pith. machine review for the scientific record. sign in

arxiv: 2604.26236 · v1 · submitted 2026-04-29 · 💻 cs.DL

Recognition: unknown

Do E-Scooter Speed Governance Policies Reduce Harsh Acceleration and Deceleration? Evidence from 19.5 Million Trips Around a Regulatory Ban

SeongJin Choi , Sunbin Yoo , Sugie Lee
Authors on Pith no claims yet

Pith reviewed 2026-05-07 12:49 UTC · model grok-4.3

classification 💻 cs.DL
keywords e-scootersspeed governanceharsh accelerationharsh decelerationdifference-in-differencesfirmware removalmicromobility safetybehavioral offsetting
0
0 comments X

The pith

Firmware removal of ungoverned e-scooter mode reduces harsh acceleration and deceleration events through mechanical limits alone.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper tests whether e-scooter speed governance, implemented by withdrawing the ungoverned high-speed option, produces safety improvements beyond the direct speed cap by also lowering rates of harsh acceleration and harsh deceleration. Researchers examine 19.5 million GPS-tracked trips across 52 South Korean cities before and after the operator's December 2023 system-wide firmware change. A two-stage approach first predicts within-user reductions from pre-ban rider data using a heterogeneous logit model, then confirms those predictions with difference-in-differences estimates that pass parallel-trends checks. The estimates match in sign and size, show Bonferroni significance, and reveal no behavioral offsetting in within-user composition, pointing to a purely mechanical transmission of the governance effect.

Core claim

The causal estimates from the difference-in-differences specification around the December 2023 removal confirm the pre-ban predictions in sign and order of magnitude for both harsh-acceleration and harsh-deceleration events; the within-user composition check finds no behavioral offsetting, indicating that firmware removal of an ungoverned mode lowers both harsh-event margins through a purely mechanical channel.

What carries the argument

Two-stage predict-then-validate design that combines a rider-heterogeneous random-parameters binary logit on pre-ban data with difference-in-differences exploiting the operator's December 2023 system-wide removal of the ungoverned mode.

If this is right

  • Speed governance policies deliver measurable safety gains on unconstrained behavioral margins such as harsh acceleration and deceleration.
  • The reduction occurs without riders offsetting through changes in trip composition or other behaviors.
  • The mechanical speed cap directly alters acceleration and deceleration patterns even on trips that never reached the former top speed.
  • Results are robust across 19.5 million trips and 52 cities with Bonferroni-corrected significance.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Fleet operators could achieve similar safety margins on other micromobility vehicles by standardizing firmware limits rather than relying on rider education.
  • The absence of offsetting implies that riders treat the new speed ceiling as a binding constraint and adjust their riding style accordingly.
  • Urban planners might extend firmware-based governance to reduce crash risks in dense scooter-sharing zones without needing additional enforcement.

Load-bearing premise

The difference-in-differences specification satisfies the parallel-trends assumption and the December 2023 system-wide removal was the only relevant change affecting harsh-event rates during the study window.

What would settle it

A post-removal dataset in which harsh-event rates show no decline or in which within-user trip composition shifts toward more high-risk segments would falsify the mechanical-channel claim.

Figures

Figures reproduced from arXiv: 2604.26236 by SeongJin Choi, Sugie Lee, Sunbin Yoo.

Figure 1
Figure 1. Figure 1: Cross-city variation in pre-ban TUB intensity and post-ban harsh-event reduction. Left, November 2023 TUB trip share by city. Right, change in trip-level harsh-event probability (Dec − Nov 2023). Circle size is proportional to √ 𝑁trips. therefore provides broad coverage of the Korean shared-e-scooter population rather than a niche or operator-specific sample. During the study period (Feb–Nov 2023) Swing op… view at source ↗
Figure 2
Figure 2. Figure 2: Acceleration distribution and harsh-event threshold calibration. (a) Point-to-point acceleration distribution by speed mode with the ±0.5 m/s2 threshold (dashed lines). (b) Exceedance rate as a function of threshold. nothing but the governor differs across modes, any behavioral difference is attributable to speed governance rather than to vehicle characteristics, providing a quasi-experimental setting for … view at source ↗
Figure 3
Figure 3. Figure 3: Distribution of the continuous treatment dose TUBNov 𝑐 across 50 of the 52 study cities. Dashed line marks the sample-mean dose TUBNov = 0.542. Two cities (Jeju and Mokpo) are excluded because they had no recorded trips in November 2023. 3.5. Composition check Aggregate DiD effects may reflect composition shifts (former TUB users entering the STD/ECO pool) rather than a within-user reduction. To verify tha… view at source ↗
Figure 4
Figure 4. Figure 4: Trip-level user-FE parallel-trends event study (Sep–Nov 2023 pre-ban, Nov as reference) for the two harsh-event outcomes, with 95% city-clustered confidence intervals. Green marks the reference month, grey the pre-treatment months, and red the post-treatment month (Dec 2023). Vertical dotted line marks the policy break. 4.1.3. Cross-sectional pattern under the TUB ban. The fitted logit lets us summarize th… view at source ↗
Figure 5
Figure 5. Figure 5: Predict-then-validate concordance on the two harsh-event outcomes. The top row of each panel (“Aggregate”) reports the full-sample comparison, and subsequent rows split riders at the median of pre-ban experience, night-trip rate, and same-route rate. Grey diamonds show the Phase I model-implied reduction, and blue circles show the Phase II causal estimate with 95% city-clustered CI. 4.2.3. Heterogeneity an… view at source ↗
Figure 6
Figure 6. Figure 6: Spatial heatmap of trip-level harsh-event probability on a 2 km grid. Left, pre-ban (Sep–Nov 2023). Right, post-ban (Dec 2023). Color scale is the sample-average √ 𝑃 (any harsh event on trip), and circle area is proportional to 𝑁trips per cell. 6.1. External validity Two features support cautious generalization. The ban was hardware-identical (same chassis across all modes), so the causal estimand isolates… view at source ↗
read the original abstract

