Election Law Journal: Rules, Politics, and PolicyVol. 10, No. 1 Original ArticlesFree AccessWho Converts to Vote-By-Mail? Evidence From a Field ExperimentNathan W. Monroe and Dari E. SylvesterNathan W. MonroeSearch for more papers by this author and Dari E. SylvesterSearch for more papers by this authorPublished Online:17 Mar 2011https://doi.org/10.1089/elj.2009.0058AboutSectionsPDF/EPUB Permissions & CitationsPermissionsDownload CitationsTrack CitationsAdd to favorites Back To Publication ShareShare onFacebookTwitterLinked InRedditEmail IntroductionOver the past 30 years, a number of states have instituted election reforms aimed at lowering the costs associated with voting. With the expansion of various alternative voting options, reformers have hoped to remedy the low levels of voter turnout that have marked modern American democracy. Following suit, scholars have increasingly turned their attention to “convenience voting” (Gronke et al. 2008) mechanisms, studying voter registration through motor voter acts (Highton and Wolfinger 1998), easing restrictive registration rules (Wolfinger and Rosenstone 1980; Nagler 1991; Highton 1997), early voting (Gronke et al. 2007; Stein 1998; Richardson, Jr. and Neeley 1996), same-day registration (Burden et al. 2009), no-excuse absentee (Leighley and Nagler 2009) and centralized vote centers (Stein and Vonnahme 2008; Brady and McNulty 2004). Among them, however, no voting reform has had more attention—from either reformers or scholars—than vote-by-mail (Gronke et al. 2007; Southwell 2004; Southwell and Burchett 2000; Kousser and Mullin 2007; Arceneaux et al. 2009).Vote-by-mail (VBM) began to gain momentum in the late 1980s and found the national spotlight in the early 1990s when Oregon citizens approved an initiative to conduct all elections by mail. Since then, VBM's popularity has expanded significantly, becoming the modal form of voting in several states, with potential for many more to follow. All counties but one in Washington State use VBM exclusively. In the 2008 general election, nearly 65 percent of all votes cast in Colorado were permanent mail ballots.1 At the time of this writing, a majority of states have a form of VBM prior to the election without the need for the voter to submit a reason to obtain a mail ballot (National Conference of State Legislatures website). This form of voting, typically referred to as “no excuse absentee voting” is currently available in thirty states. It is important to note that there are several forms of vote-by-mail that exist across states, including those (such as Oregon's) in which law mandates that voting is conducted by mail and those (such as California's) in which voters have an option to cast their votes solely by mail. For the purposes of this article, we shall limit our analysis to the latter, which has also been referred to as permanent vote-by-mail. In this case, voters make the choice to submit a one-time request to become a VBM voter, and then cast future votes by mail.Voters' choices to convert to permanent VBM status have not been studied until now. These choices may be quite complex and interesting. An easier, less costly voting mechanism that can be used at a time and place of their choice may appeal to some voters. For others, opting to use VBM may mean foregoing some significant psychological benefits of going to the polls, such as fulfilling expressive needs publicly (Fiorina 1976; Riker and Ordeshook 1968) or yielding to perceived social pressure to vote (Green et al. 2008). Moreover, some voters may feel uncomfortable voting without the assistance of poll workers, or trusting the U.S. Postal Service to deliver their mailed ballot to the correct destination.2In this article, we begin unpacking the dynamics behind the VBM choice.After a state or county adopts VBM, will lowering the cost of signing up convert significant numbers of new voters to become permanent VBM voters?When VBM does convert new voters, what types of citizens will be most likely to adopt VBM through the low-cost option?To answer these questions, we combined a field experiment with supplemental survey research. We conducted our field experiment prior to the June 3, 2008, election in California's San Joaquin County, where signing up for permanent VBM is an option for registered voters.3 We found that citizens who received information on VBM and opportunities to become VBM voters—in the form of already-filled-out postcards—are more likely to convert to that voting method.Next, we used survey data to assess whether the effect of the treatment is conditioned by a voter's turnout frequency. We first investigated conversion rates within subsets of the population based on education, income, ethnicity, language, and homeowner status, and found that traditionally “high propensity” voting groups were more responsive to the postcard. We then used a multivariate probit model to look directly at the interactive effect of our treatment and prior