Do COVID-19 vaccine mandates reduce uptake of other vaccines?

A spillover question, not a mandate-effectiveness question

The paper asks a narrow question: when a country adopted a COVID-19 vaccine mandate, did coverage of vaccines that were not targeted by that mandate change afterwards? This is a cross-vaccine spillover question. It is different from asking whether a mandate raises uptake of the vaccine it targets, and the authors are explicit that their estimates should not be read as evidence about mandates' direct effectiveness.

The concern being tested is a backlash or reactance channel: if a COVID-19 mandate hardened attitudes toward vaccination or toward coercion, coverage of other vaccines could fall even where no policy required them. The study measures realized national coverage rather than intentions or self-reported attitudes, so it tests whether any such response shows up in aggregate vaccination behavior.

Binary mandate exposure against WHO coverage series

The treatment is a binary indicator equal to one in country-years where at least one COVID-19 vaccine mandate was in force, built from the Oxford COVID-19 Government Response Tracker. First recorded mandate adoption in the sample occurred in 2021–2022, concentrated in the second half of 2021, particularly July to December, with fewer countries adopting in early 2022. Because adoption was staggered across countries but clustered in a narrow window, the identifying variation is compressed in time.

Outcomes come from the WHO Immunization Data Portal and split into two groups. Discretionary vaccines are those not de facto required or institutionalized: HPV for females and males, and influenza for children, older adults, healthcare workers, and pregnant women. Routine childhood vaccines are those not given at birth but often embedded in schedules: first and second measles doses, first rubella dose, third DTP, third polio, third Hib, third hepatitis B, completed rotavirus series, and completed pneumococcal conjugate series.

The panel covers 2018–2024, and sample sizes differ by outcome because coverage reporting differs across vaccines and sources. For example, female HPV has 611 country-year observations with mean coverage 56.93%, while first-dose measles has 1,358 observations with mean coverage 85.49%. Those denominators matter: the discretionary estimates rest on far fewer observations than the routine ones.

Estimating effects only for countries that switched

Countries adopted mandates at different times, and effects may differ across countries. A standard two-way fixed-effects comparison can be biased in that setting, so the authors use the de Chaisemartin and D'Haultfoeuille staggered difference-in-differences estimator. It produces dynamic average treatment effects on the treated for switching countries, meaning countries whose mandate exposure changed, with event time defined relative to the last period before the change.

Pre-treatment coefficients act as placebo tests for differential trends before adoption. This is the study's central diagnostic, and it is where the two vaccine groups diverge. For all six discretionary outcomes, joint placebo tests do not reject, but the authors show this non-rejection is weakly informative. For routine childhood vaccines, the placebo tests reject at the 5% level for first-dose rubella and third-dose polio, and are marginally significant at the 10% level for first-dose measles.

The authors also run power and sensitivity analyses. Following Roth, they compute minimum detectable linear pre-trends; following Rambachan and Roth, they compute robust bounds that allow post-treatment deviations from parallel trends. These tools let them say not just whether an effect was detected, but how large an effect the data can exclude.

HPV and influenza show no clear effect, but little power

For discretionary vaccines, no post-treatment estimate is statistically significant, and no joint test of post-treatment effects is significant. Female HPV point estimates are negative at every post-treatment horizon, but their confidence intervals include zero. Influenza estimates vary by target group: for children they are negative before treatment and move toward zero or slightly positive after; for older adults they are negative in the far pre-treatment period and near zero after; for healthcare workers they move from negative and insignificant toward zero or slightly positive; for pregnant women they fluctuate around zero.

The absence of a detected effect here is not strong evidence of absence. The joint-Wald 80%-power minimum detectable linear pre-trend ranges from 6.94 to 15.44 percentage points per year, and none of the six discretionary tests reaches 80% power even against a 5-percentage-point annual differential trend. Under the sensitivity analysis at the moderate departure level, robust lower bounds for the first post-treatment effect range from −25.77 to −16.94 percentage points. Sizeable adverse effects on discretionary vaccines therefore remain compatible with these data.

Routine childhood coverage stays close to zero after mandates

The routine childhood results are more stable. For most routine vaccines, post-treatment estimates are small and statistically insignificant, and no individual post-treatment coefficient or joint post-treatment test is significant. Measles first and second dose estimates stay close to zero. Rubella first dose moves modestly upward in the first two post-treatment periods before turning negative at the third, with confidence intervals including zero throughout. DTP third dose, polio third dose, Hib third dose, and hepatitis B third dose are similarly small and centered near zero.