Do e-scooter speed governance policies yield behavioral safety gains beyond the mechanical cap they impose? A firmware ceiling mechanically prevents speeding, but whether the same riders also generate fewer harsh accelerations and harsh decelerations when the ungoverned mode is withdrawn remains open. We analyze 19.5 million GPS-instrumented trips from 52 South Korean cities (February to November 2023). Our two-stage predict-then-validate design targets two trip-level binary outcomes, any harsh-acceleration event and any harsh-deceleration event. In Phase~I, we predict each outcome's within-user reduction under an ungoverned-to-governed substitution, using a rider-heterogeneous random-parameters binary logit on the pre-ban period. In Phase~II, we validate these predictions using a difference-in-differences specification that exploits the operator's system-wide December~2023 removal of the ungoverned mode. The causal estimates confirm the Phase~I predictions in sign and order of magnitude on both outcomes, are Bonferroni-significant, and satisfy a 3-month pre-ban parallel-trends test. A within-user composition check finds no behavioral offsetting, indicating that firmware removal of an ungoverned mode lowers both harsh-event margins through a purely mechanical channel. These results imply that speed governance policies can deliver measurable safety gains on unconstrained behavioral margins.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript examines whether e-scooter speed governance policies reduce harsh acceleration and deceleration events beyond their mechanical speed cap, using 19.5 million GPS trips from 52 South Korean cities (Feb–Nov 2023). It deploys a two-stage predict-then-validate design: Phase I fits a rider-heterogeneous random-parameters binary logit on pre-ban data to predict within-user reductions in harsh-event probabilities under ungoverned-to-governed substitution; Phase II validates those predictions with a DiD specification that exploits the operator’s system-wide December 2023 removal of the ungoverned mode. The DiD estimates match the Phase I predictions in sign and magnitude, survive Bonferroni correction and a 3-month pre-ban parallel-trends test, and a within-user composition check finds no offsetting behavior, supporting the conclusion of a purely mechanical channel.

Significance. If the central claim holds, the work supplies credible evidence that firmware-based speed caps can improve safety on unconstrained behavioral margins without requiring rider adaptation. The pre-registered predict-then-validate structure, large-scale instrumented data, and explicit test of behavioral offsetting are methodological strengths that make the mechanical-channel interpretation falsifiable and policy-relevant for urban mobility governance.

major comments (2)
  1. [Phase II DiD specification] Phase II DiD (Section 4.2 and Table 5): the parallel-trends test is reported only for the three months immediately preceding the December 2023 ban; this window does not address potential December-specific aggregate shocks (weather, holidays, or operator changes) that could independently shift harsh-event rates and thereby produce a spurious match between DiD estimates and Phase I predictions.
  2. [Within-user composition check] Within-user composition check (Section 5.3): while the check rules out rider-selection effects, it does not test whether the observed reduction in harsh events is driven by the mechanical firmware change versus other time-varying factors that coincided with the system-wide policy rollout.
minor comments (2)
  1. [Abstract] The abstract states that results are 'Bonferroni-significant' but does not report the exact number of hypotheses or the precise correction threshold applied.
  2. [Data and sample construction] Data section: the criteria for excluding trips from the 19.5 million sample (e.g., GPS quality thresholds, minimum trip length) are not fully enumerated; a supplementary table listing exclusion counts by criterion would improve reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive feedback on our manuscript. We address each of the major comments below and have revised the manuscript accordingly to strengthen our analysis.

read point-by-point responses

  1. Referee: [Phase II DiD specification] Phase II DiD (Section 4.2 and Table 5): the parallel-trends test is reported only for the three months immediately preceding the December 2023 ban; this window does not address potential December-specific aggregate shocks (weather, holidays, or operator changes) that could independently shift harsh-event rates and thereby produce a spurious match between DiD estimates and Phase I predictions.

