Postal network redesign and mail ballot rejections in the 2026 primaries

In 2025, the Postal Service changed how mail leaves thousands of post offices. Mail sent from those offices can now wait an extra day before beginning its journey. Voters send their mail ballots back through those offices, and unlike most mail, those ballots must arrive by a deadline. The 2026 primaries were the next federal elections after this change and are the best available guide to what that may mean. November will be the first general election since the change, with a far larger electorate.

That change is part of Regional Transportation Optimization (RTO), a U.S. Postal Service (USPS) effort to run its transportation network more efficiently. RTO covers post offices that are more than 50 miles from the plant that processes their mail. It is one part of Delivering for America (DFA), the agency’s 10-year plan to adjust the network to declining letter volumes.

Three other developments followed, each relevant to whether a mailed ballot counts. In December 2025, USPS confirmed that a postmark may no longer show the day a piece of mail was sent. In June 2026, the U.S. Supreme Court left states free to count ballots that arrive after Election Day, provided they were postmarked on time. Finally in August, at the direction of a March executive order, USPS issued a rule setting new requirements for ballot envelopes and for the data election officials must share with the agency about the ballots they mail. A federal court blocked that rule; and on September 14, the Supreme Court left the block in place, so it will not apply for the November general election.

RTO is different. It changed how mail moves, not how ballots are counted. For a ballot mailed close to the deadline, an extra day can be the difference between being counted and rejected.

In this new analysis of the 2026 primaries in California, North Carolina, and Washington, I find that RTO exposure roughly doubled the rate at which mail ballots were rejected for arriving late. Rejections for every other reason did not move differently where RTO applies, and before RTO, the places it now covers rejected late ballots at about the same rate as places it does not, in both the two preceding federal primaries and across a decade of general elections nationally.

Using data from the 2022 and 2024 general elections, I estimate that roughly 40% of mail ballots returned by voters (16.6 million) come from places where the nearest post office is covered by RTO. If the same relationship holds in November, tens of thousands of additional ballots could be rejected for arriving late, against a recent average of about 115,000 late rejections. The remedy is not complicated. Voters in covered places can mail their ballots a few days earlier, consistent with what USPS has long advised. Knowing which places those are is harder. USPS and election officials should make that easy to find out.

RTO can add a day to a ballot’s return journey to the election office

RTO is, in the Postal Service’s own words, an “effort to align transportation schedules between processing facilities and Post Office locations to improve operational efficiencies.” Each post office used to see two trucks a day, one to drop off mail in the morning and a second later in the day to pick up additional mail that had accumulated. Under RTO, post offices more than 50 miles from their regional plant were reduced to a single daily trip. Carriers still collect mail on their routes, and blue-box collections still occur on schedule. But mail arriving after that truck has left now waits for the next day’s truck.

The reason given by USPS for this change is volume: single-piece First-Class Mail fell from 57 billion pieces in FY1997 to 12 billion in FY2023, so the second trip had come to carry very little additional mail. The published service standard reflects this change. For single-piece First-Class Mail originating from these zip codes, a day is added to the first leg of travel. At the same time,  USPS shortened its promised delivery times for longer distances, so many pieces of mail may be unaffected on net. Ballots returned by voters are single-piece First-Class Mail, so a ballot mailed from a covered zip code is subject to that added day.

The places RTO covers are more rural and older than the country as a whole

Who is affected by this change depends on where the covered post offices are and who lives nearest to them. The Postal Service’s own location data identifies post offices by their exact location, and an October 2024 determination it filed with the Postal Regulatory Commission indicates which post offices are covered by RTO. I assign every census block in the country (the smallest unit of Census geography, roughly a city block in a town and larger in rural areas) to its nearest post office, and I count the voting-age population in the blocks whose nearest post office is covered by RTO. I characterize a county’s exposure as the share of its voting-age adults living in those blocks.