turnout on conversion to VBM, and found further support for the notion that frequent voters are more likely to respond to VBM conversion efforts. These findings extend the logic of Berinsky (2005; Berinsky et al. 2001) that shows that among VBM voters, turnout improved mainly amongst already high-propensity voters. Our analysis focuses on voter participation prior to the voting stage, asking whether an outreach effort draws more citizens into the voting process by encouraging them to register as permanent vote-by-mail voters.In the next section, we survey previous literature, providing the foundation for our theoretical framework and core hypothesis, which occupy the section that follows the literature review. With predictions in place, we proceed to describe our field experiment and results, then give our supplementary analysis using survey data. In the final section, we briefly summarize and conclude.Whither Vbm?As VBM has grown in prevalence, particularly in the western portion of the United States, the VBM movement has become a highly controversial electoral reform in the voting literature. Some scholars have found evidence of VBM's positive impact, for example cost savings, increased turnout, improved convenience of voting, and increasing acceptance among the electorate and election administrators. For instance, elections conducted entirely by mail are approximately 33–50% less expensive to run than elections conducted entirely at polling places (Southwell 2004). Moreover, some have found that VBM increases the percentage of registered voters turning out to vote (Gronke et al. 2007; Southwell and Burchett 2000).4 And, in her analysis of the Oregon VBM system, Southwell (2004) finds widespread majority support for the system five years after its adoption for all elections. Her findings seem to cast doubt on the notion that high initial rates of support for VBM were based on the novelty of the system rather than true satisfaction.On the other hand, skeptics suggest that VBM “has not changed who makes up the electorate, but only how they vote” (Fortier 2007). In other words, rather than adding new voters to the pool of voting Americans, VBM seems to improve the convenience of voting for those who were already inclined to vote, and this effect seems fairly robust across election types such as special elections (Southwell and Burchett 2000) as well as primary and general elections (Berinsky et al. 2001). Berinsky et al. therefore argue that any increase in the voter pool is based strictly on better retention of previous voters, rather than mobilization of new recruits (Berinsky et al. 2001).Furthermore, some scholars have found that pernicious effects may result from the institution of VBM. Berinsky, for instance, found evidence across multiple studies that VBM may contribute to widening socio-economic gaps in turnout, such that more educated, older, higher-income, and Caucasian voters would enjoy higher rates of turnout owing to VBM (Berinsky 2005). Kousser and Mullin also found a slight but statistically significant reduction in voter turnout when comparing VBM districts to traditional-polling-place districts (Kousser and Mullin 2007).5No previous research on VBM, however, has focused on voters' choices to convert to VBM. In a state where both traditional voting and VBM choices exist, which citizens will choose to vote by mail? One motivation behind the VBM movement is the notion that it will open up voting to segments of the population that currently find it difficult to vote. Does this low-cost voting option really draw in voters who find it too difficult to get to the polls? In the next two sections, we tackle these questions, first theoretically and then empirically.Cutting Costs of Signing Up For VbmDowns (1957) proposed that would-be voters consider the costs and potential benefits of voting before deciding to do so. The strictest interpretation of his framework seems to fall flat: In most cases, a single vote will not be decisive, and if Downs is correct, no “rational” person should show up and vote.Inaccurate though that prediction may be, there is still something appealing about the basic calculus of voting that it suggests. Specifically, it seems useful to assume that, in broad terms, as the costs of voting decrease, citizens are more likely to vote, all else constant. A voter who lives 30 miles from the polling place is almost certainly less likely to vote, on average, than someone who lives across the street.But how does the notion of the cost of turning out to vote relate to the decision to sign up for VBM? VBM is considered a lower-cost voting option, but VBM sign-up carries its own costs. Compared to the act of locating and traveling to a polling site for each election, becoming a permanent VBM voter carries a one-time cost that typically takes attention, effort, and knowledge. This up-front, fixed cost of signing up for VBM must be made up for in cost savings over all the future instances