Skip to content
No Vape

A Bayes factor compares specified models, not your chance of quitting

Check the direction, alternative prediction and prior assumptions before interpreting a Bayes factor in vaping-related research.

No Vape · Updated

Disposable vape resting on concrete
Illustrative image · Markus Winkler / Unsplash

Write the two competing predictions in words

A report may compare a no-effect model with a particular directional alternative. That is narrower than asking whether any possible difference exists. Read the methods for the predicted direction and magnitude assumptions. Do not silently replace a specified alternative with every outcome you can imagine after seeing the data.

Dienes’ original methodological article explains Bayes factors as relative evidence for two theories, with the alternative represented sufficiently to predict data. It distinguishes this comparison from posterior odds, which also incorporate prior model odds. A Bayes factor alone therefore does not supply an unconditional percentage that one explanation is correct.

Dienes, Frontiers in Psychology 5:781 (2014): Using Bayes to get the most out of non-significant results

Check which way the ratio runs

If the reported comparison is alternative over null, values above one favor that alternative relative to the null, while values below one favor the null relative to that alternative. A different numerator changes the reading. Look for the actual notation or explanation rather than assuming that a larger number always favors the intervention.

Imagine a fictional research note labeled model A over model B. Before describing the result, write what A predicts and what B predicts. If the caption reverses the order, your interpretation must follow that order too. The example is deliberately nonnumerical: it illustrates a direction check, not evidence about a real support program.

Dienes, Frontiers in Psychology 5:781 (2014): Using Bayes to get the most out of non-significant results

Keep the alternative’s assumptions attached

An original prospective study in England concerning dual e-cigarette use and smoking cessation calculated Bayes factors against an alternative of lower odds of quitting. Its methods described a half-normal distribution and an expected-effect assumption drawn from earlier research. We use those construction details only, without transferring its outcome findings or making a claim about quitting vaping.

That example shows why evidence relative to a particular expected negative association is not a verdict on every positive, negative or negligible effect. Your reading card can contain numerator model, denominator model, predicted direction, prior specification and stated interpretation. If one ingredient is missing from a summary, locate the original methods or leave it unresolved.

Jackson et al., Addictive Behaviors 103:106230 (2020): Associations between dual use of e-cigarettes and smoking cessation: A prospective study of smokers in England

Distinguish comparison from estimation and equivalence

A posterior credible interval describes uncertainty for a quantity under a model and prior. A Bayes factor instead compares predictive evidence between specified models. Neither label should be substituted for the other. Similarly, a nonsignificant p-value alone does not supply a Bayes factor or establish equivalence under a prespecified margin.

The methods source allows for data that do not distinguish the specified predictions clearly. Keep that possibility visible instead of forcing every analysis into a decisive winner. This worksheet does not choose priors, calculate evidence or validate a named study. Consequential health interpretation needs the full evidence and suitable expertise; a model comparison cannot become a personal success forecast.

Dienes, Frontiers in Psychology 5:781 (2014): Using Bayes to get the most out of non-significant results

Sources and scope

Sources inform the guidance. The worksheets and examples are No Vape editorial suggestions, not tested treatments.

  1. Dienes, Frontiers in Psychology 5:781 (2014) — Using Bayes to get the most out of non-significant results
  2. Jackson et al., Addictive Behaviors 103:106230 (2020) — Associations between dual use of e-cigarettes and smoking cessation: A prospective study of smokers in England
  3. No Vape — No Vape — app features and store links

Written with AI assistance and checked against the cited sources. No clinical review is claimed. This guide covers practical planning, not individual medical treatment. How we research and write