Voter ID DATA!

--by Sebastian

I found an interesting dataset from the bipartisan Election Assistance Commision.  So far as I can tell from the murky backstory, this report was not adopted by the Commission because neither side liked parts of its conclusions (there are hints of the Administration not being thrilled either).  And I can see why, there are all sorts of things for both sides of the standard story to hate.   Voter ID Report (warning big PDF file).

I learned lots of things I didn't know.  For example, were you aware that 10 states let you get by with just stating your name to vote?  I didn't either! 

I won't try to oversummarize, but there is quite a bit of information.  Essentially they tried to analyze how different voter qualification requirements change voting participation.  And there are soundbites for everyone! 

All my discussion will be from appendix C which most directly addresses the data and which forms the backbone of the report.  Appendix C begins on page 68 of the PDF.  Page number references will be from the PDF not listed page numbers. 

First the table showing mean turnout using the aggregate county level data for each state.  (footnotes to indicate some concern that the methodology--which uses voting age population estimates--fails to factor in adults who are ineligible to vote). 

Table 1 – Variation in 2004 State Turnout Based on Voter Identification Requirements

Maximum Requirement           Minimum Requirement

State Name 64.2 %                 State Name 63.0 %

Sign Name 61.1 %                   Sign Name 60.8 % 14

Match Signature 60.9 %           Match Signature 61.7 %

Provide Non-Photo ID59.3 %     Provide Non-PhotoID59.0 %

Provide Photo ID 58.1 %           Swear Affidavit 60.1 %

Average Turnout for All States 60.9 %

This appears to show a massive 3% drop from just being forced to sign your signature as a form of ID and an additional 3% drop from being forced to provide a photo ID.  Photo ID advocates probably aren't happy with this chart. 

But of course there are complications (p.24):

Voter identification requirements are just one factor that might affect voter turnout.  Multivariate models that take into account other predictors of turnout can paint a more complete picture of the relationship between voter identification requirements and turnout. This analysis estimated the effects of voter identification requirements in multivariate models that also took into account the electoral context in 2004 and demographic characteristics of the population in each county. While the model takes account of several important variables, statistical models do not capture all the messiness of the real world. It is a simplification of a complex reality, and its results should be treated with appropriate caution. The model also took into account such variables as:

• Was the county in a presidential battleground state?

• Was the county was in a state with a competitive race for governor and/or the U.S.

Senate?

• Percentage of the voting-age population in each county that was Hispanic or African-

American 22

• Percentage of county residents age 65 and older

• Percentage of county residents below the poverty line

Another contextual factor to consider is voter registration requirements, such as the deadline for registration. As states set the deadline farther away from Election Day, the task of remembering to register to vote becomes more challenging. Thus our model takes into account the number of days between each state’s registration deadline and the election. The dependent variable in each model using the aggregate data was voter turnout at the county level, with turnout calculated as the percentage of the citizen voting-age population that voted in the 2004 election. The results of this modeling suggest that the stricter voter identification requirements of matching one's signature to a signature on file with election authorities or presenting a nonphoto ID are associated with lower turnout compared to turnout in states that required voters to simply state their name, holding constant the electoral context and demographic variables. Contextual factors, such as whether the county was in a battleground state or whether that state had a competitive race for governor and/or U.S. Senate, were associated with increased voter turnout. The time between the closing date for registration and the election was correlated with a slight negative effect on turnout. As the percentage of Hispanics in the county’s population increased, turnout declined. The percentage of senior citizens in the county and household median income were associated with higher turnout. The percentage of African-Americans in the county did not have a significant effect in the model. The percentage of senior citizens in the county and household median income showed a positive correlation with turnout. In this aggregate model, the percentage of African-Americans in the county was not associated with a significant difference in turnout.

This leads to another table (p84):

Table 4. Predicted probability of voter turnout – full model

Maximum requirement           Minimum requirement

State name 0.917                  0.915

Sign name 0.899                    0.902

Match signature(N.S.)            (N.S.)

Non-photo ID 0.890                0.890

Photo ID 0.888                        ---

Affidavit    ----                      0.875

Total difference from “state name” to “photo identification” or “affidavit” 0.029 0.040 N 54,973 Figures represent the predicted probability of registered voters saying they voted as the identification requirement varies stating one’s name to providing photo identification or an affidavit, with all other variables held constant. N.S. = nonsignificant coefficient in the probit model.

