Comparative Analysis of User Behavior of Dock-Based vs. Dockless Bikeshare and Scootershare in Washington, D.C
Pith reviewed 2026-05-24 16:10 UTC · model grok-4.3
The pith
Dockless bikeshare and scootershare attract 63.8% and 69.6% member-like users in DC, close to the 73.3% in conventional docked systems.
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
Core claim
Dockless bike users exhibit member behavior in 63.8% of trips and dockless scooter users in 69.6% of trips, compared with 73.3% member trips in the conventional bikeshare system; dockless trips also display short durations similar to those of registered members with no significant weekday versus weekend difference.
What carries the argument
Logistic regression and random forest models that classify individual trips as member (commuter) or casual (recreational) behavior using available trip features such as duration and timing.
If this is right
- Dockless systems promote bicycle usage for daily urban travel at rates close to those of conventional bikeshare.
- These systems complement station-based bikeshare by attracting similar proportions of commuter-style trips.
- Short trip durations and consistent weekday-weekend patterns indicate reliable support for routine mobility needs.
- The classification approach supplies a practical method for operators to monitor behavior in new shared-mobility services.
Where Pith is reading between the lines
- Cities could apply the same modeling technique to test whether dockless fleets reduce car trips in neighborhoods lacking docking stations.
- If member-like usage persists, planners might prioritize dockless vehicles near transit hubs to strengthen first-last mile connections.
- Extending the analysis to additional seasons would reveal whether summer or winter patterns shift the reported member percentages.
- Operators in other U.S. cities could replicate the study to compare local dockless performance against their own docked baselines.
Load-bearing premise
The models correctly identify commuter versus recreational behavior from trip features alone and the three-month 2018 window represents longer-term usage patterns.
What would settle it
A validation dataset with known actual membership status or user surveys collected over a full year that shows the model's predicted member percentages differ by more than ten points from observed rates.
read the original abstract
In 2017, dockless bikeshare systems were introduced in the United States, followed by dockless scootershare in early 2018. These new mobility options are expected to complement the existing station-based bikeshare systems, which are bound to static origin and destination points at docking stations. The three systems attract different users with different travel behavior mobility patterns. The present research provides a comparative analysis of users' behavior for these three shared mobility systems during March-May 2018 in the District of Columbia. Our study identifies similarities/differences between the two systems aiming for better planning, operating, and decision-making of these emerging personal shared mobility systems in the future. It uses logistic regression and random forest modeling to delineate between "member" behavior, which aligns most closely with commuter behavior, and "casual" behavior that represents more recreational behavior. The results show that 63.8% of dockless bike users and 69.6% of dockless scooter users demonstrated "member" behavior, which is slightly lower than the actual percentage of trips made by members within the conventional bikeshare system (73.3%). Dockless systems users also showed to have short trip durations similar to conventional bikeshare system's registered members, with no significant difference between trips during weekdays and weekends. Overall, this study provides a methodology to understand users' behavior for the dockless bikeshare system and provides sufficient evidence that these new shared mobility systems can potentially make positive contributions to urban multi-modal infrastructure by promoting bicycle usage for urban daily travel.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper claims that logistic regression and random forest models trained to distinguish 'member' (commuter) from 'casual' (recreational) behavior on conventional docked bikeshare data can be applied to dockless bikeshare and scootershare trips in Washington D.C. (March–May 2018). It reports that 63.8% of dockless bike trips and 69.6% of dockless scooter trips are classified as member behavior (vs. 73.3% actual member trips in the docked system), that dockless users exhibit short trip durations similar to docked members, and that there is no significant weekday/weekend difference in dockless trip patterns. The work concludes that dockless systems promote daily urban travel and can complement existing infrastructure.
Significance. If the cross-system transfer of the member/casual classifier is valid, the study supplies concrete empirical evidence on the user composition of early dockless deployments and a reusable methodology for behavior classification in shared mobility. The March–May 2018 comparison is timely for policy on multi-modal integration, though the absence of reported model diagnostics limits immediate applicability.
major comments (3)
- [Methods] Methods (model training and application): the logistic regression and random forest classifiers are fit on conventional docked data (where labels are observed) and applied to dockless trips, yet no cross-system validation, feature-distribution shift test, or out-of-sample performance metrics on a held-out docked subset are reported. This directly undermines the central percentages (63.8 %, 69.6 %, 73.3 %) because any systematic difference in trip-duration or start-time distributions between systems will move the decision boundary.