Exposure is measured by designation as opposed to implementation. The October 2024 filing records which offices were planned for RTO, not which have already adjusted their operations. USPS reports implementation separately, by zip code, and as of September 2026, roughly 58% of designated offices nationally were operating on the new schedule.

Roughly 121 million of the country’s 257 million voting-age adults in the 48 contiguous states and Washington, D.C., live nearest to a covered post office. Table 1 compares them with those whose nearest post office is not covered. Half of them (49.2%) live in a small town or rural area, compared to just 8.8% for those not covered. They are also more elderly, poorer, and more likely to have a disability. And they are more likely to be White, which follows from their geographic distribution: rural America is Whiter than the country as a whole.

Three states report enough detail to measure late rejections

RTO activations began rolling out in 2025. The next federal elections were the 2026 primaries, and they are the only ones that can be observed before November. California, North Carolina, and Washington each held primary elections and report how many mail ballots voters in each county returned, how many were rejected, and why, with late arrival recorded as its own category. All three also report the same detail for the 2022 and 2024 primaries, both held before RTO, so 2026 rejection rates can be compared with those from before the change.

These three states differ in how a late ballot is handled. In California and Washington, which both mail a ballot to every registered voter, a ballot that is postmarked by Election Day still counts even if it arrives later. A voter in North Carolina must request an absentee ballot, and the state ended its grace period before the 2024 election. So a ballot must arrive by Election Day, or it does not count.

Ballots can be rejected for several reasons: a signature that does not match the one on file, a missing witness line, a missing date. In all three states and in each of the three federal primaries between 2022 and 2026, most rejections were for arriving too late to be counted (Figure 1).

Comparing counties based on where RTO applies reveals its effect

Estimating RTO’s effect on late-arriving ballots is complicated because rejection rates can move for many reasons that have nothing to do with the mail: how many people vote by mail, how carefully they complete the ballot envelope, how a state runs its elections. RTO’s reach, however, varies from county to county, and this variation makes it possible to separate its effect from these other forces. RTO coverage is close to all-or-nothing in most places. In these three states, for example, 84% of counties are either less than 10% covered or more than 90% covered, and only six of the 197 sit between 40% and 60%.

I divide counties into two groups: those where most residents’ nearest post office was covered by RTO and those where most were not. This division makes the comparison easy to see. If rejection rates in the two groups moved together before the change and apart afterward, the difference between them is likely to be driven by RTO itself. At the same time, few counties sit near the line, and the technical appendix shows that these results are not sensitive to where the line is drawn.

Figure 2 shows the rejection rate for each group in each of the three primaries. For primaries that were held before RTO (2022 and 2024), the two groups look alike in each state; and when rejection rates moved, they moved together. That is what makes the uncovered counties useful: they can provide a reasonable estimate of how much rejection rates would have moved in 2026 regardless of RTO. Instead of moving together in 2026, the groups separate. Rejection rates rose sharply in counties where most residents’ nearest post office was covered and held steady or fell in the rest; the difference is what is attributed to RTO. Comparing the change in one group with the change in a comparable group is an estimation strategy called “difference-in-differences.”  

RTO nearly doubled the share of ballots rejected for arriving late

Using this comparison, I estimate that across all three states RTO exposure increased the rate at which mail ballots were rejected for arriving late by 96%, or about one percentage point. The size of the effect differs by state: the rate rose about 142% in Washington (1.1 points), 86% in California (0.9 points), and 61% in North Carolina (1.9 points). North Carolina’s proportional increase is the smallest because it is scaled by its baseline rejection rate, which was already several times higher than in the other states. Its estimate is also the least statistically precise, likely because voters in two-thirds of its counties returned fewer than 100 mail ballots in the 2026 primary. Estimates for California, Washington, and the three taken together are statistically significant, meaning chance is an unlikely explanation for the observed increase. The technical appendix provides more detail on the full results, explains how they are estimated, and shows that they are insensitive to how counties are divided into two groups based on RTO exposure.

