List of Poorly Done Vaccine-Related Studies

This page contains a list of several vaccine-related studies that the anti-vaccine lobby has held up as supposed “proof” of a link between vaccines and autism/neurodevelopmental disease.

By way of information, a study can only be considered valid when several things are in place: 1) study type truly reflects a desire for unbiased study; 2) sample/cohort size is large enough to be considered truly representative of a population; 3) the study author(s) are reputable (i.e., no retracted papers, no accusations/proof of research fraud); 4) a clear and standard explanation exists of the methods; 5) the exposures/data are measured accurately and consistently in order to avoid error or deliberate misinterpretation; 6) the ability of other studies to replicate findings, and 7) no conflicts of interest on the part of the authors with respect to the focus of the study.

None of the studies below satisfy accepted study criteria, and as such, their findings are deemed invalid by the scientific community at large and are ineligible to be used as a basis for public health policy decisions.

Overall Takeaway / What to Watch For

  1. Cross-sectional vs. prospective: cross-sectional surveys and convenience samples (homeschool studies) are weak for causal inference. Prefer prospective cohort studies or well-done case-control designs with validated outcomes.
  2. Outcome validation: parental report ≠ clinical diagnosis; claims data can be better but depend on coding and access.
  3. Sparse data / subgroup instability: very small cell counts (e.g., few unvaccinated boys) can produce large, unstable ORs. Always look at raw counts.
  4. Confounding & reverse causation: children who are delayed or ill early may get different vaccination schedules — this can create spurious associations unless explicitly modeled. Administrative datasets must handle timing carefully.
  5. Reproducibility/transparency: look for shared code, pre-registration, or independent replication; papers with retractions or opaque methods deserve extra skepticism.
  6. Study funding and author conflicts of interest: often you will need to search separately for any author conflicts of interests, as unscrupulous authors and those who have had work retracted and invalidated, may attempt to disguise their associations with organizations that present a clear conflict in conducting unbiased study methods and obtaining unbiased, accurate results. Funding is often cited in the papers, but again, the reader will need to research the funding source to determine if a group is “paying for” a particular desired outcome.

This page will be reviewed periodically to reflect additional poorly-done vaccine-related studies if/when published.

Gallagher & Goodman (2010) — Hepatitis B Vaccination of Male Neonates and Autism Diagnosis, NHIS 1997–2002

  • Sample size / population: Boys aged 3–17 years, born before 1999, with vaccination records in the NHIS survey. They report n = 30 with parent-reported autism diagnosis and 7,044 without.
  • Design: Cross-sectional logistic regression using NHIS (1997–2002) data, adjusting for race, maternal education, and family structure.
  • Main finding: Reported higher odds of autism diagnosis among boys vaccinated with Hep-B at birth versus unvaccinated or vaccinated later (authors emphasized an association, not proof of causation).
  • Errors, Methodology Concerns, Controversies and Conflicts:
    • Sample Size Small and Unequal: Very small number of autism cases (n=30) is an egregious error. It introduces statistical instability, especially when stratifying by vaccination timing. No unbiased study author would dare weight their sample size so heavily to favor one group over the other; this in itself can be considered evidence of methodological fraud.
    • Parent-reported diagnosis: This is an incredibly inaccurate method, particularly as the authors did not do their own work in clinically validating the parent reporting. This makes the data subject to recall/reporting bias.
    • Potential for confounding: Although they adjusted for some variables, there may be unmeasured factors (healthcare-seeking behavior, socioeconomic variables) that influence both vaccination timing and diagnosis.
    • Cross-sectional data methodology: Cross-sectional study design is at best incredibly weak, and at worst, deliberately invites fraudulent outcomes, as cross-sectional data makes it tricky to infer causality or timing (when diagnosis occurred relative to vaccination).

Mawson et al. (2017a) — Pilot comparative study on the health of vaccinated and unvaccinated 6–12-year-old US children

