Memo: Does CA still separate riders after accounting for Q28?

Research memorandum — residual CA contrasts stratified by ride-share days

Author

Jack J. Burleson

Published

Invalid Date

Research question: After accounting for ride-share days (Q28), does group/interpersonal CA still separate regular from non-regular public-transit riders?

Parent open questions: transit_riders_ca.qmd · ca_scores_predict_transit.qmd

1 Answer, Response, + Summary of Results

On Q28-complete rows (n = 241), we (i) replicate the overall transit→CA (McCroskey, 1970) Welch contrast, (ii) compare nested RFs ((Breiman, 2001); CA only / Q28 only / CA+Q28), and (iii) run within-Q28-stratum Welch tests.

Short answer: The mean CA gap among regular riders remains real overall, but for prediction CA adds almost nothing once Q28 is present (+0.021 AUC). Within Q28 strata, CA differences are heterogeneous—sometimes reversed (Figure Figure 1).

Figure 1.. Residual CA nested AUCs

1.1 Overall transit → CA (replication)

Table Table 1 reports the overall transit → CA Welch contrasts.

Table 1.. Residual CA after controlling for Q28.
Score Mean regular Mean not Δ Cohen’s d Welch p
Group CA 13.04 15.76 −2.72 −0.46 <.001
Interpersonal CA 13.31 15.04 −1.73 −0.30 .020

1.2 Nested predictive models

Table Table 2 reports the nested predictive models.

Table 2.. Model / ROC-AUC.
Model ROC-AUC
CA + Q28 0.783
Q28 only 0.762
CA only 0.590
Chance 0.500

CA incremental over Q28 ≈ +0.021.

1.3 Within-Q28 strata (exploratory)

Table Table 3 reports the within-Q28-stratum contrasts.

Table 3.. Q28 level / Group CA Δ (reg − not) / Welch p.
Q28 level Group CA Δ (reg − not) Welch p Note
Never −4.83 .006 Regulars lower CA
0–1 days/month +4.49 .033 Regulars higher CA
2–4 days/month +0.11 .92 Null
4–8 days/month −2.62 .22 Underpowered
8+ days/month — — Nearly all regular

1.4 Interpretation

  1. Third-variable concern from the transit→CA memo is partly justified for prediction: Q28 absorbs most ranking signal.
  2. The descriptive CA gap is not a pure Q28 artifact—contrasts persist in the Never stratum—but the 0–1 days reversal warns against a single global CA–transit story.
  3. For persona-tier design, ride-share exposure remains the higher-value mobility cue; CA should not be treated as an independent transit classifier once Q28 is known.

Sources: ca-personas followup-experiments --experiments residual_ca_q28 · outputs/followup_experiments/residual_ca_q28/ · formal write-up docs/secondary_rq_followup_experiments.md

2 What questions or uncertainties remain?

Are stratum reversals stable under alternate Q28 groupings or a formal interaction model (CA × Q28)? Larger samples would be needed for confirmatory stratum-wise inference.

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