Tag: Presidential election in Pennsylvania

Look-Back Assessment of the Pennsylvania 2024 Voter Turnout Model: Findings and Interpretive Discussion

Apr 7th, 2025
As described in a previous Unrivaled Politics blog post, the Pennsylvania Voter Turnout Model is designed to forecast participation rates in an upcoming election for Pennsylvania registered voters. The model was developed in advance of the November 5, 2024 election and has been applied to generate turnout predictions for that election. This article assesses the performance of the turnout model via a look-back comparison of projected to realized outcomes from that election.

The model segments voters based on their history of participation in prior elections; on active vs. inactive status as assessed by election officials; and on other voter characteristics including party affiliation, county of residence, and age. In the current version of the model each segment’s projected turnout rate is its observed participation rate from the 2020 general election.

The model was developed using data on historical participation rates are obtained from the Pennsylvania Department of State’s Full Voter Export (FVE) file as of August 12, 2024.¹ Specifically, the calculated turnout frequencies are those observed for the subpopulation of the full FVE sample whose presence predates (whose registration dates are prior to) the registration deadline for the 2020 general election. ²

As with any projection model based on extrapolation from historical experience, realized outcomes may diverge from projections. In our context, the realized participation rate within a given voter history segment may diverge from the outcome projected by the model.

For instance, realized outcomes may diverge because of how voters react to the particular candidates or issues specific to each election cycle. Also, the composition of the voter population may change over time in ways that affect the segment participation rate. In the latter case, changes in size or demographic composition of the segment population might be observable, but the effects of these on participation rates may not be foreseeable.

That said, there are good reasons to expect projections to be reasonably accurate, especially in regard to rank ordering of participation likelihoods across voter history segments. Voters who have tended to vote sporadically in the past typically will continue to do so. Those who rarely have missed voting in an election almost surely will vote in the next one. Inactive voters have been classified as such by election officials for a substantive reason (such as failure to respond to an address verification notice), so that inactive status should correlate with relatively low likelihood of voting in the next election.

Performance analysis results by party affiliation

Projected outcomes for our assessment of model performance are obtained from applying the model to the population of registered voters in the FVE file as of October 21 (the registration deadline for the November election.) Realized outcomes–whether and how each registered voter cast a ballot in the election–are obtained from the January 20, 2025 FVE file.³

Table 1 shows the gap between projected and actual turnout in 2024 by party affiliation: turnout rate differentials in panel A and projected versus actual counts of voters in panel B. The Democrats achieved a turnout rate 1.5 percentage points higher than predicted (79.4 versus 77.9 percent), with actual turnout exceeding projected by about 56,000 voters. The Republican turnout rate (85.7 percent) was almost identical to predicted.

Although this might seem to be a small differential, in a closely contested battleground state like Pennsylvania even small turnout differences might swing an election one way or the other. In this case, the end result was that the Republican vote count in 2024 exceeded the Democrat count by about 52,000 instead of the projected 111,000 voters.

Table 1: Prediction error by party affiliation

Turnout percentage

Party
actual projected gap
Democratic 0.794 0.779 0.015
Republican 0.857 0.858 -0.001
Other 0.678 0.694 -0.014

Turnout count

Party
actual projected gap
Democratic 3,004,427 2,948,447 55,980
Republican 3,056,493 3,059,585 (21,211)
Other 925,607 946,908 (3,092)

Source: January 20, 2025 PA FVE file; Pennsylvania Voter Turnout Model

What factors lay behind the higher-than-predicted turnout of Democrat registered voters in 2024? One plausible explanation derives from the fact that between November 2020 and November 2024, the Democrats experienced a net loss of about 230,000 party-affiliated registered voters.

It is conceivable that these exiting voters had been less inclined to vote in 2020 compared to others within the same segment—a conjecture consistent with these voters’ apparent dissatisfaction with the party. With their departure, the party’s overall turnout rate would then rise.

It also is possible that some of the lift to Democratic turnout came from crossover voters—disaffected Democrats who had low turnout in prior elections but opted to vote for Trump in 2024 (while not changing their affiliation.) This conjecture is consistent with reported findings from national polls showing increased support for Trump within certain demographic segments of Democrat voters in 2024, such as younger men.

Also in Table 1, we see that turnout was lower than predicted among unaffiliated voters. The realized turnout rate for this cohort was 63.2 percent, lower than their predicted 65.4 percent, with actual turnout falling short by about 45,000 voters.

Reportedly, many voters from across the political spectrum in 2024 viewed both Harris and Trump as significantly flawed candidates. This view may have been particularly strong among independent voters and those not affiliated with either of the two major parties, providing a potential explanation for the relatively low turnout among these voters.

Drilling down further

While, the accuracy of projected turnout rates by party affiliation in an upcoming election is of foremost interest, having direct implications for the election outcome, accuracy within distinct voter segments is also of interest. Segment-level analysis may suggest ways to improve overall accuracy of the model. It also provides insight into robustness of the model–how accurate its projections would be if the composition of the population were to shift.

The model’s projections proved highly accurate for the segment consisting of voters who turned out for the primary preceding the general election. As shown in Table 2, the gap between actual and projected turnout rate for these voters is no greater than 6-tenths of a percentage point in each party affiliation category.

Table 2: Model performance for the segment of primary voters

Party
actual projected gap
Democrat 0.985 0.979 0.006
Republican 0.985 0.983 0.002
Other 0.949 0.953 -0.004

Also noteworthy, but for a different reason, is the model’s performance within the inactive voter segment (exclusive of primary participants). Here, a substantially lesser degree of accuracy is observed. As shown in Table 3, actual exceed projected turnout rate for these voters by multiple percentage points in each party affiliation category.

