The state of science’s reproducibility crisis (Part One)

Some of the most famous scientific experiments in history may be misleading. And that’s not necessarily because of a lack of strong theoretical basis, flawed methodology, or scientific misconduct.

Take the world-famous “Stanford Prison Experiment” (1971). The study, which simulated a prison with students as guards and prisoners, found that ordinary people quickly become abusive in positions of power. But modern attempts to replicate the study have failed to reproduce its results. Today, psychologists generally disregard its conclusions.

More recently, a 2010 study that argued standing in expansive “power poses” increases confidence and alters hormone levels, has likewise failed to stand up to the test of scientific reproduction.

The power pose craze has already come and gone.

But it’s not only these two famous studies. Writ-large, science has an alarming inability to reproduce the findings of many published studies.

Known as the reproducibility crisis, or sometimes the replication crisis, the phenomenon threatens the credibility of scientific research in fields from psychology to medicine and beyond.

In this article, we explore the state of the scientific reproducibility crisis and an explanation of its underlying causes.

Why it matters

Reproducibility and replicability are two different but fundamental elements of the scientific research process. (Stay tuned for a breakdown of reproducibility vs. replication.) When studies are either reproduced or replicated, the process enhances the validity of the original results and shows whether the original study suffered from research biases, ranging from selection and sampling biases to analytical or interpretive ones.

A substantial body of research shows just how essential it is to reproduce and replicate studies. Some 65% of researchers have tried and failed to reproduce their own research. The Many Labs 2 project showed that the typical replicated effect of 28 published studies was only 20% as large as the original study—in other words, that original studies overstated the size of effects by roughly five times.

Meanwhile, noteworthy replication projects in psychology and social sciences such as Reproducibility Project: Psychology (2015) have found replication rates around 40-60%: only about half of studies were successfully replicated. Projects compiled by the Metascience Observatory show the following ranges in replication rates by field: 10-87% in psychology, 54-62% in social sciences, 26% in biomedicine, 11-78% in oncology, 61% in economics, and 28% in sports/exercise science.

Researchers first began to focus on reproducibility in the 1970s and 1980s with early meta-analyses of studies. Then, in the 2000s and 2010s, groundbreaking studies introduced concerns about publication bias, selective reporting, and misaligned incentives in the scientific publishing industry.

While some studies suggest that open science and data sharing have improved reproducibility in the 2010s and 2020s, it remains one of the biggest challenges in modern science.

Image from Nature.

Causes and factors behind the crisis

The reproducibility crisis is by no means new. However, a growing body of evidence in recent years has proved its severity more conclusively than in the past.

Here are the four biggest causes:

1. Publication Bias

One of the landmark studies first positing publication bias was published in 2005 in PLOS by John Ioannidis. The study describes how “a research finding is less likely to be true when the studies conducted in a field are smaller; when effect sizes are smaller… when there is greater financial and other interest and prejudice; and when more teams are involved in a chase for scientific significance.”

Journals prioritize novel and positive results. This means that scientists are incentivized to ensure that their analyses produce positive results—analyses that sometimes fail to stand up to the test of reproduction.

Studies have showed 73% of positive trials received publication vs. only 41% of negative trials, with positive studies arriving 1-3 years sooner. Significant outcomes also had 2.2-4.7 times higher odds of being published. These are significant variances, altering the material understanding of many scientific phenomena and studies toward positive and unreproducible results.

2. Scientific and Academic Misconduct

Research retractions have increased substantially since the 1980s. Part of this perceived rise is due to greater awareness and tracking. But the proportion of retractions due not to errors but rather fraud has grown in the 2020s. Entities that promote systemic scientific fraud have undoubtedly emerged, such as sellers of mass-produced low quality and fabricated research as well as predatory journals that do not place quality controls on submissions.

Unfortunately, the number of fraudulent publications is growing at a rate far outpacing the growth of legitimate science. So long as scientific fraud remains a major part of the research world, reproducible science will suffer in turn.

3. Insufficient Data Sharing

A less malicious but equally present factor contributing to a lack of reproducible research has to do with modern research methods: namely, that scientists cannot obtain the datasets used by other studies.

When a major 2019 study attempted to reproduce research in the field of hydrology, researchers were only able to reproduce the results of 1.6% of 1,989 articles. For 89% of articles, the researchers could not find or receive directions to obtain data. Then, in 44% of articles, only partial digital artifacts were available. Out of the papers with digital artifacts completely available, 60% were fully or partly reproducible.

In the field of artificial intelligence research, sharing data alone wasn’t enough to allow scientists to reproduce findings. An 2024 arXiv preprint found that scientists needed both data and code fully shared for a study to be successfully reproduced.

4. Pressure to Publish

The “publish or perish” aphorism remains deeply true in science. The link between research productivity and research funding strengthened throughout the late 20^th and early 21^st centuries, eventually resulting in publication metrics becoming key factors in academic promotions and evaluations.

One 2023 preprint showed that Journal Impact Factor and recency of publications were two of the more important criteria in promotion at 159+ institutions worldwide. And a study of professor recruitment at the University of Oslo showed that reliance on publication metrics moderately increased from 2000 to 2017.

This pressure to publish is having a real effect on researchers, with up to 63% of respondents in some surveys citing publication pressure as the largest or second largest challenge to research integrity. The pressure can sway authors to cut corners, commit misconduct, or unintentionally be remiss when editing their work—inevitably leading to unreproducible research.

Reproducibility in 2026

Image from Wayne State University. 

As recently as February 2026, a scientific study showed that academic promotion in the field of radiology prioritized standard, rank-specific results for publication productivity. Publication bias and the pressure to publish positive results remain serious barriers to the proliferation of reproducible research.

On the flip side, has the rise of open science done its job and helped to improve the problem of insufficient data sharing?

Research says that, at this point, results are mixed. While measures like data sharing and the pre-registration of studies increase transparency, they do not necessarily discourage questionable research practices. In other words, they do solve our above #3 problem limiting reproducibility (insufficient data sharing), but not #1, #2, or #4 (publication bias, scientific and academic misconduct, pressure to publish).

Scientists are hopeful that registered reports and mega-studies can fundamentally reshape incentive structures, shifting scientists’ focus from producing statistically significant results to emphasizing methodological rigor—and hence, publishing replicable and reproducible research.

Fortunately, these types of studies are on the rise as research grows increasingly more collaborative. But there is still a lot that needs to be done on the level of the individual scientist, journal and institutional level, and policy level to address the crisis.

Stay tuned for Part Two of our series to science’s reproducibility crisis. There, we discuss the crucial difference between reproducibility and replicability, and what these two concepts offer to researchers and science as a whole.