Is Distance to Drop Box an Appropriate Proxy for Drop Box Treatment? A Case Study of Washington State

Introduction

In recent years convenience voting has increased across the United States, with several states adopting high or mandatory vote by mail (VBM) regimes, including Washington, Oregon, Colorado, Hawaii, and Utah. All states with a VBM regime have used the option of drop box voting as a means of increasing political participation and voter confidence in ballot security. A voting drop box is a metal container/box where voters slot their ballots (similar to a United States Postal Service box) and serves as an alternative to sending ballots through the mail. While only five states have completely adopted VBM and the widespread use of drop boxes, a number of other states have implemented mixed regimes. In 2018, California rolled out drop box voting in five counties, with an eye toward establishing drop box voting as a way to further increase the state’s relatively dismal voter turnout. In addition, due to the coronavirus, vote by mail in 2020 increased dramatically across the U.S., with many states implementing drop boxes for the first time.

In Washington State, voters have been casting ballots exclusively by mail since 2011 (http://blog.thenewstribune.com/politics/2011/04/05/washington-to-shift-to-all-vote-by-mail-elections). Increasingly, instead of returning ballots through the mail, voters place ballots into conveniently located, and secure drop boxes. This voting technology has become so popular that on May 16, 2017, Governor Jay Inslee signed a law mandating a 300 drop box expansion across the state (http://app.leg.wa.gov/billsummary?BillNumber=5472&Year=2017). Today, depending on the election, between 40% and 60% of Washington State voters choose to vote by drop box rather than send ballots via USPS (https://www.sos.wa.gov/elections/research/ballot-drop-box-usage-by-year.aspx).

Collingwood et al. (2018) studied whether drop boxes increased turnout in two election years (2015 and 2016) in King County, WA, by exploiting a change in the number of boxes between the two years. By moving from 10 to 40 boxes within 1 year, the distance between voters’ and their nearest drop box changed, which presented a unique opportunity to examine whether this change in distance for each individual voter increased one’s likelihood of casting a ballot. The researchers found that drop boxes did improve turnout but primarily in off-years and primaries. Instead of relying upon a change in distance between one election and the next to proxy treatment, McGuire et al. (2020) worked with a local election auditor to randomly assign five new treatment boxes and one placebo control box (a site that could have been treated with a new box but did not ultimately get one due to randomization) to geographic locations across Pierce County, WA. This tighter design revealed that a decrease of one mile to the closest drop box increases a voter’s probability of voting by 0.64%.

Other than these two papers, the academic literature on convenience voting has not analyzed the use or effects of drop boxes in American politics. In this paper, we investigate three questions (and relevant hypotheses) related to drop box voting behavior: (1) Do voters, in fact, disproportionately use the box closest to their residence? (2) Do voters disproportionately opt for voting in central locations (i.e., auditor’s offices, election headquarters)? (3) Do voters who choose to vote by drop box rather than mail physically live closer to a drop box? These questions are important to help guide state investment in drop boxes, particularly considering the increase in the number of VBM states, with UT and HI transitioning to mail elections in 2019 and 2020, respectively.

To investigate these questions we gathered publicly available individual-level voter data from Washington State’s 2020 August primary election. ¹ Washington has a top-two primary system, where all candidates are listed on a single ballot and the two candidates with the highest vote share proceed to the general election regardless of party. It is worth noting that primaries in Washington State tend to see significantly lower turnout than the general election, often by double-digits, which may make the primary electorate less representative of the voting public as a whole. High propensity voters are more likely to cast a ballot in primaries and these voters tend to be wealthier, older, more educated, and largely White. And indeed, as indicated by Figure A1 in Appendix A, primary voters are less likely to use a drop box than are general election voters. That said, it is hard to imagine how this gap in drop box usage across election type would alter whether drop box voters are more or less likely to use boxes closer to their residence—although we leave that specific question to future research.

The data include measurements for voting method (e.g., drop box, mail, etc.) and ballot return location. To our knowledge this is the only publicly available data with such drop box detail. We then geo-coded all voters’ residences and paired these data with a geo-coded list of all drop box sites across the state. We next calculated the distance in miles between each voter’s residence and all drop boxes across the state. ²

We find that: (1) Voters disproportionately use drop boxes closer to their residence relative to other drop boxes within their county; (2) Drop box voters who fail to use their nearest drop box live farther away from their nearest box relative to voters who do use their nearest box; (3) Drop box voters who fail to use their nearest drop box do disproportionately cast a ballot in a central location like an election headquarters, city hall, or courthouse; and (4) Voters who return their ballots by mail rather than by drop box live farther away from their nearest box. Coupled with evidence from a recent survey on voting behavior, we find consistent support for the argument that a convenience versus security framework guides drop box decision-making.