I find this table interesting because the politically interesting comparison is NOT the difference between "state name" and "photo identification".  What is interesting to me is that the move from "Sign Name" to "Photo ID" is a very small step in turnout and that it is far less than the step from "State Name" to "Sign Name".  Also interesting is the fact that non-photo ID has almost as strong an effect as a photo ID. 

Other interesting findings is that ethnicity can effect the results, but not as predicted.  Controlling for African Americans did almost nothing to the fit.  On page 76 we learn an astonishing to conventional wisdom:  "Voter identification requirements also influenced turnout among Black voters, but to a lesser extent relative to white voters."  Also, "Of the maximum voter identification requirements, only the non-photo identification requirement reduced turnout compared to turnout in states that required voters to state their names. 

In a statistic most likely to cause a huge fight about what it means, "Hispanic voters were 10 percent less likely to vote in non-photo identification states compared to states where voters only had to give their name."  Some will argue that this suggests that a huge number of Hispanic voters are ineligible, while others will argue that having any documentation requirement whatsoever seems to discriminate against Hispanic voters.  The data doesn't resolve the question.

The report accurately concludes:  "The data examined in this project could not capture the dynamics of how identification requirements might lower turnout.  If these requirements dampen turnout, is it because individuals are aware of the requirements and stay away because they cannot or do not want to meet the requirements."  But we now now the magnitude of the question.  Forcing people to sign their names reduces turnout by about 2%.  Photo ID reduces it an additional 1%.

That is a greater dampening effect than I would have guessed though much less than I had heard touted. 

We also learned that the racial effect is almost nil with respect to African-Americans but is huge with respect to Hispanics.  That isn't the conventional wisdom at all, and is interesting fodder for both sides of the debate. 

The most interesting question for me is, what is the marginal effect on Democrats and Republicans?

Another interesting question would be "does this persist over time?" or "do people get used to a set of rules and adapt?"

And big chicken me, I'm off to a volleyball tournament tomorrow at 4 in the morning (here I come Denver) so I won't be participating in the discussion except for later tonight.      

Comments

[Sven]

I believe this is the report that was altered without consent of the original authors (who've been prohibited from speaking about the discrepancies). The unaltered draft report is here (pdf).

[hilzoy]

Sven: I don't think it is the same report, actually. Different topic, different authors.

[Sebastian Holsclaw]

No, that is the other report. The one I linked to is on Voter ID. That is on Fraud and Supression. I'll be looking at that when I get back.

[hilzoy]

On the other hand, there are hints of a story on this one as well.

From the same blog, a possible methodology worry (since the Rutgers study appears, at least in its individual data, to rely on self-reports of voting):

"Utilizing studies which validate voter turnout, previous researchers have been able to identify a strong tendency for individuals to report voting when they in fact did not..."

[hilzoy]

Cruising around some law blogs, I find a decent comment by one of the researchers. Interesting bit:

"It's important to remember that this research relies on 2004 data. That was before strict photo ID requirements were passed in Georgia, Indiana, and Missouri. Georgia's and Missouri's were both enjoined by state courts, but Indiana's took effect in the 2006 election season. The USA Today reports that Indiana's voter turnout increased in 2% of the 2006 election compared to the last mid-term election (2002). It would be wrong to conclude from this statistic that Indiana's ID law didn't affect turnout, since there are lots of other factors that can affect turnout. The real trick in this research is to try to isolate the effect of voter ID laws by controlling for those other factors, which is what the Eagleton researchers have attempted to do."

Also: on the EAC, and an article on the general topic (haven't done more than skim the first few pages and decide that I want to read the rest.)

[Gary Farber]

"And big chicken me, I'm off to a volleyball tournament tomorrow at 4 in the morning (here I come Denver)"

If you feel like driving out to Boulder, do please let me know. It's only about a half hour or so drive (much longer by bus).

[Sebastian Holsclaw]

Sorry, won't have a car and not much time for traveling I'm afraid. Anyone who wants to visit can come to the Denver Convention center. I'll be there starting at 2:30 pm on Friday and be there all day Saturday and (hopefully) Sunday.

[Sebastian]

Ohh, and I'm on BB team San Diego Del Sol. Pool #2.

[Gary Farber]

Unfortunately, while at the moment I'm as mobile as I get these days (meaning: until I lose some weight; I'm grossly over-weight these days, to a point that severely hobbles my getting around, though at least I'm still significantly under 270), which means I can walk around Boulder just fine (up to a certain number of blocks), and take the bus, getting into Denver is about two and a half hours each way, on several buses, which makes it a fairly big, tedious, pain for me. Also, the Boulder Creek Festival (just a few blocks from me) is this weekend, and I'm looking forward to it: lots of free bands, and people-watching, and odd food, and the like.