    Authors: We appreciate the referee pointing out the potential for December-specific shocks to confound the results. While our primary parallel-trends test focuses on the three months prior to the ban to closely mirror the treatment timing, we acknowledge that this may not fully capture all possible December effects. To address this, we have extended the parallel-trends analysis to include earlier periods and incorporated additional covariates for weather conditions, holidays, and any known operator changes in the revised DiD specification. Importantly, the predict-then-validate design provides a safeguard: the Phase I predictions are based exclusively on pre-ban data and rider behavior under the ungoverned mode, so any spurious December shock would have to coincidentally match the predicted mechanical effect in both sign and magnitude, which is improbable. We have updated Section 4.2, Table 5, and the appendix with these robustness checks. revision: yes

  2. Referee: [Within-user composition check] Within-user composition check (Section 5.3): while the check rules out rider-selection effects, it does not test whether the observed reduction in harsh events is driven by the mechanical firmware change versus other time-varying factors that coincided with the system-wide policy rollout.

    Authors: The referee is correct that the within-user composition check is designed to rule out changes in the rider pool or selection into trips. To distinguish the mechanical firmware effect from other concurrent time-varying factors, we depend on the DiD estimator, which accounts for common temporal shocks through the time fixed effects and the control group (if any, or within variation). The alignment between the DiD estimates and the Phase I predictions—where Phase I models the expected change from the mechanical speed cap alone—serves as evidence that the reduction is attributable to the firmware change rather than unrelated factors. We have revised Section 5.3 to elaborate on this interpretation and added a discussion of why alternative explanations are less likely given the design. If the referee has specific suggestions for additional tests, we would be happy to incorporate them. revision: partial

Circularity Check

0 steps flagged

No circularity in the two-phase predict-then-validate design

full rationale

The paper separates Phase I (fitting a rider-heterogeneous random-parameters binary logit exclusively on pre-ban data to generate out-of-sample predictions of within-user reductions) from Phase II (a distinct DiD specification that exploits the December 2023 system-wide policy change to produce causal estimates). These estimates are then compared to the Phase I predictions for confirmation in sign and magnitude. The within-user composition check is an independent robustness test rather than a re-derivation of the fitted parameters. No load-bearing self-citations, self-definitional steps, or fitted-input-renamed-as-prediction patterns appear in the derivation chain. The structure is self-contained and relies on external validation via the policy shock and parallel-trends test rather than reducing to its own inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The analysis relies on standard econometric assumptions for binary logit and DiD rather than new axioms or invented entities. Free parameters exist inside the rider-heterogeneous random-parameters logit (e.g., distribution of random coefficients) but are not enumerated in the abstract.

axioms (2)
  • domain assumption Parallel trends assumption holds for the 3-month pre-ban window in the DiD specification
    Invoked to validate the causal estimates in Phase II
  • domain assumption No other contemporaneous policy or operational changes affect harsh-event rates during the study period
    Required for attributing the observed change solely to the ungoverned-mode removal

pith-pipeline@v0.9.0 · 5552 in / 1361 out tokens · 41717 ms · 2026-05-07T12:49:42.381051+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

61 extracted references · 56 canonical work pages

  1. [1]

    Doese-scooterownershipmatter?acomparisonofusagepatternsandmodereplacementeffectsofshared vs

    Aarhaug,J.,Egner,L.E.,Fearnley,N.,2025. Doese-scooterownershipmatter?acomparisonofusagepatternsandmodereplacementeffectsofshared vs. personal e-scooters. Research in Transportation Economics 113, 101626. URL:https://doi.org/10.1016/j.retrec.2025.101626, doi:10.1016/j.retrec.2025.101626. Abdi, A., O’Hern, S.,

    work page doi:10.1016/j.retrec.2025.101626 2025

  2. [2]

    Accident Analysis & Prevention 215, 108018

    Understanding e-scooter rider crash severity using a built environment typology: A two-stage clustering and random parameter model analysis. Accident Analysis & Prevention 215, 108018. URL:https://doi.org/10.1016/j.aap.2025.108018, doi:10.1016/j.aap.2025.108018. Agheli, A., Aghabayk, K.,

    work page doi:10.1016/j.aap.2025.108018 2025

  3. [3]

    Accident Analysis & Prevention 211, 107896

    How does distraction affect cyclists’ severe crashes? a hybrid catboost-shap and random parameters binary logit approach. Accident Analysis & Prevention 211, 107896. doi:10.1016/j.aap.2024.107896. Agheli, A., Aghabayk, K., Sadeghi, M., Das, S.,

    work page doi:10.1016/j.aap.2024.107896 2024

  4. [4]

    IATSS Research 49, 155–168

    E-scooter crash severity in the united kingdom: A comparative analysis using machine learning techniques and random parameters logit with heterogeneity in means and variances. IATSS Research 49, 155–168. doi:10.1016/j.iatssr.2025.03.004. Ahmed, S.S., Corman, F., Anastasopoulos, P.C.,

    work page doi:10.1016/j.iatssr.2025.03.004 2025

  5. [5]

    Analytic Methods in Accident Research 37, 100250

    Accounting for unobserved heterogeneity and spatial instability in the analysis of crash injury-severity at highway-rail grade crossings: A random parameters with heterogeneity in the means and variances approach. Analytic Methods in Accident Research 37, 100250. doi:10.1016/j.amar.2022.100250. Aizpuru, M., Farley, K.X., Rojas, J.C., Crawford, R.S., Moore...

    work page doi:10.1016/j.amar.2022.100250 2022

  6. [6]

