Review Reports - Analysis of Spatial and Environmental Factors Beyond Speed Limits Affecting Drivers’ Speed Choice

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The authors evaluated the influence of spatial and environmental factors on speeding at a micro-level by employing and testing different modeling techniques. The study's approach and methodology are sound overall. This is a very interesting study, and the findings confirm the commonly known importance of not focusing solely on speed limit reduction. While these results are not new, there are limited scientific approaches to confirm these findings. Most previous approaches have been based on before-and-after analyses, and as such, the innovative approach used by the authors brings relevance and makes an addition to the field. I support the publication of this interesting paper, provided the authors address the comments below.

The title of the study, especially the part “affecting drivers to slow down,” appears disconnected from the narrative and conclusion of the study. Firstly, the abstract does not summarize the results of those factors that affect drivers to slow down. This also holds true for Section 5 and the conclusion. It may be better to use alternative wording, such as “affecting drivers’ speed choice,” which could fit better, given that the speeding degree was the independent variable used. Improving the abstract by highlighting the most significant factors will also be important.
The authors interchangeably use the word “crash” and “accident” in several areas to mean the same thing, which is not correct. It is better to choose one terminology and stick to it. In this case, “crashes” could be a preferred option, since crashes are perceived as being preventable, as compared to using “accidents.”
The authors made mention that “In fact, speeding is a known contributor to approximately one-third of fatal crashes.” The citation is okay, but it is very old (2006). There are more recent studies that show how much speeding contributes to fatal crashes globally and also distinguish between HICs and LMICs. The suggestion is to use recent literature. See, for example: https://doi.org/10.1016/j.treng.2024.100259
Where speeding is first mentioned within the introduction, it is advisable to provide a definition of what speeding is. Do the authors mean excessive speeds (above the speed limit) or inappropriate speeds (below the speed limit, but not appropriate for conditions)? Or both? Generally, speeding could mean either.
Temporal variations are also another factor that affects speeding. Did the authors consider any specific time period during the data processing? Speeding violations may have an effect on the time period; for example, speeding violations in summer will differ from those in winter. If this was not done, mentioning this as a limitation and an area for further studies is useful.
Lines 339, 344, and 353 have some text that is not in English; please check.
It is not clear if the vehicle speeds used in calculating the speeding degree were congested speeds or free-flow speeds. It is important to use the free-flow speeds (with headways of at least 5 seconds) or to ensure that the speeds of vehicles are not influenced by other vehicles. Clarity on this aspect is needed in the paper, along with a justification of which speeds were used.
The authors should provide a clear definition of what a “Road Node” means from line 431.
The sample size of the vehicle speeds appears to be missing. The authors should state the total sample used for the analysis and the number of data points.
The models and methodological development used are appropriate and sound.
The findings in Figure 2 stated at the introduction are in contrast with the findings from lines 916 to 922. Figure 2 shows that there are more speeding violations at lower speed limits, but the findings from the authors show that for every 1 km/h increase in the speed limit, the speeding degree rises by approximately 10%. Improving the discussion by comparing these two results and providing a justification for the discrepancy will be necessary, as these results are all for the same locations.
The limitations of the paper and areas where future research needs to be conducted have not been stated. The authors need to include this at the conclusion.
Several before-and-after studies where speed limit interventions alone have been applied, and where these interventions have been complemented with other measures, have been undertaken in the literature, and these have reached similar findings that speed limits alone are not sufficient. Acknowledging this in the paper and then highlighting the different approach used by the authors to confirm existing knowledge will add more relevance to the paper. This is because the authors' findings dwell on the fact that reduced speed limits alone do not ensure slower speeds.
On line 995, the authors should specify clearly the definition of speeding. Instead of “speeding,” it is better to clearly say “speeding degree,” which in this case is the difference between the speed limits and the driver's speed. This is because “speeding” encompasses either both excessive or inappropriate speeds. The use of “speeding” also applies to other paragraphs. It is better to be consistent with the wording “speeding degree.”

