for his view-the regressive nature of its sales-tax funding, the importance of fighting global warming by not building roads, the failure of the measure to address the area's crumbling bridges, and the inflexibility of light rail to meet future population growth. His friends Doris and Carrie then challenged those views by arguing that turning down an imperfect proposal would replay the area's history of not being able to accomplish anything in major public policy areas like transportation.
Ben: They say we have to support all of this to get any of it-that's blackmail! Doris: But to play devil's advocate, if you're passing something that affects three counties, I think you have to have things that appeal to people; you need to compromise.
The discussion then turned to the question of whether light rail, the backbone of the transit portion of Proposition 1, is too rigid to react to future land development, and whether a better option might be the more flexible system of bus rapid transit-which was not included in the policy proposal on the ballot. The conversation wrapped up with Ben standing his ground in opposition to Proposition 1: "There is one argument that the perfect is the enemy of the good. One could make the argument that it applies here. But it just seems this plan is so bad." Even in voicing his final word on the ballot measure, Ben offered an olive branch to his fellow discussants-one that could also help them feel more justified in disagreeing with him and voting in favor of Proposition 1.
In sum, this incident illustrates a discussion showing strong signs of deliberation. Each participant was given chances to speak up in a respectful setting, and all participants seemed to be truly considering other group members' points of view and those of experts and political leaders. There were few significant disparities between the discussants-all were older, affluent, white, and concerned about environmental and transportation issues-but those differences in opinion that did arise were addressed in each person's attempts to set up common ground on those issues during the discussion. Although all of the participants were obviously politically liberal, one of the participants cited more moderate-to-conservative views to help balance out the discussion.

Summarizing the Key Features of Vote-by-Mail Discussions
Even this small sample of vote-by-mail discussions shows tremendous variation, ranging from relatively deliberative engagement to conflict avoidance to passive reliance on expertise. Many participants and groups were open to, and even relied upon, the political knowledge and expertise of other group members, which helped promote the consideration of a wider array of views and information. That benefit was double-edged, however, as evidenced when some groups were dominated by self-proclaimed experts-a finding that is in keeping with other scholars' warnings about disparities in political knowledge (Delli Carpini & Keeter, 1996). Discussions often moved past simple endorsements or appeals to expertise, as participants tried to reconcile conflicting information or at least seek corroboration of their first impressions before moving to the next item on their ballots.
As for the more relational aspects of the discussions, jokes and cynical remarks arose often in the discussions. These seemed to help relieve tension between participants and draw them closer together as a group, and they even helped people learn about issues and candidates. At the same time, they also worked to reinforce negative impressions of politicians and "politics as usual" in campaigns. Participants were often very respectful of each other's points of view and right to speak up, and none of the gatherings ever dissolved into outright arguing. This agreeable tone, however, foreshortened some discussions of controversial ballot issues and likely prevented some participants from voicing legitimate concerns about a candidate or proposal.

Hypothesizing the Antecedents and Consequences of Voter Discussions
The direct observation of ballot discussions yielded more than just focal categories of communication behavior. Mixed with prior work on political conversation generally, the first-hand observations also suggest testable hypotheses regarding the antecedents and consequences of these discussion behaviors. 3 We will examine each of these hypotheses through a survey of voters who recalled having ballot discussions, but first, we enumerate our predictions.
To simplify the analysis of the six behaviors we examine, it can help to think of three as pro-deliberative and three as anti-deliberative in relation to Burkhalter, et al.'s (2002) definition. On the one hand, absentee voter ballot discussions often emphasize deliberative analysis and disagreement, along with civility or respect.
On the other hand, they can foster fear of heated conflict, acquiescence to unchallenged expertise, and digressions into cynicism and forecasting of electoral outcomes. Table 1 summarizes this breakdown of the six discussion behaviors along these lines.  Table 1 identifies the antecedents about which we advance hypotheses. We begin by predicting that those with more formal education and/or political knowledge will bear witness to the most pro-deliberative behaviors. In the groups we observed, these individuals stimulated much of the best ballot analysis, as well as heated-but substantive-conflicts during discussion. These predictions are consistent with prior research, which has shown the importance of education and knowledge in discursive participation (Jacobs et al., 2009), as well as in political engagement more generally (Delli Carpini & Keeter, 1996;Luskin, 1987).

