Electoral reform and mobilization as responses to inequality

Existing research raises questions about the ability of Seattle’s democracy voucher program to ameliorate inequality. This research shows that election reforms which make participation easier are often insufficient to engage infrequent participants.Footnote ⁴ While voting by mail, for instance, has been shown to modestly increase turnout, this effect is largely concentrated among regular voters in low-salience races (Gronke, Galanes-Rosenbaum, and Miller Reference Gronke, Galanes-Rosenbaum and Miller2007; Gronke and Miller Reference Gronke and Miller2012). In a mail-in voucher system such Seattle’s, then, we might expect higher participation and responsiveness from frequent voters. To address the gap between reforms’ intended impact and the actual outcome, some scholars have argued that mobilization can aid electoral reforms in democratizing the active public (Mann Reference Mann, Burden and Stewart2014; Stein, Owens, and Leighley Reference Stein, Owens and Leighley2003).

However, most experimental research examining the effects of mobilization – which largely focuses on efforts to get out the vote (GOTV) – finds that GOTV is most effective in activating those who are already likely to participate in politics (Enos, Fowler, and Vavreck Reference Enos, Anthony and Vavreck2014). Personal forms of contact such as canvassing and volunteer phone calls tend to be most effective (Green and Gerber Reference Green and Gerber2015). The effects of GOTV on low-propensity populations vary by the salience of the election (Arceneaux and Nickerson Reference Arceneaux and Nickerson2009) and people’s embeddedness within social networks (Sinclair Reference Sinclair2012). In general, however, high-propensity voters are most responsive to mobilization efforts.Footnote ⁵ While Seattle’s democracy voucher program was adopted to address inequality in campaign contributions, mobilization may be necessary to make it work. We ask whether mobilization can indeed reduce inequality in one of the most unequal areas of public life.

Previous work suggests that mobilization efforts which entail building an enduring relationship with low-propensity communities might stand a better chance of activating this population around election time. This research finds that GOTV efforts among members of civic organizations can significantly increase participation among low-income, minority, and low-propensity voters (Davenport Reference Davenport2010; Garcia Bedolla and Michelson Reference Garcia Bedolla and Michelson2012; Wong Reference Wong2004). Association members may be more likely to accept a mobilization treatment or more familiar with the organization delivering the treatment. Alternatively, due to stronger political interest, efficacy, civic skills, or relational ties, members may develop a stronger predisposition to comply with requests to participate in political activity (Han Reference Han2014; Verba, Schlozman, and Brady Reference Verba, Schlozman and Brady1995; Sinclair Reference Sinclair2012; Warren Reference Warren2001).Footnote ⁶ In other words, a higher propensity to vote could be a mechanism through which organization membership increases responsiveness to mobilization. If this were the predominant mechanism, we would find an interaction between treatment assignment and vote history, but not between treatment assignment and organization membership.

We propose the following hypotheses for our study:

1. H1: Mobilization will increase the use of democracy vouchers.

2. H2: The effect of mobilization on voucher use will be greater among high-propensity voters.

3. H3: The effect of mobilization on voucher use will be greater among members of the organizations affiliated with the Win/Win Network.

In addition to testing these hypotheses regarding voucher use, we examine whether mobilization increased voter turnout, and whether this effect was greater among high-propensity voters and Win/Win organization members.

Experimental setting and design

The November 2017 general municipal election in Seattle offered a chance to estimate the effect of mobilization on participation in the city’s new democracy voucher program. Beyond the city council and city attorney races for which voters could contribute vouchers, the mayor’s race headlined the ticket. Voting occurred largely by mail, and about 49% of Seattle voters cast their ballots by Election Day. Each registered voter in the city received a set of four democracy vouchers on January 3, 2017, along with prepaid postage to encourage voters to send the vouchers in to the city government for distribution to candidates. Of the 17 primary election candidates, six qualified to receive vouchers by meeting the program’s requirements. In the general election, all but one of the six candidates used vouchers to finance their campaigns. In a sign of the program’s viability, a candidate using vouchers won each of the three races for which vouchers could be used (Friedenbach Reference Friedenbach2017).Footnote ⁷

