Calibration

Thread Master 05-08-2026

Part of Thesis summary. Research question 1: how the model was built step by step, from a simple first model to the final 12-parameter model.

How the models are compared: Model evaluation metrics, Grouped cross-validation.

The models, in the order they were tried (all in one table: Model progression):

1. Model 1: base and R_buffer: 10 m vs 15 m
2. Model 2: share-type and sigma
3. Directional scope ablation (3a vs 3b)
4. Model 4a: alternative-specific parameters
5. Model 5: alternative-specific constants (rejected)
6. Interaction term screen (α = 1)
7. Alpha calibration, which revealed the The f_distance² artifact
8. Model 6, re-derived at calibrated alpha and Stepwise VIF elimination
9. f_directional: shared vs split
10. Result: Final model (Final model correlation matrix, Multicollinearity is structural)

Meetings: MNL model development, RQ1 rebuilt with grouped CV.

Linked from Alpha calibration · Assignment framework · Directional scope ablation (3a vs 3b) · Final model · Final model correlation matrix · Grouped cross-validation · Interaction term screen (α = 1) · Model 1: base · Model 2: share-type and sigma · Model 4a: alternative-specific parameters · Model 5: alternative-specific constants (rejected) · Model 6, re-derived at calibrated alpha · Model evaluation metrics · Model progression · Multicollinearity is structural · R_buffer: 10 m vs 15 m · Research questions · Stepwise VIF elimination · The f_distance² artifact · Thesis summary · f_directional: shared vs split