Table 3: Model performance for the segment of inactive, primary non-participants

Party
actual projected gap
Democrat 0.432 0.37 0.062
Republican 0.515 0.395 0.12
Other 0.378 0.337 0.041

Because these prediction gaps are positive and large, they have a material influence on model performance within the subpopulation comprising non-participants in the primary. This is seen in Chart 1, which shows the prediction gaps for that population overall and by active versus inactive, by party affiliation. (Percentage point gaps are shown in panel A and count gaps in panel B.)

Chart 1: Actual minus predicted turnout rate by party affiliation and active versus inactive status

Turnout percentage

Vote count

These results suggest that after 2020, the inactive voter classification process became less precise at distinguishing voters who had exited the Pennsylvania voting population (among those who had not participated in the primary.) Accordingly, in 2024 relative to 2020 there would have been more “false positives” (willing and able to vote) within the inactive segment. Conversely, there would have been more “false negatives” (unable or unwilling to vote) among those classified as active.

Let us now drill down yet further within the cohort of active, primary non-participants in 2024, distinguishing between those who voted by mail-in, absentee or provisional ballot in 2020 and others. Within this cohort, those who had voted by mail-in, absentee, or provisional ballot in 2020 had weaker 2024 turnout than those who had voted at the polls in 2020, relative to what the model had projected. This was the case within each party affiliation category, as shown in Chart 2.

Specifically, among Democrats, actual 2024 turnout exceeded projected by nearly 3 percentage points for those who had voted at the polls in 2020, while actual was 6-tenths of a percentage point below projected for the mail, absentee or provisional cohort. Among Republicans, actual 2024 turnout was about equal to projected for those who had voted at the polls in 2020, while actual was a full 6 percentage points below projected for the mail, absentee or provisional cohort. The latter cohort also underperformed relative to the at-polls cohort among voters in the “other” party affiliation category.

Chart 2 similarly shows that active, primary non-participants in 2024 who had been registered but did not vote in 2020 had weaker 2024 turnout than those voting at the polls in 2020, relative to what the model had projected. Again, this was the case within each party affiliation category.

Chart 2: Prediction error by party affiliation and 2020 method of voting

(cohort of active, non-primary participants in 2024)

Turnout percentage

Vote count

Interpretive discussion

The factor or factors underlying the unexpectedly high turnout of inactive voters are not directly observable. As noted previously, their higher turnout suggests that being labelled inactive had become a less precise indicator of exit from the state’s voting population. But why might this be the case?

One plausible scenario is that the state has had less success after 2020 at identifying, and classifying as either ineligible or inactive, individuals who had relocated to other states. Hence, assignment of voters to inactive status became more idiosyncratic (such as due to voters relocating within the state and failing to update their registration data) and thus less reflective of actual exit from the voting population.

Consistent with this scenario is the fact that in early 2023, Florida, Ohio, Virginia, and West Virginia each pulled out of the cross-state data cooperative known as the Electronic Registration Information Center (ERIC).⁴ States that participate in ERIC share their voter registration data and receive notification about voters who appear to have relocated to other states. Each of these states that exited ERIC is a relatively common destination for voters relocating from Pennsylvania. Their departure could have impaired Pennsylvania’s efforts to maintain the accuracy of its voter registration database, including by labelling potentially relocated voters as inactive.

No-excuse mail-in ballots, first introduced in Pennsylvania for the 2020 election, may have further compromised the state’s ability post-2020 to identify voters who had moved out-of-state. No-excuse mail-in voting may encourage or facilitate cross-state voting, a phenomenon examined in a 2024 study from the National Bureau of Economic Research.⁵ Cross-state voting by relocated voters via mail-in ballot could preempt their being assigned to the inactive list or being removed from the registration rolls.

This line of reasoning suggests that the state’s ability to identify inactive or exited voters might have been undermined primarily among the population voting by mail during 2020. Thus, this reasoning is consistent with the finding that in 2024, active primary non-participants who had voted by mail in 2020 had substantially lower turnout rates than those who had voted at the polls in 2020, relative to the model’s projections.⁶

The preceding scenarios are clearly plausible. Confirmation of their explanatory relevance, however, requires gathering more direct evidence.

Stability of the Rank Ordering

Although, there are aspects of divergence between actual and modelled turnout rates, as described above, the model reliably differentiates turnout likelihood across voter turnout segments. This is evidenced by stability of the rank ordering of turnout rates across modelled segments by party affiliation, as seen in Chart 3.

The five pairs of bars in Chart 3 depict projected and actual turnout rates for five turnout likelihood segments of voters. The first segment from the left comprises primary participants. The next three segments classify non-participants in the primary who are active into three ranges of projected turnout likelihood: high (90 percent or greater turnout likelihood); high-mid (at least 80 but less than 90 percent); and lower (below 80 percent). The last segment comprises inactive voters.

Both projected and actual turnout rates decline across segments moving from left to right in the chart. In most cases, the changes in actual and projected turnout rates closely align across adjacent segments, demonstrating stability and reliability of the model’s rank ordering of turnout likelihoods.

For each party affiliation, the largest misalignment occurs between the lower propensity range of active voters and the inactive segment. Here, the model’s projected decline in turnout rate overstates the actual decline, consistent with the earlier discussion on model accuracy.

For Democrats and Republicans, the next largest divergence occurs between the high and high-mid propensity ranges of active voters. Here, the model’s projected decline modestly overstates the actual decline for Democrats and understates it for Republicans. For the Other affiliation category, the second largest divergence occurs between the high-mid and lower propensity ranges, where the model understates the actual decline in turnout rate.

Chart 3- Stability of rank ordering by model turnout propensity

Concluding Summary

Projected voter turnout rates by party affiliation from the Pennsylvania Voter Turnout Model were close to actual. For Democrats, the actual turnout rate exceeded projected by 1.5 percentage points, while for Republicans, actual and projected were virtually equal. The model’s understatement of Democrat turnout conceivably might be tied to an exodus of less engaged voters from the Democratic party after 2020.

The model also reliably differentiates turnout likelihood across turnout propensity segments. While the model exhibits a high degree of accuracy, one should bear in mind that even small deviations in turnout can swing an election one way or the other in a closely contested state like Pennsylvania.