A second test uses the same county groups and the same elections but studies a different outcome: ballots rejected for any reason other than late arrival, including a missing signature or missing date. These ballot rejections should not respond to a change in mail transportation. Figure 3 shows those rejection rates for both groups of counties, on the same scale as Figure 2. Rejection rates for other reasons did move between elections, and they moved in both groups in the same way. The gap that opened for late rejections in 2026 does not appear here. Across the three states, the estimated effect of RTO on other rejections is a decrease of 0.1 percentage points, about a tenth the size of the increase estimated for late rejections. This estimate is not statistically significant.

This comparison has limits, however, and the most important one is that exposure to RTO is not randomly assigned. RTO applies when a post office is far from its processing plant, which is overwhelmingly more likely in rural America. As a result, counties that are more and less exposed may differ in many ways besides the mail. The two groups moved together in 2022 and 2024, and the comparison assumes that they would have continued to. If, however, something else changed for rural voters between 2024 and 2026, it would appear here as an effect of RTO.

Counties are also grouped by whether their post offices are designated for RTO, not by whether the new schedule has been implemented. By the time these primaries occurred, 98% of designated offices in Washington were operating on the new schedule, 75% in California, and 31% in North Carolina. North Carolina’s lower implementation rate is likely another reason why its estimate is the least precise of the three states.

RTO designation is the right basis for comparison because it was determined based on distance from a processing plant and in advance of the 2026 primaries being studied, rather than based on anything related to how people vote. The order of implementation, by contrast, is the Postal Service’s own scheduling decision, which could respond to such factors. The consequence of this choice is that the exposed group includes some places where nothing has changed yet, which narrows the gap between the two groups of counties and, all else equal, biases the estimates toward zero.

Nearly a decade of general elections also shows little gap in late rejection rates before RTO

The 2026 primaries are not general elections, and anticipating what may happen in November requires accounting for the difference. A general election, for example, draws much higher ballot volumes, and it draws them from voters who participate less regularly. A general election also receives far more public attention, which could mean more voters return ballots well before the deadline rather than close to it. Those differences can push the effect of RTO in either direction.

Millions of ballots are returned by mail in every general election, and that number varies across election cycles. The Election Administration and Voting Survey (EAVS), which the U.S. Election Assistance Commission collects from local election offices after every federal election, records how many mail ballots each jurisdiction received and how many it rejected, with late arrival recorded as its own category. In the two most recent midterm elections, which are the cycles that most resemble the coming election in November, election officials nationwide reported 30.4 million mail ballots returned in 2018 and 36.7 million in 2022. Presidential years run higher, at 33.4 million in 2016 and 48.0 million in 2024. The pandemic election of 2020 stands apart, at 70.6 million.

Late-arrival rejections are uncommon in general elections. Between 0.2% and 0.4% of mail ballots returned were rejected for arriving after the deadline in each of the five most recent election cycles, with 2020 the exception at 0.1%. Averaged over the last two general elections, roughly 115,000 ballots nationally were rejected for arriving late.

They are also no more common in places that became newly exposed to RTO for the 2026 election. Figure 4 again divides counties into those where most residents’ nearest post office is now covered by RTO and those where most are not. In every cycle over the last decade, the two lines sit on top of each other, with the gap between them within a tenth of a percentage point. Whatever else distinguishes these places—and much likely does—their mail ballots arrived on time at the same rate.

40% of mail ballots in recent elections came from places RTO covers

How much of that mail comes from places RTO now covers follows from two numbers from each county: on average, how many mail ballots its voters returned in the 2022 and 2024 general elections, and what share of its voting-age adults live nearest to a post office RTO now covers. Multiplying the two together provides an estimate of the county’s exposed mail ballot volume. Nationally, two in five mail ballots are exposed to RTO: 16.6 million out of the average 41.8 million.¹ Importantly, this is an estimate of exposure, not of harm: it is the volume of mail voting that happens where the transportation change has added a day before those ballots enter the processing network, not a count of ballots delayed or rejected.

