Replication and reproducibility problems are not uniform across science. Recent large-scale replication projects in social and behavioral sciences, targeted efforts in cancer biology and preclinical biomedicine, and meta‑research in medical and clinical research give us comparative data on where replication rates are lowest and where economic and scientific stakes are highest as of 2026.

Fields with the most serious replication problems

1. Social and behavioral sciences (psychology, parts of economics, sociology, management)

• A massive DARPA‑funded SCORE project, summarized in Nature in 2026, attempted to replicate claims across 3,900 social‑science papers:

• Overall, only about half the studies replicated; a key paper reports a 49–55% replication rate across social and behavioral sciences.

• A companion Forbes overview notes replication success in 49.3% of 164 papers, and 55.1% of 274 claims, with economics showing the lowest replication rate among fields studied.

• Earlier focused work in psychology found even lower rates: some high‑profile projects report 36–47% replication success in experimental psychology.

• Management research appears to sit between psychology and economics in replication prevalence.

Why this matters: Social and behavioral findings heavily influence policy, education, business, and clinical practices. A 50% replication rate implies a large proportion of influential results are unreliable without further confirmation.

2. Psychology (especially experimental and social psychology)

• A 2026 popular overview of psychology's replication crisis reiterates earlier findings that only ~36–47% of key results replicate, based on large collaborative projects.

• A 2026 review on trust in science notes that the replication crisis in psychology has significantly undermined both internal and public trust, necessitating reforms in methods and incentives.

3. Cancer biology and preclinical biomedicine

• The Reproducibility Project: Cancer Biology replicated 158 effects from 23 papers:

• Only 46% of effects ""replicated successfully on more criteria than they failed.""

• Replication effect sizes were on average 85% smaller than original findings.

• A 2026 biopharma analysis reports that more than half of US preclinical animal‑model studies are considered irreproducible, with an estimated annual cost of $28–40B.

• A 2026 Annual Review article describes ""low rates of success at replicating studies"" across biomedical research, with up to 40% of clinical trials classified as ""zombie trials"" when raw data are scrutinized (i.e., trials whose data do not support their published conclusions).

4. Education research (moderate but still concerning)

• A 2026 education‑research commentary, drawing on the same SCORE project, reports that about 60% of education findings replicated, slightly better than the ~50% cross‑discipline rate but still leaving a large fraction unconfirmed.

• Replication studies themselves occupy only 0.2% of published education papers (2011–2020), meaning reliability is often unknown.

5. Broader biomedicine and clinical research

• Meta‑research reviews in 2026 emphasize persistent reproducibility issues across biomedical and clinical trials:

• Many studies lack sufficient detail or shared data to even attempt replication.

• High rates of ""zombie trials"" and inconsistent reporting of outcomes and protocols limit trust in published results.

• Preclinical animal research in particular is flagged as an area with systemic design flaws, underpowered studies, and poor reporting practices.

Comparative perspective

• Social/behavioral sciences (esp. psychology & parts of economics): Quantitatively, these fields have some of the lowest observed replication rates (≈36–55%), with large-scale, systematically conducted replication studies backing the numbers.

• Cancer biology and preclinical biomedicine: Replication rates are similarly troubling (≈40–50% at best), but the effect-size shrinkage (85% smaller) and financial and human stakes make the crisis particularly acute.

• Clinical/biomedical trials: Less comprehensive replication benchmarks exist, but the high prevalence of zombie trials and persistent meta‑research concerns mark this as a high‑risk area.

Counterarguments and nuance

• Some commentators argue that a ~50% replication rate is not necessarily catastrophic for social science, given noisy phenomena and evolving methods, and caution against oversimplifying the ""crisis"" narrative.

• In some areas (e.g., parts of finance), recent work suggests higher replication rates (≈75–80% for certain equity anomalies), showing that not all data‑intensive fields are equally affected.

• Reforms—Registered Reports, better reporting, open data—are improving practices, and some 2023–2026 psychology projects report higher replication in newer, more rigorous work.

Implications

• The worst-affected fields in 2026 by volume and impact are:

• Social and behavioral sciences (especially psychology and some economics/social science subfields);

• Cancer biology and preclinical/animal biomedicine;

• Broader biomedical and clinical trial research where ""zombie trials"" and poor reporting remain common.

• These areas are therefore the primary targets for reforms in design, reporting, incentives, and funding of dedicated replication work.