In their excellent article, Daniel Thompson, Jennifer Wu, Jesse Yoder and Andrew Hall show that the adoption of universal vote by mail (VBM) does not significantly advantage either party. However, they examine policy changes that, on average, only caused a modest increase in mail voting. Many counties in their study that adopted universal vote by mail already had a large share of ballots cast by mail before making it universal. Could the effects be different if states that adopt universal VBM that had low levels of preexisting mail voting? (My previous post lays out reasons why one might expect effects of universal VBM to differ based on preexisting levels of mail voting.)

      To make progress on this question, I reanalyze the Thompson et. al. data. Specifically, I check if universal VBM has a different effect in counties that had a below the median proportion of voters using VBM before the universal policy than counties that had higher than median proportion of voters using VBM. In other words, is the effect of universal VBM different when it causes a larger change in the proportion of voters using mail ballots?

      Thompson et. al. use data on county level adoptions of universal vote by mail in California, Utah, and Washington state. Unfortunately, data on the proportion of voters casting mail ballots was not available in Utah, and in California, only 5 counties adopted universal vote by mail. Therefore, my analysis concentrates on Universal VBM in Washington State. Washington has 39 different counties, all of which eventually adopted universal vote by mail.

      The graph below shows the distribution of vote by mail rates for different counties in the election before they adopted universal vote by mail. The median county had 65% of ballots cast by mail before making it universal. However, there is considerable variation, with some counties with vote by mail rates less than 30% and others above 85%.

      To assess how the effects of Universal VBM varies by preexisting mail voting rates, I test whether the effects of Universal VBM differ between the counties below the median pre adoption rate compared to those above the median.1 I re-estimate the same baseline model used by Thomson et. al. with county and election-year fixed effects. However, I interact the treatment variable with an indicator if a treated state had below the median share of mail voters pre-treatment.

      First, I show how much more the adoption of universal VBM increased VBM rates in below median counties compared to above median counties. Column 1 of the table below shows how much adopting universal VBM increases the share of ballots cast by mail compared to counties that did not universal VBM by about 30 percentage points.2 Averaging over all counties, universal VBM increases the share of ballots cast by mail by 30%. In column 2, I interact the universal VBM indicator with an indicator if a county had a below the median pre-implementation share of mail voters. The universal VBM indicator can be interpreted as the effect of universal VBM on high pre-implementation VBM share counties, and the interaction can be interpreted as the differential effect of VBM in low pre-implementation VBM counties. The total effect of VBM in low pre-implementation VBM counties is the sum of the two coefficients. In low pre-implementation VBM share counties, universal VBM caused an 15 percentage point greater increase in the share of voters using mail ballots than high pre-implementation counties, for a total effect of 37 percentage points. While the ideal setting would provide more variation between high and low pre-implementation VBM counties, this is enough of a difference to get some idea of universal VBM effects may differ depending on how many voters had experience with VBM before making it universal.

     

Effect of Universial VBM on Proportion of Ballots Cast By Mail [0-1]
Baseline Interaction
Universal VBM 0.300*** 0.224***
(0.031) (0.022)
Universal VBM X Low Pre-Treat VBM 0.146***
(0.035)
N 312 312
R2 0.889 0.903
* p < 0.05, ** p < 0.01, *** p < 0.005
All models include county and year fixed effects and standard errors clustered at the county level
Data from Washington state, 1996-2010

     

      Next, I test if effects of universal VBM on democratic voteshare differs between high and low VBM pre-implementation counties. Thomson et. al. found weak a small partisan effect in their baseline specification (just county and election year fixed effects) that I reproduce (limiting the data to Washington State) in column 1 below.3 The second column shows that this effect is entirely driven by counties with high preexisting vote by mail rates. Adding the interaction term with the main VBM term yields an estimate of 0.00 percentage points. This suggests that the lack of partisan effects in Thomson et. al. is not due to the average county’s high preexisting rates of vote by mail.

     

Effect of Universial VBM on Democratic Presidential Voteshare [0-1]
Baseline Interaction
Universal VBM 0.014** 0.027***
(0.005) (0.007)
Universal VBM X Low Pre-Treat VBM -0.027*
(0.013)
N 234 234
R2 0.919 0.922
* p < 0.05, ** p < 0.01, *** p < 0.005
All models include county and year fixed effects and standard errors clustered at the county level
Data from Washington state, 1996-2016

     

      Finally, I test the effects of universal VBM on turnout. The first column reproduces the small positive effect of universal VBM on turnout of about 1 percentage point.4 The second column suggests that the turnout effect of universal VBM is slightly higher in counties with lower mail voting shares. However, this estimate is small (less than half a percentage point) and not precisely estimated. Overall, the evidence suggests that counties with low preexisting levels of VBM do not have very different turnout effects.

     

Effect of Universial VBM on Voter Turnout [0-1]
Baseline Interaction
Universal VBM 0.009* 0.007
(0.004) (0.006)
Universal VBM X Low Pre-Treat VBM 0.004
(0.008)
N 312 312
R2 0.950 0.950
* p < 0.05, ** p < 0.01, *** p < 0.005
All models include county and year fixed effects and standard errors clustered at the county level
Data from Washington state, 1996-2010

     

      Overall, I find weak evidence that universal vote my mail differs much between counties that had lower levels of vote by mail before making it universal compared to those with higher levels. Any partisan effects are even smaller in low pre-implementation VBM counties, and turnout might be slightly higher. All things equal, this should add to the evidence that universal VBM will not cause dramatic changes in turnout or composition of the electorate.

      However, even low VBM counties in this context have significant pre-existing levels of mail voting before making it universal. It is possible that states that implement universal VBM that currently have a very small fraction of ballots cast by mail will see different effects. There are also administrative issues with the rapid expansion of vote by mail that Washington State’s experience cannot speak to.

  1. There may be slightly better ways of doing this. Vote by mail rates were increasing overall throughout this period in Washington, so some counties with above the median vote by mail rates that adopted universal VBM late might have a larger “bite” than counties below the median.↩︎

  2. Thomson et. al. only use data from California for their estimate of how much universal VBM increases the share of ballots cast by mail (columns 3-6 of table 3), which produce slightly smaller estimates.↩︎

  3. They argue that the small partisan effect is a result of a preexisting trend towards higher democratic voteshare in counties that adopted universal vote by mail.↩︎

  4. Pooling all states, they find an effect of two percentage points, though at least some of this appears to be a result of a pre-trend, see figure S6.↩︎