Behavioral Antecedents
We also expected that discussants who are more partisan (i.e., strongly identified with a major party) will be more likely to make focused deliberation difficult. In our initial observational research, it was the most passionate speakers who foreclosed debate and spoke more cynically about the electoral process. Even though popular accounts ascribe such political behavior to the wider public, such emotions and strident speech come most readily for strong partisans (Fiorina, Abrams, & Pope, 2010). In terms of the behaviors measured in our survey, this means that partisan respondents will be more likely to recall cynical talk and speculation on election outcomes. They were expected to preside over discussions in which participants uncritically accept expert advice, but given the strength of their partisan convictions, they would also be less likely to keep quiet to avoid tension.

Discussion Outcomes
As for the outcomes of such talk, the discussions we observed showed signs of voters changing their minds as a result of discussion, and it is important to understand which aspects of the discussion affect voters' confidence in their choices. Research on political discussion has shown that citizens can get greater clarity in their voting choices by talking through their political choices with others (Huckfeldt, et al., 2004;Jacobs, et al., 2009), but more specifically, we expect that only the pro-deliberative behaviors will promote confidence in voting choices (Gastil, Black, & Moscovitz, 2008). By contrast, two anti-deliberative behaviors-the suppression of conflict and uncritical acceptance of advice-were expected to undermine voter confidence in voting choices.
Though ballot discussions convene for the overt purpose of making voting choices, an apparent secondary purpose is strengthening the relational bonds among those taking part. Thus, the discussions may often have an impact on voters' attitudes toward one another-one of the hallmarks of both everyday political conversations and public forums (Cramer Walsh, 2004. Here, the relationships are not as straightforward. We expect that recollections of respect and civility will correlate with stronger bonds with fellow discussants, but the other pro-deliberative features (playing the devil's advocate and heated argument) may have the opposite effect, owing to the unfortunate tension that often exists between conflict and connection (Pearce & Littlejohn, 1997). Meanwhile, the anti-deliberative behaviors that suppress substantive disagreement might be thought to promote group bonding by avoiding tension (Bales, 1970), but in the context of discussions designed to yield information to improve voting choices, we expect such behaviors would make participants feel more distant from their fellow discussants.
Absentee voter discussions also could have implications for participants' attitudes toward the broader electoral process. Just as citizens often feel more respect for democracy after engaging in a formalized jury deliberation (Gastil, Black, Deess, & Leighter, 2008) or other forms of deliberative engagement (Jacobs et al., 2009), so might these informal discussions have the same effect, to the extent that they promote deliberation and avoid anti-deliberative tendencies, particularly cynical talk about elections and politics.

Low-Knowledge Voters and Deliberation
Finally, past research on political discussion has underscored the difference in the experiences of those with varied levels of political expertise (Huckfeldt, 2001;Huckfeldt et al., 2004). Our first-hand observations of absentee voter discussion showed the importance of this difference, with the most politically sophisticated participants taking the lead and exerting influence more than receiving advice on their ballot choices. Moreover, recent research has found that the impact of deliberative experiences can be greater for those less knowledgeable about conventional politics (Gastil, Black, Deess, et al., 2008). This leads to our final hypothesis: Relative to their highly sophisticated peers, those participants with low political sophistication will show a stronger association between their reported discussion behaviors and outcomes.

Research Method
To test the hypotheses summarized in Table 1, we analyzed data from a pair of statewide public opinion polls. The 2008 Washington Poll surveyed by phone 1,203 registered voters from across the state, including an oversample of African-American voters. 4 This was combined with data from a 2012 online survey of 1,539 Oregon voters, collected by mass emailing to a list of registered voters created by a commercial mailing-list firm. 5 The aim was to create a diverse and broadly representative sample of people voting by mail in these two states, and the resulting sample met that aim. Roughly half of the voters were female (49%), nearly two-thirds (62%) were 35-65 years of age, 81% were white, and the median ideological identification was moderate (4.03 on a 1-7 scale). These roughly parallel the demographic and political profile of these two Pacific Northwestern states. Survey question wordings are provided in the sections below, in Table 2 (the descriptive statistics table), or in the included appendix.

Participation in Absentee Voting Discussion
Forty-two percent of those surveyed reported that they had already voted absentee, and of those, 30% reported that they had talked "to friends or family members to get advice on one or more of [their] voting choices." 6 This figure suggests that a substantial portion of the electorate is engaging in absentee voting discussion. All subsequent analyses refer to the subsample of 295 respondents (including 164 from Washington and 131 from Oregon) who reported having discussions with others while filling out their ballots. The size of this subsample limits our quantitative analysis somewhat, as it provides only enough statistical power to detect medium to large effects using regression analysis with multiple predictor variables (Jacob Cohen, 1992). As such, our results below should be taken as a preliminary study of a relatively under-examined context of political deliberation.