Author Response

Comment: The title of the study, especially the part “affecting drivers to slow down,” appears disconnected from the narrative and conclusion of the study. Firstly, the abstract does not summarize the results of those factors that affect drivers to slow down. This also holds true for Section 5 and the conclusion. It may be better to use alternative wording, such as “affecting drivers’ speed choice,” which could fit better, given that the speeding degree was the independent variable used. Improving the abstract by highlighting the most significant factors will also be important.
Response: Thank you for your comment. The title has been revised to: “Analysis of Spatial and Environmental Factors beyond Speed Limits Affecting Drivers’ Speed Choice.”
Abstract has been updated to highlight the most significant factors influencing drivers’ speed choice, such as road geometry (e.g., curvature, number of lanes), node-level features (e.g., intersections, property change points), and enforcement measures. For example, the abstract now states:
“Key factors influencing drivers’ speed choice include road geometry (e.g., curvature, number of lanes), node-level features (e.g., intersections, property change points), and the presence of enforcement measures. Importantly, reduced speed limits alone do not ensure slower speeds.”
Conclusion has been revised to consistently use drivers’ speed choices instead of slowing down and to explicitly mention the study’s limitations and future research directions. Specifically, we added:
“… These findings offer valuable insights for policymakers but have several limitations. First, temporal variations in speeding—such as seasonal or time-of-day effects—were not examined, though patterns may differ across seasons and peak hours. Future work should incorporate temporal dimensions for a fuller understanding.”

Comment: The authors interchangeably use the word “crash” and “accident” in several areas to mean the same thing, which is not correct. It is better to choose one terminology and stick to it. In this case, “crashes” could be a preferred option, since crashes are perceived as being preventable, as compared to using “accidents.”
Response: We have revised the manuscript to consistently use the term “crash,” as it implies preventability, rather than “accident.”

Comment: The authors made mention that “In fact, speeding is a known contributor to approximately one-third of fatal crashes.” The citation is okay, but it is very old (2006). There are more recent studies that show how much speeding contributes to fatal crashes globally and also distinguish between HICs and LMICs. The suggestion is to use recent literature. See, for example: https://doi.org/10.1016/j.treng.2024.100259
Response: We have updated the citations to include the recent literature you suggested and have strengthened the introduction by incorporating statistics reflecting the global and high-income/low- and middle-income country distinctions.

Comment: Where speeding is first mentioned within the introduction, it is advisable to provide a definition of what speeding is. Do the authors mean excessive speeds (above the speed limit) or inappropriate speeds (below the speed limit, but not appropriate for conditions)? Or both? Generally, speeding could mean either.
Response: We appreciate the reviewer’s suggestion. To clarify our study’s terminology, we have now explicitly defined “speeding degree” in the objectives section. In the revised manuscript, speeding degree is defined as the difference between a driver’s actual speed and the posted speed limit. While previous studies have employed varying definitions of speeding—such as binary indicators of limit violations or crash-related metrics—we emphasize the magnitude of deviation from the limit as a continuous measure, which can be directly incorporated into spatial econometric modeling. This revision ensures conceptual clarity and consistency throughout the manuscript. Please refer to the revised manuscript’s page 4 line 128.

Comment: Temporal variations are also another factor that affects speeding. Did the authors consider any specific time period during the data processing? Speeding violations may have an effect on the time period; for example, speeding violations in summer will differ from those in winter. If this was not done, mentioning this as a limitation and an area for further studies is useful.
Response: We thank the reviewer for this important suggestion. In the revised manuscript, we have added a clear statement in the conclusion to acknowledge temporal variations as a limitation and to propose it as a direction for future research. Specifically, we note that seasonal or time-of-day effects were not explicitly examined, and that future studies should incorporate temporal dimensions to provide a more comprehensive understanding of speed choice dynamics. Please refer to the revised manuscript’s page 29 line 1056.

Comment: Lines 339, 344, and 353 have some text that is not in English; please check.
Response: This was an error that occurred during file conversion. It has been corrected.

Comment: It is not clear if the vehicle speeds used in calculating the speeding degree were congested speeds or free-flow speeds. It is important to use the free-flow speeds (with headways of at least 5 seconds) or to ensure that the speeds of vehicles are not influenced by other vehicles. Clarity on this aspect is needed in the paper, along with a justification of which speeds were used.
Response: We thank the reviewer for this insightful comment. The vehicle speed data were collected from on-board units installed in rental cars traveling across Jeju Island, yielding 99,851 point-level observations. In practice, these data inevitably include a mixture of free-flow and congested conditions, and it was not feasible to separate the two states for each observation. We have clarified this in the Methods section as follows:
“Vehicle speed data collected from the ADAS-equipped rental cars inevitably reflect a mix-ture of free-flow and congested traffic conditions. Due to data limitations, it was not feasi-ble to explicitly distinguish between these two states for each of the 99,851 point-level ob-servations. As a result, the constructed models incorporate both free-flow and congested speeds. This scope is explicitly acknowledged as a limitation of the study, but it also re-flects the reality of mixed traffic conditions on Jeju Island’s road network.”
Furthermore, in the Spatial Factors section, we explained how the network centrality measures partially capture these dynamics. Please refer to revised manuscript’s page 7 line 268.
Finally, we also added this point explicitly to the Conclusion (limitations): “Third, the data did not allow clear separation of free-flow and congested speeds; richer trajectory data and headway-based methods could address this.”

