Spearman's Rank Correlation

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

• Spearman's rank correlation among different alignment metrics (CKA, SVCCA, Mutual k-NN, CKNNA) over 78 vision models is high across variants, with all p-values below 2.24×10^-105finding0.782
  Validates robustness of alignment metric choice
• Kendall's tau rank correlationmethod0.780
  Used to measure alignment between human judgments and LLM-based scores in validation
• Average Spearman correlation of Elo trait rankings between all three models increases to 0.87 after loving constitution character trainingfinding0.773
  Post-training inter-model convergence in trait preferences demonstrating persona convergence
• Rank-Order Correlation (Kendall's rho)method0.764
  Statistical method used by Yodan Rose to measure agreement between different people's neighborhood diagnoses.
• Full-vector regression over 32 traits reaches leave-one-out Spearman correlations ~0.58-0.61 depending on layer windowfinding0.725
  Vector geometry features provide moderate secondary signal beyond elicitation-only screen
• Claude achieves significantly higher Spearman correlation predicting feature activations vs neuron activationsfinding0.724
  Automated interpretability analysis of activations confirms features are more interpretable than neurons
• GPT-4.1 verbalized probabilities for US states show Pearson r=0.741, Spearman ρ=0.896 correlation with corpus-based reference frequenciesfinding0.711
  Indicates verbalized probabilities contain meaningful distributional information for constrained answer spaces
• Five independent LLM scorers from four labs produce identical rankings (Spearman ρ > 0.8).finding0.706
  Scorer bias validation: Claude Haiku, Gemini Flash, GPT-5.4, Grok 4, Kimi K2.5 all converge on same model ordering.