Background and Expectations

There is a robust literature on the effect of convenience voting on turnout. In general, convenience voting increases participation, though by how much varies by election and voter demographics. Recently, COVID-19 may have generated long-term changes in convenience voting, with states investing in vote by mail (Baringer et al., 2020). While this is generally welcomed by election administration scholars, it is important that researchers investigate both the pros and cons of VBM (e.g., ballot rejections (Shino et al., 2021), postal service disruptions (Herron & Smith, 2021)). In particular, there are a number of studies that have shown that geography plays a significant role in determining the likelihood that someone will cast a ballot. Dyck and Gimpel (2005) found that the greater the distance between a precinct location and a voter’s home, the less likely that individual was to cast a ballot. Similarly, Brady and McNulty (2011) found that if you increased the distance between a voter and their nearest precinct voting location, their likelihood of voting decreased: Twenty-five percent for every tenth of a mile increase up to 0.40 of a mile. Building on Brady and McNulty (2011), Amos et al. (2017) found that redistricting in Manatee County, FL, led to a decrease in turnout for those who were assigned a new precinct, which they posit is due to the information and possible transportation costs on voters as a result of the change in location of the precinct. Distance to voting location was found to have the largest effect for those in metropolitan areas, with a much smaller impact on rural voters (Gimpel & Schuknecht, 2005). These studies were, however, on distance to precinct locations rather than to the nearest drop box.

To date, just two papers investigate drop box voting in American politics (Collingwood et al., 2018; McGuire et al., 2020). Both papers test whether decreasing the distance to the nearest drop box increases voter turnout. Using different methods but still instrumenting drop box treatment by distance to nearest drop box, both papers find a positive and statistically significant effect for distance to drop box (how close one lives to a drop box) and probability of voting. Still the authors proxy drop box treatment based on distance to nearest drop box, crucially assuming that voters are most likely to use the box closest to their address available on the voter file. While both studies are illuminating, the assumption that voters use the drop box closest to their house, ³ however, may not always be true.

As more states consider moving to a VBM system, it is important to have a better understanding not only of the effect that drop boxes have on voting behavior, but also how voters interact with these boxes. With each box costing an estimated $6,000, a better understanding of how voters use these boxes can inform states how to install them in a more strategic manner, potentially cutting costs in the process (https://www.eac.gov/sites/default/files/electionofficials/vbm/Ballot_Drop_Box.pdf). If voters are just using the box closest to their home, then states would be encouraged to place boxes in a variety of locations. If voters are using boxes in a central location, such as at a courthouse or city hall, then states may be able to install few boxes while still reaping the positive benefits that drop boxes have on turnout. Finally, if increased distance to a box leads to a greater likelihood of using the USPS to return a ballot, states may have to think strategically in terms of the location and number of boxes installed or whether to include pre-paid postage.

We theorize that choosing to vote by drop box, then which one, versus returning a ballot by mail is largely a matter of a convenience versus security trade-off. In general, the closer one lives to a drop box voting location, the more likely one is to use a drop box relative to returning a ballot by mail. First, the process of even walking or driving down the street may at times be more convenient than placing a ballot in the mail (especially if the state or county does not provide a stamp). Second, voters may reason that a drop box is more secure than USPS, where, theoretically, postal workers and election workers can tamper with ballots. On the other hand, if a voter must drive far to drop off a ballot, the convenience of returning a ballot by mail is likely to override ballot security concerns—especially when a voter casts their ballot more than one or two weeks prior to election day and can verify their county auditor has received it.

In a study of Colorado’s VBM system, Menger and Stein (2020) examined why a voter would choose to return their ballot in person at a ballot drop-off location, instead of by mail. They found that distrust in the Postal Service had the largest overall effect, but that there were also only a small proportion of voters in their study who expressed unease with sending their ballot via the USPS. They did find that among both frequent and infrequent voters, a desire to avoid a change in routine made them less likely to cast a ballot in person or at a drop box location. This desire to avoid costs associated with travel to vote in person, necessitating a change in routine, is likely to also be influenced by the kind of interruption voting in person will entail. In other words, in the context of the current study, we would expect voters who choose to return their ballot by drop box to choose the box closest to their residence, as this will require a smaller change in their daily routine.

