Memo: What predicts Q27 intensity among regular riders?

Research memorandum — within-rider transit-day intensity classifiers

Author

Jack J. Burleson

Published

Invalid Date

Research question: Among weekly+ public-transit riders, which covariates predict high transit-day intensity (Q27 ≥ 3–4 rides on a typical use day)?

Parent open question: q27_q28_predict_transit.qmd

1 Answer, Response, + Summary of Results

Among 101 regular riders, 35 (34.7%) report high intensity (3–4 / 5–6 / 7+ rides); 66 report the modal 1–2 rides. We treat high intensity as the binary target and evaluate candidate Random Forests ((Breiman, 2001); seed=42).

Short answer: Nothing predicts intensity well. The best family (Age+Sex+Student, n=95) reaches only AUC ≈ 0.549—near chance. CA (McCroskey, 1970), Q28, car access, and employment all fail as within-rider intensity classifiers (Figure Figure 1).

Figure 1.. Q27-among-riders comparison

1.1 Q27 distribution among weekly+ riders

Table Table 1 shows the Q27 distribution among weekly+ riders.

Table 1.. Q27 intensity among regular riders.
Q27 n
1–2 rides in a typical day 66
3–4 rides in a typical day 31
5–6 rides in a typical day 3
7 or more rides in a typical day 1

1.2 Candidate predictors of high intensity

Table Table 2 reports the candidate models for high intensity.

Table 2.. Model / n / ROC-AUC.
Model n ROC-AUC
Age + Sex + Student 95 0.549
Q28 days 101 0.529
Car access (Q21) 58 0.506
Group + interpersonal CA 101 0.496
Employment status 101 0.447
Chance — 0.500

1.3 Interpretation

  1. Answers the Q27/Q28 memo’s within-subgroup question: Q27 is not a useful secondary forecasting target among already-regular riders in this cohort.
  2. Combined with wave-1 results (Q27 alone AUC ≈ 0.589 for the weekly+ threshold), intensity items are weak both as predictors of and as outcomes within regular ridership.
  3. Persona prompts that lean on “rides per day” without ride-share days are emphasizing a low-signal mobility facet.

Sources: ca-personas followup-experiments --experiments q27_among_regular · outputs/followup_experiments/q27_among_regular/ · formal write-up docs/secondary_rq_followup_experiments.md

2 What questions or uncertainties remain?

Would continuous or ordinal models of Q27 (instead of a high/low split) recover more signal, or is intensity simply noisy self-report in this instrument?

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