In collaboration with the Win/Win Network, we conducted a field experiment which randomly assigned a multifaceted mobilization effort providing information about the democracy vouchers. We assembled the experimental universe through a two-stage sampling procedure. First, we identified the precincts in the southern half of the city – which is more racially diverse than the northern half of the city – with at least 20% people of color. Figure 1 maps the precincts included in the sample. Second, within these precincts, we sampled people of color and people under the age of 35 from a voter file provided by Win/Win.Footnote ⁸ This sampling procedure facilitated Win/Win’s efforts to increase participation among underrepresented communities in the city. As the intervention was designed to encourage people to use their vouchers, we dropped from the experimental universe anyone who had used a voucher before we selected our sample on July 31, 2017.

Figure 1

Precincts in the Experimental Universe.

Notes: Pink shading indicates that a precinct was included in the experimental sample. This map was provided by the Win/Win Network.

We used block random assignment, clustered by household, to assign voters to treatment and control groups.Footnote ⁹ To combine clustered random assignment with blocking, we first randomly selected one individual per household to stand in for the household in the random assignment process. We then stratified this subsample into blocks based on a set of voter characteristics. To help assess our core hypotheses, the random assignment process blocked on voting history and membership in a Win/Win partner organization.Footnote ¹⁰ To improve precision, we also blocked on whether the voter shared a household with other voters in the experimental universe, reachability by cell phone, reachability by e-mail, and – among members of the Washington Community Action Network (Washington CAN) – reachability by landline or cell phone.Footnote ¹¹ In total, voters in the subsample were sorted into 124 separate strata representing distinct combinations of voter characteristics, and random assignment to the treatment and control groups occurred separately within each stratum. Following random assignment, we conducted balance checks on age, race, sex, income, and education level, re-randomizing if imbalance reaching at least borderline levels of significance (p < 0.1) was found. Next, we assigned each individual sharing a household with a subject in the subsample to the same experimental group as their selected housemate. After removing households with eight or more voters, we were left with a sample of 41,414 voters. Our final sample is 30.88% Asian/Pacific Islander, 21.63% Black, 8.24% Latinx, and 34.18% White; 59.69% of our sample is under 35.Footnote ¹²

Win/Win coordinated several organizations to deliver four modes of outreach to voters in the treatment group.Footnote ¹³ The voter contact methods included door-to-door canvassing, text messages, e-mail messages, and digital advertisements delivered through Facebook.Footnote ¹⁴ Each of these methods provided information about the democracy vouchers and reminded voters to contribute their vouchers to municipal campaigns. The door-to-door canvassing script also included a message intended to frame the voucher program as a means of empowering ordinary voters:

As is increasingly common in mobilization efforts, the canvassing script incorporated questions intended to prompt the voter to make a plan to send in their vouchers (Nickerson and Rogers Reference Nickerson and Rogers2010). Canvassers also delivered literature to each household they visited regardless of whether a voter answered the door. Win/Win followed up with voters who responded to the text messages to provide information about how to use the vouchers.Footnote ¹⁵

Win/Win provided us with data on covariatesFootnote ¹⁶ and contact rates.Footnote ¹⁷ Fourteen percent of voters had a conversation with a canvasser, 25% received texts, and 1% received e-mails, while the contact rates for the digital ads ranged from 23% to 37%.Footnote ¹⁸ After the election, Win/Win obtained data on voucher use and voter turnout from the Public Disclosure Commission and the Washington Secretary of State, respectively (Henderson and Han Reference Henderson and Han2020).

Using publicly available data, we construct four categories of voting history. We tally how often each person voted over the four previous general elections (2013, 2014, 2015, and 2016) and divide this figure by four. We code voters with perfect voting records or who missed one election as high-propensity voters, while those who voted in two of four elections are medium–high-propensity voters. Those who voted just once are medium–low-propensity voters, while those who did not vote at all are low-propensity voters.Footnote ¹⁹ We drop voters ages 21 and younger from the analyses involving voting history, as these voters were ineligible for at least one of the past four elections.