The model appears to be least accurate at projecting turnout rates of inactive, primary non-participants. This outcome suggests that Pennsylvania has had less success since 2020 at identifying and classifying as inactive, individuals who had relocated to other states.

One factor that may have impeded the state’s ability to identify these voters is the departure from ERIC in early 2023 of Florida and a few other states that Pennsylvanians commonly relocate to. No-excuse mail-in ballots, first introduced in Pennsylvania for the 2020 election, may have further hindered the state’s ability to identify voters who had moved out-of-state.
1. The FVE data can be accessed here . ↩︎
2. Thus, the estimation sample comprises only voters who were present in 2020 and have survived through 2024. We are unable to estimate a model using the full population that was present in 2020 because we lack access to archived FVE data from 2020. ↩︎
3. The FVE data distinguishes between four methods of voting: at polls (in -person machine voting), provisional ballot, mail-ballot (no excuse mail-in), and absentee ballot (including military and other traditional absentee ballots). A missing value indicates non-participation. ↩︎
4. See Miles Park, “Republican States Swore Off a Voting Tool. Now They’re Scrambling to Recreate It.” NPR.org, October 20, 2023. (Link) ↩︎
5. See Gordon B. Dahl, Joseph Engelberg, Runjing Lu, and William Mullins, “Cross-State Strategic Voting.” National Bureau of Economic Research, Working Paper 30972, September 2024. (Link) ↩︎
6. Similarly, the departures from ERIC may have curtailed the state’s ability to identify inactive or exited voters among the cohort who were registered but did not vote in 2020. As noted, this cohort had weaker 2024 turnout than those voting at the polls in 2020. Alternatively, this cohort may simply have become even less engaged as of 2024 than they were in 2020. ↩︎
Swing or Trajectory? The Shifting Political Landscape of Pennsylvania

Mar 16th, 2025
In the 2020 Presidential Election in Pennsylvania, Joe Biden prevailed by about 80,000 votes over Donald Trump. In 2024, against Kamala Harris, Trump prevailed by about 120,000 votes, thus gaining about 200,000 votes on net.

Might the 2024 outcome signify a fundamental shift in Pennsylvania’s political landscape? This note delves into this question from a data-driven perspective.

Inferences from registration and participation data

Two key factors indicate that at least a part of the Republican gains in 2024 are more than just fleeting and will be challenging for the Democrats to turn back.

First, between November 2020 and November 2024, the number of voters registered as Democrats declined substantially, while the counts of registered Republicans and of those unaffiliated with either major party increased, as seen in chart 1. Moreover, those increases were a continuation of gains accrued between 2016 and 2020 by those groups.

Chart 1: Pennsylvania registered voter counts by party affiliation

Source: PA Department of State 2016, 2020, and 2024 Voter Registration Statistics

Going into the November 2020 election, the Democrats had enjoyed an advantage of 686,000 more registered voters compared to the Republicans (rounding to the nearest 1,000). Because of each party’s gains and losses over the interim period, as of November 2024, the Democrats’ registered voter advantage had fallen to 286,000.

Second, based on extrapolation from 2020 turnout rates (using statistically modeled turnout likelihoods), the Republicans had a substantial turnout rate advantage over the Democrats going into the November 2024 election.¹ Democrat registered voters had an expected turnout rate of 77.9 percent, compared to 85.8 percent for the Republicans and 69.4 percent for unaffiliated and other voters.

Consequently, the Democrats’ shrinking margin in raw number of registered voters enabled the Republicans in 2024 to surpass the Democrats in participation-weighted count—the number of voters expected to participate in the November 2024 election. As shown in chart 2, the expected count of participating Republican voters, based on modeled turnout likelihoods, exceeded that of the Democrats by about 110,000, presaging a Republican victory.

Chart 2: Predicted (turnout-weighted) 2024 counts by party affiliation

Source: Unrivaled Politics data turnout model, from PA voter registry

Democrat turnout in 2024 ended up higher than expected

Model-based predictions, of course, are subject to error. Chart 3 shows the gap between predicted and actual turnout in 2024 by party affiliation: the turnout rate differentials in panel A and predicted versus actual counts of voters in panel B.

The Democrats achieved a turnout rate of 79.4 percent, higher than the predicted 77.9 percent, with actual turnout exceeding predicted by about 56,000 voters. The Republican turnout rate (85.7 percent) was almost identical to predicted. The end result was that the Republican vote count exceeded the Democrat count by about 50,000 (instead of the predicted 110,000).

What factors lay behind the higher-than-expected turnout of Democrat registered voters in 2024? One plausible explanation is a “selection effect” from voters who had left the Democratic Party (by switching to unaffiliated or Republican) between 2020 and 2024. It is conceivable that these exiting voters had had been less inclined to vote in 2020 compared to others within the same segment—a conjecture consistent with these voters’ apparent dissatisfaction with the party. With their departure the party’s overall turnout rate would rise.

It is also conceivable that the lift to turnout came from disaffected Democrats who had low turnout in prior elections but opted to cross-party lines and vote for Trump in 2024, without changing their affiliation. This conjecture is consistent with reported findings from national polls showing increased support for Trump within certain demographic segments of Democrat voters in 2024, such as younger men.

Chart 3: Prediction error by party affiliation

Chart A: Turnout Percentage Chart B: Vote Count

Source: Unrivaled turnout model (from PA voter registry)

Independent and cross-over voters necessarily account for the remaining margin

The analysis thus far has established that if voting had been strictly along party lines, with Trump and Harris tied among independent voters or those not affiliated with either major party, then Trump would have won by 50,000 votes. It then follows purely mathematically that the residual 70,000 vote portion of Trump’s winning margin must have come from getting more cross-over votes from Democrats than Harris received from Republicans, or from dominance among unaffiliated voters, or from some combination thereof.

