Eric-Jan Wagenmakers

Introduction

Eric-Jan Wagenmakers is a Dutch mathematical psychologist at the University of Amsterdam who has probably done more than anyone else to move Bayesian inference from a specialist interest to a routine option in psychological research. His argument is that null hypothesis significance testing, as psychologists actually use it, cannot do what researchers want it to do — it cannot express evidence for a null hypothesis, it is asymmetric in its treatment of non-significant results, and it rewards the analytical flexibility that produced the replication crisis. His proposed remedy, Bayes factor hypothesis testing in the tradition of Harold Jeffreys, has been accompanied by a second campaign: making the remedy usable. The JASP software he founded is now how most psychologists first encounter Bayesian analysis. He is also one of the most visible figures in the replication and open science movements.

Early Life and Education

Wagenmakers was born on 21 May 1972 in Baarland, a village in the Dutch province of Zeeland. He studied psychology at the University of Groningen from 1994, working with Ritske de Jong on ageing, task switching, and reaction times, and took his master's degree in August 1996. He then moved to the University of Amsterdam as a doctoral student under Jeroen Raaijmakers, modelling human memory and visual word recognition. A Fulbright scholarship took him to Indiana University for the 1998–99 academic year to work with Richard Shiffrin. His dissertation, Priming in Visual Word Recognition: Empirical Studies and Computational Models, was defended in September 2001.

The training was in mathematical psychology rather than statistics, and it shows: Wagenmakers arrived at questions of inference through cognitive modelling — comparing formal accounts of memory and response time — and the move to Bayes factors was an extension of model comparison rather than a conversion from a different field.

Academic and Professional Career

After his doctorate he spent two years as a postdoctoral fellow with Roger Ratcliff at Northwestern University, working on lexical decision, response time modelling, and inferential statistics, then a further year at Amsterdam with Han van der Maas and Peter Molenaar on catastrophe-theoretic models of phase transitions. The Dutch funding system then carried him through a sequence of personal grants — a Veni (2004–2007), a Vidi (2007–2012), and a Vici (2017–2023) from the Netherlands Organisation for Scientific Research — interleaved with European Research Council support: a Consolidator grant (2012–2017) and two Advanced grants (2018–2023 and 2025–2030).

In 2012 he was appointed full professor of Bayesian Methodology at the University of Amsterdam's Psychological Methods Unit, with an honorary chair at Groningen until 2016. That year also saw the first release of JASP (Jeffreys's Amazing Statistics Program), which his lab has led ever since; he chaired the JASP Foundation from 2023 to 2025 and, since 2025, has been CEO of JASP Services BV, a company providing commercial support around the open-source program. In 2025 he helped found the diamond open-access Journal of Robustness Reports and became a senior fellow of its editorial college. He chairs Stichting Skepsis, the Dutch scientific skepticism foundation. His research group received the Ammodo Science Award in 2024.

Major Contributions to Psychology

Bayes factors as a replacement for the p-value

Wagenmakers' central technical claim is that the Bayes factor — the ratio of the marginal likelihood of the data under two competing hypotheses — answers the question researchers actually have. It can quantify evidence for a null hypothesis, not merely failure to reject it; it can be monitored as data collection proceeds without the inferential penalty that optional stopping imposes on frequentist tests; and it makes the analyst's assumptions explicit in the form of a prior over effect size. His 2007 paper proposed the BIC approximation as a practical bridge, and a long series of tutorials, the most cited of which is Bayesian Inference for Psychology, Part I: Theoretical Advantages and Practical Ramifications (2018), set out how to use the machinery and how to report it. The default specification he advocates, a Cauchy prior on effect size sometimes called the JZS prior after Jeffreys, Zellner, and Siow, is the one JASP produces.

The Bem reanalysis

The paper that made his reputation outside methodology was Why Psychologists Must Change the Way They Analyze Their Data: The Case of Psi (2011), with Ruud Wetzels, Denny Borsboom, and Han van der Maas. Daryl Bem had published nine experiments supporting precognition in a top-tier journal, each analysed with conventional t-tests. The Dutch group reanalysed them with default Bayes factors and found the evidence far weaker than the p-values suggested — in several studies, the data were barely diagnostic at all. The paper's larger argument was about editorial standards: a result that a field cannot believe should not be publishable because a p-value crossed a threshold.

JASP

JASP is a free, open-source program that presents both classical and Bayesian analyses through a graphical interface, with the Bayesian options sitting alongside the familiar ones rather than behind a separate toolchain. It is the main reason a generation of graduate students met Bayes factors before meeting any argument about them, and Wagenmakers has been explicit that lowering the barrier was the point.

Replication and open science

Wagenmakers was one of the coordinators of the Reproducibility Project: Psychology, whose 2015 report in Science found that only 36 of 100 original findings replicated significantly, a result that reframed the discipline's conversation about its own evidence base. He co-authored A Manifesto for Reproducible Science (2017), which catalogued the practices that inflate false positives, and was one of the 72 authors of Redefine Statistical Significance (2018), which proposed lowering the default alpha threshold for new discoveries from .05 to .005. He has advocated pre-registration of analysis plans and, more recently, the registered report format, and served as an action editor at Psychological Science and associate editor at Psychonomic Bulletin & Review.