    The American Journal of Emergency Medicine 37, 1133–1138

    Motorized scooter injuries in the era of scooter- shares: A review of the national electronic surveillance system. The American Journal of Emergency Medicine 37, 1133–1138. URL: https://doi.org/10.1016/j.ajem.2019.03.049, doi:10.1016/j.ajem.2019.03.049. Alnawmasi, N., Mannering, F.,

    work page doi:10.1016/j.ajem.2019.03.049 2019

  7. [7]

    Analytic Methods in Accident Research 48, 100408

    A note on random parameters models of crash injury severities with k-means clustering for data preprocessing. Analytic Methods in Accident Research 48, 100408. doi:10.1016/j.amar.2025.100408. Arikan Öztürk, E., Karaçor, F., Bayirtepe, H.,

    work page doi:10.1016/j.amar.2025.100408 2025

  8. [8]

    Traffic Injury Prevention 25, 1089–1097

    The characteristics of e-scooter accidents reported by police in türkiye. Traffic Injury Prevention 25, 1089–1097. URL:https://doi.org/10.1080/15389588.2024.2363478, doi:10.1080/15389588.2024.2363478. Asensio, O.I., Apablaza, C.Z., Lawson, M.C., Chen, E.W., Horner, S.J.,

    work page doi:10.1080/15389588.2024.2363478 2024

  9. [9]

    Nature Energy 7, 1100–1108

    Impacts of micromobility on car displacement with evidence from a natural experiment and geofencing policy. Nature Energy 7, 1100–1108. URL:https://doi.org/10.1038/s41560-022-01135-1, doi:10.1038/s41560-022-01135-1. Baby,T.,Yoon,S.H.,Lee,S.C.,2024. Developmentandvalidationofe-scooterridingbehaviorquestionnaire(erbq)amongkoreanriders. Ergonomics 68, 1660–1...

    work page doi:10.1038/s41560-022-01135-1 2024

  10. [10]

    Transportation Research Part F: Traffic Psychology and Behaviour 16, 117–126

    Assessing safety critical braking events in naturalistic driving studies. Transportation Research Part F: Traffic Psychology and Behaviour 16, 117–126. URL:https://doi.org/10.1016/j.trf.2012.08.006, doi:10.1016/j.trf.2012.08.006. Bloom, M.B., Noorzad, A., Lin, C., Little, M., Lee, E.Y., Margulies, D.R., Torbati, S.S.,

    work page doi:10.1016/j.trf.2012.08.006 2012

  11. [11]

    The American Journal of Surgery 221, 227–232

    Standing electric scooter injuries: Impact on a community. The American Journal of Surgery 221, 227–232. URL: https://doi.org/10.1016/j.amjsurg.2020.07.020, doi:10.1016/j.amjsurg.2020.07.020. Caggiani, L., Camporeale, R., Di Bari, D., Ottomanelli, M.,

    work page doi:10.1016/j.amjsurg.2020.07.020 2020

  12. [12]

    Journal of Intelligent Transportation Systems 29, 401–416

    A geofencing-based methodology for speed limit regulation and user safety in e-scooter sharing systems. Journal of Intelligent Transportation Systems 29, 401–416. URL:https://doi.org/10.1080/15472450.2023. 2201681, doi:10.1080/15472450.2023.2201681. Callaway, B., Goodman-Bacon, A., Sant’Anna, P.H.C., 2024a. Difference-in-differences with a Continuous Trea...

    work page doi:10.1080/15472450.2023 2023

  13. [13]

    Industrial and Labor Relations Review 46, 22–37

    Using regional variation in wages to measure the effects of the federal minimum wage. Industrial and Labor Relations Review 46, 22–37. URL:https://doi.org/10.2307/2524736, doi:10.2307/2524736. Cho, E., Yun, Y., Oh, C., Lee, G.,

    work page doi:10.2307/2524736

  14. [14]

    Accident Analysis & Prevention 190, 107186

    Derivation of riding risk precursors using 100 delivery motor scooter naturalistic riding study. Accident Analysis & Prevention 190, 107186. URL:https://doi.org/10.1016/j.aap.2023.107186, doi:10.1016/j.aap.2023.107186. Choi, S., Jin, Z., Ham, S.W., Kim, J., Sun, L.,

    work page doi:10.1016/j.aap.2023.107186 2023

  15. [15]

    Transportation Research Part C: Emerging Technologies 176, 105145

    A gentle introduction and tutorial on deep generative models in transportation research. Transportation Research Part C: Emerging Technologies 176, 105145. URL:https://doi.org/10.1016/j.trc.2025.105145, doi:10.1016/j.trc.2025.105145. Choi, S., Kim, J., Yeo, H.,

    work page doi:10.1016/j.trc.2025.105145 2025

  16. [16]

    Transportation Research Part C: Emerging Technologies 128, 103091

    TrajGAIL: Generating urban vehicle trajectories using generative adversarial imitation learning. Transportation Research Part C: Emerging Technologies 128, 103091. URL:https://doi.org/10.1016/j.trc.2021.103091, doi:10.1016/j.trc. 2021.103091. Cicchino, J.B., Kulie, P.E., McCarthy, M.L.,

    work page doi:10.1016/j.trc.2021.103091 2021

  17. [17]