Comment: The authors should provide a clear definition of what a “Road Node” means from line 431.
Response: We appreciate the reviewer’s helpful comment. To improve clarity, we have added a clear definition of “road node” in the revised manuscript. Specifically, we now state:
“A road node, in this study, refers to a topological point within the road network that represents intersections, junctions, or other critical points where traffic conditions or road attributes may change.”
This definition appears at the beginning of the section describing the road node variables, before explaining how downstream nodes were identified and how node-level attributes were attached to the vehicle data. We believe this addition clarifies the concept of “road node” for readers and strengthens the consistency of our methodological explanation.

Comment: The sample size of the vehicle speeds appears to be missing. The authors should state the total sample used for the analysis and the number of data points.
Response: We thank the reviewer for this comment. To make the sample size more explicit, we have revised the Dependent Variable (Speeding Degree) section to clearly state the number of data points and vehicles used for the analysis. The section now begins as follows:
“The final dataset used for the analysis consisted of 99,851 point-level speed observations collected from 100 ADAS-equipped rental vehicles. The primary outcome of interest in this paper is the speeding degree, defined as the difference between a vehicle’s observed speed and the posted road speed limit.”
This clarification ensures that the total sample size is unambiguous and clearly visible to readers.

Comment: The models and methodological development used are appropriate and sound
Response: We sincerely thank the reviewer for the positive feedback. We appreciate the acknowledgement that the models and methodological development used in this study are appropriate and sound.

Comment: The findings in Figure 2 stated at the introduction are in contrast with the findings from lines 916 to 922. Figure 2 shows that there are more speeding violations at lower speed limits, but the findings from the authors show that for every 1 km/h increase in the speed limit, the speeding degree rises by approximately 10%. Improving the discussion by comparing these two results and providing a justification for the discrepancy will be necessary, as these results are all for the same locations.
Response: We thank the reviewer for this valuable observation. We agree that the relationship between Figure 2 and the regression results needed further clarification. We have revised the Discussion section to explicitly address this issue. Specifically, Figure 2 shows that speeding violations are more frequent in zones with lower speed limits (e.g., 20–50 km/h), reflecting the frequency of violations. In contrast, the regression analysis captures the magnitude of speeding, showing that in camera-enforced segments, the speeding degree increases by approximately 10% for every 1 km/h rise in the posted limit. This reflects the behavioral tendency of drivers to exploit the known tolerance margin (around 10%).
We have added the following text to the Discussion to make this distinction clearer. Please refer to the revised manuscript’s page 27 line 944. We believe this revision resolves the perceived inconsistency and strengthens the interpretation of our findings.

Comment: The limitations of the paper and areas where future research needs to be conducted have not been stated. The authors need to include this at the conclusion.
Response: We thank the reviewer for this important suggestion. In the revised manuscript, we have added a clear statement in the Conclusion section to acknowledge the study’s limitations and propose directions for future research. Specifically, we now highlight that temporal variations in speeding behavior—such as seasonal or time-of-day effects—were not explicitly examined. We also emphasize that future research should incorporate temporal dimensions to provide a more comprehensive understanding of speed choice dynamics. Please refer to the revised conclusions in page 29.

Comment: Several before-and-after studies where speed limit interventions alone have been applied, and where these interventions have been complemented with other measures, have been undertaken in the literature, and these have reached similar findings that speed limits alone are not sufficient. Acknowledging this in the paper and then highlighting the different approach used by the authors to confirm existing knowledge will add more relevance to the paper. This is because the authors' findings dwell on the fact that reduced speed limits alone do not ensure slower speeds.
Response: We thank the reviewer for this insightful comment. In the revised manuscript, we now explicitly acknowledge prior before-and-after studies showing that speed limit interventions alone are insufficient, and we highlight how our approach differs from these studies. To ensure logical flow, we placed this addition in the Conclusion section, directly after the paragraph discussing the insufficiency of speed limit reduction alone, and before the final concluding paragraph. Please refer to the revised paper’s page 29, line 1044. This addition acknowledges the relevance of existing literature and clarifies the unique contribution of our study.

