Blog

One of the more serious problems the National Popular Vote interstate compact (NPV) has with recounts is that it would be impossible to conduct a full nationwide recount if the national margin were close enough to warrant one, because recount laws vary from state to state. (By “recount” I mean both pre-certification and post-certification counts after an initial count.) Some states have automatic recounts if the margin is below a certain threshold (typically, between 0.2 and 1 percent) while others allow candidates or voters to request recounts if a threshold is met. In some states, recounts can be requested no matter the margin. 

This means that if you had a close national margin – say, three-tenths of a percent (which would be somewhere around 475,000 votes) – and even if every state were to interpret its recount laws to use the national margin to determine if a recount could be triggered or requested (highly unlikely), several would still not meet the threshold (Connecticut, Delaware, New Hampshire, New Mexico, New York, North Carolina, Rhode Island, South Dakota, and Utah) and therefore would not conduct recounts. 

I recently was reading the fifth edition of the NPV organization’s book, Every Vote Equal, and found its updated response to this problem (beginning on page 1035). After some impressive statistical gymnastics, NPV claims that not being able to conduct a nationwide recount is no big deal because recounts are unlikely—according to their math, a recount should be required just once in every 324 presidential elections, or every 1,294 years.

The statistical analysis is fine as far as the math goes (don’t worry, I’m not going to try here to explain Gaussian curves, the Central Limit Theorem, or the importance of using absolute values), but that this is “garbage in, garbage out” becomes apparent once you see that NPV’s implied threshold for determining if a recount is needed is a margin between the candidates of about fifteen-one thousandths of a percent (0.015%).*

This absurdly-low threshold is not only far lower than what states currently require, it would also have prevented recounts in races where just about everyone agrees recounts were needed.

For example, under the 0.015% standard NPV thinks is appropriate the initial Florida 2000 margin (0.03%, 1,784 votes out of 5,816,486 total) in favor of George W. Bush would not have triggered a recount (Florida’s actual standard, 0.5%, did trigger an automatic machine recount). 

And in Vermont’s state auditor race in 2006, the Republican candidate was originally certified the winner with a 0.04% margin.  A recount followed, and the outcome was reversed when the Democrat gained enough votes in the recount to eke out a narrow win. Had Vermont been relying on NPV’s threshold of when recounts are needed, the wrong candidate would have held office.

And there have been larger shifts as well, including one of roughly .22 percent (Pennsylvania Commonwealth Court race in 2021), and several other races in recent decades also saw shifts larger than NPV’s 0.015% margin.

NPV’s basic mistake here is forgetting that a recount serves two purposes. First, obviously, is to ensure that the correct candidate has been named the winner. The second purpose may be just as important: to assure the public that the correct candidate has been named the winner. 

The history of recounts shifting a greater share of votes than NPV’s 0.015% threshold shows how it fails to satisfy the first purpose, and it’s hard to imagine the public being satisfied if the top two candidates were separated by 50,000 votes, or even 500,000 votes, but no recount was possible. 

There’s just no getting around it – if NPV took effect, there would need to be some process for a recount should the national margin fall under the thresholds of 0.5 or 1 percent used by most states, as it has in three of the last seventeen presidential elections. But because each state has different laws on what triggers or permits a recount (not to mention how that recount is conducted), it will be impossible to assure the public that the correct candidate has won. 

Having NPV’s lobbyists wave Gaussian curves around just isn’t going to cut it.

* Determined by dividing NPV’s estimate of the number of votes that would need to shift in order to reverse an election (24,294) by total turnout in the 2020 presidential election (158,429,631).