    Journal of Safety Research 76, 256–261

    Severity of e-scooter rider injuries associated with trip characteristics. Journal of Safety Research 76, 256–261. URL:https://doi.org/10.1016/j.jsr.2020.12.016, doi:10.1016/j.jsr.2020.12.016. Cittadini, F.,Aulino,G.,Petrucci, M.,Valentini, S.,Covino,M.,2022. Electric scooter–relatedaccidents: apossibleprotectiveeffectof helmetuseon the head injury severi...

    work page doi:10.1016/j.jsr.2020.12.016 2020

  18. [18]

    European Economic Review 160, 104593

    Shared e-scooter services and road safety: Evidence from six european countries. European Economic Review 160, 104593. URL:https://doi.org/10.1016/j.euroecorev.2023.104593, doi:10.1016/j.euroecorev.2023.104593. Cohen, A., Einav, L.,

    work page doi:10.1016/j.euroecorev.2023.104593 2023

  19. [19]

    Review of Economics and Statistics 85, 828–843

    The effects of mandatory seat belt laws on driving behavior and traffic fatalities. Review of Economics and Statistics 85, 828–843. URL:https://doi.org/10.1162/003465303772815754, doi:10.1162/003465303772815754. Currie, G., Delbosc, A., Cox, R., Jayawardhena, M., Reynolds, J.,

    work page doi:10.1162/003465303772815754

  20. [20]

    Multimodal Transportation 4, 100205

    Exploring shared e-scooter trip patterns and links to public transport service level. Multimodal Transportation 4, 100205. URL:https://doi.org/10.1016/j.multra.2025.100205, doi:10.1016/j.multra.2025. 100205. Elvik, R.,

    work page doi:10.1016/j.multra.2025.100205 2025

  21. [21]

    Accident Analysis & Prevention 50, 854–860

    A re-parameterisation of the power model of the relationship between the speed of traffic and the number of accidents and accident victims. Accident Analysis & Prevention 50, 854–860. URL:https://doi.org/10.1016/j.aap.2012.07.012, doi:10.1016/j.aap. S. Choi, S. Yoo and S. Lee:Preprint submitted to ElsevierPage 18 of 25 Behavioral Responses to E-Scooter Sp...

    work page doi:10.1016/j.aap.2012.07.012 2012

  22. [22]

    Accident Analysis & Prevention 123, 114–122

    Updated estimates of the relationship between speed and road safety at the aggregate and individual levels. Accident Analysis & Prevention 123, 114–122. URL:https://doi.org/10.1016/j.aap.2018.11.014, doi:10.1016/j. aap.2018.11.014. Fischer, A., pyfixest contributors,

    work page doi:10.1016/j.aap.2018.11.014 2018

  23. [23]

    Difference-in-differences with variation in treatment timing

    Difference-in-differences with variation in treatment timing. Journal of Econometrics 225, 254–277. URL: https://doi.org/10.1016/j.jeconom.2021.03.014, doi:10.1016/j.jeconom.2021.03.014. Gu, Y., Cho, E., Oh, C., Lee, G.,

    work page doi:10.1016/j.jeconom.2021.03.014 2021

  24. [24]

    Accident Analysis & Prevention 211, 107871

    Riding safety evaluation of food delivery motor scooters based on associating sensor-based riding behavior and road traffic characteristics. Accident Analysis & Prevention 211, 107871. URL:https://doi.org/10.1016/j.aap.2024.107871, doi:10.1016/j.aap.2024.107871. Haworth,N.,Schramm,A.,Twisk,D.,2021. Changesinsharedandprivatee-scooteruseinbrisbane,australia...

    work page doi:10.1016/j.aap.2024.107871 2024

  25. [25]

    Injury Prevention 6, 82–89

    Risky business: safety regulations, risk compensation, and individual behavior. Injury Prevention 6, 82–89. URL: https://doi.org/10.1136/ip.6.2.82, doi:10.1136/ip.6.2.82. Hoveidaei, A., Mosalamiaghili, S., Keshtkar, A., Suresh, S.J., Adolf, J., Conway, J.D.,

    work page doi:10.1136/ip.6.2.82

  26. [26]

    Injury56,112361

    Orthopaedic fractures in skateboard, scooter, and e- scooterinjuries:Anationwidestudyintheu.s.(2010–2022). Injury56,112361. URL: https://doi.org/10.1016/j.injury.2025.112361, doi:10.1016/j.injury.2025.112361. Kazemzadeh, K.,

    work page doi:10.1016/j.injury.2025.112361 2010

  27. [27]

    Accident Analysis & Prevention 211, 107874

    Assessing e-scooter rider safety perceptions in shared spaces: Evidence from a video experiment in sweden. Accident Analysis & Prevention 211, 107874. doi:10.1016/j.aap.2024.107874. Kimpton, A., Loginova, J., Pojani, D., Bean, R., Sigler, T., Corcoran, J.,

    work page doi:10.1016/j.aap.2024.107874 2024

  28. [28]

