SECOND MEASUREMENT SM-028
Pennsylvania law gives counties 48 hours to mail a ballot once a mail-in or absentee application is approved. In the 2024 general election, 227,378 of 702,985 went out later
The claim: Pennsylvania mail ballots: the law gives counties 48 hours to mail a ballot once an application is approved; in the 2024 general election 227,378 of 702,985 went out later
What we found: 227,378 of 702,985 mail-in and absentee ballots approved after a county’s main mailing day were recorded as mailed three or more days after approval (32.3%); 17,863 more than a week. 64 of the 65 counties counted had late ballots.
Status: Published 10 Nov; checked against the exhibits before publication; question to the Pennsylvania Department of State not yet sent
In Pennsylvania any voter can vote by mail. The county election board approves the application and mails the ballot. Before the first ballots exist, approved applications wait. Once the county has started mailing, the Election Code gives it 48 hours for each new approval, for mail-in and absentee ballots alike. The Department of State publishes a record of 2024 mail ballot requests, with the date the county approved it and the date the county recorded the ballot as mailed. We counted mail-in and civilian absentee applications approved after the county’s main mailing day and by 29 October 2024, the last day to apply. 227,378 of 702,985 ballots (32.3%) were recorded as mailed three or more calendar days after approval, which is always more than 48 hours. Most were three or four days late; 17,863 took more than a week. Some recorded dates trail the real mailing, and we show how many. Ballots mailed late came back less often, but our test could not separate the delay from the shorter time left before Election Day.
- The rule, Election Code §1305-D (25 P.S. §3150.15): “As additional applications are received and approved, the board shall deliver or mail official mail-in ballots to the additional electors within 48 hours.” §1305(b)(1) (25 P.S. §3146.5) sets the same 48 hours for absentee ballots.1 The Department of State’s guidance to counties for 2024 says the same: “within forty-eight (48) hours of receipt of approved applications.”2
- The data: the Department’s 2024 General Election Mail Ballot Requests on data.pa.gov, final version pulled 24 January 2025, 2,404,091 rows.3 Dates are days, not times, so a ballot counts as late only when it was mailed three or more calendar days after approval. Two days may or may not exceed 48 hours; we count it on time.
- Mail-in and absentee ballots approved after the county’s main mailing day, through 29 October 2024: 227,378 of 702,985 late (32.3%). Median gap two days; 127,522 of the late ballots were three or four days. 17,863 more than seven days. Mail-in alone: 223,792 of 693,209; absentee alone: 3,586 of 9,776.
- Largest counts: Allegheny 44,850 of 76,650 (58.5%), Philadelphia 24,780 of 91,668 (27.0%; 3,826 more than a week), Montgomery 22,570 of 68,465, Delaware 16,957 of 29,842 (56.8%; 2,699 more than a week), Berks 15,529 of 22,403 (69.3%).
- Six counties with at least 1,000 ballots in the count recorded under 1% late, among them Lawrence (0.3% of 4,127) and Lackawanna (0.8% of 10,591). The data cannot tell whether they mailed faster or recorded an earlier date; see what we can’t be sure of.
- Some of the late dates may be late entries, not late ballots. 2,330 late ballots came back on the day they were recorded as mailed, impossible by mail unless the voter collected the ballot at the office; 3,455 more came back the next day. Treating all 5,785 as on time leaves 221,593 late (31.5%).
- For applications approved 15–21 October, 89.1% of ballots mailed on time were returned by Election Day, against 77.1% of those mailed more than a week late. For approvals in the last week before the deadline, the return rate is not ordered by the length of the delay. That is a comparison, not a measured cost: see what the delay cost.
Disclosure: Markovian Protocol holds no position in any campaign or party, was paid by no one for this work, and did not publish or share these numbers before the 3 November 2026 election. The paper covers 2024 only. How we work.