In this sense, most of Trump’s margin appears to have come from swing voters, who tend to be less wedded to a partisan point of view and comparatively responsive to individual candidates and the candidates’ stands on key issues. Underscoring this point is the fact that in the Pennsylvania senatorial race, the Republican candidate Dave McCormick garnered only about 15,000 more votes than the incumbent, Democratic candidate Bob Casey Jr., mostly due to Trump voters who had opted not to support McCormick.²

On the role of mail-in voting in 2024

Reportedly, many voters from across the political spectrum in 2024 viewed both Harris and Trump as significantly flawed candidates. As was seen in Chart 3, turnout was lower than expected among unaffiliated voters, perhaps for that reason. However, Republicans did not have noticeably lower turnout, and Democrats had higher turnout relative to model predictions.

Potential explanations for Democrats’ higher-than-expected turnout have already been offered. But what might explain the sustained high turnout of Republican voters?

One possibility is that Republican voters who may have disliked Trump still voted for him because of concerns about what a Harris presidency might bring (in terms of a doubling down on Biden’s most divisive policies). Another is that the unrelenting, brutal lawfare and perceived weaponization of justice against Trump motivated Republican voters to circle the wagons, so to speak.

Another important factor to consider is stronger embrace of mail-in voting by the Republican party in 2024 relative to 2020. Mail-in voting can bolster participation by making the process more convenient for some voters and less susceptible to Election Day attrition. Thus, greater receptivity toward mail-in voting may have helped boost Republican turnout and offset lack of enthusiasm for Trump among some elements of the party.

During the 2020 election, the ongoing coronavirus pandemic spurred popularity of mail-in voting, due to perceived risks of social contact at polling places. However, the mail ballot option proved far more popular among Democrats in 2020 compared to Republicans. The latter held far less confidence in the security and reliability of mass mail-in voting, which had been newly introduced into Pennsylvania.³

This attitudinal gap between Democrats and Republicans in 2020 was evidenced by the fact that about 58 percent of the votes for Biden were by mail-in or absentee ballot, compared to just 18 percent of Trump votes. In 2024, however, this attitudinal gap had moderated, likely due in part to organized efforts, such as an initiative known as the “Pennsylvania Chase”, that promoted mail-in voting among Republicans.⁴

This change in attitude is demonstrated in Chart 4, which shows the ratio of Democrat to Republican mail-in ballot share (frequency of voting by mail-in ballot) across four segments of the 2024 participating voter population.⁵ The first bar from the left shows the ratio of mail-in ballot shares in 2020 for the subpopulation comprised of voters who participated in both the 2020 and 2024 elections. The second bar from the left shows the ratio of mail-in ballot shares in 2024 for the same subpopulation.

The third bar shows the ratio of Democrat to Republican mail-in ballot shares in 2024 for participants in 2024 who were present but did not vote in 2020. The final bar shows the ratio of Democrat to Republican mail-in ballot shares in 2024 for participants in 2024 who were not present in 2020.

Thus, in 2020 ,the frequency of mail-in voting by Democrats was 2.8 times that of Republicans among voters who participated in both elections. The ratio of Democrat to Republican mail-in voting frequency within this cohort declined to 1.7 in 2024. The other two cohorts had ratios in 2024 of 1.4 and 1.7, respectively.⁶

This analysis suggests that organized efforts to normalize mail-in voting among the Republican electorate had succeeded, whereby they became more receptive to the mail-in option than they had been in 2020. As noted, such increased openness toward mail-in voting could have bolstered turnout.

Chart 4: Ratio of Democrat to Republican mail-in ballot share by voter history segment

Source: Unrivaled voter turnout calculations using the PA Voter Export data file

Concluding summary

Since 2016, there has been a steady increase in the number and percentage of registered voters in Pennsylvania choosing to affiliate with the Republican party. The number and percentage of unaffiliated voters has also grown. There are roughly 220,000 fewer registered Democrats in the state today compared with 2020 (or with 2016).

The Democrats still enjoy a raw numerical advantage in Pennsylvania—the number of registered Democrats exceeds the count of registered Republicans by over 300,000 voters. However, the Republican share of the voter population has clearly been on a upward trajectory.

Moreover, the Republican Party in 2024 benefitted from a substantially higher turnout rate compared to Democrats. Their turnout rate advantage provided them with a slim, turnout-weighted margin, whereby they outvoted Democrats by about 50,000 votes cast. Efforts to make Republican voters more receptive to vote-by-mail plausibly helped bolster the party’s turnout in 2024.

In 2024, Trump prevailed over Harris by 120,000 votes, from which we can deduce that he netted some 70,000 votes from cross-over Democrats and independent voters. However, these should be considered swing voters who might readily reverse course in the next election.

The current, slim, turnout-weighted difference in number of Republican versus Democrat registered voters signifies that Pennsylvania remains a swing state—the decisions of cross-over and independent voters can still swing an election one way or other. Yet there is no question that the Republican Party today is much better positioned for winning in Pennsylvania compared to the recent past and appears to be on a trajectory to strengthen its position.
1. An ex-ante probability of turnout is assigned to each registered voter in 2024 based on a statistical model fit to data from previous elections. These modeled turnout probabilities are then applied to obtain estimates of the number of participants in the November 2024 election, by party affiliation. ↩︎
2. Their lack of support was reflected in their opting not to cast a vote or voting for a minor party candidate. ↩︎
3. Before 2020, vote-by-mail was limited to traditional absentee ballots, which were available only to military and oversees civilian voters or those professing to have a valid reason for being away from their home precinct on election day. Beginning in 2020, no excuse was required to vote by mail, although voters could still opt for the traditional absentee ballot. The vast majority of voting by mail has relied on the new, no-excuse ballot format. ↩︎
4. The PA Chase is an initiative of the of the Citizens Alliance of Pennsylvania. ↩︎
5. Here, and through the discussion below, the term mail-in ballot includes both the traditional absentee and the newer, no-excuse vote-by-mail formats. ↩︎
6. Mail-in voting might be most effective at increasing turnout rates among voters with lower turnout propensities. If voters with high modeled likelihood of participation (90 percent or higher) are excluded from this analysis, then these ratios become 2.1, 1.5, 1.3, and 1.6, respectively. ↩︎
Bonus Ballots for Biden: A Case Study of Election Lawfare Excess

Aug 27th, 2024
As the 2024 Presidential campaigns ramp up, election lawfare is underway as well, with the political parties seeking to exploit ambiguities in voting rules or bend them in their favor, and we can only hope that its role will remain limited and not affect trust in the election. This article highlights the problematic nature of election lawfare, focusing on a controversial case from the 2020 Presidential election in Pennsylvania.