when the voter uses VBM instead of paying the extra cost to go the polls.The flow chart in Figure 1 illustrates this point. The voter, represented in the oval on the far left, faces three choices: Pay no cost and do not vote; pay cost C to vote in person; or pay cost V to sign up for permanent VBM status, and then pay cost M to vote by mail. Assuming, however, that there is a latent preference to cast a ballot, the relative size of the costs outlined here are instructive. Of course, where the cost of signing up for VBM plus the cost of casting a ballot by mail is lower than the cost of showing up to vote (i.e., V+M<C), even for one election, then voters obviously have an incentive to sign up for permanent VBM. But notice the permanent part of VBM. Even where it would be cheaper in the short run (one or two elections, perhaps) to simply vote in person, the voting discount from VBM can eventually offset the initial cost, V, of signing up.FIG. 1. Costs of voting in person and by mail. VBM, vote-by-mail.Either way, if citizens face lower costs of signing up for VBM, they will find it more appealing to convert. Thus,Hypothesis 1:By reducing the costs of signing up for vote-by-mail, more voters will convert to permanent VBM status, c.p.As noted earlier, individuals with lower socioeconomic status (SES) tend to vote at significantly lower levels, even when offered lower-cost options like VBM (Berinsky 2005; Berinsky et al. 2001). Indeed, such lower-cost options may actually lead to wider gaps in turnout if higher-SES individuals take advantage of these options while low-SES individuals do not. However, what if there is an even lower-cost way of signing up for a low-cost option such as VBM? Despite an overall increase in the number of individuals signed up for VBM, if Berinsky (2005) is correct, certain subgroups of the population will be less motivated by sign-up cost reduction efforts such as an already filled-out postcard. Thus, Berinsky's argument implies the following about the decision to convert to VBM:Hypothesis 2:The effect of receiving the postcard will be conditioned by a voter's prior likelihood of turning out, c.p.In the next section, we evaluate this hypothesis with the results of a field experiment and survey supplement.Undoubtedly, this rational voter framework excludes certain other factors that drive individual decisions to turn out, such as the psychological benefits of behaving in an altruistic manner (Jankowski 2002), fulfilling one's civic duty (Hardin 1982), satisfying one's expressive needs in a public setting (Fiorina 1976), or acquiescing to social pressure (Green et al. 2008). Indeed, the presence of such psychological motivations to vote—and to vote publicly—may imply our null hypothesis that offering a low-cost alternative to voting in person will have no effect on the decision to vote by mail. In other words, inasmuch as psychological motivations drive the choice of voting method, variation in conversion costs should be less relevant.Research Design: A Field ExperimentTo test our hypotheses, we ran a field experiment6 using the entire registered voting population of San Joaquin County, California, as our subject pool.7 Unlike observational studies, experimental research boasts stronger causal inference about treatment effects “through the transparency and content of experimental procedures, most notably the random assignment of observations to control and treatment groups” (Druckman et al. 2006, 627). Random assignment thus provides a more lucid test of the explanatory variable of interest, alleviating persistent methodological problems pertaining to inference (Bergan 2009; Ha and Karlan 2009; Bedolla and Michelson 2009).We focused our efforts on the June 2008 statewide primary election. Using this as our election of interest is ideal insomuch as field experimental studies on voter turnout have been shown to be more effective in less salient non-presidential elections (Gosnell 1927; Gerver et al. 2003). We began by using an algorithm to randomly identify 60 percent of the voting precincts within the county as our treatment group.8 This group consisted of approximately 250 precincts, containing a total of 101,553 registered voters, leaving a control group of approximately 160 precincts and 82,903 voters.9 On April 15, 2008, all registered voters in the treatment group who were not already signed up for permanent VBM were sent a postcard. The postcard informed them about the option of permanent VBM status. It was pre-printed with their name and other information, already had the box check-marked to indicate desire to become a permanent VBM voter, and included return postage (see Figure 2 for a picture of an example postcard). Voters needed only to sign their name and put the postcard back in the mail in order to gain permanent VBM status.FIG. 2. Vote-by-mail postcard mailed to treatment group.Of course, not all the voters in the treatment group actually got the postcards, paid attention