Features of Absentee Discussion
Respondents answered a series of questions about their ballot conversations to measure six discussion behaviors.  Minimum N = 290.
The survey began by asking how often (very often, occasionally, or never) participants engaged in five behaviors: "Heated arguments about issues or candidates" (the modal response [61%] was "never" seeing this); "Polite and respectful discussion of issues and candidates" (63% saw this "very often"); "Cynical talk and jokes about elections and politics as usual" (42% saw this only "occasionally"); and "speculation about the outcome of the election" (56% saw this "very often"). In addition, the survey asked two yes/no questions: "Did anyone play devil's advocate to present the other side of an issue or debate? (49% said that this happened); and "Did you keep quiet on an issue or candidate just to avoid creating tension or starting an argument?" (29% reported doing so).
The intent of this analysis was not to derive multi-item behavioral scales, but rather to test hypotheses against each item individually. Nonetheless, Table 3 provides the zero-order correlations among these items for those interested in considering the ties among them. The items were more often moderately correlated than not, but none of the associations was strong (maximum r = .31). That said, it was notable that the perception of cynical talk and speculation about outcomes were positively associated with the three pro-deliberative behaviors in five out of six possible instances, with an average correlation of r = .22. Note. + p < .10, * p < .05, ** p < .01, two-tailed tests.

Perceived Outcomes of Discussion
The last four items in Table 2 show the discussion outcome measures included in the survey. One question asked whether, as a result of their ballot conversations, respondents felt more confidence in any of their ballot choices, and a separate item asked whether they felt less confidence in any other choices. That two-item pair was intended to capture the possibility of a discussion having countervailing effects across different issues on the ballot. Table 2 also includes descriptive statistics for the items asking whether respondents, "as a result of your discussions," felt closer to or farther from their fellow discussants, and/or felt more or less "respect for the larger electoral process."

Demographic and Political Background Measures
Education was measured on a six-point scale (M = 4.58, SD = 1.01). Forty-nine percent had completed some college or less, 29% had completed college but not gone further, and 22% had more advanced graduate or professional education.
Knowledge of politics was measured by totaling a respondent's score on several federal and state political knowledge questions (e.g. "Do you happen to know what job or political office is now held by John Roberts?"; see appendix for full list of questions), α = .65, M = 2.26, SD = 1.29. The scores were standardized to allow for analysis across the two surveys. For the final hypothesis test, this scale was split into a low-versus high-knowledge contrast, with approximately 60% of respondents (i.e., those with standardized political knowledge scores below the mean) being placed in the low-knowledge group.
Strength of partisan identity was arrayed on a four-point scale, including no major party affiliation (0), independent leaning toward a major party (1), major party members with weak ties (2), and major party members with strong ties (3). The mean response was 2.02 (SD = 1.06), with a modal response of "strong" partisanship (47% of respondents).
Finally, interest in politics was measured with a question asking the respondent their level of interest in "what's going on in government and politics," with responses coded on a five-point scale from "not at all interested" (1) to "extremely interested" (5).
Conventional survey items measured sex, age, and ethnicity. For the main analyses, these and all other continuous variables were mean-centered to reduce multicollinearity. Also, for the purpose of testing directional hypotheses, onetailed p-values were used to assess statistical significance (Kimmel, 1957).

Predictors of Discussion Elements
When linear regression equations were created to predict each discussion element using demographic and political background variables, three of the models failed to reach even the more modest .10 significance level (see Table 4). 7 As Abelson (1995) has emphasized in relation to effect size generally, however, the important question is whether there is a pattern of significance that has a substantive meaning. In this case, such a pattern exists for two variables we foregrounded in our theoretical discussion. More often than not, education and political knowledge had significant positive relationships with pro-deliberative behaviors, as predicted. This was clearest in the case of playing Devil's advocate, which was positively associated with both education (β = .13) and knowledge (β = .15).
Partisanship had only one of the four expected associations, with the others near zero or contrary to prediction. As hypothesized, the more partisan the respondent, the less likely they were to keep quiet to avoid conflict (β = -.19). The survey findings contradicted our prediction regarding cynical talk and joking, a behavior that was less frequent the stronger the respondent's partisanship (β = -.16). 7 Because some of our variables (e.g., frequency measures) could be interpreted as ordinal variables, we also ran logistic regressions in addition to the linear models presented herein. The substantive results of these alternative equations were essentially the same. Note. + p < .10, * p < .05, ** p < .01, two-tailed tests. One-tailed tests were used to test directional predictions.

