The Vaccine Evidence Ledger

How to use this without losing the argument

You will not win by volume. Pick one claim, ask them to commit to it, then answer that one with a citation. If they change subject before conceding, name it and go back.

If the person is someone you love, stop here and read Talking to people you love instead. This page is built for public argument, and in a private conversation almost all of it is the wrong tool.

Concede the real things — there are real things, and they're in What’s actually true. A cheat sheet that admits nothing is a cheat sheet nobody believes. The strongest position available to you is the accurate one: vaccines have measurable, rare, catalogued harms, which is exactly how we know the surveillance is working, and those harms are orders of magnitude smaller than the diseases.

The three flu concessions worth making first

Effectiveness genuinely is mediocre in some seasons. A 19–40% season happens, usually when H3N2 drifts after strain selection. Nobody who works on this pretends otherwise, and the honest framing is a seatbelt that works most of the time, not armour.

Repeat vaccination may blunt the response. A 2022 systematic review estimated roughly 18% lower effectiveness against H3N2 in people vaccinated two seasons running versus the current season only. It's a live research question — and every combination still beat no vaccination.

Rotavirus vaccine really does shed. Routinely, in stool, for about a week. The guidance is to wash your hands after nappy changes if someone in the household is severely immunocompromised. That's the complete, unexciting truth of the one vaccine where shedding is an everyday occurrence.

The one thing every agency agrees on

Every US intelligence agency — including those favouring a lab origin — assesses that SARS-CoV-2 was not developed as a biological weapon, and almost all assess that it was not genetically engineered. The mainstream lab-leak hypothesis is an accidental release of a natural or field-collected virus during research, which is a claim about biosafety practice. It is not the same claim as "they built it," and the two get merged constantly. Separating them is usually the most useful thing you can do in this argument.

Where the law actually stands

Jacobson v. Massachusetts (1905) upheld a state's power to compel smallpox vaccination — though what it actually upheld was a $5 fine, not forcible vaccination. Zucht v. King (1922) held unanimously that schools may exclude unvaccinated students even absent an outbreak. Those remain good law, and they are why school mandates exist.

What is changing is the religious-exemption question. For over a century no appellate court held that a mandate must contain one. In December 2025 the Supreme Court vacated the Second Circuit's judgment in Miller v. McDonald — the Amish challenge to New York's 2019 repeal — and sent it back for reconsideration in light of Mahmoud v. Taylor. West Virginia's own supreme court is weighing a parallel dispute after the governor created religious exemptions by executive order and the state board of education refused to honour them. If you are arguing that mandates without religious exemptions are legally settled, you are describing 2019, not now.

The dead ends — including the one that did real damage

From the 1940s into the 1970s the leading theory was the “refrigerator mother”: that autism was caused by cold, withholding parenting. It was wrong, it was never supported, and it inflicted decades of blame on mothers of autistic children. It is worth knowing about, because it is the clearest demonstration that a confident, institutionally endorsed causal story about autism can be completely false — and because the people harmed by it were the same people the vaccine theory now recruits.

Also tested and unsupported: thimerosal, the MMR, the schedule as a whole, screen time, and “leaky gut.” Acetaminophen is the live one — observational studies found associations, but the largest and best-designed analysis, a sibling-control study of 2.4 million Swedish children, found the association disappeared once shared family factors were accounted for (hazard ratio 0.98). When HHS announced a label change in September 2025, the Coalition of Autism Scientists and the Autism Science Foundation objected that the evidence was “limited, conflicting, and inconsistent.” The honest statement is not “disproven” but “the best-designed evidence points away from it.”

The part that’s actually useful: real allergic reactions

Gelatin — used as a stabiliser in MMR, varicella and the nasal flu spray — is a genuine cause of vaccine anaphylaxis, and people with known gelatin allergy should say so before vaccination. PEG in mRNA vaccines is associated with anaphylaxis at roughly 5 to 8 cases per million doses. Neomycin appears in trace amounts and more often causes contact dermatitis than anything severe. Some vial stoppers contain latex. Overall vaccine anaphylaxis runs on the order of one to two cases per million doses, which is why providers observe you for fifteen minutes and stock epinephrine.