Analysis

Our analysis estimates the degree to which mobilization increased participation in the voucher program in the aggregate. The purpose of this research is not to examine the effects of particular kinds of treatments (e.g., door-to-door canvassing versus texting), but rather to understand whether mobilization in general can significantly increase participation in the program. Consistent with similarly designed mobilization studies, our estimand is the intent-to-treat effect (ITT) on voucher use (see e.g., Garcia Bedolla and Michelson Reference Garcia Bedolla and Michelson2012; Green and Gerber Reference Green and Gerber2015).Footnote ²⁰ Rather than the effect of receiving the treatment – known as the treatment-on-treated effect (TOT) – the ITT represents the effect of being assigned to receive the treatment. An advantage of estimating the ITT is that it incorporates noncompliers into the estimate, thus accounting for the challenges that mobilization efforts face in reaching voters. A TOT estimate would indicate the intervention’s effect only on compliers, which is less informative from the perspective of evaluating the intervention’s effect on the composition of the active public.

First, we estimate the mobilization campaign’s effect on voucher use. To test H1, we fit a linear probability model with standard errors clustered by household and block fixed effects. Model 1 takes voucher use as the dependent variable and treatment assignment as the explanatory variable, and Model 2 adds a series of demographic covariates. These covariates include a discrete variable for age (A), an indicator for female (F), ordinal variables for education (E) and income bracket (I), and indicators for Asian American/Pacific Islander (P), Black (B), and Latinx (L).

We test H2 and H3 by adding interactions between treatment assignment and an indicator of high-propensity voter, and between treatment assignment and an indicator of Win/Win partner organization membership, to the two previous models (see Model 3). Model 4 adds demographic covariates; the full model is shown below:

$$\rm{Y_i} = {\rm{ }}{\beta _0} + {\rm{ }}{\beta _1}*{D_i} + {\rm{ }}{\beta _2}*{H_i} + {\rm{ }}{\beta _3}*{W_i} + {\rm{ }}{\beta _4}*{D_i}*{H_i} + {\rm{ }}{\beta _5}*{D_i}*{W_i} + {\rm{ }}{\beta _6}*{A_i} + {\rm{ }}{\beta _7}*{F_i} + {\rm{ }}{\beta _8}*{E_i} + {\rm{ }}{\beta _9}*{I_i} + {\rm{ }}{\beta _{10}}*{P_i} + {\rm{ }}{\beta _{11}}*{B_i} + {\rm{ }}{\beta _{12}}*{L_i} + {\rm{ }}{\gamma _i} + {\rm{ }}{\varepsilon _i}$$

In the above model, D_i is a binary indicator of treatment assignment, H_i is a binary indicator of high-propensity voter, and W_i is a binary indicator of Win/Win organization membership. The ITT estimate is provided by β₁, β₄ provides an estimate of the interactive effect of treatment assignment and high-propensity voter on voucher use, and β₅ provides an estimate of the interactive effect of treatment assignment and organizational membership. Block fixed effects are represented by γ_i and ε_i represents the error term.

For each hypothesis test, we conduct randomization inference as a robustness check to account for the effect of re-randomization on the sampling distribution from which our permutation of the random assignment was drawn. As further robustness checks, we estimate heterogeneous treatment effects by voting history using both OLS regression and randomization inference. Finally, we replicate our analyses with voter turnout as the dependent variable.

Mobilization increases voucher use largely among frequent voters

The rate of participation in the voucher program among voters in our experimental universe was 1.45%, well below the city’s overall rate of voucher use, 3.3% (Kliff Reference Kliff2018). However, voucher use appears to increase with past voting. Among high-propensity voters in our sample, the rate of voucher use was 3.18%, more than twice the rate among the experimental universe as a whole. Medium–high-propensity voters contributed vouchers at a rate of 2.31%, compared to 1.23% among medium–low-propensity voters and 0.52% among low-propensity voters. While 1.83% of newly registered voters contributed their vouchers, none of the more than 3,000 unregistered voters in the experimental universe contributed a voucher. The rate of voucher use was also higher among members of the organizations comprising the Win/Win Network – 2.8% of members used their vouchers, more than twice the rate (1.2%) among non-members (Table 1).