The rule, in their words
| Where | What it says |
|---|---|
| Election Code §1305-D, 25 P.S. §3150.15 (mail-in ballots) | “As additional applications are received and approved, the board shall deliver or mail official mail-in ballots to the additional electors within 48 hours.” |
| Election Code §1305(b)(1), 25 P.S. §3146.5 (absentee ballots) | “As additional applications are received and approved after the time that the county board of elections begins delivering or mailing official absentee and mail-in ballots, the board shall deliver or mail official absentee ballots to such additional electors within forty-eight hours.” |
| Department of State, Guidance Concerning Civilian Absentee and Mail-in Ballot Procedures, version 3.1, 10 September 2024 | “Once a county begins delivering their balloting materials, as additional applications are received and approved, the county must deliver or mail balloting materials to such additional voters within forty-eight (48) hours of receipt of approved applications.” |
§1305(b)(2) also lets a voter apply in person at the county office and take the ballot away; the guidance calls these over-the-counter requests. Both statute texts and the guidance are saved in sm-028/exhibits/.
What we counted
- Clock start: “Application Approved Date”, defined by the Department as “the date the application was approved by the county.”
- Clock stop: “Ballot Mailed Date”: “the date the county queued a ballot label to mail the ballot materials to the voter. This column may be updated when the county indicates they have mailed the ballot. If the county has not mailed the ballot or does not indicate the ballot was mailed, the date in which the ballot label was queued will remain. This data may be updated based on county data entry, not necessarily when the ballot is scanned by USPS. There can be delays between when the ballot is mailed and the entry of the Ballot Mailed Date.” The error runs both ways. A queued label that was never updated makes a ballot look earlier than it was mailed. A late entry makes it look later. The second kind shows up when a ballot comes back within a day of its recorded mailing; we count those below.
- When the 48 hours start to run: the rule applies once a county has begun mailing. For each county we took its main mailing day, the date on which it recorded the most ballots mailed, counting mail-in, absentee, emergency and alternative ballots (18 September to 21 October 2024; over-the-counter ballots count, because the rule says “deliver or mail”), and counted only applications approved after it. Applications approved earlier, which waited for ballots to be printed, are left out.
- Which ballots: mail-in application types (MAILIN, OLMAILV, OLMAILNV, PMI) and civilian absentee types (CIV, REG, PER, OLREGV, OLREGNV) approved after that day and on or before 29 October 2024, the last day to apply (the Tuesday before Election Day).2 Not counted: military, overseas, veteran, federal-only and remote-civilian ballots, which the Department’s data describes as “UOCAVA (Uniformed and Overseas Citizens Absentee Voting Act) or UMOVA (Uniform Military and Overseas Voters Act) ballots”; the guidance says overseas and military voters “are subject to separate procedures”. §1305(a) has its own timetable for the voters it covers (mailing must begin no later than 45 days before the election, or 70 for voters in remote or isolated areas) and also gives 48 hours after approval; we did not map each of these ballot types onto it and left them out of scope. Also not counted: emergency (C) and alternative (ALT) ballots, whose rules we did not check.
- Left out: two counties whose mailed dates are not mailing dates. In Franklin, 2,682 of the 9,313 ballots that would be counted are recorded as returned before they were mailed. In Indiana, 5,011 of 9,600 mail-in ballots are dated as mailed before 1 September 2024; its busiest “mailing” day is 26 June. No other county has more than 405 such dates. In the other 65 counties, rows whose mailed date cannot be right: 1,522 recorded as returned before they were mailed, and 206 with a mailed date after the data’s final pull (Bucks rows dated 25 October 2025, Susquehanna rows dated October 2027). Putting those rows back changes the late share from 32.345% to 32.349%. Also left out: 6,916 approved applications with no mailed date at all (5,412 of them recorded as returned), 93 with a mailed date before approval, and 8 with no approval date.