    Journal of Transport Geography 104, 103439

    Weather to scoot? how weather shapes shared e-scooter ridership patterns. Journal of Transport Geography 104, 103439. URL:https://doi.org/10.1016/j.jtrangeo.2022.103439, doi:10.1016/j. jtrangeo.2022.103439. Kwon, N., Chang, I., Lee, J., Ahn, S.,

    work page doi:10.1016/j.jtrangeo.2022.103439 2022

  29. [29]

    KSCE Journal of Civil Engineering 28, 3533–3542

    Analysis of e-scooter risk factors by road types on different speed levels. KSCE Journal of Civil Engineering 28, 3533–3542. URL:https://doi.org/10.1007/s12205-024-1335-6, doi:10.1007/s12205-024-1335-6. Lawrence, B., Fildes, B., Thompson, L., Cook, J., Newstead, S.,

    work page doi:10.1007/s12205-024-1335-6

  30. [30]

    Traffic Injury Prevention 21, S96–S101

    Evaluation of the 30 km/h speed limit trial in the city of yarra, melbourne, australia. Traffic Injury Prevention 21, S96–S101. URL:https://doi.org/10.1080/15389588.2021.1895990, doi:10.1080/15389588. 2021.1895990. Lerman, S.R., Manski, C.F.,

    work page doi:10.1080/15389588.2021.1895990 2021

  31. [31]

    Transportation Research Part A: General 13, 29–44

    Sample design for discrete choice analysis of travel behavior: The state of the art. Transportation Research Part A: General 13, 29–44. URL:https://doi.org/10.1016/0191-2607(79)90084-0, doi:10.1016/0191-2607(79)90084-0. Liu, H.C., Lin, J.J.,

    work page doi:10.1016/0191-2607(79)90084-0

  32. [32]

    Transport Policy 126, 107–119

    Associations of built environments with spatiotemporal patterns of shared scooter use: A comparison with shared bike use. Transport Policy 126, 107–119. URL:https://doi.org/10.1016/j.tranpol.2022.07.012, doi:10.1016/j.tranpol.2022.07.012. Liukkonen, R., Aarnikko, H., Stenman, P., Ovaska, S., Reito, A.,

    work page doi:10.1016/j.tranpol.2022.07.012 2022

  33. [33]

    JAMANetworkOpen6,e2320868

    Association of nighttime speed limits and electric scooter–related injuries. JAMANetworkOpen6,e2320868. URL: https://doi.org/10.1001/jamanetworkopen.2023.20868,doi: 10.1001/jamanetworkopen. 2023.20868. Lu, Y., Zhang, L., Corcoran, J.,

    work page doi:10.1001/jamanetworkopen.2023.20868 2023

  34. [34]

    Journal of Cycling and Micromobility Research 2, 100036

    How weather and built environment factors influence e-scooter ridership: Understanding non-linear and time varying effects. Journal of Cycling and Micromobility Research 2, 100036. URL:https://doi.org/10.1016/j.jcmr.2024.100036, doi:10.1016/j.jcmr.2024.100036. Ma, Q., Yang, H., Mayhue, A., Sun, Y., Huang, Z., Ma, Y.,

    work page doi:10.1016/j.jcmr.2024.100036 2024

  35. [35]

    Accident Analysis & Prevention 151, 105954

    E-scooter safety: The riding risk analysis based on mobile sensing data. Accident Analysis & Prevention 151, 105954. URL:https://doi.org/10.1016/j.aap.2020.105954, doi:10.1016/j.aap.2020.105954. Mannering, F.L., Shankar, V., Bhat, C.R.,

    work page doi:10.1016/j.aap.2020.105954 2020

  36. [36]

    Mannering, V

    Unobserved heterogeneity and the statistical analysis of highway accident data. Analytic Methods in Accident Research 11, 1–16. doi:10.1016/j.amar.2016.04.001. Manski, C.F., Lerman, S.R.,

    work page doi:10.1016/j.amar.2016.04.001 2016

  37. [37]

    Econometrica 45, 1977–1988

    The estimation of choice probabilities from choice-based samples. Econometrica 45, 1977–1988. URL: https://doi.org/10.2307/1914121, doi:10.2307/1914121. Morton, C.,

    work page doi:10.2307/1914121 1977

  38. [38]

    Case Studies on Transport Policy 20, 101431

    Braving the elements: A time series analysis of e-scooter ridership assessing the impact of weather and seasonality across different climate regions. Case Studies on Transport Policy 20, 101431. URL: https://doi.org/10.1016/j.cstp.2025.101431, doi:10.1016/j.cstp.2025.101431. Mundlak, Y.,

    work page doi:10.1016/j.cstp.2025.101431 2025

  39. [39]

    Econometrica 46, 69–85

    On the pooling of time series and cross section data. Econometrica 46, 69–85. doi:10.2307/1913646. Nellamattathil, M., Amber, I.,

    work page doi:10.2307/1913646

  40. [40]