A survey by Charles Stewart on the “Performance of American Elections, 2016,” provides further context for why some voters might prefer voting by drop box versus returning a ballot by mail (Stewart, 2017). Surveying n = 10,200 respondents who cast a ballot in the 2014 general election, Stewart finds that 21.6% of voters cast an absentee ballot. Of these, about 22% voted via some type of drop box—whether an official box, an election headquarters, or some type of voting center. A follow up open-ended question asks why voters decided to use a drop box reveals answers consistent with our theoretical framework ⁴ : Responses generally referenced either convenience or security as a reason for using a drop box. Common responses are: “I didn’t have to pay postage and it was conveniently located”; “It was more convenient and the place was near my home”; “That way I was sure it wasn’t lost in the mail”; and “I wanted to make sure it was received and I personally put it in the container.”

The survey also asked voters who mailed their ballots back why they chose to do this rather than return their ballot personally. ⁵ Here, the majority of the responses reflect convenience: “easier and more convenient”; “It is convenient. We are Senior citizens (my wife and me). This is great for us”; and “Convenience and to save time and gasoline.”

Based on the available studies and the survey evidence just presented, conditional on voting via drop box, voters should be more likely to use the drop box closest to their physical residence compared to other drop boxes. We operationalize this by testing whether more than 50% of drop box voters use the box closest to their house. In this way, we can, at the mean, test the assumption in earlier studies that proxy drop box treatment by closest box. Here we can test whether at least a majority—the lowest burden of proof—of drop box voters do indeed line up with this assumption thereby demonstrating the assumption statistically legitimate. We therefore deduce hypothesis 1:

However, there are ample reasons to expect why voters might not use the drop box closest to their house, but instead use one closer to a job, their children’s school, their own school, some other place that is more convenient to one’s daily movements, or a central voting location. All of this seems more plausible the further one lives from their nearest drop box. From this we deduce our second hypothesis:

Many voters may feel more confident their ballots will be counted if they drop ballots off at a central voting location than in a drop box say on a shopping center corner. The reasoning here is that by delivering a ballot at an auditor’s office, courthouse, or city hall, the ballot likely has shorter distance to travel to its final counting location than it would if dropped off at a more distant neighborhood drop site. Thus, the ballot likely faces fewer security threats. Given this, a voter might reason a central facility is simply more secure and opt to take their ballot there. From this, we deduce the following hypothesis:

Finally, we move beyond only drop box voters to better understand why voters might opt to place their ballots in the mail as opposed to into a drop box. Based on the extant literature that voters are more likely to vote the closer they live to a drop box, we theorize that voters who live farther from a drop box are more likely to cast a ballot by mail. We therefore deduce our final hypothesis:

Data

To test our above hypotheses, we rely on individual-level voter data from Washington State’s August 2020 primary (https://www.sos.wa.gov/elections/research/2020-primary-election.aspx). These data include each individual who voted in the August primary election, their residential address, county, voting method, and if voting by drop box, the box’s name. We pair these data with a list of all 419 drop boxes in the state as generated from the 2019 general election. ⁶ We cross-referenced this list against the secretary of state’s 2020 general election ballot drop box list then against ballot return data in the individual-level voting data. Of the 419 drop boxes only 330 are present in the individual level data. We therefore subset our initial list of available drop boxes to 330. ⁷ Unfortunately, King and Clallam Counties did not adequately record drop box voting data (either method or location) and we thus had to omit them from our analysis. While Clallam is a relatively small county, ranking 18th in terms of population, and unlikely to significantly affect our findings, King County is another matter. King County is Washington’s most populous and thus could have affected our findings had it been included. However, past research has found that the expansion of drop boxes in King County in 2016 increased the likelihood of voting, and that voters who lived closer to a drop box were more likely to vote (Collingwood et al., 2018). Therefore, it seems most likely that our results would not substantively change had King County data been available for this study.

To generate our measure of drop box centrality, we ran each of the 330 drop box names through the following algorithm such that if any of the below phrases appeared in the drop box name, we coded the box as central: city hall OR town OR hall OR auditor OR auditor’s OR clerk OR vote OR center OR election OR elections OR civic OR courthouse OR administration OR admin. We then validated each of the centrally located drop boxes via Google maps to ensure the box indeed is centrally located. We classified 93 of the 330 (28.2%) boxes as central locations.