Direct Associations between Discussion Elements and Outcomes
The next set of regression equations treated the voter discussion elements as predictors of outcome measures. The two left-hand columns in Table 5 show some support for our hypotheses, but one surprising finding, as well. Consistent with predictions, polite talk was positively associated with more confident voting choices (β = .15), and keeping quiet to avoid conflict was negatively associated with such confidence (β = -.12). The other expected finding was a positive link between uncritical acceptance of advice and voters' losing confidence in their choices (β = .10), but heated argument was a stronger predictor of that drop in voter confidence (β = .21). The latter finding suggests that lively argument may have less value than expected and could even undermine voters' ability to make confident choices on their ballots. Another possibility, though, is that a voter who has their mind made up on their ballot choices but then encounters differing perspectives could leave that discussion feeling less confident in their choices.
The third column in Table 5 shows the relationships between the various discussion behaviors and voters' sense of feeling closer or farther apart from their fellow discussants as a result of talking over their ballots together. All five of the predicted associations were in the expected directions but only two reached (or approached) statistical significance. Those who kept quiet to duck potential conflicts felt less connected afterward (β = -.13), but those who found themselves in heated conflicts had the same fate (β = -.11, one-tailed p = .056).
The final regression equation in Table 5 tested the relationship between each of the elements of voter discussions and respondents' sense that their ballot talk had decreased or increased their respect for the "larger electoral process." All of the predicted associations ran in the expected direction, except that choosing to keep quiet to avoid conflict had no association whatsoever. The strongest predictor was heated argument (β = .21), with polite talk adding a marginally significant additional boost (β = .10, p = .056). As expected, anti-deliberative talk undermined respondents' respect in elections, with negative associations coming from spending time on election forecasting (β = -.14), accepting advice without question (β = -.11), and engaging in cynical talk and joking (β = -.10, p = .07). Note. + p < .10, * p < .05, ** p < .01, two-tailed tests. One-tailed tests used for directional predictions of effects from discussion elements. Values are upon-entry regression coefficients for block indicated.

The Moderating Effect of Political Knowledge
The final hypothesis revisited all these outcome associations to consider whether the effects were stronger for those with low versus high levels of previous political knowledge. To test for a general interaction between political knowledge and discussion elements, the same regression equations shown in Table 5 were augmented with interaction terms between each element and the political knowledge dichotomy. The results are shown in Table 6, with each significant interaction term corresponding to bold text for the corresponding semi-partial correlations at two levels of political knowledge. Note. + p < .10, * p < .05, ** p < .01, two-tailed tests. One-tailed tests used for directional predictions. Minimum n = 163 (low-PK) and 128 (high-PK). Control variables included sex, age, ethnicity, education, ideology, and partisanship. Bold text indicates a statistically significant interaction term in regression (two-tailed p < .05).
The clearest support for this hypothesis concerns voters' confidence in their ballot choices. Three of the five predictors interacted with political knowledge, and Table 6 shows that their associations were stronger (and in the expected direction) for those with less knowledge. For voters with high political knowledge, in fact, the performance of the devil's advocate role had a negative effect on their confidence in voting decisions (partial correlation [pr] = -.18). By contrast, lowknowledge voters had the hypothesized positive association between devil's advocacy and confident choices (pr = .18). Polite/civil talk boosted lowknowledge voters' confidence (pr = .36), whereas going so far as to engage in conflict-avoidant silence undermined confidence (pr = -.29). For high-knowledge voters, the same two associations were non-significant (and in the opposite direction).