Comment: On line 995, the authors should specify clearly the definition of speeding. Instead of “speeding,” it is better to clearly say “speeding degree,” which in this case is the difference between the speed limits and the driver's speed. This is because “speeding” encompasses either both excessive or inappropriate speeds. The use of “speeding” also applies to other paragraphs. It is better to be consistent with the wording “speeding degree.”
Response: We appreciate the reviewer’s valuable suggestion. In the revised manuscript, we have carefully replaced ambiguous uses of “speeding” and “speeding behavior” with the more precise term “speeding degree.” This variable is consistently defined as the difference between a driver’s actual speed and the posted speed limit.
To ensure conceptual clarity, the following revisions were made throughout the manuscript:
Abstract, Objectives, and Contributions: replaced speeding behavior with speeding degree.
Introduction: explicitly added a definition of speeding degree.
Literature Review, Methodology, Results, and Conclusion: updated wording to consistently use speeding degree.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

First of all, thank you for involving me in this review. I apologize for the delay in my response, which was necessary to thoroughly and accurately examine the manuscript.

The paper investigates the factors influencing driver speeds beyond simple speed limits, using data collected from the Cooperative Intelligent Transport System (C-ITS) on Jeju Island, South Korea.

The main objective is to identify the spatial factors that influence speed and to evaluate the effectiveness of spatial autocorrelation models compared to traditional regression models, such as OLS (Ordinary Least Squares).

The authors used point-level speed data obtained from rental vehicles equipped with ADAS (Advanced Driver Assistance Systems). They analyzed the relationship between speed and variables such as road geometry, node characteristics (e.g., traffic lights, speed cameras), and network connectivity.

The models employed showed greater effectiveness in capturing spatial dependencies compared to the OLS models, which produced counterintuitive results.

Factors such as the presence of speed cameras, the absence of parking, the presence of school zones, and the complexity of the road network (measured with closeness centrality) have a negative impact on vehicle speed, causing drivers to slow down.

The research also found that lower speed limits (< 50 km/h) are less effective in reducing speed, with a speed violation rate reaching up to 100% for the 20 km/h limit.

This suggests that regulations alone are not sufficient to change driver behavior unless they are supported by environmental and design factors that encourage more cautious driving.

Overall, the manuscript is well-written and structured. The data and methodology used are of high quality, and the results are relevant to the discussion on road safety and the effectiveness of traffic management policies.

However, before publication, the article could benefit from some minor additions:

For greater completeness, it would be useful to elaborate on the discussion regarding the choice of the 991.07-meter distance threshold used for the spatial weight matrix. Although the authors justify it based on the characteristics of the dataset, a more detailed explanation could strengthen the validity of the methodology.
Although the study is robust, it is important that the authors acknowledge and discuss its limitations. A key point to raise is the generalizability of the results. The research is based on data collected on Jeju Island, South Korea. The authors should discuss to what extent the results can be applied to different geographical and infrastructural contexts. This discussion would add academic honesty and pave the way for future comparative research.
The article focuses primarily on spatial factors. It might be interesting to suggest, as a future development, an analysis that also includes temporal factors, such as time of day, day of the week, or weather conditions, to see how they influence speed.
Given that the study focuses on sustainability, further research could analyze the impact of the presence of infrastructure dedicated to cyclists and pedestrians on vehicle speed.
The manuscript mentions policy implications. It would be appropriate to strengthen this part, in particular, by highlighting how the results can directly inform policymakers in designing safer and more sustainable roads. For example, specific traffic calming interventions based on their findings could be suggested, such as the installation of cameras or the modification of road geometry at critical points.

Finally, I would like to point out the following aspects:

Page 1, Line 10 (Abstract): There is a comma between "Factors" and "- beyond". For greater flow, it could be considered to remove it or replace it with other punctuation. The phrase "spatial factors- beyond speed limits-affecting speeding behavior" is a bit heavy.
Page 1, Line 18 (Abstract): The phrase "reduced speed limits alone do not ensure slower speeds" is somewhat imbalanced. It could be rephrased for greater clarity, for example: "The findings suggest that the reduction of speed limits alone may not guarantee a corresponding decrease in vehicle speed".

I remain at your disposal.

