Should voters need to bring identification (ID) to vote? On the one hand, one could argue that verifying the identity of every voter prevents election fraud. It ensures that noncitizens don’t vote and that each citizen can’t cast their vote more than once. On the other hand, one could argue that not everyone has access to an ID due to geographical or financial constraints, and putting a requirement in place would suppress such voters.
This is a hotly debated topic in American politics. According to a Gallup poll done in 2024, the majority (84%) of people favor voter ID laws. However, the popularity of this policy differs by political party. Nearly all Republicans (98%) advocate for ID requirements, whereas one-third of Democrats oppose them.
Understanding the impact of strict voter ID laws on turnout is important for helping settle this debate. If the laws decrease turnout, it could mean that a significant number of prospective voters are turned away for not having ID. But if the laws don’t affect turnout, this provides an argument that the laws should stay. This is the question I try to answer in this blog post; the full details are available in my paper.
As of the 2024 election, 36 states have some sort of identification requirement, but these voter ID laws vary in their stringency. Most of these states have non-strict laws, which still allow some voters without ID to cast their ballot. However, about a dozen of these states have strict laws, which means that it is virtually impossible for a citizen without ID to vote. I was particularly interested in studying strict laws because I thought they would be the most likely to affect turnout.
Disentangling correlation from causation
Determining the impact of strict voter ID laws is difficult. I can’t simply compare the turnout of states with and without the laws. The map below depicts which states had strict voter ID laws during the study period, which spanned from 1984–2020. Notice that many of these states are battleground states, which typically have higher turnout. This means that a simple comparison between Georgia and California could imply that voter ID laws increase turnout, even though the effect is just coming from Georgia being a swing state.
One way that many political scientists have tried to adjust for this is a strategy called difference-in-differences. If California’s turnout and Georgia’s turnout follow a similar trend for a long time, then Georgia’s sharply diverges after implementing a voter ID law, it provides some evidence that the law had an effect. If we assume that California’s trend represents the counterfactual if Georgia had not implemented a law, we can even calculate the impact on turnout.
However, most states don’t follow parallel trends. This is because of factors that affect each state differently but vary over time. Senate elections occur every six years, but on three different cycles so that states don’t all have their elections at the same time. Suppose that these tend to have higher turnout than other elections. If the states that implemented strict voter ID laws are on the same Senate election cycle, their turnout trends will differ from states on other cycles.
The two examples named above—battleground states having higher turnout and Senate election cycles—are a non-exhaustive list of the ways that states with strict ID laws could differ from those without. So how can we actually separate out the impact of the laws?
Modeling the data
My solution was to model state turnout in a more flexible way. The synthetic difference-in-differences method is similar to the aforementioned difference-in-differences method, but it also adds weights on more predictive time periods and states in a similar way to the synthetic control method. These weights also ensure that states are only compared to other states that have similar factors that vary with time.
I also realized that this was useful for studying midterm elections. One reason that many researchers had previously avoided studying these elections was that congressional, gubernatorial, and senatorial races vary across years and affect turnout. The synthetic difference-in-differences method can better model this variation.
Another idea I had, unrelated to synthetic difference-in-differences, was determining how the laws affected states that were eager to adopt them versus states that adopted them later. There was a pivotal Supreme Court case in 2008, Crawford v. Marion County Election Board, which made it significantly easier to adopt voter ID laws. There was also a clear gap in 2010 when no state implemented voter ID laws. It seemed plausible that the late adopters (adopting after 2008) could be affected differently from the early adopters (adopting before 2008).
What I found
First, I needed to establish that the synthetic difference-in-differences (henceforth SDID) method was an improvement. I compared it to the fixed effects counterfactual estimator (FEct), which is similar to difference-in-differences. I found that SDID yielded more precise estimates. Additionally, during the period before states adopted voter ID laws, the difference between predicted and actual turnout was nearly zero, meaning SDID had superior predictive powerpower the probability that a study will correctly reject the null hypothesis when it is false, indicating the sensitivity of a test to detect an effect if it truly exists .
When combining all election types—or when examining only presidential elections—I found no impact on turnout. However, when I narrowed my focus to midterm elections, I found, counterintuitively, that turnout increased by 2.9 percentage points. It is important to note that I did not always obtain this result when making minor adjustments to the model. Still, the findings strongly suggest that strict voter ID laws do not reduce midterm turnout.
Next, I compared early- and late-adopting states. Here, I found evidence that the laws decreased turnout. States that adopted their laws after Crawford v. Marion County Election Board (2008) experienced a 1.9 percentage point drop in turnout, driven largely by a 2.7 percentage point decrease in presidential elections.
What are the implications?
When combining all states and elections, I found no effect of strict voter ID laws on turnout, similar to what most researchers have concluded. I have particularly strong evidence against voter suppression in midterm elections, where turnout even appeared to increase slightly. However, I did find evidence of turnout decreases when focusing on late-adopting states. This may explain why some researchers found negative effects: if the laws’ impacts vary across states or time periods, researchers examining different states will naturally reach different conclusions.
My theory for this variation is that voter mobilization efforts were stronger in early-adopting states, where the laws were more salient. These efforts may have counteracted any negative effects on turnout. I also suspect that the lack of an effect on midterm turnout may be because midterm voters are generally more politically engaged and already have ID. However, I can’t test this with my data, so these will remain conjectures for now.
Besides the effects on turnout, I also showed that the synthetic difference-in-differences estimator outperforms the difference-in-differences method. Additionally, its ability to control for factors that vary with time made studying midterm election turnout more tractable.
If you want to learn more, you can read the full paper, Strict voter identification laws and turnout: Differential effects by election type and adoption timing.
