Memo: Does employment status predict regular transit?

Research memorandum — employment follow-up to the geography memo

Author

Jack J. Burleson

Published

Invalid Date

Research question: Does employment status predict whether a matched respondent takes public transportation regularly?

Parent memo: geo_predicts_transit.qmd
Comparison memo: transit_covariate_followups.qmd

1 Answer, Response, + Summary of Results

Using the full matched analytic cohort (n = 241; 101 regular / 140 not regular), we evaluated Employment status (Full-Time / Part-Time / Other) as a single-feature Random Forest ((Breiman, 2001)) predicting weekly+ public transit (Q26 threshold as in the geo and CA memos). Stratified 5-fold CV, random_state=42.

Short answer: Barely. Employment status yields CV ROC-AUC = 0.528 — only +0.028 above chance and below both the geo (0.551) and CA (0.590) benchmarks. It is not a competitive stand-alone predictor in this sample. Figure Figure 1 shows prevalence by level and the ROC curve.

Figure 1.. Employment prevalence by level and ROC curve

Descriptive associations. Table Table 1 reports employment prevalence.

Table 1.. Random Forest performance for employment predicting regular transit.
Employment status n % regular
Part-Time 31 48.4%
Full-Time 148 45.3%
Other 62 30.6%

Part-time and full-time rates are close; the “Other” category is less often regular. With only three coarse levels, discrimination is inherently limited.

Random Forest (stratified CV). Table Table 2 reports the CV ROC-AUCs.

Table 2.. Model / n / ROC-AUC.
Model n ROC-AUC
Employment status RF 241 0.528
Geo RF benchmark 241 0.551
CA RF benchmark 241 0.590
Chance — 0.500

Balanced accuracy = 0.560 and Brier = 0.247. Permutation importance mean \(\Delta\)AUC = 0.059.

Conclusion. Employment status, as coded in the Prolific export, recovers AUC = 0.528 for weekly+ transit. It remains useful as a control/covariate in multi-feature models, but it does not outperform geography (0.551) or CA (0.590) on its own.

Sources: notebooks/secondary_rq_employment_transit_rf.ipynb · src/ca_personas/transit_covariate_rf.py · ca-personas covariate-transit-rf --specs employment · github.com/Exios66/psych755-jjb · formal write-up docs/secondary_rq_transit_covariate_followups.md

2 What questions or uncertainties remain?

Would finer employment categories (student, gig work, remote vs on-site) recover more signal? Does employment interact with car access or urbanicity rather than acting as a main effect?

3 What other features may also well-predict regular public transit use?

Ride-share (Q28/Q29) recovers AUC = 0.745 and car access (Q20/Q21) recovers AUC = 0.607; see rideshare_predicts_transit.qmd and car_access_predicts_transit.qmd.

References

Breiman, L. (2001). Random forests. Machine Learning, 45(1), 5–32. https://doi.org/10.1023/A:1010933404324