Egg allergy is the example worth citing when someone claims guidance never changes. For years, people with severe egg allergy were told to get flu vaccines only in medical settings. In June 2023 ACIP removed every egg-related precaution for every influenza vaccine — including for people with prior anaphylaxis to egg — because the accumulated evidence showed the residual ovalbumin content was too low to matter. The rule was loosened because the data said so, which is precisely what critics claim never happens.

The criticisms of this process that are actually fair

Phase 3 trials genuinely cannot detect rare events. A 40,000-person trial has reasonable power to find something occurring at 1 in 1,000 and essentially none to find something at 1 in 100,000. Myocarditis was not missed because anyone hid it; it was missed because it is rarer than the trial could resolve. That is a real limitation, and the correct response is not to deny it but to point at what the post-licensure systems are for.

Trial populations underrepresent the people who most need the data — pregnant women, the very elderly, the immunocompromised — which is why guidance for those groups often lags and rests on observational evidence.

Most trials are sponsored by the manufacturer. Independent replication happens less than it should, and publication bias is real across all of medicine. The strongest counterweight is that the large post-licensure cohorts — Danish, Swedish, UK, Finnish — are run by government agencies on compulsory registries, which is exactly why those studies carry so much of the weight on this page.

The whole mechanism, in one line

Every infection has a basic reproduction number, R₀ — the average number of people one case infects in a population with no immunity. If R₀ is above 1 the outbreak grows; below 1 it dies out. So you don't need to make everyone immune. You only need to make enough people immune that the average case infects fewer than one other person.

That gives the herd immunity threshold: HIT = 1 − 1/R₀. For a disease with R₀ of 2, you need half the population immune. At R₀ of 6, 83%. At R₀ of 15, 93%. The relationship is steep at first and then flattens, which is why highly transmissible diseases are so unforgiving — the last few percentage points are where all the difficulty lives.

What the eight have in common

The fuse is long. Wakefield published in 1998; England lost endemic-free status in 2006 and a man died in Swansea in 2013. The people who got sick were the babies whose parents read the coverage fifteen years earlier. Nobody experiences the consequence close enough in time to connect it to the cause, which is exactly why the argument never settles itself.

The trigger is often real. Samoa began with two infants killed by a genuine vaccine administration error. The Philippines began with Dengvaxia, which genuinely did harm dengue-naive children. In both cases the damage was done less by the original error than by how badly the response to it was handled — which is the strongest practical argument for acknowledging real adverse events quickly rather than minimising them.

Outbreaks do not reliably change minds. This is the uncomfortable one. Reporting from Gaines County after the 2025 Texas outbreak found the experience had hardened local Mennonite attitudes against vaccination rather than softening them. If your theory is that people will come round once they see a child die, the evidence does not support you.

The honest summary of the risk picture

Serious vaccine adverse events are real, rare, and usually identified within months to a few years by exactly the surveillance systems critics say don't exist. Between 2006 and 2024 the US compensated 10,633 injury claims against 5.65 billion doses distributed — 1.89 per million. Measles, by contrast, hospitalises about 1 in 5 unvaccinated people who catch it and kills 1 to 3 per 1,000 infected children.

That asymmetry is the whole argument. It does not require pretending the numerator is zero.

When to stop

You may not change their mind, and that is a real possible outcome rather than a sign you argued badly. If the conversation is becoming the defining feature of the relationship, stop — not as a concession, but because the most valuable thing you can be is the person they come to when something changes. A cousin who stayed warm is reachable in two years. A cousin who blocked you is not.

Two things are worth holding onto if you get nowhere. Partial is not nothing: a child who gets some vaccines late is better protected than one who gets none, and “would you consider just the MMR?” is a question people say yes to far more often than “will you do the whole schedule?” And clinicians outrank relatives — a provider's clear recommendation is the single strongest predictor of uptake, so getting someone to a conversation with their own doctor may accomplish more than anything you say.