Reading focus
Syllabus: What Can Statistical Evidence Tell Us?
Distinguish the alternative’s predicted effect distribution from prior odds between models.
Why you’re reading it. Explain what relative support adds to the interpretation of an uncertain result.
A question to consider. What range and direction of gains would the school’s intervention theory predict, and how could changing that prediction change its comparison with a null model?
Why this reading
Dienes distinguishes evidence favoring a null model from data that do little to discriminate between models. The examples make the alternative’s predictions an explicit part of the evidential comparison.
Before you read
A Bayes factor compares how well two models predict the observed data. Prior distributions within a model describe possible effect sizes; prior odds between models are a different ingredient needed to obtain posterior model odds.
Read for
- Examine how substantive expectations motivate the alternative’s predicted range, direction and scale.
- Ask whether reasonable changes to those predictions would change the interpretation of a Bayes factor.
Consider afterwards
- Two analysts use the same data but different plausible effect distributions. What would you ask them to show before accepting either evidential conclusion?
Editorial context
This is a tutorial and an argument for Bayesian analysis, not proof that it is always preferable. Its conventional evidence thresholds are not universal decision rules. A Bayes factor favors one specified model relative to another; both may be inadequate. The author cautions against choosing Bayesian analysis merely after obtaining a nonsignificant p value. Historical software examples may be obsolete.