But I'll think about it. I'd get to watch you play vollyball? (Free of charge?) That might make me really regret not having a digital camera....

(And, naturally, that I'm mobile in the last few weeks doesn't guarantee how I'll be tomorrow, even if only in distractions such as, say, toothache, which is acting up a lot again at present.)

It's been rainy and cool all week, by the way, but, to be sure, weather is incredibly changeable around here, so it might be cool, or it might be warm, or more likely it will be both.

Oh, and avoid dead squirrels: lots of bubonic plague. A monkey at the Zoo died of it this weekend.

[Sebastian Holsclaw]

I'll avoid the squirrels. And do come over if your able. Friday or Saturday would be best. Sunday is only 'til we lose, and who knows when that will be. (I'm hoping to play all day of course). And getting a chance to see the big black drag queen, Denise, play college-level volleyball is really something not to be missed (best chance on Saturday).

[Gary Farber]

At risk of stating the obvious, do factor into your exertions that you'll be getting less oxygen here than you're used to. People sometimes forget that. (And that UV cuts through a lot less here, as well, so keep that also in mind if it does turn sunny again.)

Hey, I got an identical comment [with nouns changed] from my friends at ebonyfriends on a post at my blog yesterday! They're so friendly! And they love to leave their URL available!

[Sebastian Holsclaw]

I have played in tournaments in Denver before. The oxygen issue is noticeable! I suspect the 3rd day will be won entirely on stamina.

[Hartmut]

At last a concrete base for discussion.
That's one of the things I like about ObWi, even "hard conservatives" are willing and able to present something solid instead of mere talking points here.

[aimai]

I believe that in my ward you simply state your name and are crossed off a list of residents by house. They either already know who you are or the assumption is that everyone votes and the chances of someone walking in and voting under your name are slim. I don't think we show any ID at all. Its a very stable neighborhood with residents who've been voting in the same ward and the same room for 50 years.

aimai

[Dantheman]

Thanks for bringing some data to the table, Sebastian.

[Matt Weiner]

What is interesting to me is that the move from "Sign Name" to "Photo ID" is a very small step in turnout and that it is far less than the step from "State Name" to "Sign Name".

It's early in the morning so I may be getting this wrong, but that's what I get from Table 4. In the first column the distance from Sign Name to Photo ID is 0.011 and the distance from State Name to Sign Name is 0.018; in the second column the distance from non-photo ID to Sign Name is 0.012 and the distance from Sign Name to State Name is 0.013.

[Matt Weiner]

that's not what I get from Table 4

And I forgot to say, thanks for posting this data.

[John Doe]

Forcing people to sign their names reduces turnout by about 2%. Photo ID reduces it an additional 1%.

Well, then, the implication is obvious, since we all know that illicit voting is the least likely explanation for anything: Making people sign their names is twice as much of an "awful inconvenience" and "bureaucratic difficulty" as an ID requirement.

By the way, the supposed "methodology worry" raised by hilzoy is irrelevant.
Over-reporting has no effect unless it's more likely to occur in some places rather than others. If the over-reporting is equally likely to occur in any state, then that would just mean that the real voter turnout in name-signing states is 53.1% (rather than 61.1%) and in voter-ID states is 50.1% (rather than 58.1%). The margin of difference between the two types of states would remain the same. Thus, unless you have additional evidence on differential rates of over-reporting, there is NO "methodology worry."

On page 76 we learn an astonishing [what?] to conventional wisdom: "Voter identification requirements also influenced turnout among Black voters, but to a lesser extent relative to white voters."

If this finding is true, it should be spread far and wide. Of course, if an ID requirement hurts voters who are more likely to be Republicans, the purported objections to ID requirements are going to fade away from the current crowd and be picked up by Republicans.

[OkieByAccident]

Like aimai, our rural Okla precinct just asks my name and crosses me off a computer-printed list. I've never been asked to show any kind of ID. We also have what imho is the best system - optical scan. This combines the efficiency of computer tabulation with the crucial existence of actual ballots which can be recounted by hand if necessary.

Sebastian, good luck in Denver! I played one year of B in the mid 90's and had the fun of winning a regional tournament. I scrimmaged with a BB team for a year but never competed.