Author Response

Reviewer 2
Comment: For greater completeness, it would be useful to elaborate on the discussion regarding the choice of the 991.07-meter distance threshold used for the spatial weight matrix. Although the authors justify it based on the characteristics of the dataset, a more detailed explanation could strengthen the validity of the methodology.
Response: In the revised manuscript, we have elaborated on the rationale for the construction of the spatial weight matrix. The threshold of 991.07 m was derived as the minimum distance required to ensure that every observation has at least one neighbor, thereby preventing isolated points and ensuring connectivity. This approach follows established practice in spatial econometrics, where distance thresholds are commonly defined in this way [79, 80]. We also acknowledge that Euclidean distance is a proxy for network-based distance; however, Euclidean distance is widely adopted in large-scale point-level transportation studies because it provides a tractable and consistent approximation of local spatial dependencies. This clarification strengthens the validity of our methodological choice while acknowledging its limitation.

Comment: Although the study is robust, it is important that the authors acknowledge and discuss its limitations. A key point to raise is the generalizability of the results. The research is based on data collected on Jeju Island, South Korea. The authors should discuss to what extent the results can be applied to different geographical and infrastructural contexts. This discussion would add academic honesty and pave the way for future comparative research.
Response: We appreciate the reviewer’s point and agree that discussing generalizability is essential for transparency and usefulness. To address this, we have added the following paragraph to the Conclusions section:
“Beyond Jeju Island, applying these findings to other regions requires attention to both infrastructural and contextual differences. For example, the relative influence of design features (e.g., medians, bus lanes) and enforcement strategies may vary depending on road hierarchies, traffic compositions, and cultural driving norms. Future research should therefore test the framework in diverse geographical and temporal settings to establish the extent of generalizability and to guide context-specific policy design.”

Comment: The article focuses primarily on spatial factors. It might be interesting to suggest, as a future development, an analysis that also includes temporal factors, such as time of day, day of the week, or weather conditions, to see how they influence speed.
Response: In the revised manuscript, we have acknowledged this limitation in the Conclusions, noting that temporal variations in speeding—such as seasonal or time-of-day effects—were not examined and should be incorporated in future research. We further emphasize that extending the framework to include factors such as day-of-week patterns and weather conditions would enrich the explanatory power of the models and support more generalizable insights.

Comment: Given that the study focuses on sustainability, further research could analyze the impact of the presence of infrastructure dedicated to cyclists and pedestrians on vehicle speed.
Response: In the revised Conclusions, we have added the following statement to address this point: “Future research should also consider the role of pedestrian and cycling infrastructure—such as sidewalks, crossings, and bike lanes—in shaping vehicle speeds, thereby linking traffic safety outcomes more directly to sustainability objectives.”

Comment: The manuscript mentions policy implications. It would be appropriate to strengthen this part, in particular, by highlighting how the results can directly inform policymakers in designing safer and more sustainable roads. For example, specific traffic calming interventions based on their findings could be suggested, such as the installation of cameras or the modification of road geometry at critical points.
Response: In the revised Conclusions, we have strengthened the policy implications by explicitly highlighting how the results can inform concrete interventions as follows: “Despite these limitations, policymakers and planners are encouraged to adopt a holistic perspective, implementing integrated strategies that recognize the complex interplay between infrastructure, environment, enforcement, and human behavior. In practice, this may include targeted traffic calming interventions such as the installation or relocation of speed cameras at identified hot spots, the redesign of median barriers or intersections to reduce excessive speed, and the incorporation of visually salient features like bus lanes, pedestrian crossings, and bike facilities. By aligning these interventions with the spatial factors identified in this study, policymakers can design safer and more sustainable road environments.”

Comment: Page 1, Line 10 (Abstract): There is a comma between "Factors" and "- beyond". For greater flow, it could be considered to remove it or replace it with other punctuation. The phrase "spatial factors- beyond speed limits-affecting speeding behavior" is a bit heavy.
Response: In the revised Abstract, we have rephrased the sentence for smoother readability, now stating: “This paper investigates how spatial and environmental factors beyond speed limits affect drivers’ speed choice.”

