Sequential hypothesis testing with Bayes factors: Efficiently testing mean differences.
read more
Citations
Bayesian inference for psychology. Part II: Example applications with JASP
Bayesian inference for psychology. Part I: Theoretical advantages and practical ramifications.
The Bayesian New Statistics: Hypothesis testing, estimation, meta-analysis, and power analysis from a Bayesian perspective
Bayesian Statistical Inference for Psychological Research
Bayes factor design analysis: Planning for compelling evidence
References
R: A language and environment for statistical computing.
Statistical Power Analysis for the Behavioral Sciences
Conducting Meta-Analyses in R with the metafor Package
False-Positive Psychology: Undisclosed Flexibility in Data Collection and Analysis Allows Presenting Anything as Significant
Related Papers (5)
Frequently Asked Questions (10)
Q2. What have the authors stated for future works in "Sequential hypothesis testing with bayes factors: e ciently testing mean di↵erences" ?
Among these, the authors wish to stress that it makes a commonly used procedure perfectly acceptable, which has been considered as questionable so far: While in NHST this option is taboo, using the SBF it can be done without any guilt. Not only it can be done, but doing so results in a more e cient research strategy, provided that some rules are followed. Meta-analysis of clinical trials with early stopping: An investigation of potential bias.
Q3. Why do the authors underestimate the true eect size?
Due to the Bayesian shrinkage of early terminations, meta-analytic aggregations of multiple SBF studies underestimate the true e↵ect size by 5-9%.
Q4. How many participants did the authors increase the sample size in order to keep the simulation time manageable?
In order to keep simulation time manageable, the authors increased the sample in several step sizes: +1 participant until n = 100, +5 participants until n = 1000, +10 participants until n = 2500, +20 participants until n = 5000, and +50 participants from that point on.
Q5. How many participants can be used to calculate a BF?
Run a minimal number of participants (e.g., nmin = 20 per group), increase sample size as often as desired and compute a BF at each stage (even after each participant).
Q6. What is the often-heard critique of Bayesian approaches?
One of the most often-heard critiques of Bayesian approaches is about the necessity to choose a prior distribution of the parameters (e.g., Simmons et al., 2011).
Q7. What is the way to fine-tune the rate of misleading evidence?
the choice of the minimal sample size before the optional stopping procedure is started is another parameter for fine-tuning the expected rate of misleading evidence.
Q8. What was the mean posterior eect size in the final sample?
The mean posterior e↵ect size in the final sample was Cohen’s d = 0.72, with a 95% highest posterior density (HPD) interval of [0.22; 1.21].
Q9. What was the critical BF for stopping the sequential sampling?
The minimum sample size was set to nmin = 20 in each group, and the critical BF10 for stopping the sequential sampling was set to 10 (resp. 1/10).
Q10. What is the problem with a p value of.08?
But as journals tend to reject non-significant results, a p value of .08 can pose a real practical problem and a conflict of interest for researchers.