  • Sample size: 666 children (homeschool), of which 261 (39%) were unvaccinated.
  • Design: Cross-sectional, convenience sampling. Parents (mothers) of 6–12-year-old biological children completed an anonymous online questionnaire about vaccination status, birth history, physician-diagnosed illnesses (including neurodevelopmental disorders), medications, etc.
  • Main finding: The authors reported higher rates of allergies and NDD in vaccinated children compared with unvaccinated homeschool children.
  • Outcome measure: NDD (neurodevelopmental disorder) defined as learning disability, ADHD, or ASD.
  • Errors, Methodology Concerns, Controversies and Conflicts::
    • Self-Selected Convenience sample: Homeschool families who respond to vaccine-health surveys (the data-gathering methodology in this study) is highly nonrepresentative and thus selection bias is extreme. Homeschooling families are a very specific subpopulation, thus generalizability is extremely limited.
    • Sample Size: As noted, only 39% of the 666 children enrolled in this study were reported to be unvaccinated. Thus, the unequal distribution of groups and the extremely small sample size (less than 1000) is neither representative nor valid for accurate outcome reporting.
    • Self-report of vaccination and diagnoses: This is another study that relies on self-reporting without any attempt to clinically verify diagnoses, and as such is subject to recall/reporting bias and deliberate misreporting.
    • Potential for strong selection bias: Families who choose to homeschool and respond to such a survey may differ systematically (e.g., health beliefs, medical care access).
    • Statistical Error: Wide confidence intervals, especially for interaction terms (preterm + vaccination) equals less precision. Odds ratios in this study are not stable, lending to justified criticism of the validity of the data and the intention of the authors (conclusion predetermination).
    • Publication issues / retractions: Versions of Mawson’s vaccine-homeschool paper have been retracted or republished amidst controversy; this raises red flags about editorial and quality control.

Mawson et al. (2017b) — Preterm birth, vaccination and neurodevelopmental disorders: a cross-sectional study of 6- to 12-year-old vaccinated and unvaccinated children

  • Sample size: Same 666-child cohort from the 2017a homeschooling survey. In that sample, ~7.7% were reported as preterm.
  • Design: This is a “Secondary analysis”, reviewing the same dataset as the 2017a study that was retracted — they looked specifically at interaction between preterm birth and vaccination in relation to NDD. Cross-sectional survey data.
  • Main finding: The authors report that preterm vaccinated children had a much higher odds of NDD compared to several comparison groups (e.g., vaccinated non-preterm, unvaccinated non-preterm). In one comparison, OR = 14.5 (95% CI: 5.4, 38.7) when comparing preterm vaccinated vs non-preterm unvaccinated.
  • Errors, Methodology Concerns, Controversies and Conflicts::
    • Statistical error: Very small subgroup sizes (especially preterm + unvaccinated) led to extremely unstable estimates and very wide confidence intervals, which is a serious statistical error which in itself would be grounds for invalidating the entire study.
    • Same limitations as the 2017a study: Convenience sample, parent self-reporting without any attempts at clinical validation, extreme selection bias and extremely high risk for recall bias with no controls to limit.
    • Risk of confounding and reverse causation: Preterm birth itself is strongly associated with developmental disorders; disentangling that from vaccination is very hard in a small, biased survey.
    • Multiple testing / interaction analysis: This increases the risk of spurious associations, statistical instability, and false positives.

Hooker & Miller (2020) — Analysis of health outcomes in vaccinated and unvaccinated children: Developmental delays, asthma, ear infections and gastrointestinal disorders

