In an essay for STAT, Mark Louie F. Ramos, an assistant research professor at Pennsylvania State University, pushes back against the NIH's decision to scale back its replication agenda. NIH Director Jay Bhattacharya had emphasized replication as part of the gold standard for science, and in March 2025 he launched the $850,000 Replication Prize. Health Secretary Robert F. Kennedy Jr. even suggested a replication effort worth 20% of the NIH's $47 billion budget. But after a Council of Councils report raised concerns about cost and unfairly undermining confidence in individual scientists, the plans were reduced to $174 million.
Ramos argues that both concerns miss the point. A Type 1 error—concluding a claim is true when it is not—does not reflect poor science; it reflects the limits of statistical inference. If all studies use a 5% significance threshold, then 5% of studies where no real effect exists will still produce false positive evidence, even when conducted perfectly. Replication fixes this: the chance of two independent studies both producing false positives at that threshold is 0.25%, a 20-fold reduction.
Cost is a legitimate issue, Ramos acknowledges, but replication does not mean blindly repeating expensive studies. Because researchers already have data on observed effect size and variability, they can design replications that are more targeted and efficient than the original work. He also notes that scientific incentives push researchers to push the boundaries of evidence, making false positives inevitable in aggregate. Replication, he concludes, is the only real way to determine which important findings are robust.