To calculate our key distance measure—between voters’ residences and each drop box—we geo-coded both datasets. Of the 1,785,535 voters in our initial dataset, 3,279 returned missing latitude/longitude coordinates, so these records are dropped from our analysis. Another 3,054 records—primarily PO Boxes return inaccurate geo-codes. In addition, a handful of records (0.003%) are out of state based on voter file records. We drop these voters too. Finally, our geo-coder placed 3,745 respondents out of state, when clearly they are not. We therefore drop these records as clear geo-coding errors. All told, our geo-coding accuracy is greater than 99%.

For most of our analyses we are only concerned with voters who cast ballots via drop box; whereas our last hypotheses pits drop boxes versus returning a ballot by mail. We therefore subset our data to only voters who the secretary of state records as voting by drop box or voting by mail. Finally, upon inspection of Table A2, we observe that just one box (Elections Office—Ballot Drop Box) is used for the entire county of Spokane. It seems highly plausible that the county auditor simply tallied all drop box votes as coming in just this one box. This data collection procedure disqualifies Spokane from our analysis. This generates a final dataset of 1,554,857 rows. Of this, 851,149 (54.7%) people voted by drop box, whereas 703,708 (45.3%) voted by mail. ⁸

We pair these data with a geo-coded list of all drop box sites. Following Collingwood et al. (2018), we calculate the nearest drop box for each voter based on a straight-line (i.e, as the crow flies) calculation. Figure A2 presents the variable’s distribution (overall, mail voters, and drop box voters). On average, people in our data live 2.88 miles (SD, 3.13 miles) from the closest drop box. The median distance is 1.78 miles, and the inter-quartile range 0.96 to 3.74 miles. We then compare the drop box return location name in the voter database against the drop box name in the drop box database (after correcting for minor typos across the two datasets, and adjusting for scenarios where drop boxes are labeled the same but are in different counties). Voters are given a 1 when the drop box names match between the two databases, and a 0 otherwise. This unique dataset lets us to test our hypotheses presented in the previous section.

Results

To test hypotheses 1 to 3 we rely on the subset of voters who cast a ballot by drop box. Recall our first hypothesis states that voters who vote via drop box are more likely to use the box closest to their residence relative to other boxes in that voter’s county.

First, we find that of the 851,149 drop box voters, 445,581 (52.4%) voted via the box closest to their residence. ⁹ To test hypothesis 1, we conducted a difference of means t-test against a µ value of 50. Our results produce a t-statistic of 43.53, p-value <.001, indicating that a majority of voters did indeed use the box closest to their house. This provides clear support for hypothesis 1.

We think that if a voter uses a box other than the one closest to their home it is most likely that they still use a box in their own county. The question now becomes whether an individual is more likely to use the box closer to their house than they are any other nearby drop box—where we define nearby as within the voter’s county. We therefore calculate the number of boxes per county and calculate the probability that any one voter will vote in any one box within that county. This value ranges from 0.02 to 1, with a mean value of 0.098. Thus, to test whether voters disproportionately use boxes closer to their house we conduct a difference of means t-test between the proportion of voters who do (1) and do not (0) use the box closest to them against the µ value of 0.098. We find a t-statistic of 785.52, which is statistically significant at the p-value <.001 level. This provides overwhelming support for hypothesis 1a.

Our second hypothesis begins to intimate a causal mechanism as to why voters might use the box closest to their house versus any other box. And the answer is quite simple. We hypothesize that voters who use a box other than the one closest to their residence live farther from their nearest drop box. To evaluate this hypothesis we conduct a difference of means t-test between our two groups (use box closest to residence versus do not use box closest to residence) and distance to nearest drop box. The mean distance in group 1 (voted in box nearest) is 2.28 miles to the nearest drop box, whereas the mean distance in group 2 (did not vote in box nearest) is 2.82 miles to the nearest drop box. This difference is statistically significant (t = 91.93, p < .001), substantively the difference is a bit more than half a mile. These results support hypothesis 2.