to them, understood them, and considered the VBM option. However, it seems safe to assume that since no one in the control group got a postcard, the rate of voters who received the treatment was much higher in the treatment group. In this way we are estimating the intent-to-treat (ITT) effect (Gerber and Green 2000; Arceneaux 2005).We turn first to consider Hypothesis 1, that by reducing the start up costs of VBM, more voters will obtain permanent VBM status. For those in the control group, becoming a permanent VBM voter would have required considerably more effort than simply returning a pre-completed postcard with their signature.10 Certainly, then, voters in our treatment group who received our postcard had a significantly lower cost of signing up for VBM. Thus, based on Hypothesis 1, we should expect to see a significantly larger proportion of voters signing up for permanent VBM status in the treatment group than in the control group.As shown in Figure 3, that is just what we found. Of the 101,553 individuals who received the postcard treatment, 20,400 (20%) subsequently converted to permanent VBM status before the June election. The control group of 82,903 subjects, on the other hand, yielded only 8,151 (10%) new permanent VBM voters during the same time period. The conversion rate in the treatment group was approximately double that of the control group, and the difference is highly statistically significant (p<.001). At first glance, this result is encouraging. It seems that, ceteris paribus, by lowering the start-up cost of entry into permanent VBM status, voters are substantially more likely to take advantage of this option. However, results obtained from our survey supplement demonstrated that not all else is equal. In the next section, we review these findings.FIG. 3. Voters converted to vote-by-mail by postcard mailer. Difference significant at better than 0.001.Survey SupplementTo supplement our field experiment, we ran a survey of voters in San Joaquin County in May 2008, leading up to the June election, to help us gain more information about our population of interest. The Social Science Research Center (SSRC) at California State University, Fullerton, conducted a telephone survey completing interviews with 542 registered voters in San Joaquin County. Five hundred nine (93.9%) interviews were completed in English and 33 (6.1%) in Spanish.11 The actual survey instrument can be found in Appendix 1. By surveying a sample of registered voters in San Joaquin County, we can better evaluate our second hypothesis, that the impact of the postcard treatment on conversion to VBM will be conditioned on a voter's likelihood of turnout and on their socioeconomic status.12Conditioning the treatment on individual characteristicsIt is well known that there is a significant turnout gap based on a number of socioeconomic factors, with members of lower-SES groups less likely to turn out. Contrary to the hopes of reformers, Berinsky's work (2005; Berinsky et al. 2001) shows that these turnout differences are sustained—or even increased—by VBM. Here, we move one step back in the process, asking whether the VBM turnout gap might begin to widen at the juncture where citizens choose how they will vote—in person or by mail. If citizens who are already likely to turn out respond to the low-cost VBM conversion option, while low-propensity voters fail to respond, it would indicate a selection effect that precedes the choice to turn out but has important implications for the makeup of the active electorate.Table 1 shows the effect of the postcard treatment on conversion to VBM, broken down by education level. Among those with the lowest levels of formal education—specifically, those with a high school degree or less—the decision to convert to VBM was unaffected by the receipt of a VBM postcard reminder, in line with our second hypothesis. In contrast, among those having at least some college education or an associate's degree, receiving a postcard made an individual significantly more likely to sign up for VBM, and the result was substantial. The receipt of a postcard among those with some college education more than doubled the probability of signing up. Among those with an associate's degree, the card more than tripled the probability of converting to VBM. Though we see similarly large effects in more educated groups, in substantive terms, they fall short of statistical significance (these effects are not great enough to be differentiable from pure chance).Table 1. The Effect of the Postcard Treatment on Conversion to VBM by Education Level Control (No Postcard)Postcard TreatmentPr(|Z|<|z|)Less than High School6.25% (1/16)5.26% (1/19).90High School/GED14.71% (10/68)19.72% (14/71).43Some College12.12% (8/66)26.03% (19/73).03**Associate's Degree11.11% (4/36)40.74% (11/27).01**Bachelor's Degree10.53% (4/38)20.51% (8/39).22Graduate or Professional Degree12.2% (5/41)40% (4/20).41Cell percentages represent the proportion