In 2020, in states with newly expanded mail-in voting processes where the parties battled in court over what would be permissible under the new rules and what would not be, with Pennsylvania a particularly active lawfare battleground. The case discussed and analyzed here resulted in extension of Pennsylvania’s originally mandated deadline for receipt of mail-in ballots, and likely had a material effect on the final vote tally.

The analysis, which is in part data-driven, calls into question the primary, stated rationale for extending the deadline, which was to ensure sufficient time for return of mail-in ballots for those who had been relatively late in requesting a mail-in ballot. It also offers a rough quantification of the net vote gain for Joe Biden arising from the extension.

While this number may have been material–possibly as many as 9,000 votes, it was far from enough to swing the election. Nonetheless, in combination with other perceived shortcomings of the process such as lax security around ballot drop boxes, the case contributed to a broader erosion of trust in the election outcome.

The Democrats’ legal challenge to the ballot return deadline

No-excuse mail-in voting was introduced into Pennsylvania by PA Act 77, which was legislated in 2019 with bipartisan support in the state legislature and the blessing of the Democrat governor (Tom Wolf). Although the Democrats in the state’s General Assembly in 2019 mostly opposed Act 77, the party had a change-of-heart by the summer of 2020. Conversely, at the national level during 2020, President Trump and other Republican leaders increasingly were expressing fears that no-excuse mail-in voting might facilitate election fraud.

Observations on voter turnout during the 2020 Primary in Pennsylvania might have helped motivate the Democrats’ embrace of mail-in voting. No-excuse mail-in ballots accounted for almost 40 percent of total votes in the 2020 Primary Election, underscoring the preference of many voters for the mail-in option. Although data on mail-in ballot shares by party are not publicly available, it is a reasonable presumption that Democrats in the Primary were more likely to vote by mail than Republicans, as turned out to be the case in the general election, and that this fact was known to strategists in each party. ¹

Moreover, it seems that, because of the perception that liberalized vote-by-mail favored their party, Democrats were not content with a literal or narrow reading of Act 77. They adopted a strategy of lawfare to try and stretch the limits of mail-in voting as far as possible in order to maximize their perceived vote potential.

On July 10, 2020, the state Democratic Party, joined by some individual Democratic officials and candidates, filed a petition for review in the Commonwealth Court against Secretary of State Kathy Boockvar and all 67 county boards of elections. The petition requested declaratory and injunctive relief “to protect the franchise of absentee and mail-in voters.”² Specifically, the plaintiffs asked the Court to rule in favor of a number of requested flexibilities, among which was to grant an exception to the deadline in Act 77 for receipt of mail-in ballots, such that any ballot postmarked by 8:00 p.m. on election day (November 3) and received by 5:00 p.m. on Tuesday, November 10 can be counted.

Technically, Boockvar was an opponent of the petitioners in the case. However, she herself was a Democrat, and on August 13, she filed a response that objected only to the requested seven-day extension of the received-by deadline, primarily on logistical grounds, proposing instead a three-day extension. Her response strongly concurred with the petitioners’ other requests.

On August 16, Boockvar filed an application requesting that the state’s Supreme Court exercise extraordinary jurisdiction over the petition for review; that is, take over for the intermediate appellate Commonwealth Court, in order that the resolution of these matters be expedited to preclude any disruption of the 2020 General Election. On August 19, the petitioners filed an answer to the Secretary’s application, indicating no objection to the Supreme Court exercising extraordinary jurisdiction. The Pennsylvania Supreme Court, where 5 of the 7 justices are Democrats, then took over the case.

The state Supreme Court ultimately agreed to an extension of the ballot receipt deadline by three days so that any ballot cast by 8 p.m. on Election Day, (Nov. 3, 2020) and received by 5 p.m. on Nov. 6 would be counted. The Court’s decision, which was based on majority opinion, was announced on September 17.

Review and critique of the petitioners’ arguments

The petitioners had argued that special circumstances arising from the Covid-19 pandemic justified a seven-day extension to the received-by deadline for mail-in ballots. Essentially, they contended that, due to the preference of many voters to stay away from polling places on account of the ongoing pandemic, the volume of mail-in ballot applications would likely overwhelm county resources, such that bottlenecks and delays in the processing of ballot applications and distribution of ballots would likely arise. Secondarily, they noted the potential for delays in the United States Postal Service (USPS) delivery system as an aggravating factor.

They argued that under the state constitution’s Free and Equal Elections Clause, which prohibits interference in or prevention of “the free exercise of the right to suffrage”, the state was obligated to take action to remedy the anticipated disruption to mail-in voting. As described in the Court’s written opinion:

Petitioner interprets this provision as forbidding the Boards from interfering with the right to vote by failing to act in a timely manner so as to allow electors to participate in the election through mail-in voting.

A key component of the petitioners’ argument was a claim that the kinds of obstacles to the timely processing of ballot applications that were encountered prior to the June 2^nd Primary Election would recur and be even more severe. As summarized in the Court’s written opinion:

Petitioner avers that the difficulties encountered by Boards processing the ballot applications prior to the June Primary will only be exacerbated in the November General Election. It emphasizes the continued grip of the pandemic and a potential second wave of infections…. It highlights that the Secretary estimates that 3 million electors will seek mail-in or absentee ballots for the General Election in contrast to the 1.5 million votes cast by mail at the Primary.

