Voter ID

What the evidence says

The largest and most recent quasi-experimental studies find no measured effect of voter ID laws on aggregate turnout. Alvarez, Bailey and Katz (2008) found no evidence of reduced turnout, in the aggregate, across the 2002–2006 presidential and midterm elections. Cantoni (2019), using administrative voter-file data, estimates a turnout effect of -0.1 percentage points (95% CI: -3.0 to 2.8pp) and a registration effect of -0.1 percentage points (95% CI: -2.3 to 2.1pp) — both statistically indistinguishable from zero, overall and for every group defined by race, gender, age, or party. Hoekstra (2019) supplies a mechanism for why the aggregate effect stays small: at most 0.10% of votes cast in Florida and 0.31% in Michigan were cast without ID even where the option existed, so the population an ID requirement could exclude is itself tiny; the implied maximum turnout reduction is 0.06 percentage points in Florida and 0.2 in Michigan. Epperly (2019) states plainly that "it remains unclear whether they effectively reduce voting at all," attributing this to the fact that most Americans already possess ID.

A separate body of state-panel comparisons, synthesized by Highton (2017), finds small but nonzero declines concentrated in states with the strictest requirements. The turnout gap between a state with a strict photo ID requirement and one with the least restrictive requirement runs about 2 percentage points. States where ID laws changed in a more restrictive direction saw turnout drop 2–3 percentage points more than states with no change or a permissive change. Comparing specific election pairs, turnout declined by nearly 4 percentage points in two of four pairs (2006/2010 and 2004/2010) with "virtually no observable effects" in the other two. GAO's (2014) comparison of Tennessee and Kansas to control states found declines of 1.9 to 3.2 percentage points, averaging 2.6. Across the four studies Highton reviews, none finds an overall effect exceeding 4 percentage points, leading to his conclusion that "the claim that voter identification laws depress turnout to a substantial degree is difficult to sustain based on existing evidence."

Racial and ethnic disparities in turnout are not established by the evidence here, even though disparities in ID possession are. Cantoni's point estimate for non-white turnout relative to white is +1.4 percentage points (95% CI: -0.5 to 3.2pp), and for Black voters specifically the study rules out a turnout reduction larger than 1.3 percentage points relative to white voters. Hood and Bullock (2012), studying Georgia's 2008 election after its 2007 ID statute, found individuals lacking ID were less likely to vote than those with ID, but found no evidence the law disproportionately affected racial or ethnic minorities. Separately, Barreto (2022) measures ID possession itself in North Dakota: Native Americans held valid ID at 82% versus 89% for white residents in 2015, and 81% versus 88% in 2017 — a persistent 7-point gap that remains statistically significant after controlling for other covariates. Part of that gap traces to a specific mechanism: the address requirement accounts for 20% of Native Americans' lack of valid ID versus 8.2% of non-Native residents, and Native Americans are less likely to hold the underlying documents needed to obtain ID (a 19-point gap in 2015 widening to 30 points in 2017).

Individual-level analysis finds a small effect that aggregate figures can mask. At the individual level, "very strict" laws — those requiring both an ID card and signature matching — reduced turnout slightly, and did so more for less-educated and lower-income citizens, though the source describes the overall effect as "surprisingly small." Consistent with this, exit polls show almost no voters reporting difficulty accessing the ballot after being asked for ID.

A psychological mechanism runs in the opposite direction: exposure to voter ID laws, especially framed as disenfranchising Black voters, increases anger, and anger increases self-reported intention to participate. Total anger about voter ID laws rises from 0.16 in a control condition to 0.24 in a basic voter-ID condition, 0.31 in a disenfranchisement-frame condition, and 0.41 in a Black-disenfranchisement-frame condition. That anger is associated with a 0.15-point higher participation-intention score (0–1 scale, p<0.01) for anger about the laws, and a 0.32-point higher score (p<0.001) for anger about voter fraud. Exposure to the Black disenfranchisement frame produces a direct effect on turnout among Democratic respondents (B=0.08, p<0.05), mediated by anger. Consistent with a backlash mechanism, Milyo (2007) detected a small increase in Democratic turnout in places with new voter ID laws — though a separate panel study finds that Democratic-advantage effect weakened in subsequent elections, "leveling off near zero."

Where the evidence splits

Large administrative-data studies (Cantoni's -0.1pp estimate with a confidence interval crossing zero; Alvarez, Bailey and Katz's finding of no aggregate effect 2002–2006; Hoekstra's sub-0.2pp upper bounds) disagree with the state-panel comparisons Highton reviews, which find 2–4 percentage point declines in states that adopted stricter ID requirements (GAO's 1.9–3.2pp, averaging 2.6; up to nearly 4pp in two of four election-pair comparisons). Highton notes only that these four studies differ from one another without specifying which design feature — data source, comparison group, or period — produces the larger estimates, so the split is unexplained by the sources on this page beyond that general acknowledgment.

Sources disagree on whether voter ID laws affect Democratic turnout in a positive direction as a subgroup, or produce no measured effect at all. Milyo (2007), as cited by Valentino, finds a small increase in Democratic turnout in places with new ID laws. Alvarez, Bailey and Katz, and Cantoni's overall estimates, find no aggregate effect in either direction. The claims here do not state whether the positive Democratic-specific finding is compatible with a null aggregate (a subgroup effect offset elsewhere) or reflects a different measurement approach; no source reconciles the two.

Within Highton's own review, one passage states that racial differences "appear minimal at most" across three studies that tested for them, while another states "it is possible that these laws might lower turnout, and do so disproportionately among those for whom the burden is greater." This tension is not resolved within the source itself — the review flags both the absence of a measured disproportionate effect and the possibility of one without stating which studies support the latter claim.

Thresholds and dose-response

Effect size tracks the strictness of the requirement. States with a strict photo ID requirement show turnout about 2 percentage points lower than states with the least restrictive ID requirement (Highton 2017). States whose laws changed toward stricter requirements saw declines of 2–3 percentage points more than states with no change or a permissive change, and up to nearly 4 percentage points in two specific election-pair comparisons, against "virtually no observable effects" in states with unchanged or permissive laws in two other pairs. At the individual level, the same pattern holds by requirement type: "very strict" laws — requiring both a photo ID card and signature matching — reduced turnout slightly, while the source does not report a comparable effect for less strict ID requirements.

Open empirical questions

• No claim quantifies the monetary or time cost of obtaining qualifying ID for those who lack it, only that most people already possess it.
• No claim measures the interaction between ID requirements and other voting restrictions (registration deadlines, early or absentee voting rules), which Epperly (2019) explicitly flags as unaddressed.
• The Democratic-turnout increase found by Milyo (2007) is shown fading toward zero in later elections by one panel study, but no source measures what happens beyond that horizon or in states adopting ID laws more recently.
• No claim traces whether the small population lacking ID who are refused a ballot (as in the UK pilots, or the sub-1% shares in Florida and Michigan) are able to cure that refusal through provisional ballots or same-day remedies.