[bernard Yomtov]

I'm a little puzzled by the methodology, which tey don't explain very thoroughly. If I understand the results cited by Sebastian, they got the numbers by using county level turnout as the depnedent varaible in a regression that uses ID requirements and several other things as independent variables.

If that's right, it seems to me that turnout in a large urban county will have the same weight as turnout in a thinly populated rural one. Unless I'm missing something, that looks like a serious problem.

[Gromit]

John Doe: Over-reporting has no effect unless it's more likely to occur in some places rather than others. If the over-reporting is equally likely to occur in any state, then that would just mean that the real voter turnout in name-signing states is 53.1% (rather than 61.1%) and in voter-ID states is 50.1% (rather than 58.1%). The margin of difference between the two types of states would remain the same. Thus, unless you have additional evidence on differential rates of over-reporting, there is NO "methodology worry."

The article in question says:

Moreover, our analysis reveals that with the new question wording, estimates of the turnout rate for those usually thought to be the least likely to vote are considerably lower than estimates using the traditional question.

This means that the over-reporting phenomenon isn't uniform across the population, and might tend to cast suspicion on some of the more "astonishing to conventional wisdom" results of the study Sebastian cites.

[Mike3550]

Sebastian, thank you for this post - it was a very interesting read. There are two points, though, that I feel should be addressed:

1. In the first model, the author uses a hierarchical model. This means that the data are at two different "levels" - in this case the county and the state. The problem is that it looks like Vercellotti included the state-level variables (i.e. voter identification, battleground state, competitive governor/senate race and days from Election Day to register). This means that some of the effects at the state level are probably "bloated" because they are correlated with each other within states. The random and fixed effects that he discusses only work if the variables are controlled at the proper level.

2. Also, I find it wierd that he does not run the same type of model in the final analysis. What I would be interested in knowing is where blacks and whites, poor people and richer people have different voting patterns. What are the effects of having a county composition on more poor people voting or more blacks voting? What types of counties are more likely to see black voter drop-off - ones with more blacks or ones with less blacks? That would be a really interesting account to tell. My guess is that there wasn't enough data within each county to run that kind of analysis, but if someone could get the money to do it, I think that it would be fascinating.

[gwangung]

Data is always good. May even change my mind.

[Jack Okie]

Let me second OkieByAccident's postive comments about our optical scanner voting system. At the voting location for the 2006 general election, I spent some time discussing the voting machines with the poll official supervising them. There are several things to like, such as

1) Heavy paper (cardboard) ballot with easily readable printing, a paper trail and straightforward manual re-counts.

2) A mis-marked ballot is backed out of the scanner so the voter can correct it or complete a new ballot.

3) In the event of a power failure, the ballots are deposited and held in the stand supporting the machine, to be retrieved and scanned under observation by the poll watchers when power is restored.

BTW, in my precinct we sign beside our printed name on the pre-printed list of registered voters for that precinct.

[Gary Farber]

At risk of repetition, I'll note yet again that leaving comment spam standing is an open invitation to the spammer to return and spam some more, as it marks a blog that doesn't delete it, and thus is a great place to spam.

[Gary Farber]

"(And that UV cuts through a lot less here, as well, so keep that also in mind if it does turn sunny again.)"

More. i meant it cuts through "a lot more," not a "lot less."

[Gary Farber]

"(And that UV cuts through a lot less here, as well, so keep that also in mind if it does turn sunny again.)"

More. I meant it cuts through "a lot more," not a "lot less."

[trilobite]

I am very dubious that meaningful comparisons can be drawn between different states as to the effect of a given variable on voter turnout. There aren't many states, and most of them differ from most others in many other respects that could affect voter turnout, with different amounts of variation for each possible variable, and some of those variables probably interact so that the same change in two different states would have a different effect on voter turnout. I never really learned the math of multivariable comparisons, but my impression is that any result you get from a dataset like this is going to be an artifact of data massage.

[hilzoy]

Gary: I was at work. Now it's fixed.

[Anarch]

Let me second OkieByAccident's postive comments about our optical scanner voting system.

Thirded, from up here in Wisconsin.

[Michael Cain]

And big chicken me, I'm off to a volleyball tournament tomorrow at 4 in the morning (here I come Denver)

Excellent timing wrt the weather. Last night the snow line was at 7,000 feet, but it's warming up nicely today and is supposed to get better as we go into the holiday weekend.

[Joe Thomas]

2) A mis-marked ballot is backed out of the scanner so the voter can correct it or complete a new ballot.