Discussion and Conclusion
Looking across both qualitative observations and quantitative survey results, we found that discussions between people voting by mail can have fairly positive effects, even if they left some people feeling less confident in their decisions. All of the participants from the qualitative observations appeared glad to have chosen to talk through their choices together while poring over their ballots with official state voting guides, campaign ads, and newspapers in hand (or on laptops). Our survey data showed that, on balance, voters gained confidence in their decisions far more often than they lost it. In general, discussants also grew closer to one another and gained more respect for the electoral process.
Some of those benefits appear to distribute themselves widely across the population, though as we hypothesized, pro-deliberative features of political discussion were reported more often by voters with more formal education and political knowledge. We had predicted that the most partisan voters would report participating in less deliberative talk, but this was not the case. Strong partisans and more independent voters appear equally likely to have deliberative conversations, or the converse. Our data suggest that many people were engaging in conversations with people who disagreed with them, or at least played devil's advocate, but we did not explicitly ask about the partisan make-up of discussants' conversations. In light of research results showing important differences between like-minded and mixed political discussion (Huckfeldt et al., 2004;Mutz, 2006), as well as the rise of polarized political discourse in general (Mutz & Young, 2011), future research on talking while voting could examine the deliberativeness of homogeneous and heterogeneous conversations.
As a result of vote-by-mail discussions, most voters reported greater respect for the electoral process and a stronger connection with their discussion partners, but those changes were somewhat less related to pro-and anti-deliberative behaviors than expected. Some anti-deliberative features of discussion, such as keeping quiet to avoid conflict and sharing unquestioned voting advice, did result in voters feeling more distant from their discussion partners and more ambivalent toward the electoral process. The effect of pro-deliberative features was less substantial. Both heated argument and polite talk were positively related to feeling more respect for the electoral process (though polite talk only approached significance as a predictor), and arguing was marginally related to feeling more distant from one's discussion partners. For some citizens, it may be that a more adversarial style of democracy (Mansbridge, 1983;Tracy, 2010) better suits their conception of how elections should unfold, though it comes at a cost of disconnection from one's fellow citizens.
The benefits of deliberative voting discussions appear more pronounced for less politically knowledgeable voters, who also feel more acutely the sting of being silenced or seeing expert advice go unchallenged. These voters, traditionally labeled as less "sophisticated" in political science (e.g., Luskin, 1987), were likely to lose confidence in their voting choices when discussions turned into arguments, a finding that appeared in both the survey data and qualitative observations. This suggests the need for balancing the deliberative emphasis on respect (Burkhalter et al., 2002) and agonistic debate (Goi, 2005;Tracy, 2010) to get maximum benefit from these conversations, an approach consistent with Mansbridge's (1983) aim of fusing adversarial and unitary democracy.
At a time when many citizens feel disconnected from politics and government (Capella & Jamieson, 1997;Mathews, 1994) and have a limited grasp of politics and public life (Delli Carpini & Keeter, 1996), voting discussions appear to convey useful information about the candidates and policies on the ballot. These benefits were found for both partisan and independent voters, with the effects more pronounced for those with less political knowledge. Making it easier for people to cast a ballot with confidence may ultimately increase both the frequency and deliberative quality of electoral participation.
That said, some of what we initially saw as anti-deliberative features of discussion-cynical talk and joking about politics and speculation about electoral outcomes-may be beneficial at times. We initially conceived of outcome speculations as a focus on horse-race politics at the expense of substantive policy debate (Hallin, 1992), but such talk made less knowledgeable voters more confident in some of their choices-though perhaps less confident in others. Lowknowledge voters may benefit from any kind of political analysis, even one deliberative theorists disparage.
Likewise, cynical talk and jokes about politics had an unexpected positive effect on voter confidence, an effect seen most clearly for more politically knowledgeable voters. Perhaps these voters are so well-versed in politics that they can joke about corrupt and ethically questionable political figures without feeling a need to disengage from politics entirely. Also, cynical talk and joking was negatively related to strength of partisanship, meaning that stronger partisans were less likely to engage in or witness this behavior. Perhaps politically knowledgeable independents, who by definition resist any party identification, also like to mock "party politics as usual" in a way that helps them blow off steam and build confidence in their political decisions. Further research should address these seemingly non-or anti-deliberative features of discussion to give greater insight into their purposes and effects. Such work may go "beyond deliberation" (Eveland et al., 2011), but it also reveals deliberative (and anti-deliberative) moments that occur in informal settings.
This study has its limitations, though. Our regression models were likely missing important unmeasured variables; our analysis relied on fairly standard political and communication measures often related to voting and discussion behavior, but future research may uncover variables of greater predictive value than these. Our modest sample sizes reflected the fact that this study used a subset of a full sample. Now that one can estimate the frequency of this form of discussion behavior, it appears feasible to conduct a survey focusing entirely on those who talk while voting.
Limitations aside, this study makes clear that ballot conversations can have a range of impacts, with the greatest benefits generally stemming from those discussions replete with pro-deliberative features. Such discussions represent more than a gathering of heuristic cues. They can provide more thoughtful analysis, move into pointed argument, or devolve into silencing-or selfcensorship-of dissenting views. The findings presented here suggest that voteby-mail discussion is an important context of informal deliberation, with substantial effects on citizens. Researchers should continue to investigate this process of vote-by-mail discussion and its benefits to understand better its relevance for traditional theories of political communication and voting. This will also help us better estimate the net impact of state and county governments granting voters permission-or even requiring them-to cast their ballots by mail.