  • Sample size/population: Total sample size: 4,821 children; sub-cohort (claimed continuous followup ≥ 3 years old) = 2,047. In the the ≥3-year cohort, ~52% male.
    • Data came from three pediatric medical practices in the U.S., of children born between 11/2005 and 6/2015.
    • Vaccination status:
      • Unvaccinated by age 1 (plus 15 days): 633 of the ≥3-year-old sub-cohort (30.9%)
      • Vaccinated by age 1: 1,414 (69.1%) of the sub-cohort.
  • Design: Observational cohort study (retrospective), using pediatric medical record data.
    • Stratified by medical practice, year of birth, and gender.
    • Used conditional logistic regression to compare vaccinated vs unvaccinated in relation to the outcomes diagnoses (below).
    • They also did a “quartile analysis” by number of vaccine doses in the first year — children divided into quartiles based on how many vaccine doses they got, then compared odds of health outcomes across quartiles.
    • Temporal analyses: They looked at developmental delay odds by whether vaccine exposure was cut off at 6, 12, 18, or 24 months to see how timing related to later diagnosis.
    • Power calculation: They said their ~2,000 sample gives “80% power” to detect ORs of ~1.8 (α = 0.05), though for rarer diagnoses their detectable OR was higher (i.e., lower power).
    • No maternal/birth covariate control: They explicitly state they did not adjust for maternal or birth data because it was not available.
  • Outcomes measured
    • The study claims to seek to measure developmental delays, asthma, ear infections, and gastrointestinal (GI) disorders.
    • Diagnoses were based on chart review, using ICD-9 and ICD-10 codes from the medical records.
  • Main finding: the study authors claim that vaccination before age 1 was associated with:
    • Developmental delays: OR = 2.18 (95% CI 1.47–3.24)
    • Asthma: OR = 4.49 (95% CI 2.04–9.88)
    • Ear infections: OR = 2.13 (95% CI 1.63–2.78)
    • Gastrointestinal disorders: in some analyses they saw elevated ORs in higher-dose quartiles and with more time allowed for diagnosis.
    • In a temporal cut-off analysis, odds for developmental delay rose as they extended the “vaccine cut-off age” from 6 to 24 months.
    • When they extended the window for diagnosis (i.e., allowed diagnoses up to age ≥ 5 instead of ≥ 3), the ORs increased a bit for some conditions.
  • Errors, Methodology Concerns, Controversies and Conflicts:
    • Premise Not Proven: While the authors at least used clinical diagnoses from medical charts (and not inaccurate and prone-to-bias parental reporting), the statistical data on the 4 diagnoses they included in the study call their obvious intention to “prove” vaccines are associated with autism, into doubt – as the OR for asthma is over twice as high as for NDDs and ear infections.
    • Poor overall design: As with many “vaccinated vs unvaccinated” studies, this is not a randomized trial, so causation cannot be firmly established. Without random assignment, there is a significant risk that unmeasured variables explain the observed associations.
    • No adjustment for key confounders: They did not adjust for maternal health, socioeconomic status, birth complications, or other early-life factors. Without these, associations may be confounded: for example, children with developmental risk may also have different patterns of healthcare use or vaccination timing.
    • Definition of “unvaccinated”: They classify “unvaccinated” as “no vaccine by age 1 (plus 15 days).” But children could have been vaccinated later, and that group is partially included. This does not necessarily represent never-vaccinated children in their later years, which complicates interpretation.
    • Potential for reverse causation or selection bias: Children who are slower to develop or have early health problems might have their vaccine schedules delayed, which could produce a “vaccinated” group that is systematically different. The study only includes children with continuous follow-up; children lost to follow-up may differ in meaningful ways.
    • Multiple comparisons without corrections: They tested multiple health outcomes and multiple temporal cut-offs but did not correct their significance threshold for multiple statistical tests. This significantly raises the risk of false positives.
    • Power limitations for rarer diagnoses: The authors noted that for some less common outcomes, their sample was underpowered to detect small-to-moderate ORs. That means some non-significant findings could be due to limited power, not lack of effect — or vice versa.
    • Conflict of interest: “Conflict of interest” is a massive red flag in the research community, but the authors have made no secret of their well-documented affiliations with known anti-vaccine advocacy groups, which raises concerns about potential bias. Authors with conflicts of interest do not automatically invalidate a study’s data, but it makes a study much more likely to be invalid, used as propaganda for the cause of the conflict of interest. Given all of the serious flaws that this and their subsequent study suffer from, the conflicts of interest are much more suspect, and objectively must make this study completely invalid.
      • Brian S. Hooker: Paid scientific advisor for Focus for Health, on the Board of Trustees for Children’s Health Defense.
      • Neil Z. Miller: Director of Thinktwice Global Vaccine Institute, was a paid consultant to Physicians for Informed Consent.

Hooker & Miller (2021) — Health effects in vaccinated versus unvaccinated children, with covariates for breastfeeding status and type of birth