of registered voters who converted to permanent VBM status during our target period. The proportion in parentheses represents the number of converts in that group (numerator) over the total number of individuals in that group (denominator), where “groups” are defined by the combination of an education level and the treatment/control condition. The final column (Pr |Z|<|z|)) indicates the results of a difference of proportions test between the control and treatment conditions in a given row. **indicates p<.05, for a two-tailed test.In Table 2, we see that postcard treatment effects are discernable only among those with the highest level of income. Receiving a postcard reminder has no statistically significant effect on the decision to register for VBM except among those earning more than $100,000 per year. That said, once a voter reaches an annual income of at least $30,000, increasing levels of income substantively increase the probability that a postcard reminder will induce conversion to VBM. These effects fail to reach statistical significance because of small sample sizes in each category.13 Among those earning more than $100,000 per year, the receipt of a postcard increases the probability that one will convert to VBM nearly seven-fold.Table 2. The Effect of the Postcard Treatment on Conversion to VBM by Income Control (No Postcard)Postcard TreatmentPr(|Z|<|z|)Less than $30,00032.26% (10/31)19.35% (6/31).25$30,000–$60,0005.26% (2/38)13.16% (5/38).23$60,000–$100,00010.91% (6/55)23.40% (11/47).09More than $100,0005.45% (3/55)37.50% (15/40).0001**Cell percentages represent the proportion of registered voters who converted to permanent VBM status during our target period. The proportion in parentheses represents the number of converts in that group (numerator) over the total number of individuals in that group (denominator), where “groups” are defined by the combination of an income level and the treatment/control condition. The final column (Pr(|Z|<|z|)) indicates the results of a difference of proportions test between the control and treatment conditions in a given row. **indicates p<.05, for a two-tailed test.Racial/ethnic identification held the pattern for responsiveness to the postcard treatment: Groups with historically lower turnout were not converted at higher rates. We see in Table 3 that receiving a postcard reminder has no statistically significant effect on the decision to register for VBM among African Americans or Latinos. In contrast, Asians receiving a VBM postcard were more than ten times more likely to register for VBM, and Caucasians were more than twice as likely to register.Table 3. The Effect of the Postcard Treatment on Conversion to VBM by Race/Ethnicity Control (No Postcard)Postcard TreatmentPr(|Z|<|z|)Asian4% (1/25)50% (6/12).001**Black/African American26.32% (5/19)20% (3/15).67Latino/Hispanic12.20% (5/41)14.29% (9/63).76Caucasian10.49% (17/162)28.03% (37/132).0001**Cell percentages represent the proportion of registered voters who converted to permanent VBM status during our target period. The proportion in parentheses represents the number of converts in that group (numerator) over the total number of individuals in that group (denominator), where “groups” are defined by the combination of racial/ethnic group and the treatment/control condition. The final column (Pr(|Z|<|z|)) indicates the results of a difference of proportions test between the control and treatment conditions in a given row. **indicates p<.05, for a two-tailed test.Among non-English speakers who are registered to vote, VBM should theoretically be a preferred option for casting a ballot, as it allows individuals to seek additional time and/or assistance outside the polling place. This would enable them, for instance, to find a family member or friend who could help translate election options in a less-pressured environment than the polling place. Results in Table 4, however, demonstrate otherwise. Among individuals who speak a language other than English at home, we found that receiving a postcard mailer had no statistically significant effect on registering for VBM. In contrast, among English-only speakers, receipt of a postcard doubled the probability that a voter would register for VBM. The lack of findings here for non-English speakers is likely due to the fact that the costs of signing up for VBM at the outset may actually be higher for them, given the difficulty in comprehending and completing the VBM application.Table 4. The Effect of the Postcard Treatment on Conversion to VBM by Language Spoken at Home Control (No Postcard)Postcard TreatmentPr(|Z|<|z|)English Only Home11.94% (24/201)27.62% (50/181).0001**Non-English Only Home14.29% (7/49)19.30% (11/57).49Cell percentages represent the proportion of registered voters who converted to permanent VBM status during our target period. The proportion in parentheses represents the number of converts in that group (numerator) over the total number of