However, the petitioners offered no substantive analysis to support this claim—only gripes about difficulties during the Primary and a poorly supported, inflated estimate of increased processing burden for the General Election.

During the Primary, many counties had indeed been inadequately prepared for the unexpectedly large volume of ballot applications they received, as this was the state’s first encounter with no-excuse mail-in voting. Thus, in some counties, most notably in Bucks and Delaware Counties, a significant number of ballots were mailed out too late for voters to return them by the deadline.

But clearly, county boards would have learned from their experiences in the Primary and taken action to shore up their capabilities in anticipation of needs for the General Election. The Court did not adequately consider what remedies might already have been implemented or what preparations were currently underway.

At least anecdotally, such ongoing efforts were substantive. A press report in late August contained the following assessment:³

Democratic leaders also contend they’ve remedied the issues that plagued them in the June primary when election officials were overwhelmed by a deluge of mail-in ballot requests during the coronavirus pandemic. It was the first time the state had conducted a mail-in election since a new law authorized the procedure last year.

For example, according to this report, Montgomery County had “used federal funds from the Cares Act to add eight staffers and more technology to help process the vote faster” since the Primary.

Another press report in early September included the following snippet:⁴

We were learning on the fly from February to June,’ says Jeff Greenburg, the director of elections in Mercer County until August, when he stepped down to work for The National Vote at Home Institute. ‘I really think Pennsylvania is in a better position now. … To me, we have a much better chance of succeeding because we now know what to do.’

Not only did the petitioners fail to back their claim that county boards remained inadequately prepared, but they overstated the anticipated increase in application processing volume. As evidence of increased burden, the petitioners had alluded to the gap between 1.5 million “votes cast by mail” in the Primary and 3 million anticipated applications for mail-in or absentee ballots for the General Election. This comparison, referenced in the Court’s written opinion, is misleading, as it is not apples-to-apples. The 3 million applications anticipated for the General Election is more appropriately measured against the total number of applications for mail-in or absentee ballots received for the Primary, which was 2 million.

Even that corrected numerical comparison, however, materially overstated the potential increase in processing burden, because much pre-processing of applications had already been completed. In fact, about 1.1 million individuals who had requested and been approved for a mail-in ballot in the Primary had signed up to be automatically sent a mail-in ballot for the upcoming general election, a highly pertinent fact missing from the discussion.

No further processing was required for the latter requests aside from the eventual automated printing and mailing of the ballots. Netting these out from the projected 3 million implies 1.9 million new applications for the General Election, not surpassing the 2 million processed for the Primary.

Review and critique of Boockvar’s arguments

Secretary Boockvar did not concur with the petitioners’ claim regarding the lack of preparedness of county boards, nor their request for a one-week extension. Instead, she advocated for a three-day extension based on the contention that those applying for mail-in ballots on or close to the statutory deadline for doing so, October 27, would face a high likelihood of missing the receive-by deadline for returned ballots if they chose to return them by mail.

The USPS delivery standards in effect at the time specified 2-5 day delivery for domestic First-Class Mail and 3-10 day delivery for domestic Marketing Mail. As described in the Court’s written opinion, “the USPS had advised that voters should mail their ballots no later than October 27 in order to meet the received-by deadline”. Boockvar argued that under these standards, a voter applying for a mail-in ballot on Oct. 27 would face a substantial risk of not meeting the November 3^rd deadline:

…even if a county board were to process and mail a ballot the next day by First Class Mail on Wednesday, October 28th, according to the delivery standards of the USPS, the voter might not receive the ballot until five days later on Monday, November 2nd, resulting in the impossibility of returning the ballot by mail before Election Day, Tuesday, November 3rd…This mismatch [between the USPS’s delivery standards and the Election Code deadlines] creates a risk that ballots requested near the deadline under state law will not be returned by mail in time to be counted…

Further, Boockvar argued that mail delays associated with the pandemic would make the problem worse, selectively disadvantaging those who are applying for mail-in ballots shortly before the October 27 deadline and therefore necessitating a one-time extension of the received-by deadline:

The Secretary emphasizes that the remedy sought here is not the invalidation of the Election Code’s received-by deadline, but rather the grant of equitable relief to extend temporarily the deadline to address ‘mail-delivery delays during an on-going public health disaster’.

This argument, however, is specious. To begin with, it rests on the illusory notion that the application and received-by deadlines in Act 77 were originally flawed because of inconsistency with standard USPS delivery times.

The legislature could have opted for an earlier application deadline, whereby voters returning their ballot by postal mail would be more assured of meeting the received-by deadline. Instead, the legislature opted to provide more flexibility on application timing with the later application deadline.

Act 77 grants voters the choice to wait up to a week before Election Day to apply for a mail-in ballot, with the implicit, common sense understanding that voters will bear an increased risk of missing the received-by deadline the longer they wait to apply. To mitigate this risk, Act 77 also provides voters the option of hand delivering their mail-in ballots, including the option of applying for and submitting ballots in-person at an election office on the same day. To claim the deadlines were flawed is disingenuous.

Moreover, Boockvar’s argument presumes that voters have little or no agency — that those who apply close to the application deadline and consequently miss the receive-by deadline do so by fate — suggesting an obligation to provide such voters with flexibility on the receipt deadline when mail delays occur. However, voters can choose when to submit a ballot application. The possibility of pandemic- related mail delays should just change their calculation, motivating them to apply sooner rather than later.

Additionally, the argument fails to consider factors that would mitigate the effect of any pandemic-related mail delays. Drop boxes are one obvious mitigant. Surely, the availability of drop boxes for returning a mail-in ballot limits the risk that late applicants will miss the receive-by deadline. Another is to inform voters about the potential for mail delays and to encourage them to apply early, which was, in fact, the sensible advice being disseminated by the USPS at the time.