  • Sample size: According to the paper, 1,565 children from three U.S. pediatric practices.
    • In their second questionnaire / control-diagnosis cohort: 831 children (466 unvaccinated, 290 partially vaccinated, 75 “up-to-date” vaccinated).
  • Design: Survey and medical chart review. They collected data from three pediatric practices; parents filled out a questionnaire (including breastfeeding status and birth type), and diagnoses (autism, GI disorders, ear infections, asthma, ADHD) were confirmed via chart (ICD codes). Logistic regression was used, adjusting for sex, birth type (vaginal vs cesarean), breastfeeding, and vaccination status.
  • Main finding: Reported associations between early vaccination and higher odds of developmental delays, asthma, ear infections and some GI disorders; suggested breastfeeding and birth type modify associations.
  • Errors, Methodology Concerns, Controversies and Conflicts:
    • Selection bias: The sample is small, and comes from only three practices. As such it is not representative of the population and is not generalizable.
    • Data source clarity / replication: Some versions of this study (various versions of this and the previous Hooker & Miller study appeared in small journals and preprint/outlet chains) rely on limited or nontransparent datasets; reproducibility concerns have been raised.
    • Potential confounders not fully controlled: The authors neglected to account for confonders such as socioeconomic status, parental health behavior, access to care, or provider practices. As with the other cross-sectional/home-practice samples, vaccinated vs unvaccinated groups differ on many axes (socioeconomic, health-seeking behavior) that are difficult to fully adjust for.
    • Self-reported data + chart diagnoses: Chart data helps, but there may be variability in how accurately diagnoses are coded; also, survey responses may be biased, either by the responder misreporting or misunderstanding survey questions, or by the authors providing improper questions (leading, misreporting, too general, etc.).
    • Risk of reverse causation: Children with early health problems might influence timing of vaccines and/or number of visits.
    • Statistical error: Wide confidence intervals were noted for many reported ORs, which is a statistical error leading to imprecision of data and inaccuracy of findings.
    • Multiple outcomes & analytic choices: many outcomes and subgroup analyses increase type I error, unless pre-specified and corrected.
    • Author credibility, prior controversies, conflicts of interest: Brian Hooker has authored retracted vaccine-autism work in the past; that history invites careful scrutiny of methods and data handling. As noted previously, both authors carry significant conflicts of interest with respect to the topic of this study, and also as noted previously, with the significant errors in this study, it is considered invalid.

Mawson & Jacob (2025) — “Vaccination and Neurodevelopmental Disorders: A Study of Nine-Year-Old Children Enrolled in Medicaid” (Florida Medicaid study, 2025)

  • Sample size: 47,155 children continuously enrolled in Florida Medicaid from birth to age 9.
    • Of these, 42,032 (89.1%) had at least one vaccination-related claim by age 9; 5,123 (10.9%) had no such claim.
    • Preterm births: 5,009 (10.6%) of the cohort.
  • Design: Considered in 2 parts: cross-sectional and a retrospective cohort.
    • Cross-sectional part: they computed prevalence odds ratios (Aim 1 & 2) for NDDs vs vaccination / preterm status.
    • Retrospective cohort part: to assess “dose-response” (number of vaccination visits) vs ASD risk, they computed relative risks based on vaccination-visit counts.
    • Vaccination was measured via claims data (i.e., Medicaid claims) including a procedural, diagnostic of drug code for a vaccine; not a parent-reporting.
  • Main Finding: Authors reported an association between vaccination and higher odds of NDDs in this Medicaid cohort and called for further investigation.
  • Errors, Methodology Concerns, Controversies and Conflicts:
    • Confounding: Administrative claims data often lack detailed confounders (e.g., prenatal exposures, maternal health, social determinants) that might influence both vaccination and NDD risk.
    • Reverse causation / healthcare-seeking bias: Children who develop early developmental problems may interact more with the healthcare system, leading to more documented “vaccination visits” or coding artifacts.
    • Outcome misclassification / ascertainment bias: NDD diagnoses based on Medicaid claims, which depend entirely on coding practices, frequency of visits, and appropriate documentation. Some children may be underdiagnosed or miscoded.
    • Definition of “unvaccinated”: Using “no vaccination-related claim” without evaluating medical records for vaccine administrations, can deliberately misclassify children (e.g., those vaccinated out-of-state, or whose vaccination visits weren’t coded as such).
    • Survivor / enrollment bias: The study only includes children continuously enrolled for 9 years. This is a serious error, as it excludes children who left Medicaid, died, or had coverage gaps — biasing the sample.
    • Transparency / analytic reproducibility: Open peer reviews have questioned whether all model choices, sensitivity analyses, and data cleaning steps were fully reported. Given a full review of the study text,
    • Author credibility and prior/present controversies:
      • As noted above, Mawson’s vaccine-homeschool paper had been retracted or republished amidst controversy; this raises red flags about editorial and quality control.
      • At the time of initial publishing of this page (November 23, 2025), this paper appeared in preprint in the journal Science, Public Health Policy & the Law. Its editor-in-chief, James Lyons-Weiler, is a well-documented anti-vaccine advocate, and has been strongly critical of mainstream public-health policy and vaccines. The journal’s stated mission: to influence public health policy and law and explicitly to question “orthodox” vaccine-policy assumptions, is a clear example of ideological bias.
      • The study was funded by the National Vaccine Information Center (NVIC), which is a leading voice in the anti-vaccine activism movement.
      • The authors are affiliated with the Chalfont Research Institute, which is not a well-known academic institution, and there are questions about its transparency.