County by county
| County | Ballots counted | Late | Share | More than 7 days | Median days | Main mailing day |
|---|---|---|---|---|---|---|
| Allegheny | 76,650 | 44,850 | 58.5% | 2,315 | 3 | 2024-09-26 |
| Philadelphia | 91,668 | 24,780 | 27.0% | 3,826 | 2 | 2024-09-24 |
| Montgomery | 68,465 | 22,570 | 33.0% | 794 | 2 | 2024-09-24 |
| Delaware | 29,842 | 16,957 | 56.8% | 2,699 | 4 | 2024-10-02 |
| Berks | 22,403 | 15,529 | 69.3% | 83 | 3 | 2024-10-04 |
| Lancaster | 26,093 | 14,939 | 57.3% | 125 | 3 | 2024-10-01 |
| York | 18,802 | 12,518 | 66.6% | 814 | 5 | 2024-10-10 |
| Bucks | 45,648 | 10,058 | 22.0% | 402 | 2 | 2024-10-02 |
| Dauphin | 16,441 | 8,713 | 53.0% | 127 | 3 | 2024-09-27 |
| Monroe | 11,064 | 7,909 | 71.5% | 1,297 | 4 | 2024-10-01 |
| Westmoreland | 16,662 | 6,847 | 41.1% | 101 | 2 | 2024-10-11 |
| Northampton | 21,140 | 4,790 | 22.7% | 84 | 1 | 2024-10-01 |
| Erie | 11,007 | 4,151 | 37.7% | 1,231 | 1 | 2024-10-09 |
| Lehigh | 16,434 | 3,296 | 20.1% | 51 | 0 | 2024-10-07 |
| Beaver | 8,119 | 3,262 | 40.2% | 451 | 2 | 2024-10-02 |
| Fayette | 4,434 | 2,996 | 67.6% | 582 | 4 | 2024-10-04 |
| Luzerne | 15,029 | 2,709 | 18.0% | 150 | 2 | 2024-10-08 |
| Butler | 12,743 | 2,270 | 17.8% | 11 | 0 | 2024-10-01 |
| Lebanon | 7,430 | 1,999 | 26.9% | 140 | 1 | 2024-10-04 |
| Carbon | 2,768 | 1,575 | 56.9% | 447 | 3 | 2024-10-01 |
Mail-in and absentee ballots together. The 20 counties with the most late ballots. All 65 counties counted: summary_2024.json. Late = mailed three or more calendar days after approval. Main mailing day = the county’s busiest mailing date, after which the 48-hour clock is counted.
What the delay cost
Pennsylvania counts a mail ballot only if the county receives it by 8 p.m. on Election Day, which in 2024 was 5 November.2 A ballot that leaves the county late has less time to reach the voter and come back.
| Approved (2024) | Mailed after approval | Ballots | Returned by Election Day |
|---|---|---|---|
| 15–21 Oct | 0–2 days (on time) | 90,484 | 89.1% |
| 15–21 Oct | 3–4 days | 28,790 | 82.0% |
| 15–21 Oct | 5–7 days | 26,697 | 79.9% |
| 15–21 Oct | 8 days or more | 4,437 | 77.1% |
| 22–29 Oct | 0–2 days (on time) | 210,135 | 78.3% |
| 22–29 Oct | 3–4 days | 46,453 | 66.2% |
| 22–29 Oct | 5–7 days | 10,226 | 73.2% |
| 22–29 Oct | 8 days or more | 1,234 | 75.7% |
Ballots counted in the headline, by approval dates, leaving out ballots approved, recorded mailed and returned on the same day (15,758 and 43,974), the pattern of a ballot handed over at the county office rather than mailed. “Returned” is the county’s “Ballot Returned Date”, on or before 5 November 2024. In the second week, ballots mailed three or four days late came back less often than those mailed more than a week late.
In most counties the pattern holds. Using each county’s own on-time return rate as the benchmark (counties with at least 50 on-time and 50 late ballots in the week), late ballots came back less often in 22 of 31 counties for approvals of 15–21 October and in 19 of 28 for 22–29 October. Had late ballots in the 31 and 28 counties come back at their county’s on-time rate, about 8,400 more would have arrived by Election Day (5,534 and 2,887).
This does not show that lateness caused those ballots not to come back. Voters whose ballots were mailed late may differ from the rest in other ways. To test this we compared voters whose applications were approved from 1 to 29 October on a Thursday or Friday, when 48 hours runs into the weekend, with voters approved on the Monday or Tuesday of the same county and week. Averaged across county-weeks, weighted by size, Thursday and Friday approvals were mailed late 8.5 points more often and came back by Election Day 2.1 points less often. But those voters also had fewer days before the election. Compared instead with the next week’s Monday and Tuesday approvals, who had less time, Thursday and Friday approvals were mailed late 21.9 points more often and came back by Election Day 2.5 points more often. Both results fit time left alone; neither separates the delay. Ballots handed over in person are left out of this test too, and each county-week needs at least 30 voters in each group.