The Court decides

The Court’s written opinion indicates no awareness of any of these weaknesses in the arguments presented by the petitioners and Boockvar. The judges apparently were swayed by those arguments:

In light of these unprecedented numbers and the near-certain delays that will occur in Boards processing the mail-in applications, we conclude that the timeline built into the Election Code cannot be met by the USPS’s current delivery standards…The Legislature enacted an extremely condensed timeline, providing only seven days between the last date to request a mail-in ballot and the last day to return a completed ballot. While it may be feasible under normal conditions, it will unquestionably fail under the strain of COVID-19 and the 2020 Presidential Election, resulting in the disenfranchisement of voters…this Court can and should act to extend the received-by deadline for mail-in ballots to prevent the disenfranchisement of voters.

In short, the Court undertook to alter the deadline ensconced in plain print in the legislation, potentially skewing the vote outcome in favor of one party. The decision lacked factual support, and instead relied on an inaccurate and misleading estimate of processing volume and on nonsensical griping about a “condensed timeline.”

Quantifying the effect on the vote tally

The Court’s decision to grant a three-day extension to the ballot return deadline favored Biden due to the overwhelming ratio of Democrats relative to Republicans anticipated to vote by mail. It can also be seen as implicitly disfavoring those who were planning to vote at the polls, insofar as they were not extended any comparable flexibility, such as expanded hours at polling places or more than one day of in-person voting.

How many mail-in and non-overseas civilian absentee ballots did the Democrats manage to gain during the extended timeframe, net of ballots returned by Republicans? To address this question, we turn to a dataset from the Pennsylvania Department of State which has detailed information on each approved application for an absentee or mail-in ballot for the 2020 General Election. A link to the dataset, which is housed within the Department’s online Open Data Portal, was shared with the author in response to a “Right-to-Know Law” (RTKL) request.⁵

Individual data items include the date the application was received by the county in which the voter is registered. In addition, the data distinguish between traditional absentee ballots available to voters and the newly available, no-excuse mail-in ballots (henceforth to be referred to simply as absentee and mail-in, respectively) and further differentiate among types of traditional absentee ballots, such as military, overseas civilian, regular civilian, and so on.

The database also provides the date the ballot was received back by the county election office (the ballot return date). Non-return of a mail-in ballot that was sent to a requester is indicated by missing information in the return date field: if the ballot was not recorded as returned, then no return date is present.

Chart 1 shows the counts of returned mail-in and non-overseas civilian absentee ballots received in the two weeks prior to and including November 3, the statutory deadline for returning these ballots, and the return counts for the three-day extension period granted by the Court. The counts are separated by party affiliation of the voters. During these three days, 9,432 more ballots were returned by Democrats compared to the number returned by Republicans.

While some of these ballots from Democrats might have been votes for Trump and some from Republicans might have been for Biden, there is no way to determine how many of each, and a best guess is to assume that such cross-party votes are few and mutually offsetting. Under the further assumption that Independent and other voters split equally between Trump and Biden, the net additional vote count for Biden is simply this difference between the returned ballot count of Democrat versus Republican-affiliated voters.

Does the extension seem justified in retrospect?

Chart 1 also indicates that the number of ballots returned from October 21 through November 3 swamps the number returned during the added three days. The number returned during the added days sums to 45,761, just 2.3 percent of the total number (1,987,793) returned on or after October 21.⁶ This small share seems inconsistent with the notion that pandemic-related mail delays would present a major obstacle to the timely return of mail-in ballots—the primary rationale put forth to justify the extension.

The three-day extension was motivated by a perception that those applying close to the application deadline would require this flexibility due to the potential for “systematic mail delays”. Chart 2 shows how ballots with November 4-6 return dates were distributed by application period. As seen in the chart, over half of the mail-in and non-overseas civilian absentee ballots with extension period (November 4-6) return dates were associated with applications filed prior to October 15; that is, applications filed well in advance of the deadline. Thus, most of those benefiting from extension of the deadline had not been late filers.

Another question worth asking is, how many of the ballots with application dates close to the October 27 application deadline would not have arrived in time had the extension not been granted? If there were many such ballots, then this could provide some ex-post justification for the Court’s decision.

Chart 3 shows the percent returned during the November 4-6 extension period by date of application for each application date in the late October (October 15-27) application period. Not quite 10 percent of individuals applying within three days of the deadline had return dates on November 4-6, only a slightly higher percentage compared to individuals applying four to seven days before the deadline.

These outcomes seem inconsistent with the anticipated “systematic mail delays” put forth to justify the extension, and quite plausibly can be attributed instead to procrastination on the part of some individuals or to idiosyncratic factors affecting ballot return times. In retrospect, there seems to have been little justification for extending the deadline, especially as some individuals intending to vote in person may have missed their opportunity to vote due to idiosyncratic factors, and it seems unfair to have provided extra accommodation only for the mail-in voters.

Not all ballots with recorded return dates in November 4-6 are votes that would have been lost

If the Court had left the statutory deadline unchanged, then it seems quite plausible that some of the individuals whose votes came in during the added three days would have behaved differently, returning their ballots sooner.

In other words, voters’ behavior cannot be presumed independent of the stated deadline—voters would have adapted their behavior in response to changing the deadline. It is not possible, however, to determine how many voters with return dates on November 4-6 would have met the original deadline had it not been extended.

Likewise, the actions of those whose duty it is to collect and record ballots may have adapted to the extension of the deadline. In some cases, those responsible for collection may have left some ballots uncollected as of the close of business on November 3, knowing that there would be additional time for completing the task. Even if all ballots had been collected on the 3^rd, some officials might have delayed opening and recording them until the next day, at which point they might be assigned a November 4 return date.⁷ The actual net gain to Biden due to the extension would depend on the total number of ballots incorrectly attributed to the extension period and how they divide up between Biden and Trump, which are also unknown quantities.

Enabling underperformance?