individuals in that group (denominator), where “groups” are defined by the combination of language spoken at home and the treatment/control condition. The final column (Pr(|Z|<|z|)) indicates the results of a difference of proportions test between the control and treatment conditions in a given row. **indicates p<.05, for a two-tailed test.For those who own their homes versus those who rent, the effects of the postcard are somewhat less clear. On one hand, homeowners tend to turn out at a higher rate and are more likely to have consistent addresses. Both these factors imply that they may be more likely to convert to VBM. On the other hand, renters may be unfamiliar with polling locations and may prefer the mobility afforded by VBM.Empirically, however, the homeowners' propensity toward higher voter turnout bore out in VBM conversion rates (Table 5). Homeowners who received the postcard were three times more likely to convert than homeowners in the control group. For renters, however, the treatment and control groups were statistically indistinguishable.Table 5. The Effect of the Postcard Treatment on Conversion to VBM by Homeowner Status Control (No Postcard)Postcard TreatmentPr(|Z|<|z|)Renter21.43% (12/56)18.64% (11/59).71Homeowner9.05% (18/199)26.78% (49/183).000**Cell percentages represent the proportion of registered voters who converted to permanent VBM status during our target period. The proportion in parentheses represents the number of converts in that group (numerator) over the total number of individuals in that group (denominator), where “groups” are defined by the combination of homeowner status and the treatment/control condition. The final column (Pr(|Z|<|z|)) indicates the results of a difference of proportions test between the control and treatment conditions in a given row. **indicates p<.05, for a two-tailed test.Conditioning the treatment on prior turnoutTo more directly evaluate Hypothesis 2, we conducted a multivariate analysis to assess the effect of prior turnout and our postcard treatment on conversion to VBM (Table 6). Using probit (with robust standard errors), we estimate three iterations of the following model:14Table 6. The Effect of Postcard Treatment and Prior Turnout Percentage on Conversion to Permanent VBM Status Model 1Model 2Model 3Postcard (Treatment).4581*** (.1281).0971 (.2202).1440 (.2667)Prior Turnout %−.0031 (.0019)−.0081*** (.0031)−.0089** (.0037)Prior Turnout % x Postcard .0080** (.0040).0088* (.0047)High Cost .8757*** (.2294)Married −.1751 (.1791)Education .0962* (.0554)English .1390 (.2476)Ideology .5017* (.3035)Homeowner −.0160 (.2022)Caucasian .0328 (.2093)Religiosity −.0254 (.2200)Constant−1.0213*** (.1350)−.8091*** (.1612)−1.5782*** (.3622)Pseudo R2.03.04.09N542542404Table displays coefficients, with standard errors in parentheses.*Significant at .10, **Significant at .05, ***Significant at .01, two-tailed tests.Dependant Variable: Switched to VBM.Estimation Technique: Probit with robust standard errors.ConvertVBMi = α + β1Postcardi + β2PriorTurnout%i + β3Postcard*PriorTurnout%i + [β4.nControlsi]+ɛikThe dependant variable for all three models, ConvertVBMi, equals 1 if an individual converted to VBM during the treatment period, and 0 otherwise. The first model is the most basic, including just two independent variables:Postcardi , which equals 1 if an individual received the postcard treatment (as described above), and 0 otherwisePriorTurnout%i, which measures the frequency of turnout (as a percentage) for an individual for all seven elections held in San Joaquin County from March 2004 through June 2008. This is verified vote data obtained from the San Joaquin Country Registrar of Voters office.The second model adds our key variable, Postcardi*PriorTurnout%i, which is an interaction of the two previous variables.15 Based on Hypothesis 2, if high propensity voters are more likely to convert to VBM based on the postcard, then the interaction term should be positive and significant.In the third model, we add several control variables to check the robustness of our results. The controls are as follows:HighCosti, a dummy variable that equals 1 if an individual reported lack of transportation, child care issues, or too many hours of work on Election Day as difficulties related to going to the pollsMarriedi, a dummy variable that equals 1 if a respondent is marriedEducationi, an ordinal variable ranging from 1 (“less than high school diploma”) to 6 (“a graduate or professional degree”)Englishi, a dummy variable equal to 1 if only English is spoken in the respondents homeIdeologyi, a five point scale ranging from 0 to 1, where 0 is very liberal and 1 is very conservativeHomeowneri, a dummy variable equal to 1 if the respondent owns his/her homeCaucasiani, a dummy variable equal to 1 if the respondent is CaucasianReligiosityi, an ordinal variable indicating the frequency of attending religious services, ranging from 0 (“less than a few times per year”) to 1 (“every w