Do Q27 and Q28 predict regular public transit use?

Secondary research question — traditional ML write-up for transit intensity & ride-share days

Project: PSYCH 755 — CA persona / PRCA framework
Analysis code: src/ca_personas/transit_covariate_rf.py
CLI: ca-personas covariate-transit-rf --specs q27_intensity q28_days q27_q28 --join inner --seed 42
Artifacts: outputs/transit_covariate_rf/{q27_intensity,q28_days,q27_q28}/
Research memo: memos/q27_q28_predict_transit.qmd
Manuscript: index.qmd

1 Research question

Among matched Prolific↔︎Qualtrics respondents, do Q27 (public-transit rides on a typical use day) and Q28 (ride-share days in the last three months) predict regular public-transit use (Q26 weekly+) in a traditional Random Forest classifier—and which item carries the usable signal?

2 Methods

Sample: File A + File B inner-joined to File C; analytic base n = 241 with complete PRCA and usable Q26. Complete-case n = 239 (Q27), 241 (Q28), 239 (joint).
Outcome: regular_transit = Q26 ∈ {4-8 days a month, 8 or more days a month}.
Models: One-hot categorical features → balanced RandomForestClassifier (500 trees, min_samples_leaf=3), stratified 5-fold cross_val_predict, seed 42; Gini + permutation importance (scoring="roc_auc").
Benchmarks: chance AUC = 0.500; geo RF = 0.551; CA RF = 0.590; rideshare family Q28+Q29 = 0.745.

3 Results

Model	n	ROC-AUC	AP	Bal. acc.	F1
Q28 only	241	0.762	0.689	0.730	0.702
Q27 + Q28	239	0.761	0.690	0.696	0.667
Q28 + Q29	233	0.745	0.655	0.731	0.709
Q27 only	239	0.589	0.513	0.623	0.467
CA benchmark	241	0.590	—	—	—
Geo benchmark	241	0.551	—	—	—
Chance	—	0.500	—	0.500	—

Regular-transit prevalence rises from 15.7% (Q28 Never) to 94.1% (Q28 8+ days/month). Q27 shows a milder intensity gradient concentrated in the 1–2 rides bin. Joint-model permutation importance: Q28 ≫ Q27.

4 Interpretation

Q28 recovers CV ROC-AUC = 0.762 (+0.211 vs geo, +0.172 vs CA). Q27 recovers AUC = 0.589 and is redundant once Q28 is included (joint AUC = 0.761). Full prose: memos/q27_q28_predict_transit.qmd.

5 Reproducibility

ca-personas covariate-transit-rf \
  --specs q27_intensity q28_days q27_q28 \
  --join inner --seed 42 \
  --figures-dir memos/figures

--- title: "Do Q27 and Q28 predict regular public transit use?" subtitle: "Secondary research question — traditional ML write-up for transit intensity & ride-share days" --- **Project:** PSYCH 755 — CA persona / PRCA framework **Analysis code:** [`src/ca_personas/transit_covariate_rf.py`](../src/ca_personas/transit_covariate_rf.py) **CLI:** `ca-personas covariate-transit-rf --specs q27_intensity q28_days q27_q28 --join inner --seed 42` **Artifacts:** `outputs/transit_covariate_rf/{q27_intensity,q28_days,q27_q28}/` **Research memo:** [`memos/q27_q28_predict_transit.qmd`](../memos/q27_q28_predict_transit.qmd) **Manuscript:** [`index.qmd`](../index.qmd) --- ## Research question Among matched Prolific↔Qualtrics respondents, do **Q27** (public-transit rides on a typical use day) and **Q28** (ride-share days in the last three months) predict **regular** public-transit use (`Q26` weekly+) in a traditional Random Forest classifier—and which item carries the usable signal? ## Methods - **Sample:** File A + File B inner-joined to File C; analytic base n = 241 with complete PRCA and usable `Q26`. Complete-case n = 239 (Q27), 241 (Q28), 239 (joint). - **Outcome:** `regular_transit` = `Q26` ∈ {`4-8 days a month`, `8 or more days a month`}. - **Models:** One-hot categorical features → balanced `RandomForestClassifier` (500 trees, `min_samples_leaf=3`), stratified 5-fold `cross_val_predict`, seed 42; Gini + permutation importance (`scoring="roc_auc"`). - **Benchmarks:** chance AUC = 0.500; geo RF = 0.551; CA RF = 0.590; rideshare family Q28+Q29 = 0.745. ## Results | Model | n | ROC-AUC | AP | Bal. acc. | F1 | |---|---:|---:|---:|---:|---:| | Q28 only | 241 | **0.762** | 0.689 | 0.730 | 0.702 | | Q27 + Q28 | 239 | **0.761** | 0.690 | 0.696 | 0.667 | | Q28 + Q29 | 233 | 0.745 | 0.655 | 0.731 | 0.709 | | Q27 only | 239 | **0.589** | 0.513 | 0.623 | 0.467 | | CA benchmark | 241 | 0.590 | — | — | — | | Geo benchmark | 241 | 0.551 | — | — | — | | Chance | — | 0.500 | — | 0.500 | — | Regular-transit prevalence rises from 15.7% (Q28 Never) to 94.1% (Q28 8+ days/month). Q27 shows a milder intensity gradient concentrated in the 1–2 rides bin. Joint-model permutation importance: Q28 ≫ Q27. ## Interpretation Q28 recovers CV ROC-AUC = **0.762** (+0.211 vs geo, +0.172 vs CA). Q27 recovers AUC = **0.589** and is redundant once Q28 is included (joint AUC = 0.761). Full prose: [`memos/q27_q28_predict_transit.qmd`](../memos/q27_q28_predict_transit.qmd). ## Reproducibility ```bash ca-personas covariate-transit-rf \ --specs q27_intensity q28_days q27_q28 \ --join inner --seed 42 \ --figures-dir memos/figures ```