    Clinical Imaging 60, 200–203

    An evaluation of scooter injury and injury patterns following widespread adoption of e-scooters in a major metropolitan area. Clinical Imaging 60, 200–203. URL:https://doi.org/10.1016/j.clinimag.2019.12.012, doi:10.1016/j. clinimag.2019.12.012. Oster, E.,

    work page doi:10.1016/j.clinimag.2019.12.012 2019

  41. [41]

    Journal of Business & Economic Statistics 37, 187–204

    Unobservable selection and coefficient stability: Theory and evidence. Journal of Business & Economic Statistics 37, 187–204. doi:10.1080/07350015.2016.1227711. Pakarinen, O., Kobylin, A., Harjola, V.P., Castrén, M., Vasara, H.,

    work page doi:10.1080/07350015.2016.1227711 2016

  42. [42]

    JAMA Network Open 6, e2341194

    Speed and nighttime usage restrictions and the incidence of shared electric scooter injuries. JAMA Network Open 6, e2341194. URL: https://doi.org/10.1001/jamanetworkopen.2023.41194, doi:10.1001/jamanetworkopen.2023.41194. Peltzman, S.,

    work page doi:10.1001/jamanetworkopen.2023.41194 2023

  43. [43]

    Journal of Political Economy 83, 677–725

    The effects of automobile safety regulation. Journal of Political Economy 83, 677–725. URL:https://doi.org/10.1086/ 260352, doi:10.1086/260352. Petraki, V., Ziakopoulos, A., Yannis, G.,

    work page doi:10.1086/260352

  44. [44]

    Accident Analysis & Prevention 225, 108253

    State-of-the-art review on sustainable driving behavior: trade-offs between road safety, fuel consumption and emissions. Accident Analysis & Prevention 225, 108253. URL:https://doi.org/10.1016/j.aap.2025.108253, doi:10.1016/j.aap.2025.108253. Quddus, M., Theofilatos, A., Feng, M., Elvik, R.,

    work page doi:10.1016/j.aap.2025.108253 2025

  45. [45]

    Accident Analysis & Prevention 221, 108210

    Evaluating the safety and speed impacts of the 20mph speed limit in the uk: Evidence and insights. Accident Analysis & Prevention 221, 108210. doi:10.1016/j.aap.2025.108210. S. Choi, S. Yoo and S. Lee:Preprint submitted to ElsevierPage 19 of 25 Behavioral Responses to E-Scooter Speed Governance Sadeghi, M., Aghabayk, K., Quddus, M.,

    work page doi:10.1016/j.aap.2025.108210 2025

  46. [46]

    Accident Analysis & Prevention 206, 107696

    A hybrid machine learning and statistical modeling approach for analyzing the crash severity of mobility scooter users considering temporal instability. Accident Analysis & Prevention 206, 107696. doi:10.1016/j.aap.2024.107696. Scarano, A., Sadeghi, M., Mauriello, F., Riccardi, M.R., Aghabayk, K., Montella, A.,

    work page doi:10.1016/j.aap.2024.107696 2024

  47. [47]

    Journal of Safety Research 93, 373–398

    Cyclist crash severity modeling: A hybrid approach of xgboost-shap and random parameters logit with heterogeneity in means and variances. Journal of Safety Research 93, 373–398. doi:10.1016/j.jsr.2025.04.003. Sikka, N., Vila, C., Stratton, M., Ghassemi, M., Pourmand, A.,

    work page doi:10.1016/j.jsr.2025.04.003 2025

  48. [48]

    The American Journal of Emergency Medicine 37, 1807.e5–1807.e7

    Sharing the sidewalk: A case of e-scooter related pedestrian injury. The American Journal of Emergency Medicine 37, 1807.e5–1807.e7. URL:https://doi.org/10.1016/j.ajem.2019.06.017, doi:10.1016/ j.ajem.2019.06.017. Su, L., Yan, X., Zhao, X.,

    work page doi:10.1016/j.ajem.2019.06.017 2019

  49. [49]

    Transport Policy 145, 25–36

    Spatial equity of micromobility systems: A comparison of shared e-scooters and docked bikeshare in washington dc. Transport Policy 145, 25–36. URL:https://doi.org/10.1016/j.tranpol.2023.10.008, doi:10.1016/j.tranpol.2023.10.008. Tian, R., Li, L., Chien, S., Chen, Y., Sherony, R.,

    work page doi:10.1016/j.tranpol.2023.10.008 2023

  50. [50]

    Springer Nature Switzerland, pp

    E-scooter riding behaviors and risks from naturalistic driving study and crash data analysis, in: Transportation Mobility in Smart Cities. Springer Nature Switzerland, pp. 213–237. URL:https://doi.org/10.1007/ 978-3-031-64769-7_8, doi:10.1007/978-3-031-64769-7_8. Wang, L., Huang, Y., Li, Z.,

    work page doi:10.1007/978-3-031-64769-7_8

  51. [51]

    Transportation Research Interdisciplinary Perspectives 34, 101721

    Temporal instability and unobserved heterogeneity of random parameters in single-vehicle crash severity analysis. Transportation Research Interdisciplinary Perspectives 34, 101721. doi:10.1016/j.trip.2025.101721. Weschke, J.,

    work page doi:10.1016/j.trip.2025.101721 2025

  52. [52]