Comment: Page 1, Line 18 (Abstract): The phrase "reduced speed limits alone do not ensure slower speeds" is somewhat imbalanced. It could be rephrased for greater clarity, for example: "The findings suggest that the reduction of speed limits alone may not guarantee a corresponding decrease in vehicle speed".
Response: In the revised Abstract, we have modified the sentence as recommended: “The findings suggest that the reduction of speed limits alone may not guarantee a corresponding decrease in vehicle speed”.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

The manuscript aims to investigate spatial and environmental factors influencing vehicle speeding behavior beyond posted speed limits on Jeju Island, South Korea. The authors utilize point-level ADAS data from a C-ITS pilot project, integrate it with road GIS and POI data, and apply spatial econometric models (SLM and SEM) to examine spatial autocorrelation in speeding behavior. The study claims that spatial lag models outperform OLS, and highlights the role of road geometry, land use, and enforcement measures in affecting speed choice. In my opinion, this manuscript has significant flaws that render it unsuitable for publication in its current form. Without substantial revisions to address these major issues, it does not meet the standards expected for publication.

First of all, the manuscript overstates its novelty. Although it claims to be among the first to apply spatial autocorrelation to point-level speed data, this is misleading. Numerous transportation studies have already applied spatial lag/error models to micro-level data, including speed, congestion, and crash severity. The authors provide no systematic comparison to prior spatial micro-level studies, nor a justification of how their study truly advances the state of the art.
There are Korean characters in the text of the manuscript, and this is not allowed.
The dependent variable “speeding degree” is defined as (vehicle speed – speed limit), but the authors missed to explain how negative values (i.e., driving below the limit) are handled. Are they retained in the model? Are they truncated? This is crucial, because including negative values (as the histogram shows) means the model is not exclusively about “speeding,” but about deviations in both directions. Justify whether the analysis is truly about speeding or speed deviation. If the former, remove negative values or model them separately.
It is recommended that the authors strengthen the literature review by incorporating studies on portable and automated speed enforcement systems as the study “Distributing portable excess speed detectors in AL Riyadh city”, which would provide additional empirical support and a broader international context to the discussion on speed limit compliance.
I believe that the choice of a fixed Euclidean threshold of 991.07 m for the spatial weight matrix is arbitrary and poorly justified. The authors fail to test sensitivity to different thresholds or to explore more realistic network-based distances, despite acknowledging that Euclidean distance is a proxy.
No sensitivity analysis is done for other factors in the methodology design. It is recommended to systematically assess the relative importance and interaction of spatial and environmental variables on speeding behavior using global sensitivity analysis. f not possible, refer to this as future work, citing the global sensitivity analysis definition and techniques found in studies: “ Global sensitivity analysis for studying hot-mix asphalt dynamic modulus parameters” and “ Global sensitivity analysis for seismic performance of shear wall with high-strength steel bars and recycled aggregate concrete”.
Although the authors claim that SLM performs better than OLS and SEM, no formal statistical model comparison is provided (e.g., AIC/BIC tables across models). Furthermore, they do not report key spatial diagnostics such as:

Lagrange Multiplier (LM) test results

Robust LM test outcomes

Pseudo R² or log-likelihood comparisons across models

Several coefficients are misinterpreted or presented without nuance. For example:

The presence of medians is associated with higher speeding, which is counterintuitive.

Road property change points show a positive effect on speeding, which may reflect omitted variables rather than a causal relationship.

These findings are not critically interrogated. The discussion often treats coefficients as causal when the design is purely observational and cross-sectional. Adopt a more cautious tone when interpreting results. Clarify that these are associations, not causal effects.
The dataset is drawn from only 100 rental vehicles, mostly driven by tourists. This population is not representative of the general driving population on Jeju Island. The authors acknowledge this but then generalize results to all drivers, which is methodologically flawed. Acknowledge and limit the scope of inference. Consider stratifying by tourist vs. local driver behavior if possible.

Comments on the Quality of English Language

Figures such as Figures 3 and 5 are referenced in the text; however, their captions are vague, and some labels are in Korean. This indicates a lack of thorough revision of the manuscript.