11 ballots in the count are recorded as mailed after Election Day, when no return could be counted.
How this ends
The Department of State can say whether counties must update the Ballot Mailed Date when they mail a ballot, which tells readers whether the dates mean mailing or labelling. Counties can publish their own mailing logs. The Department publishes the same dataset for each general election; when the final 2026 file is out we will run the same count and add it here.
Our questions
The Department of State’s description of the 2024 file says the Ballot Mailed Date is “the date the county queued a ballot label”, that it “may be updated when the county indicates they have mailed the ballot”, and that “there can be delays between when the ballot is mailed and the entry”. One question decides how the 32.3% should be read:
- When a county mails a ballot, is updating the Ballot Mailed Date required, or optional?
We will send it to the Department of State on publication day and print the answer as written.
What we can’t be sure of
- Recorded date, not postmark. The mailed date is the date the county queued the label, sometimes updated to the mailing date. A county that mailed on time but recorded a later date would look late. We found this in Franklin County and left it out. Elsewhere, 0.2% of rows show a ballot returned before it was mailed, and we left those out too. Shorter delays show up when a ballot returns within a day of its recorded mailing: 5,785 of the late ballots (2,330 the same day, which is impossible by mail; the rest the next day, which is possible if the voter took the ballot back in person), including 2,000 of the 14,707 returned ballots that were more than a week late. Treating all 5,785 as on time leaves 221,593 late. Delays on ballots that never came back cannot be seen. The opposite error also exists: Indiana recorded label dates months before ballots existed, and we left it out. A county that records the label date at approval and mails days later would look on time; the counties under 1% late may include some.
- Days, not hours. A ballot mailed two calendar days after approval may have taken more than 48 hours. We count it on time.
- When mailing began. We start each county’s clock on its busiest mailing day. A county that started mailing earlier would have more late ballots than we count.
- Ballots handed over in person. 64,982 ballots in the count were approved, recorded mailed and returned on the same day. That is the pattern of a ballot handed over at the county office. They were delivered within 48 hours and count as on time. Leaving them out would raise the late share to 35.6%.
- Weekends. The rule says 48 hours and makes no exception for weekends; neither do we.
- The cost. See above: an association, with a test that could not separate cause.
Run it yourself
curl -sL -o pa_2024_mail_ballot_requests.csv 'https://data.pa.gov/api/views/3q5t-ddp8/rows.csv?accessType=DOWNLOAD' # about 712 MB shasum -a 256 pa_2024_mail_ballot_requests.csv # ours: d45c7ad19c82848dae1f853f68c2af48c38c5a97e66a06bef9c6289bafba9dac python3 -I sm028.py pa_2024_mail_ballot_requests.csv # writes summary_2024.json, consequence_2024.json, weekday_2024.json
Standard-library Python 3. The Department’s metadata shows the dataset updated on 14 April 2025, after the pull date it states, so a fresh download may not match our hash. Scripts, outputs and checksums: sm-028/.
References
- Pennsylvania Election Code §1305-D, 25 P.S. §3150.15, and §1305, 25 P.S. §3146.5, palegis.us, read 28 October 2026. Saved copies: §1305-D, §1305.
- Pennsylvania Department of State, Guidance Concerning Civilian Absentee and Mail-in Ballot Procedures, version 3.1, 10 September 2024, pa.gov. Saved copy dos_guidance_civilian_mail_ballot_2024-09-10_v3.1.pdf. The September 2026 guidance keeps the same sentence: dos_guidance_civilian_mail_ballot_2026-09.pdf.
- Pennsylvania Department of State, 2024 General Election Mail Ballot Requests, data.pa.gov/d/3q5t-ddp8, final version pulled 24 January 2025, downloaded 28 October 2026. Column definitions saved as dataset_3q5t-ddp8_metadata_2026-10-28.json.