Cognitive modelling and other work

Alongside the methodological work he has continued to publish on response time modelling, reinforcement learning, and long-range correlations in psychological time series. With František Bartoš and colleagues he ran a crowdsourced study of 350,757 coin flips showing that a tossed coin is very slightly more likely to land on the same face it started on — work that won an Ig Nobel Prize in 2024.

Major Publications

Wagenmakers, E.-J. (2007). A practical solution to the pervasive problems of p values. Psychonomic Bulletin & Review, 14(5), 779–804.
Wagenmakers, E.-J., Wetzels, R., Borsboom, D., & van der Maas, H. L. J. (2011). Why psychologists must change the way they analyze their data: The case of psi. Journal of Personality and Social Psychology, 100(3), 426–432.
Open Science Collaboration. (2015). Estimating the reproducibility of psychological science. Science, 349(6251), aac4716.
Wagenmakers, E.-J., Marsman, M., Jamil, T., et al. (2018). Bayesian inference for psychology. Part I: Theoretical advantages and practical ramifications. Psychonomic Bulletin & Review, 25(1), 35–57.
Benjamin, D. J., Berger, J. O., Johannesson, M., et al. (2018). Redefine statistical significance. Nature Human Behaviour, 2(1), 6–10.
Lee, M. D., & Wagenmakers, E.-J. (2013). Bayesian Cognitive Modeling: A Practical Course. Cambridge University Press.

Impact and Legacy

Bayesian methods were marginal in psychology twenty years ago and are now ordinary, taught in graduate methods courses and available in point-and-click software, and Wagenmakers is the single largest reason for both changes. The wider effect has been to shift the discipline's default posture toward its own evidence: pre-registration, registered reports, and data sharing moved from eccentric to expected during the same decade, and he was a consistent and unusually constructive voice in that transition.

The default prior is the weak point, and it has been criticised precisely. A Bayes factor is a comparison of two models, and the answer depends on what is assumed about effect size under the alternative. Bem, Utts, and Johnson (2011) showed that the default JZS prior Wagenmakers applied to Bem's data assigned roughly half its mass to precognition working in the opposite direction and a quarter to effect sizes above one standard deviation, leaving only a quarter for the small effects in the predicted direction that any proponent would expect. Specify a prior that takes the hypothesis seriously, they argued, and the Bayes factors show more support for psi than the original t-tests did. Uri Simonsohn made the general point in a paper titled "The Default Bayesian Test Is Prejudiced Against Small Effects." Ulrich Schimmack's powergraph analyses pushed further, arguing that to reproduce observed replication rates the default prior has to imply that over 90% of hypotheses tested in psychology are false with a false-positive rate near 50%, and that it predicts far more very large test statistics than real data contain.

The Bem reanalysis is not the demonstration it is often taken for. Schimmack's related point is uncomfortable: pooling Bem's nine studies and applying Wagenmakers' own default prior yields a Bayes factor rejecting the null by roughly a billion to one. If the data are taken at face value, the Bayesian analysis does not rescue the field from Bem — it convicts him more strongly than the p-values did. The reanalysis is best read as an argument about what a single small experiment can establish, not as a proof that the effect is absent.

Bayes factors do not, by themselves, license a claim that an effect is zero. Because a Bayes factor compares the null with one specified alternative, evidence favouring the null is evidence against that alternative, not evidence that the true effect size is exactly zero — there is always another alternative that fits better. Tendeiro and Kiers (2019) make the point directly. The tools that do the job people want are equivalence tests and region-of-practical-equivalence procedures, and Wagenmakers' own writing sometimes blurs the distinction in the direction of his preferred statistic.

"Redefine statistical significance" drew immediate organised opposition. Wagenmakers was a co-author of the 2018 proposal to move the default alpha to .005. Lakens and colleagues responded with Justify Your Alpha, signed by dozens of methodologists, arguing that a new universal threshold simply relocates the ritual rather than ending it; the broader "retire statistical significance" movement argued that dichotomised inference is the problem and that no alpha level fixes it. A decade on, .05 remains overwhelmingly dominant.

The Reproducibility Project's headline number is contested. The 36% replication rate was challenged by Gilbert, King, Pettigrew, and Wilson (2016), who argued that many replications were underpowered and that some deviated from the original protocols; the project's authors replied that the deviations were minor and documented. The exchange has not fully settled, and the estimate is often cited with more confidence than it warrants.

Making the software easy has a cost. JASP's success means that many users run default analyses without engaging the prior specification at all, which reproduces with Bayes factors the same unreflective routine that significance testing encouraged — except that the hidden assumption is now a distribution over effect size rather than an alpha level.

Learn More

References

Wagenmakers, E.-J. (2007). A practical solution to the pervasive problems of p values. Psychonomic Bulletin & Review, 14(5), 779-804.