Author Response

Reviewer 3
Comment: First of all, the manuscript overstates its novelty. Although it claims to be among the first to apply spatial autocorrelation to point-level speed data, this is misleading. Numerous transportation studies have already applied spatial lag/error models to micro-level data, including speed, congestion, and crash severity. The authors provide no systematic comparison to prior spatial micro-level studies, nor a justification of how their study truly advances the state of the art.
Response: We appreciate the reviewer’s insightful comment and agree that our original description of the study’s novelty was overstated. In the revised manuscript, we have strengthened the Literature Review section to provide a more systematic discussion of prior research at the micro level. Specifically, we now acknowledge existing studies that have applied spatial econometric techniques to road speeds (Hackney et al., 2007; Toulouse et al., 2020). We also highlight the advantages and limitations of aggregated approaches such as TAZ-based analyses, which remain valuable for policy interpretation but may obscure micro-level variations in driver behavior.
At the same time, we have clarified how our study advances the literature. Unlike prior micro-level studies, our analysis focuses explicitly on speeding degree (actual speed minus the posted limit) as the dependent variable, utilizes large-scale point-level ADAS data rather than survey or camera-based sources, and covers the entire road network of Jeju Island, offering a broad and heterogeneous spatial context. Furthermore, by analyzing data from predominantly non-local rental car drivers, we reduce potential bias from habitual driving patterns. Finally, our study integrates network centrality measures (closeness, betweenness) to capture structural dynamics within a closed island network—an element rarely incorporated into previous micro-level spatial models.
We believe these revisions more accurately situate our contribution within the existing body of work, while clearly distinguishing the ways in which our study extends prior research.

Comment: There are Korean characters in the text of the manuscript, and this is not allowed.
Response: This was an error that occurred during file conversion. It has been corrected.

Comment: The dependent variable “speeding degree” is defined as (vehicle speed – speed limit), but the authors missed to explain how negative values (i.e., driving below the limit) are handled. Are they retained in the model? Are they truncated? This is crucial, because including negative values (as the histogram shows) means the model is not exclusively about “speeding,” but about deviations in both directions. Justify whether the analysis is truly about speeding or speed deviation. If the former, remove negative values or model them separately.
Response: We thank the reviewer for raising this important point. In the revised manuscript, we have clarified how negative values of the speeding degree are treated. Specifically, immediately following the definition, we now state:
“Positive values indicate instances of speeding, while negative or zero values reflect compliance or driving below the speed limit. We retain both positive and negative values in the baseline analysis to examine speed deviation around the posted limit, as sub-limit observations capture meaningful slowdowns due to traffic and control features (e.g., congestion, intersections, protected zones).”
This addition makes explicit that our analysis is best understood as addressing speed deviation around the posted limit, while speeding corresponds to the positive portion of the distribution. By keeping the full range of values, we capture both above-limit and below-limit driving behaviors, which provides a more complete understanding of how spatial and environmental factors influence speed choice.

Comment: It is recommended that the authors strengthen the literature review by incorporating studies on portable and automated speed enforcement systems as the study “Distributing portable excess speed detectors in AL Riyadh city”, which would provide additional empirical support and a broader international context to the discussion on speed limit compliance.
Response: We thank the reviewer for this constructive suggestion. In the revised manuscript, we have expanded the literature review to incorporate international evidence on portable and automated enforcement. Specifically, in the section on micro-level studies, we now state:
“Related work also indicates that portable and automated enforcement can improve speed-limit compliance; for example, Owais et al. [50] found that strategically positioned portable excess-speed detectors in Riyadh reduced localized speeding and increased adherence to posted limits.”
This addition provides broader empirical context beyond survey-based compliance studies and highlights the important role of enforcement design in shaping driver behavior across different international settings.

Comment: I believe that the choice of a fixed Euclidean threshold of 991.07 m for the spatial weight matrix is arbitrary and poorly justified. The authors fail to test sensitivity to different thresholds or to explore more realistic network-based distances, despite acknowledging that Euclidean distance is a proxy.
Response: We thank the reviewer for this valuable comment. In the revised manuscript, we have clarified the rationale for the construction of the spatial weight matrix. The threshold of 991.07 m was not chosen arbitrarily but was derived as the minimum distance required to ensure that every observation has at least one neighbor, thereby preventing isolated points and ensuring network connectivity. This approach follows established practice in spatial econometrics, where distance thresholds are commonly defined in this way [79, 80].
We acknowledge that Euclidean distance is a proxy for network-based distance. However, Euclidean distance measures are widely adopted in large-scale point-level transportation studies because they provide a tractable and consistent approximation of local spatial dependencies. To avoid ambiguity, we have revised the manuscript to explicitly state:
“In this paper, the spatial weight matrix is constructed using a Euclidean distance rule with a threshold of 991.07 m. This threshold was selected because it is the minimum distance required to ensure that each observation in the dataset has at least one neighbor, thereby avoiding isolated units and ensuring network connectivity. This approach follows common practice in spatial econometrics, where distance-based thresholds are often determined to guarantee connectivity among all observations [79, 80]. While Euclidean distance is a proxy for actual network distance, it is widely used in large-scale point-level transportation studies due to its tractability and consistency.”
This clarification strengthens the justification of our methodological choice and acknowledges the limitations, while aligning our approach with common practice in spatial econometric analysis.

