If the dozens of new Bluesky followers I've gotten in the past few weeks didn't make it clear enough, the amount of engagement with my recent Twitter and Bluesky posts has solidified in my head that Bluesky is alive and Twitter is dead
Across the whole county that's true, but in the City of Pittsburgh, things were more similar to Philly than the rest of PA. Harris (76.9%) got 2300 fewer votes than Biden (77.6%). And the number of ballots in Pgh was 1% lower in '24 than '20. This was -2.9% in Philly and +2.3% in the rest of PA
This article follows up on our piece from Election Week, where we looked at how precincts in Pennsylvania shifted their votes toward Trump: www.nbcnews.com/politics/202...
Ballot counts in Philly, Detroit, and Milwaukee all lagged considerably behind their states
The number of votes (relative to 2020) from Republican countries far out-paced the number in Democratic counties
Check out my new @nbcnews.bsky.social article (with Dylan Radley and Makenzie Kerneckel) about turnout in Philly, Detroit, and Milwaukee. Turnout lagged considerably in those cities compared to the rest of their states, contributing to Trump winning PA, MI, and WI www.nbcnews.com/politics/202...
Definitely would be interesting. SF has been using it for 15 years, and the evidence there suggests that these errors rates don't go down with time. But perhaps the story is more complicated and only time will tell
It's certainly possible, although SF's been using it since 2007 and they had really high error rates in the past couple cycles. We also saw that the overvote rate hasn't changed much in those 15 years, although we don't have cast vote records to fully extend our analysis all the way back
We hope work like her's and our's can bring empirical evidence to the policy discussions happening in states and cities that are considering adopting ranked choice voting.
Here's a link to her paper: https://journals.sagepub.com/doi/10.1177/1532673X231220640
There's a lot of unanswered questions about RCV and lots of smart people working on answers. Just last week, @dcinbox.bsky.social published an article showing that lower income New Yorkers may have been more likely to have their RCV ballots rejected due to overvoting.
The difference is that ballot exhaustion may result from a voter's sincere lack of preference for unranked candidates, whereas mistakes like overvotes or overrankings suggest the voter did not fully understand the proper way to mark their ballot.
It's also worth pointing out that ballot rejections are only part of the story of vote attempts that are not included in final tabulation. Rejection rates are dwarfed by the percent of ballots uncounted in the final round tabulation due to ballot exhaustion (which is often >10%).
But in non-RCV races that rate tends to be less than 0.1%. And in a place like SF or NYC, that 10x difference can mean thousands of additional people attempting to cast a vote for a race, but having their vote be rejected due to mis-markings.
So what's the take-away? At the end of the day, most RCV races tend to have fewer than 1% of ballots be rejected due to ballot mis-markings. Maybe that level is an acceptable downside, if RCV has other benefits.
All of these results point to a clear trade-off between giving voters the opportunity to express more nuanced electoral preferences and the mechanism for doing so (an RCV ballot) creating voter confusion that results in votes not being counted.
SF's ballot rejection rate was 4x higher in the first round of tabulation for RCV races than the rate in non-RCV races. And in RCV races that required multiple rounds of tabulation, the rejection rate was 8x higher than non-RCV.
In addition to having some of the highest error rates of any of the 4 jurisdictions in 2020 and 2022, SF's ballot rejection rates in RCV races were significantly higher than those in every other category of non-RCV races.
Given that RCV is fairly new in 3 of our jurisdictions, we also asked whether these error rates would decrease as voters became more familiar with RCV. SF provided us a useful chance to test this question, and its data suggests that the answer is a resounding "no".
This is true, even if you make the apples-to-apples comparison of non-RCV (red) to round 1 of RCV tabulation (blue). Because R1 rejections usually result from overvoting, this suggests that simply presenting a voter with an RCV ballot makes them more likely to overvote.
These errors have a huge impact on how many ballots get thrown out during tabulation. In every jurisdiction and election date we analyzed, the ballot rejection rate is significantly higher in RCV races than non-RCV ones. On average, the RCV rejection rates are 10 times higher.
We also find that rates of over-voting, which can occur on ranked and non-ranked offices, are significantly higher in ranked choice races than non-ranked choice ones *that appear on the same ballot*
We find that in a typical ranked-choice race, about 5% of marked ballots contain at least one type of ballot marking error. There is variation between jurisdictions--AK's error rate is about 2%, while the other 3 jurisdictions hover around 6 or 7%. Each dot represents an RCV race
We believe our ranked choice ballot database is the largest ever assembled. The data describe exactly how votes were "bubbled-in", allowing us to assess the types of mistakes made and whether they were included in tabulation (at least prior to ballot exhaustion).
We applied this typology to data from ~3 million ballots from 165 ranked choice races in two states (AK, ME) and two large cities (NYC, SF). These jurisdictions make up about 80% of American voters who can vote in RCV races.
"Over-ranking" is when a voter selects the same candidate for multiple rankings. In the example above, whether the voter's preference for Murkowski is counted depends on the jurisdiction's RCV law.
Whether skips are counted depends entirely on the specifics of the RCV law. In NYC and SF, the tabulators pretend the skips aren't even there. In AK and ME, counting is terminated if a voter skips multiple consecutive rankings.
"Skipping" occurs when a voter leaves a ranking blank but fills in subsequent rankings. The example here shows a "front skip"--no first-choice preference expressed, as well as an "interior skip"--no choice for ranking 3.
"Overvotes" occur when a voter selects multiple candidates for the same ranking. Overvotes are the biggest contributor to rejected ballots. The example here is a real ballot cast in the Nov '22 AK House race. The vote was uncounted because of an overvote for ranking 1.
We develop a 3-category typology of ballot-marking error. This extends election admin terminology that was developed around single-mark ballots into a more general framework for nearly any type of ranked ballot--not just instant runoff ones which are most common in the US.
