Memo: Do CA scores add to Q28 and car access?

Research memorandum — joint psych + mobility Random Forest (complete-case)

Author

Jack J. Burleson

Published

Invalid Date

Research question: On a complete-case frame, how do group/interpersonal CA ((McCroskey, 1970)), ride-share days (Q28), and car access (Q21) jointly and separately predict regular transit?

Companions: ca_scores_predict_transit.qmd · q28_conditioned_on_car.qmd · residual_ca_after_rideshare.qmd

1 Answer, Response, + Summary of Results

Complete cases for CA + Q28 + Q21 yield n = 149. Nested mixed Random Forests ((Breiman, 2001)) on this shared frame (seed=42):

Short answer: The joint model is strong (AUC ≈ 0.736), but it is a mobility model—Q28 and Q21 dominate. CA alone on this subset is below chance (≈ 0.471), showing how sample restriction and mobility overlap can erase the modest full-cohort CA signal (≈0.590 on n=241). Figure Figure 1 shows the nested AUC comparison.

Figure 1.. CA + Q28 + car nested AUCs

1.1 Nested CV performance (common n = 149)

Table Table 1 reports the nested CV ROC-AUCs.

Table 1.. Joint CA and mobility predicting regular transit.
Model ROC-AUC
CA + Q28 + Q21 0.736
Q28 only 0.665
CA + Q28 0.657
Q21 only 0.631
CA only 0.471
Chance 0.500

Permutation importance (joint model): Q28 ≈ 0.20 · Q21 ≈ 0.14 · interpersonal CA ≈ 0.08 · group CA ≈ 0.07.

1.2 Interpretation

  1. Answers the follow-ups memo’s psych-vs-mobility question without multiple imputation: mobility wins decisively on the overlapping frame.
  2. CA’s full-cohort AUC should not be compared naively to rideshare AUCs computed on different complete-case sets—CA collapses when the sample is the car-complete subset.
  3. Adding CA to Q28 alone does not beat Q28 (0.657 < 0.665); CA helps only once car access is also present (joint 0.736).

Sources: ca-personas followup-experiments --experiments ca_q28_car · outputs/followup_experiments/ca_q28_car/ · formal write-up docs/secondary_rq_followup_experiments.md

2 What questions or uncertainties remain?

Is the CA collapse on n=149 selection (who answers car items) or true redundancy with mobility? See also residual-CA strata in residual_ca_after_rideshare.qmd.

References

Breiman, L. (2001). Random forests. Machine Learning, 45(1), 5–32. https://doi.org/10.1023/A:1010933404324
McCroskey, J. C. (1970). Measures of communication-bound anxiety. Speech Monographs, 37(4), 269–277. https://doi.org/10.1080/03637757009375677