[mikeash]

It's trivial to find if you look for it. It's difficult to do with 100% accuracy on election day, but it's easy to find large-scale instances of it afterwards. Consider the various scenarios for the sort of voter fraud that ID requirements would prevent:

1. A person votes using the name of another registered voter. This will be detected any time the registered voter in question also casts a vote. Typically, the second person voting with the same registration will immediately be flagged and told they can't vote because they already voted. Worst case, you find it afterwards by seeing that they "voted twice." This can also be detected in cases where the registered voter didn't cast a vote by asking them whether they actually voted, and seeing who says "no" when the records say "yes."

2. A person votes using the name of a stale registration for a voter who moved away. This can be detected by seeing if the person still actually lives in the area.

3. A person votes using the name of a dead person. Detected by comparing with death records.

4. A non-citizen registers and votes using their own name. You can see who voted and check to see if they really are citizens.

Despite this, there is no evidence of any large-scale voter fraud. Known instances number in the single or at most double digits per year, nationwide. If it's happening, then it must be because nobody has ever gone looking for it. Note that the necessary evidence is all public information, so it's not just a matter of governments not looking for what they don't want to see. You'd have to propose that no university political science research group, no think tank, no public policy center, no lobbyist group, and no political party has ever gone looking for it either. And even if somehow that were true, that voter fraud is a major issue that nobody has ever gone searching for, then the best thing you can say is that these onerous and often discriminatory laws are being passed on the basis of no data, when data could be readily obtained!

[kevingadd]

I apologize for being inaccurate, your final point is the argument I was trying to make. There are examples, but they are double digits. Effectively a rounding error.

[andrewla]

I'm not so sure that gathering this data is as easy as you're making it out to be.

> 1. A person votes using the name of another registered voter

I'm not aware of any comprehensive analysis of how common this is. Ballots are frequently rejected or require provisional ballot (I had to fill out a provisional ballot once in NYC because someone had voted my ballot; no fraud in this case (probably) since the signature in my box was one of the neighboring lines in the book) but I can't find anything definitive saying how often this particular event happens.

> 2. A person votes using the name of a stale registration for a voter who moved away. This can be detected by seeing if the person still actually lives in the area.

This analysis I think is never done; analyses of voter registration routinely show huge numbers of voters who are no longer at the registered address. Once again, no proof here that anything is problematic with this, but I'm not aware of any comprehensive analyses of the scope of the problem (that is, some notion of how often these voted or had a second registration that voted, effectively voting twice. This could theoretically be targeted since voter rolls are semi-public-ish.

> 3. A person votes using the name of a dead person. Detected by comparing with death records.

This is harder than it sounds because people often share names with deceased people, and even addresses for familiar relations. Death records are not always comprehensive, especially if the individual has been deceased for a while. I've seen a couple of one-off analyses on these [1] but nothing that makes me confident that we have a grasp on the scope of the issue.

> 4. A non-citizen registers and votes using their own name

I'm not sure how you'd measure this one at all - in some ways this is the least significant because there are plenty of circumstances where aliens are permitted to vote in local, municipal, or state elections.

We have plenty of laws around preventing electioneering, voter intimidation, vote selling, etc., a voter id law does not seem substantially worse than these for identifying specific voting problems that are difficult to otherwise detect. Motor voter laws ensure that almost every licensed driver is a registered voter. I've seen estimates as low as .3-.6% of registered voters lack a photo id (in states that have instituted an affidavit-based exception to voter id laws); this seems like a small enough number to make me comfortable with voter id laws.

[1] https://chicago.cbslocal.com/2016/10/27/2-investigators-chic...

[mikeash]

I'm sure it is as easy as I'm making it out to be. I don't mean that it's trivial in the sense that some random person like you or I could go out and figure this out for ourselves. But a research team with decent resources absolutely could. Pick 1,000 voter registrations at random. Check each one for all the things I listed. Come up with a good estimate for how pervasive voter fraud actually is. How much do you suppose such a study would cost?

I can't help but notice that you completely ignored my last paragraph. Do you really believe that nobody has ever attempted what I describe above? And if so, do you really believe it's wise to be passing these laws without doing the research first?

[andrewla]

If you selected a thousand voters at random from Texas, there's a good chance you would not find a single one that doesn't have ID, since only 16,000 of the 9 million voters didn't have ID. I think that's pretty compelling evidence that voter ID laws don't result in widespread voter disenfranchisement.

If you're aware of any research, I'd love to see it. I've seen lots of research into the rate of convictions and prosecutions for detected voter fraud, but never anything as comprehensive as what you are suggesting. If had I had to guess, selecting 1,000 voters from the rolls would result in 900 solid "yes I voted", and 100 failures to validate, everything from not answering the phone to refusing to participate to incorrect records to faulty memory to they've moved away since the election, thus making the resultant study kind of worthless because the error bars are so large.

Trying to track a cohort with a good response rate is hard in any sort of study, but if your cohort is not selected by your team based on initial response (where you have some prior belief that they are not utterly opposed to participation) then I suspect that the response rates are so terrible as to make the research useless.

(As a note, my failure to reply to your last paragraph is not through any desire to avoid the question, but just because my post had already gotten too long)

[mikeash]

Why would you give up and say "failure to validate" just because someone didn't answer the phone or had moved away?

I'm proposing to actually put some real effort into this. If they no longer live in the area, find out when they moved. If they don't answer the phone, track them down. If they're dead, find out when and where they died. For the difficult cases, get some boots on the ground and figure it out. Let's take your numbers and say that 900 of the cases are easy, and 100 are tough. If you dedicate $10 million to the study, then you can spend almost $100,000 to figure out each of the tough cases.

This is well within the resources of a university research team or a lobbying group. If a major political party is convinced that voter fraud is a big problem and needs to be addressed, they could easily front the money needed to come to a definitive conclusion. An electoral commission could do it on a non-partisan basis.

I'm not aware of any such research either. If it's been done, the result has been too boring to report on. If it hasn't been done, then my question remains: why do you want to enact such a major law without doing the research first?