As currently required as part of the computation and canvass of returns, counties must complete the statistical sample required by law (25 P.S. § 3031.17).
County boards of elections are required to conduct a statistical recount of a random sample of at least 2% of the ballots cast or 2,000 ballots, whichever number is fewer.
In 2019 and 2020, the Department of State piloted additional risk-limiting audits at the county level and across the commonwealth. One of the goals of these pilots was to test models and best practices for enhanced post-election audits that are scientifically designed to strengthen election integrity, confirm the accuracy of election outcomes, and bolster voter confidence.
About Risk-Limiting Audits
Risk-limiting audits are scientifically designed procedures that use statistical methods to confirm election outcomes and to detect possible interference. RLAs examine a random sample of paper ballots, comparing the votes on paper to the totals reported by the vote-counting machines to ensure that the reported outcome of the contest being audited is correct. These types of audits can confirm that voting systems tabulated the paper ballots accurately enough that a full hand count would produce the same outcome.
County election officials, Department of State staff, and election experts from the Brennan Center for Justice at NYU School of Law, National Vote at Home Institute, Verified Voting and VotingWorks participated in developing and implementing Pennsylvania’s RLA pilot program, which began in 2019.
"We're proud to partner with state and local election officials to help them implement the gold standard of post-election audits to protect voter confidence in the integrity of our elections." Liz Howard, senior counsel, Brennan Center for Justice at NYU School of Law.
Each county's certified voting system provides a voter-verifiable paper record of each vote cast, meets the latest standards of security and accessibility, and can be thoroughly audited.
Every voting system and paper ballot in Pennsylvania must include plain text that voters can read to verify their choices before casting their ballot, and every system has successfully completed penetration testing, access-control testing and testing to ensure that every access point, software and firmware are protected from tampering. Many other important recommendations by national security and cybersecurity experts are in place in Pennsylvania, including mandatory pre-election testing of all voting equipment.
Risk-Limiting Audit
After three years of performing pilots of risk-limiting audits (RLAs), the Department of State in September 2022
directed all Pennsylvania counties to participate in a statewide RLA for every Primary and General Election beginning with the Nov. 8, 2022, General Election.
RLAs are scientifically designed procedures that use statistical methods to confirm election outcomes and to detect possible interference. RLAs examine a random sample of paper ballots, comparing the votes on paper to the totals reported by the vote-counting machines to ensure that the reported outcome of the contest being audited is correct.
These types of audits – which are conducted after an election and before certification of those election results – can confirm that voting systems tabulated the paper ballots accurately enough that a full hand-count would produce the same outcome.
Nov. 7, 2023, Risk-Limiting Audit Report
For the 2023 general election, the race for Pennsylvania Superior Court Judge was randomly selected for review. The results of the audited sample compared to the initial reported results
confirmed that the outcome of the election was accurate.
To conduct this specific audit:
- Counties created ballot manifests, which are spreadsheets showing the number of ballots counted for the selected contest and details on how the ballots are organized and stored. This collection of information allows batches of ballots to be identified, retrieved, and examined if selected for the audit.
- After counties uploaded their ballot manifests to an open-source audit software tool called Arlo, Department of State staff generated a random 20-digit seed number during a livestreamed dice roll.
- That seed number was then entered into the audit software, which selected the random list of ballot batches for certain counties to retrieve. In total, 105 batches of ballots in 33 counties were randomly chosen to be audited.
- Officials in the selected counties retrieved the randomly selected batches of ballots and verified the ballots cast for the Pennsylvania Superior Court Judge race.
- The audit software system then tallied and analyzed the results.
The audit identified 24 discrepancies among the 201,715 total votes they reviewed, which is a margin of 0.01% -- or about one-100th of 1% of all ballots audited. Such discrepancies are typically the result of human error when manually tabulating audit results or stray and unclear marks on a ballot, which can lead to subjective decisions about a voter’s intent.
Download detailed results of the 2023 General Election risk-limiting audit: 2023 General Election RLA (xlsx)
May 16, 2023, Risk-Limiting Audit Report
For the second statewide RLA, the Democratic race for Pennsylvania Supreme Court was randomly selected for review. The results of the audited sample matched the initial reported results, decisively confirming the accuracy of the outcome of the May primary.
To conduct this specific audit:
- Sixty-five counties created ballot manifests, which are spreadsheets showing the number of ballots counted for the selected contest and details on how the ballots are organized and stored. This collection of information allows batches of ballots to be identified, retrieved, and examined if selected for the audit.
- After counties uploaded their ballot manifests to an
open-source audit software tool called
Arlo, Department of State staff generated a random 20-digit seed number during a
livestreamed dice roll.
- That seed number was then entered into the audit software, which selected a random list of ballot batches for certain counties to retrieve. In total, 25 batches from 14 counties – a total of 2,637 ballots – were randomly chosen to be audited.
- The 14 selected counties retrieved the randomly selected batches of ballots and verified the vote cast for the Democratic race for Pennsylvania Supreme Court on each ballot.
- The audit software system then tallied and analyzed the results.
No variations in vote totals were found during the execution of this RLA.