A definite advantage, considering how easy it is to screw up a ballot (machines are finicky readers). The only time this has been a problem is in poor counties where there weren't enough scanners for individual voters to check their ballots, so a faulty ballot meant your vote didn't count (see: Florida, 2000).

[Jay C]

OT, but Andrew is blogging again ; alas, not in the hallowed columns of Obsidian Wings but at least the RMN provides us with a picture!

[Gary Farber]

Kewl.

Going with the Iraqi Men Respect Beards theory, I take it, Andrew? (I'm not dissing; I was just surprised for a moment.)

[Gary Farber]

Best White House. Press. Conference. Evah.

[John Snape]

I'm wondering what publius has to say about this. I remember on his other site every few weeks he'd have a post about how showing ID would significantly reduce voter turnout in poor and rural areas, and he was very upset for some reason that Georgia(?) was going around giving people free IDs so they could vote.

[hilzoy]

Andrew: totally different hair! Beard! Yow! Does your wife still recognize you?

[Gary Farber]

"...and he was very upset for some reason that Georgia(?) was going around giving people free IDs so they could vote."

I'm not familiar with publius' previous blogging, but I find myself strongly doubting this account.

[John Snape]

--"I'm not familiar with publius' previous blogging, but I find myself strongly doubting this account."

That's how I remember it, but I don't know where his old blog is (if it's still even there). He'd made a post about Georgia (I think) giving poor people IDs so they could vote, because they passed a law saying you needed a photo ID to vote. And I remember he was very upset about it.

I could be wrong (it wouldn't be the first time), but that's how I remember it.

[Gary Farber]

"...I don't know where his old blog is (if it's still even there)...."

The first page of Google results for "publius" shows this.

"He'd made a post about Georgia (I think) giving poor people IDs so they could vote, because they passed a law saying you needed a photo ID to vote. And I remember he was very upset about it."

I'll bet a nickel that this is not a characterization publius will regard as accurate. Possibly because publius was stressing other points about the bill, and what it required. If it was Georgia, those would be points that were found unconstitutional. That's after the professionals in the Civil Rights Division of the Department of Justice found it discriminated against black voters, but were then over-ruled by the political leadership.

Those IDs cost $20, so perhaps you're thinking of another Georgia law, of course.

Such as the one the courts threw out this year.

But you said publius posted on this "every few weeks," and I wasn't aware publius was still posting on the old blog in the past few months.

I say all this without troubling myself to go through publius' old blog. I'm purely using my amazing psychic abilities to guess that publius wasn't actually objecting to free IDs being given out, out of opposition to free IDs. It's a pretty wacky reach of me, I know. But sometimes "Danger" is my middle name.

[dr ngo]

OT, but not entirely -

Gary, nice of you to give us links to your own blog, which is often interesting I would comment there (occasionally, and probably trivially) *IF* I could.

But it requires (apparently) a Blogger or Google ID and password, which I ain't got.

I suppose if I really really really wanted to say something there, I'd figure out how to do so. But as it is . . .

Sayang.

[Gary Farber]

"But it requires (apparently) a Blogger or Google ID and password, which I ain't got."

As it says in the sidebar, it takes a few seconds to "register."

It's a heck of a lot easier to comment on blogger than it is to comment on Typepad, at least in my opinion. But if there's some specific problem, I'd be happy to try to help.

[Anarch]

Since I don't think I've actually voiced this here, I have no problem in requiring a meaningful voter ID. In fact, I have no problem in creating a national ID card -- one of whose purposes would be to replace the archaic, decrepit, and basically worthless SS-as-identifier system -- on the understanding that this would be at least a generational effort requiring significant advances in biometric technologies. I just have grave suspicions about the options currently on the table, sufficient for me to oppose them for the duration.

Also -- though this really belongs in the previous thread -- same-day registration in Wisconsin is supposed to require some kind of meaningful ID at the registration desk. The one they always advise us to bring is a paycheck plus a utility bill (photocopied or otherwise), though there are myriad other options.* How stringently this requirement is enforced is, of course, another matter -- but one which devolves to the training and preparation of the polling places and not AFAICT to the defining law.

[I'll grant that some of the forms of ID seem a little dubious but them's the breaks, I guess.]

* This is the 2006 model; the 2004 version was similar, though I don't recall the exact details.

[bemused]

I would think that a photo-id requirement would be much more onerous in an urban locality where driving was avoidable than in a rural area where it is almost mandatory to drive a car. Was this distinguished in the study? (Yes I should go read the study myself.)