Hypothesis 3 begins to get at a mechanism for why some voters who vote by drop box choose not to vote at the one closest to their residence. H3 states that voters who do not use their nearest drop box are disproportionately likely to use a central voting location relative to drop box voters overall. We theorize this is most likely due to issues of security although more work needs to be done to explicate this possibility. To test this hypothesis we conducted a χ 2 analysis between proportion using central location and whether one used their nearest drop box. We find a statistically significant difference between the two populations (33% [use nearest box] vs. 40.3% [do not use nearest box], χ 2  = 4896.72, <.001). These findings support hypothesis 3. Thus, voters who do not case a ballot at their nearest location are more inclined to place their ballot in a centrally located venue like an election headquarters or courthouse. This provides support for our overarching framework of convenience versus security.

Our final hypothesis further elaborates on the distance mechanism but inquires as to why this might explain voting by drop box versus returning a ballot by mail. H4 states that voters who return their ballot by mail live farther from their nearest drop box than do voters who vote by drop box. We find support for this hypothesis: people voting by drop box live on average 2.53 miles from their closest drop box, whereas voters opting to return by mail live 3.30 miles to their closest drop box (t = 148.55, p-value <.001). The substantive effect is about 3/4 of a mile.

Discussion and Limitations

While many states in recent years have turned to measures associated with lower voter turnout, some states have gone in a different direction by pursuing policies designed to enhance voter turnout. The premier non-partisan method for increasing turnout is by making voting easier (Dyck & Gimpel, 2005; Highton, 1997; Mitchell & Wlezien, 1995; Neiheisel & Burden, 2012; Rhine, 1995). Colorado, Utah, Hawaii, Oregon, Washington, and now California have or are turning to implementing some form of drop box voting whereby voters place their otherwise would be VBM ballot into a secure drop box. With a number of states installing drop boxes for the 2020 general election, it is very likely that drop box usage will continue to expand as states, and voters, realize their convenience and the positive effects on participation. With increase boxes, voters are more likely to actually see the box in the course of their daily activity, thereby increasing the election’s saliency and possibly inducing voting habits (Aldrich et al., 2011; Coppock & Green, 2016; Gerber et al., 2003).

Turning back to our analysis, of the 851,149 drop box voters in our data, 445,581 (52.3%) use a box nearest their residence. We also find that voters who opt for another box or mail do so likely for one of two reasons: (1) Distance (convenience, or lack thereof) or (2) Ballot security.

In particular, we find that voters who do not vote in the drop box closest to their residence live farther away from their nearest drop box. This suggests that as county auditors add in more drop boxes, and as voters’ distance to their closest box continues to drop, more voters will opt to vote via drop box. In a climate with rising concerns over USPS delivering ballots on time, ballot drop boxes likely add an additional avenue in which voters can feel secure in casting their ballot.

On the other hand, a healthy share of voters choose to vote in a central voting location (i.e., city hall, auditor’s office, etc.). This suggests that many voters are still nevertheless willing to forego the convenience of voting in a nearby drop box or putting one’s ballot in the USPS. We think this is due to concerns over ballot security but future research should investigate this working hypothesis further by not only using similar methods as presented in this paper but even via survey research.

One major take away from this research is that voters will use the box closest to their home, and this is most likely the case when the box is located very close to their home. Perhaps this is not surprising, but the implication is that election administrators may want to consider further increasing the number of drop boxes within their jurisdictions as they drop the overall mean distance to nearest drop box thereby potentially driving up turnout.

Future research should consider how drop box usage varies by demographic and contextual characteristics. It is well known that political participation varies by race/ethnicity, income, and education (Bowler & Segura, 2011; Fraga, 2018; Gilens, 2012; Griffin & Newman, 2007; Martin, 2003; Wolfinger & Rosenstone, 1980). As the U.S. continues to diversify so do its elected officials; however racial/ethnic gaps between population size and political influence remains large. The same goes for class (Carnes, 2013, 2016). Unfortunately in our case, our data remain limited. But, as drop boxes are rolled out in states with more demographic characteristics on voters files (i.e., in the South), more detailed analysis will be possible.

Additionally, this study would be worth replicating if King County begins collecting data on the drop box used by voters to return their ballot. While we would expect our findings to hold, King does have some unique demographic characteristics as the most populous county in Washington, the home to Seattle, and a tech hub for companies like Amazon and Microsoft. Data from an earlier study by Collingwood et al. (2018) found that the expansion of drop boxes in King County in 2016 did increase the likelihood of voting for those who saw a decrease in distance to their nearest box but the researchers did not know which box individuals used. However, the tracking of the specific box used is relatively new in Washington State, so we are hopeful that King will follow in collecting this data some time in the not-too-distant future.