- [Data and Results] Data and Results: the March–May 2018 window is the first months of dockless operation; the paper does not test or discuss whether the observed member-behavior shares are confounded by early-adopter effects versus stable system-type differences. This is load-bearing for the claim that dockless systems attract commuter-like users at rates comparable to the docked system.
- [Results] Results: the statements that dockless users show 'short trip durations similar to conventional bikeshare system's registered members' and 'no significant difference between trips during weekdays and weekends' are presented without accompanying statistical tests, confidence intervals, or effect-size measures, making it impossible to assess whether the similarity is substantive or merely descriptive.
minor comments (2)
- [Abstract] Abstract and text: the phrase 'demonstrated member behavior' is used for the model output on dockless trips; clarify that these are predicted labels, not observed membership status.
- Notation: 'member' and 'casual' are used both for observed labels in the docked system and for model predictions on dockless data; consistent terminology or explicit distinction would improve readability.
Simulated Author's Rebuttal
We thank the referee for the detailed and constructive report. The comments identify important gaps in validation, temporal context, and statistical rigor. We address each point below and indicate where revisions will be made. The core methodology relies on transferring a behavior classifier trained on labeled docked data to unlabeled dockless trips; we defend the approach on substantive grounds while acknowledging limitations that can be mitigated in revision.
read point-by-point responses
-
Referee: [Methods] Methods (model training and application): the logistic regression and random forest classifiers are fit on conventional docked data (where labels are observed) and applied to dockless trips, yet no cross-system validation, feature-distribution shift test, or out-of-sample performance metrics on a held-out docked subset are reported. This directly undermines the central percentages (63.8 %, 69.6 %, 73.3 %) because any systematic difference in trip-duration or start-time distributions between systems will move the decision boundary.
Authors: The models were trained exclusively on the docked system, where member/casual labels are observed from registration status, and then applied to dockless trips using identical features (trip duration, start hour, day of week). Because dockless operators did not provide equivalent registration labels, direct cross-system validation on labeled dockless data is not possible. However, we agree that reporting out-of-sample accuracy, precision, and recall on a held-out portion of the docked data, together with explicit comparisons of feature distributions (e.g., Kolmogorov-Smirnov tests on duration and start-time) between docked and dockless samples, would strengthen confidence in the transfer. These diagnostics can be added from the existing docked dataset without new data collection. revision: yes
-
Referee: [Data and Results] Data and Results: the March–May 2018 window is the first months of dockless operation; the paper does not test or discuss whether the observed member-behavior shares are confounded by early-adopter effects versus stable system-type differences. This is load-bearing for the claim that dockless systems attract commuter-like users at rates comparable to the docked system.
Authors: The study period was deliberately chosen as the initial deployment window to provide timely evidence on early dockless behavior. We did not possess later-period dockless data at the time of analysis, so a formal test separating early-adopter effects from stable system differences could not be performed. We will revise the discussion section to explicitly flag this as a limitation, note that the reported percentages reflect the March–May 2018 snapshot, and suggest that longitudinal comparisons with subsequent years would be valuable for future research. revision: partial
-
Referee: [Results] Results: the statements that dockless users show 'short trip durations similar to conventional bikeshare system's registered members' and 'no significant difference between trips during weekdays and weekends' are presented without accompanying statistical tests, confidence intervals, or effect-size measures, making it impossible to assess whether the similarity is substantive or merely descriptive.
Authors: The original comparisons were descriptive, relying on mean durations, histograms, and weekday/weekend proportions. We agree that formal inference is needed. In revision we will add two-sample t-tests (or non-parametric equivalents) for trip-duration differences, chi-squared tests for weekday/weekend membership proportions, 95% confidence intervals, and effect-size measures (Cohen’s d or Cramér’s V) to quantify the observed similarities. revision: yes
Circularity Check
No circularity: purely empirical classification with no self-referential derivations or fitted predictions.
full rationale
The paper applies logistic regression and random forest models to observed trip features (duration, timing, weekday/weekend) from March-May 2018 data to classify dockless trips as 'member' vs. 'casual' behavior, then reports the resulting percentages (63.8%, 69.6%) alongside the known 73.3% for the conventional system. No equations, ansatzes, uniqueness theorems, or self-citations are invoked; the outputs are direct model applications to separate datasets rather than any quantity that reduces to its own inputs by construction. The work contains no derivation chain that could be circular.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.