    Transportation Research Part A: Policy and Practice 178, 103868

    Scooting when the metro arrives—estimating the impact of public transport stations on shared e-scooter demand. Transportation Research Part A: Policy and Practice 178, 103868. URL:https://doi.org/10.1016/j.tra.2023.103868, doi:10.1016/j.tra.2023. 103868. White, E., Guo, F., Han, S., Mollenhauer, M., Broaddus, A., Sweeney, T., Robinson, S., Novotny, A., Bu...

    work page doi:10.1016/j.tra.2023.103868 2023

  53. [53]

    Journal of Safety Research 85, 182–191

    What factors contribute to e-scooter crashes: A first look using a naturalistic riding approach. Journal of Safety Research 85, 182–191. URL:https: //doi.org/10.1016/j.jsr.2023.02.002, doi:10.1016/j.jsr.2023.02.002. Wilde, G.J.S.,

    work page doi:10.1016/j.jsr.2023.02.002 2023

  54. [54]

    Risk Analysis 2, 209–225

    The theory of risk homeostasis: Implications for safety and health. Risk Analysis 2, 209–225. URL:https://doi.org/10. 1111/j.1539-6924.1982.tb01384.x, doi:10.1111/j.1539-6924.1982.tb01384.x. Wooldridge, J.M.,

    work page doi:10.1111/j.1539-6924.1982.tb01384.x 1982

  55. [55]

    Empirical Economics 69, 2545–2587

    Two-way fixed effects, the two-way mundlak regression, and difference-in-differences estimators. Empirical Economics 69, 2545–2587. URL:https://doi.org/10.1007/s00181-025-02807-z, doi:10.1007/s00181-025-02807-z. Yang,H.,Ma,Q.,Wang,Z.,Cai,Q.,Xie,K.,Yang,D.,2020. Safetyofmicro-mobility:Analysisofe-scootercrashesbyminingnewsreports. Accident Analysis & Preve...

    work page doi:10.1007/s00181-025-02807-z 2020

  56. [56]

    Analytic Methods in Accident Research 36, 100242

    Addressing endogeneity in modeling speed enforcement, crash risk and crash severity simultaneously. Analytic Methods in Accident Research 36, 100242. URL:https://doi.org/10.1016/j.amar.2022.100242, doi:10.1016/j.amar. 2022.100242. Ziakopoulos,A.,2024. Analysisofharshbrakingandharshaccelerationoccurrenceviaexplainableimbalancedmachinelearningusinghigh-reso...

    work page doi:10.1016/j.amar.2022.100242 2022

  57. [57]

    ̂Δ𝑃1→uni

    The analogous calculation for the firmware-validation speeding outcome is reported in Appendix B. Table A1 Sample re-weighting decomposition of the Phase I→ Phase II predict-then-validate gap. All entries in percentage points. “̂Δ𝑃1→uni” is the Phase I prediction rescaled bȳ 𝜏uni∕̄ 𝜏𝑃1 = 1.590. “Re-weighted share” iŝΔ𝑃1→uni ∕̂Δ𝑃2 in percent. Outcome ̂Δ𝑃...

    2019

  58. [58]

    The harsh-event coefficients remain significant in every panel

    restricted to pre-ban data; Panel C reports the same coefficient on five substantively meaningful subsamples (full, night trips, weekend trips, long trips, and urban cities) using a city-level aggregate DiD for comparability across subsamples; Panel D replaces the baseline city-level cluster with month-level and two-way (city× month) clustering; Panel E r...

    2019

  59. [59]

    Panel C uses a city-level aggregate DiD (no user FE, scaled byTUBNov = 0.542) following Callaway et al. (2024a). Its “Full” row therefore does not reproduce Table 4’s−6.24 and −5.29pp headline estimates and is included only to examine cross-subsample heterogeneity under an alternative estimator. Panel F uses the Oster (2019) bound with𝑅 2 max = 1.3 ̃𝑅2 and𝛿=

    2019

  60. [60]

    S. Choi, S. Yoo and S. Lee:Preprint submitted to ElsevierPage 24 of 25 Behavioral Responses to E-Scooter Speed Governance Placebo-test robustness A residual concern is that the Phase II estimate could be driven by unmodeled city-specific nonlinear trends rather than by the December TUB ban. We address this with two simultaneous robustness specifications i...

    2023

  61. [61]

    Standard errors in parentheses

    ∗ 𝑝 <0.05, ∗∗ 𝑝 <0.01, ∗∗∗ 𝑝 <0.001. Standard errors in parentheses. Linear trends Quadratic trends Outcome Placebo date ̂𝛽(SE)𝑝 ̂𝛽(SE)𝑝 Speeding Sep 2023+0.0446 ∗ 0.011 −0.0136 0.500 (0.0169) (0.0201) Oct 2023−0.1221 ∗∗∗ <0.001 −0.0117 0.649 (0.0347) (0.0256) Harsh accel Sep 2023+0.0155 0.216 +0.0114 0.356 (0.0124) (0.0122) Oct 2023+0.0002 0.976 +0.0078 ...

    2023