Two outcomes are less clean. Completed rotavirus series estimates are more variable, especially before treatment, without a consistent post-treatment decline. Pneumococcal conjugate series shows positive post-treatment point estimates, particularly at later horizons, but confidence intervals remain wide enough that this is not statistically significant evidence of a positive effect.

The routine estimates are also more precise than the discretionary ones, with narrower confidence intervals. That precision is what makes the sensitivity analysis informative for this group, but the authors caution that precision alone does not establish causal identification, and the pre-treatment diagnostics for measles first dose, rubella first dose, and polio third dose limit how firmly those particular nulls can be read.

Roughly 2–4 percentage points at the first horizon

The most decision-relevant number in the paper is an exclusion bound rather than a point estimate. Under the sensitivity assumption that post-treatment deviations from parallel trends can be as large as the largest pre-treatment deviation, robust lower bounds for the first post-treatment effect range from −4.06 to −1.91 percentage points across seven better-powered routine vaccines. In plain terms, immediate declines larger than about 2–4 percentage points can be ruled out for those outcomes under that assumption.

The bounds loosen with time. They widen to roughly 3–7 percentage points at the second post-treatment horizon and 5–12 percentage points at the third, so delayed effects are less constrained. They are substantially wider for rotavirus and, to a lesser extent, pneumococcal conjugate, and wider still for the discretionary vaccines. The paper's claim is therefore asymmetric: a broad, immediate routine-vaccine backlash is excluded, while outcome-specific and delayed backlash is not.

Alternative treatments and exclusions preserve the pattern

The conclusion survives several redefinitions of treatment. Using annual mandate exposure share, the number of mandated groups, group-specific mandates, and general-population mandates produces the same overall conclusion. Adding controls for development, demographics, governance, health financing, pandemic severity, policy stringency, and COVID-19 assistance does not overturn it, nor does restricting the comparison group to never-treated countries.

Two sample exclusions target specific confounders. Excluding eight countries with substantial subnational vaccination policy variation, then a broader set adding Italy, Switzerland, and the United Kingdom, changes routine estimates only modestly: first post-treatment estimates in the extended-exclusion sample were −1.46 percentage points for measles first dose, −1.93 for measles second dose, −0.83 for rubella first dose, 0.39 for DTP third dose, and 0.28 for polio third dose, against −1.20, −1.54, −0.33, 0.63, and 0.48 in the full-sample controlled models. No systematic negative pattern emerged, while discretionary estimates became less precise as switchers fell.

Excluding the 57 countries exposed to Gavi support during 2021–2025 tests whether externally financed vaccine access masked adverse spillovers. Across the nine routine outcomes, first post-treatment estimates ranged from −1.72 to 0.45 percentage points, second-horizon estimates from −1.34 to 1.49, and third-horizon estimates from −1.66 to 1.86, with all confidence intervals including zero. Official and administrative coverage data likewise showed no systematic negative post-treatment effects, although the official pneumococcal series produced a significant positive effect at the second horizon.

Country averages cannot see who changed behavior

The treatment records whether a mandate was in force and for how long, not whether individual choices were constrained, how strictly it was enforced, or what sanctions followed non-compliance. The coefficients are reduced-form policy effects, similar in spirit to intention-to-treat estimates, averaging over countries where mandates had real practical force and countries where implementation was limited. A near-zero coefficient does not identify the effect on people who vaccinated only because of the policy.

Country-year averages also cannot reveal within-country differences. Coverage may have fallen among particular occupational, political, socioeconomic, or demographic groups while staying flat or rising among others. The estimates apply to countries that changed mandate exposure during the study period and should not be generalized to all countries. And national coverage is an aggregate measure of realized behavior: it cannot show whether mandates affected trust in institutions, political attitudes, polarization, or perceptions of coercion.

Separate routine stability from discretionary uncertainty

The paper's most useful contribution is a split verdict rather than a single null. For routine childhood vaccines, the combination of narrow confidence intervals and sensitivity bounds supports a concrete statement: under a moderate departure from parallel trends, immediate declines larger than about 2–4 percentage points can be excluded for seven better-powered outcomes. That is the threshold a reader should carry away, and it is specific to the first post-treatment horizon and to those routine outcomes.

For HPV and influenza, the same data cannot support that statement. The discretionary estimates are too imprecise, with robust lower bounds reaching −25.77 to −16.94 percentage points at the moderate sensitivity level, so a sizeable adverse spillover on these vaccines is not excluded. The paper therefore does not show that mandates are costless or that their design is unimportant; it shows that a generalized cross-vaccine backlash is absent from aggregate routine coverage, while the discretionary evidence remains open.

Sources used in this answer