The Replication Crisis and Epistemic Justification
A psychology department reviews a widely cited 2011 study reporting that a brief intervention improved exam performance. The original sample was 200 students; a 2023 replication with 1,200 students finds an effect close to zero. The original authors say the replication used a different population and a modified intervention, so it does not count as a failed replication. A junior researcher, Sam, argues that the original finding should now be treated as unjustified. A senior professor, Dr. Vogel, replies that scientific knowledge is cumulative and that one failed replication cannot overturn a published result. A third colleague asks what it would mean for a belief to be justified in science at all, given that no single study is decisive.
1. Using a reliabilist account of justification, assess Sam's and Dr. Vogel's positions.
Read the worked answer
Reliabilism holds that a belief is justified when it is produced by a process that reliably yields true beliefs. A single small study is a weak process: low power, flexible analysis, and publication bias make it unreliable. A large pre-registered replication is a more reliable process. On reliabilist grounds Sam has the stronger case: the belief that the intervention works is now produced by a process (weighing a failed large replication against a small original) that is less truth-conducive than before.
Dr. Vogel's appeal to cumulativeness is not itself a reliability argument; it is a claim about how science as a social institution converges. That may be true over time, but it does not show that the current belief is justified. Reliabilism is also sensitive to the modified-intervention objection: if the replication tested a different intervention, its evidential weight falls, and Sam's conclusion would need qualification.
2. The third colleague asks what justification means in science given that no single study is decisive. Offer a coherent account and apply it to this case.
Read the worked answer
One coherent account is evidence-based justification: a scientific claim is justified to the degree that the total available evidence, weighted by study quality, supports it. Justification is graded and defeasible, not all-or-nothing. On this account the question is not whether the 2023 study alone refutes the 2011 result, but how the body of evidence now stands.
Applied here, the original study carries less weight because of its sample size and the replication's null result; the replication carries more weight if it was pre-registered and well powered, less if the intervention genuinely differed. The honest conclusion is that the claim is now weakly justified at best and should not be cited as established. This also explains Dr. Vogel's point without conceding it: cumulation is real, but it can move downward as well as upward.