πŸ”„ Flip Point Calculator

Bayes Factor Robustness Diagnostic

Based on "The Bayes Factor Flip Point" β€” Lovric (2025)

πŸ“Š Input Parameters

The observed z-statistic from your test

πŸ“š Load Example Dataset:

Click an example to see different robustness scenarios

πŸ“š Load Example Dataset:

Click to load a real-world example scenario

πŸ“š Load Example Dataset:

Click to load a real-world example scenario

πŸ“š Load Example Dataset:

Click to load a real-world example scenario

⚠️ Approximation Note: This calculator uses a t-test approximation (t = √F) for ANOVA, which is exact only for k = 2 groups. For k > 2 groups, the flip point is approximate. Future versions will implement the exact Bayes factor for multi-group ANOVA.

πŸ“š Load Example Dataset:

Click to load a real-world example scenario

πŸ“š Load Example Dataset:

Click to load a real-world example scenario

πŸ“š Load Example Dataset:

Click to load a real-world example scenario

πŸ“š Load Example Dataset:

Click to load a real-world example scenario

πŸ“– Understanding the Flip Point

The flip point (r* or Ο„*) is the critical prior scale where the Bayes factor equals exactly 1:

BF₀₁(r*) = 1

This means:

  • When r < r* β†’ BF₀₁ < 1 β†’ Evidence favors H₁ (alternative)
  • When r = r* β†’ BF₀₁ = 1 β†’ Evidence is perfectly balanced
  • When r > r* β†’ BF₀₁ > 1 β†’ Evidence favors Hβ‚€ (null)

Why does this happen?

  • Small r = tight prior β†’ concentrates probability near observed effect β†’ H₁ looks good
  • Large r = diffuse prior β†’ spreads probability over implausible values β†’ H₁ wastes probability mass β†’ Hβ‚€ wins (Lindley's paradox)

Interpreting the flip point (thresholds based on published Bayesian analyses, van Doorn et al. 2021; Heck et al. 2022):

  • r* < 1: Fragile for H₁ β€” even conservative defaults would change conclusion
  • 1 < r* < 3: ⚠️ Sensitive β€” lies within range of defensible prior choices
  • r* > 3: Robust for H₁ β€” only implausibly diffuse priors favor Hβ‚€

Justification: Surveys of JASP usage show most analyses use r ∈ [0.5, 1.0]; r > 2 implies expecting |d| > 2 as typical, which is unrealistic for most psychology research.

πŸ“ˆ Results

β€”
Flip Point
Test Statistic
β€”
P-value (two-tailed)
β€”
BF₀₁ at default (r=0.707)
β€”
Degrees of Freedom
β€”