Exposure is not spread evenly across the country. Figure 5 maps exposed mail ballots under a fully implemented RTO per 100 voting-age adults, which combines how much of a county RTO covers with how much that county votes by mail.

If RTO has the same effect in November, national late rejections could rise by tens of thousands

Three numbers give a sense of what to expect this November: how much late rejections rose in exposed counties in the 2026 primaries, how many mail ballots come from exposed places, and how often mail ballots are rejected for arriving late in a general election. Combining them requires two assumptions. The first is that the effect of RTO estimated from the three 2026 federal primaries also holds in the general elections. The second is that it applies proportionally, doubling the rate at which ballots are rejected due to late arrivals. The primary effect nearly doubled late rejections, but from a much higher baseline: about 1% of mail ballots across California and Washington and several times that in North Carolina, against roughly 0.3% in a general election.

A doubling of late rejections in exposed places could raise the national count of late rejections by 38% to 48%. Table 2 walks through this calculation. About 0.27% of mail ballots returned are rejected for arriving late, and that rate is roughly the same in exposed and unexposed places. Applied to the 16.6 million mail ballots returned by voters from exposed places, this implies roughly 45,000 late rejections. The difference-in-differences estimate from the 2026 federal primaries suggests that RTO increases late rejections by 96% to 122%. This implies an increase of 43,000 to 55,000 ballots rejected for arriving late.

These projections should be interpreted with caution. California, North Carolina, and Washington were chosen for the RTO study because they publish county-level rejection rates with reasons for three consecutive federal primaries. They were not, however, chosen because they are a representative sample of mail voting across the country, although the 197 counties in those states account for more than one-third of mail ballots voters return nationally. Late rejection is also much less common in the general election than in the primaries, where the effect was measured. How RTO affects a general election is an open question, and November is the first test.

Conclusion

RTO added a day to the start of a ballot’s return journey at more than two-thirds of the country’s post offices. Those places account for roughly 40% of the country’s mail ballots. In the 2026 primaries, exposure to RTO roughly doubled the rate at which mail ballots were rejected for arriving late. RTO was designed around postal operations. Elections run on the same network. November is the first general election held under the new schedule.

The most reliable remedy belongs to the voter. RTO’s schedule is fixed, and the day it adds applies to any ballot that reaches a post office after the day’s dispatch has left. A voter controls when their ballot is mailed, and mailing it a few days earlier more than offsets the extra day RTO adds. USPS advises sending a completed ballot at least seven days before it must be received. The difficulty for the voter is knowing whether a given post office is one that RTO covers.

Voters also have more than usual to keep track of this year. Since last December, the Postal Service has affirmed that a postmark may no longer reliably show the day mail was sent, the U.S. Supreme Court has left the postmark as the proof that a voter met a deadline, several states have ended the grace periods that once cushioned late-arriving ballots, and a federal rule that newly requires election officials to file voter mailing data with the Postal Service when ballots are sent was issued in August and blocked by the courts.

Voters in covered places need to know about the change before they mail their ballots. Election officials can tell them—as can state and local governments, civic groups, campaigns, and notices posted where people vote. Importantly, this message can be targeted, county by county, to ensure that those voters who are most likely to be affected are also the ones receiving the messaging.

The Postal Service should also make those places easier to identify. The RTO determination is public, filed with the Postal Regulatory Commission, and USPS reports regular updates on implementation to the commission as it proceeds. Both are filed with regulators, not published for voters. USPS’s own service-standards tool tells customers how long mail is expected to take from their zip code. It does not say whether that expectation changed under RTO. A voter who does not know that something changed would have no reason to mail a ballot earlier than usual. Adding that to the tool would give them one.

Everything needed to act on this is already public. Whether it reaches voters is the difference between allowing for an extra day in transit and having a ballot arrive too late to count.