Table 1

ITT Effects of Voter Mobilization on Voucher Use

NOTES: We fit four linear probability models to estimate intent-to-treat (ITT) effects – the effects of treatment assignment (rather than receipt of the treatment) on the likelihood of voucher use. We cluster standard errors by household and add block fixed effects to account for the treatment assignment process, which randomized households within blocks. Models 3 and 4 exclude voters below the age of 22, as these voters were ineligible to vote in at least one of the elections that we use to operationalize the high-propensity voter variable. ***p < 0.01; **p < 0.05; *p < 0.1, two-tailed test.

Our analysis estimates the effect of mobilization on voucher use and how this effect varied by voting history and organizational membership. First, we find that assignment to the mobilization effort significantly increased voucher use. Model 1, without covariates, shows an ITT of 0.37 percentage points. Given that the baseline rate of voucher use in the control group was 1.26%, this effect represents a substantial 29.57% increase in voucher use. Controlling for age, gender, race, education, and income (Model 2), assignment to treatment increased voucher use by 0.35 percentage points. We thus find support for H1. As a robustness check, we test H1 using randomization inference. This procedure adjusts the standard errors on our point estimates by accounting for the permutations of the random assignment procedure which were precluded by our design. Replicating the test of H1 using this procedure does not substantially alter our results.

Second, the effect of mobilization was 1.11 percentage points greater for high-propensity voters. Consistent with H2, we find a positive interaction between treatment assignment and high-propensity voter (Model 3). This effect is significant at conventional levels using a one-sided test, which is appropriate given that we hypothesized a positive relationship. Adding demographic covariates (Model 4) does not substantially alter the results. As a robustness check, we estimate the interaction between treatment assignment and high-propensity voter with regression-adjusted randomization inference, using the specification from Model 3. Using this approach, the interaction of treatment assignment and high-propensity voter becomes significant at conventional levels using a two-sided test.

Our analysis of heterogeneous treatment effects offers further support for our hypothesis that mobilization is more effective among more frequent voters.Footnote ²¹ We find that mobilization increased voucher use by 1.51 percentage points among high-propensity voters, roughly four times the effect among the whole experimental universe. While treatment assignment may have increased voucher use among medium–high-propensity voters, we lack the precision to be certain of this result. However, perhaps because there are more medium–low-propensity voters in the dataset, we find a significant treatment effect for this subgroup. Among these voters, mobilization increased voucher use by 0.35 percentage points, comparable to the effect among the broader experimental universe. Finally, we do not find a significant effect among low-propensity voters (Figure 2).Footnote ²²

Figure 2

ITT Effects by Voting History.

Notes: This figure illustrates heterogeneous treatment effects by voting history category. Each model includes block fixed effects and clusters standard errors at the household level. The dots in the figure represent point estimates. The thick error bars span one standard error above and below the point estimate, while the thin error bars represent the 95% confidence interval estimated using a two-tailed test.

Third, we examine whether treatment effects varied by organizational membership. Contrary to H3, we do not find an interaction between treatment assignment and membership in a Win/Win partner organization (see Model 3). Adding controls to the interactive model does not substantially alter the results (see Model 4).

Finally, our analysis of voter turnout finds a pattern resembling the results for voucher use. While we do not find an effect of the mobilization effort on turnout overall, mobilization was significantly more effective among high-propensity voters, by nearly 4 percentage points (Table AJ1). Mobilization significantly increased voter turnout only among high-propensity voters; those assigned to treatment were 3.68 percentage points more likely to vote (Table AK1). We again fail to reject the null hypothesis that organization membership increased responsiveness to mobilization.