A far disproportionate share — nearly half — of the mail-in and non-overseas civilian absentee ballots with return dates on November 4-6 were from voters in just two counties, Bucks and Delaware. Chart 4 shows the distribution of ballots with November 4-6 return dates across the top 10 counties ranked by number of such ballots. Delaware County alone accounted for a third of all ballots returned during the three-day extension period. Bucks had about the same share of these ballots as much larger Philadelphia.⁸

These same counties, Bucks and Delaware, were notorious for having had especially severe delays in distributing mail-in ballots for the Primary in June. Uniquely among all Pennsylvania counties, these two had gone to their county courts to seek an extension to the ballot return deadline for the Primary. In fact, the experience of these two counties in the Primary was highlighted by proponents of the November extension as portending more systematic delays in the General Election. As described in the state Supreme Court’s written opinion regarding the November extension:⁹

…in Delaware County, thousands of ballots were ‘not mailed out until the night’ of the Primary, making timely return impossible… Bucks County apparently experienced similar delays. To remedy this situation, the Election Boards of Bucks and Delaware Counties sought relief in their county courts… the Courts of Common Pleas of Bucks and Delaware Counties extended the deadline for the return of mail-in ballots for seven days, so long as the ballot was postmarked by the date of the Primary.

Hence, one might reasonably conclude that the primary accomplishment of the Supreme Court’s granting the three-day extension was to relieve these two counties from responsibility for making needed improvements to their ballot distribution and collection processes. Surely, in retrospect, this is insufficient justification for the Court’s decision.

Concluding remarks

The troubling nature of election lawfare is exemplified by a particular case from the 2020 Presidential election in Pennsylvania in which the Democrats convinced the state’s Supreme Court to extend the deadline for receipt of mail-in ballots. In short, the Court undertook to alter the deadline ensconced in plain print in the legislation, potentially skewing the vote outcome in favor of one party.

The decision lacked factual support. Rather, the judges (the majority of whom were Democrats) were swayed by inaccurate and misleading information that was presented to them, and by a nonsensical, emotionally driven argument about a “condensed timeline” for mail-in voters arriving late to the game.

The Court’s decision was wrongheaded, because of the weak reasoning brought to bear. Overriding the explicit requirement of the legislation without adequate justification amounted to judicial overreach.

A bit of irony is evident here, considering the lawfare conducted in 2020 by the Democrats against the Green Party in Pennsylvania. A Green Party official chose to fax (as opposed to hand deliver or mail) a candidate affidavit, which was technically a violation of candidacy requirements. This minimal infraction, during the state’s Covid lockdown when office hours were limited and direct contact between persons discouraged, could have merited forbearance, but the Democrats succeeded in expelling the Green Party candidate from the ballot. Whereas the Democrats would grant no leeway to the Green Party, they self-servingly pursued a far greater degree of forbearance regarding the statutory ballot receipt deadline, based on flimsy justifications using the pandemic as a pretext.¹⁰

Analysis of mail-in voting data from this election casts further doubt on the notion that extending the deadline was necessary to ensure sufficient time for return of mail-in ballots for those who had been relatively late in requesting a mail-in ballot. In particular, the data show that most of the ballots that arrived during the three-day extension period were not from late applicants.

The extension of the ballot receipt deadline yielded a material net vote gain for Joe Biden resulting from the extension–possibly as many as 9,000 votes–and alongside the Green Party’s expulsion and other factors contributed to a broader lack of trust in the election outcome. Let us hope that in 2024, experiences such as this, where lawfare is used as a means of acquiring an unfair advantage, are not repeated.

Footnotes
1. In the 2020 Presidential Election, 57.7 percent of votes cast for Biden, but only 17.6 percent of votes cast for Trump, were by mail-in or absentee ballot, according to the state’s official election results. ↩︎
2. The case history discussion is drawn from the Opinion recorded by Justice Baer for the Pennsylvania Supreme Court’s September 17, 2020, decision on this case. ↩︎
3. Roarty, Alex, and David Catanese. August 27, 2020. “Despite uncertainty, Democrats bet big on mail voting in Pennsylvania.” McClatchyDC.com. Link. ↩︎
4. Bryant, Christa Case. September 4, 2020. “Why Pennsylvania is ground zero for mail-in voting debate.” The Christian Science Monitor. Link ↩︎
5. The link was provided in a RTKL response letter dated March 25, 2021; however, the dataset itself might already have been publicly available. The version currently accessible through this link may incorporate some updates relative to the dataset used here. ↩︎
6. Broken out by party affiliation, the shares received during November 4-6 were 1.9, 3.5, and 2.7 percent for Democrats, Independents or other, and Republicans, respectively. ↩︎
7. In fact, it is almost certain that some ballots recorded as returned on the 4^th actually were returned on the 3^rd. For five outlier counties, the data show zero or near zero ballots returned on the 3^rd but a far larger number on the 4^th, which is not credible. These five outlier counties and their ballot counts for November 3^rd compared with the 4^th are: Clinton (6 versus 169); Luzerne (5 vs. 2,076); Sullivan (1 vs. 33); Lebanon (0 vs. 729); and Mifflin (0 vs. 513). In the rest of the state, more than three times as many ballots had a return date of the 3^rd relative to the 4^th. ↩︎
8. Philadelphia had about 2⅓ times as many registered voters as Bucks County. Montgomery had less than a third the number of ballots with November 4-6 return dates relative to Bucks, despite having about 40 percent more registered voters. Also note that the shares for Luzerne and Lebanon shown in the chart are inflated for the reason described in the previous footnote – zero or near-zero ballots with November 3 reported return date. ↩︎
9. See pages 21-22 the Pennsylvania Supreme Court’s September 17, 2020, decision. ↩︎
10. For a wider ranging discussion of lawfare cases initiated by Democratic Party activists during the 2020 election around the nation, see chapter two of Fund, John, and Hans von Spakovsky. 2021. Our Broken Elections: How the Left Changed the Way You Vote. New York: Encounter Books. ↩︎