From Epistemic Justification to Methodological Decision: The Epistemicization of Hypotheses and the Status of H₀–H₁ in Scientific Research

Science is often imagined as a straight line leading from a question to a single, undeniable truth. In the laboratory, a researcher proposes an idea, gathers data, and hopes to prove that idea correct. For over a century, philosophers have debated how we can ever be sure that our scientific ideas are actually true, focusing on concepts like verification—checking if something matches reality—and falsification—trying to prove something wrong. These debates have shaped how we think about knowledge, but they have left a gap when it comes to the actual, daily work of scientists. While philosophers argue about the nature of truth, researchers in fields ranging from psychology to agriculture are busy making decisions based on data that is never perfectly clear. They face uncertainty, small sample sizes, and random noise, and they need a way to move forward without waiting for absolute certainty. This is where the tools of statistical decision-making come in, specifically a system developed in the twentieth century that allows scientists to act even when they cannot be 100% sure of the answer.

A new paper by Bülent Bayat, a researcher at Nuh Naci Yazgan University, investigates a deep confusion that has grown between these two worlds: the philosophical search for truth and the practical need to make decisions. The paper argues that scientists and textbooks have been making a fundamental mistake by treating statistical tools as if they were philosophical proofs. For decades, researchers have used a standard method to test their ideas, often labeling their predictions as "hypotheses" and then declaring them "proven" or "falsified" based on a statistical test. Bayat suggests that this is a category error. The statistical system in question, known as the Neyman–Pearson decision theory, was never designed to tell us what is true in the universe. Instead, it was built to help groups of scientists make consistent, rational choices about what to do next when they are facing uncertainty.

The core of the problem lies in how we talk about the two main players in this statistical game. In the standard approach, a researcher sets up a "null hypothesis," which is essentially a default assumption that nothing interesting is happening, and an "alternative hypothesis," which is the researcher's own idea that something is happening. When the data comes in, the researcher decides whether to reject the default assumption in favor of their own idea. Bayat points out that in the original design of this system, these two ideas were not meant to be competing claims about reality. They were not meant to be symmetric opposites where one is true and the other is false. Instead, they were designed as two different paths for a decision-making process. The goal was not to find the truth, but to control the risk of making a mistake. The system allows scientists to say, "If we assume nothing is happening, the chance of seeing data this extreme is very low, so we will choose to act as if something is happening." This is a rule for behavior, not a statement of fact.

Despite this clear design, the way the method is taught and used has drifted far from its original intent. Textbooks and research papers frequently describe the rejection of the default assumption as "evidence" that the researcher's idea is true. They speak of "accepting" a hypothesis as if it has been verified by the universe. Bayat traces this confusion back to a long history of philosophical debates that focused heavily on how to justify knowledge. Because philosophers spent so much time arguing about verification and falsification, scientists began to assume that their statistical tests were the practical application of those philosophical theories. They started treating the statistical output as a direct verdict on the truth of their theories. The paper argues that this is a misunderstanding. The statistical test does not tell you if your theory is true; it only tells you whether the data you collected is unusual enough to trigger a pre-agreed rule for changing your course of action.

This confusion has led to a ritualized way of doing science that can be misleading. Researchers often write out long lists of numbered hypotheses, from H1 to H14, treating each one as a separate claim to be proven or disproven. They use language like "the hypothesis was supported" or "the null hypothesis was falsified," as if the statistical test has delivered a final judgment on reality. Bayat explains that this language is dangerous because it hides the fact that the test is actually just a tool for managing risk. When a scientist says they have "falsified" a null hypothesis, they are not saying they have proven the alternative to be true. They are simply saying that they have decided to follow a specific path because the data was unlikely under the default assumption, and they are willing to accept a small, pre-defined chance that they might be wrong. The system is designed to work well over many, many studies, ensuring that the group of scientists as a whole makes few errors, not to guarantee that any single study has found the absolute truth.

The paper also looks at how this confusion affects the way we understand the role of the researcher. In the philosophical view, the scientist is a seeker of truth, trying to build a picture of the world that is as accurate as possible. In the decision-theory view, the scientist is a manager of risk, trying to make a choice that keeps the long-term error rate low. These are two different jobs. By mixing them up, we ask statistical tools to do a job they were never built for. We ask a decision rule to tell us what is real, when it was only designed to tell us when to act. This leads to a situation where we demand more from our data than it can give, and we end up with scientific conclusions that feel certain but are actually just the result of a mechanical process.

Bayat's work does not suggest that we should stop using these statistical tools. On the contrary, the paper argues that these tools are essential for modern science because they allow us to move forward despite uncertainty. The problem is not the tool itself, but how we interpret its results. The solution proposed is to stop trying to force statistical decisions into the box of philosophical truth. Instead, we should recognize that these tools belong to a different category: they are methods for regulating how we act when we don't know the answer. If we can separate the question of "what is true" from the question of "what should we do next," we can use these tools more effectively. We can stop pretending that a statistical test has proven a theory and start understanding that it has simply helped us make a rational choice under difficult conditions.

The paper concludes that the history of science has created a divide between the philosophers who worry about the meaning of knowledge and the statisticians who worry about the mechanics of decision-making. For a long time, these two groups have been talking past each other, leading to the current confusion where scientists think their statistical tests are philosophical proofs. Bayat suggests that the way forward is not to try to fix the statistics to make them more philosophical, but to clearly distinguish between the two levels. We need to understand that the H0 and H1 structure is a method for organizing our decisions, not a map of reality. By accepting this distinction, scientists can use their tools with more honesty and clarity, acknowledging that they are making decisions based on risk and uncertainty, rather than claiming to have discovered absolute truths. This shift in perspective does not weaken science; it strengthens it by aligning our methods with their actual purpose.