Comment: No sensitivity analysis is done for other factors in the methodology design. It is recommended to systematically assess the relative importance and interaction of spatial and environmental variables on speeding behavior using global sensitivity analysis. If not possible, refer to this as future work, citing the global sensitivity analysis definition and techniques found in studies: “ Global sensitivity analysis for studying hot-mix asphalt dynamic modulus parameters” and “ Global sensitivity analysis for seismic performance of shear wall with high-strength steel bars and recycled aggregate concrete”.
Response: We thank the reviewer for this helpful suggestion. We agree that a systematic global sensitivity analysis (GSA) could provide further insights into the relative importance and interactions of spatial and environmental variables. However, given the scope of the present study, implementing such an analysis was beyond our capacity. Following the reviewer’s advice, we have revised the Conclusion section to explicitly note this as a limitation and a direction for future work. Specifically, we now state:
“Finally, while this study has advanced the understanding of spatial and environmental determinants of speeding degree, we did not conduct a global sensitivity analysis to systematically evaluate the relative importance and interactions of these factors. Global sensitivity analysis techniques, as discussed in prior studies (e.g., [98, 99]), provide a powerful framework for quantifying parameter influence. Incorporating such methods could further strengthen future investigations by offering a more rigorous assessment of how spatial and environmental variables jointly shape speeding behavior.”

Comment: Although the authors claim that SLM performs better than OLS and SEM, no formal statistical model comparison is provided (e.g., AIC/BIC tables across models). Furthermore, they do not report key spatial diagnostics such as: Lagrange Multiplier (LM) test results, Robust LM test outcomes, Pseudo R² or log-likelihood comparisons across models
Response: We appreciate the reviewer’s careful reading. Due to a document conversion issue, some cross-references and summary statistics were not displayed properly in the previous version. In the revised manuscript, we have corrected these issues and explicitly consolidated the model-comparison metrics and spatial diagnostics in the Results section. Specifically, Table 3 now reports log-likelihood, AIC/BIC, and (Pseudo) R² for OLS, SLM, and SEM, and Table 4 presents the LM and robust LM diagnostics (lag and error). These results consistently indicate that the SLM provides the best overall fit. We have also ensured that all in-text references point unambiguously to the relevant tables.

Comment: Several coefficients are misinterpreted or presented without nuance. For example: The presence of medians is associated with higher speeding, which is counterintuitive.
Road property change points show a positive effect on speeding, which may reflect omitted variables rather than a causal relationship. These findings are not critically interrogated. The discussion often treats coefficients as causal when the design is purely observational and cross-sectional. Adopt a more cautious tone when interpreting results. Clarify that these are associations, not causal effects.
Response: We appreciate the reviewer’s helpful observation. In the revised manuscript, we have clarified the interpretation of these coefficients to avoid implying causality. Immediately following the discussion of road geometry and design-related variables, we now explicitly state that the coefficients represent associations rather than causal effects. Specifically, we note that the positive coefficient for medians may reflect co-occurring design characteristics (e.g., wider cross-sections, fewer interruptions) rather than the physical presence of medians per se, and that the positive effect of road property change points may be influenced by unobserved factors such as adjacent land use or traffic composition. This addition ensures a more nuanced interpretation and emphasizes that the results highlight associations that warrant further investigation rather than definitive causal claims.

Comment: The dataset is drawn from only 100 rental vehicles, mostly driven by tourists. This population is not representative of the general driving population on Jeju Island. The authors acknowledge this but then generalize results to all drivers, which is methodologically flawed. Acknowledge and limit the scope of inference. Consider stratifying by tourist vs. local driver behavior if possible.
Response: We appreciate the reviewer’s comment regarding the representativeness of our dataset. In the revised manuscript, we have clarified that the analytic sample consists of 100 rental vehicles predominantly driven by non-local visitors, and thus is not representative of the entire Jeju driving population. We have emphasized that our findings should be interpreted as characterizing speeding degree among rental/non-local drivers, while noting that this focus helps reduce habitual driving biases common among resident commuters. In addition, we explicitly acknowledge in the Data Sources and Conclusion sections that this limits external validity and that future research should compare visitor and resident cohorts when richer data become available. Stratification by driver type is not possible with the current dataset because residency status is not observed at the trip level, but we highlight this as a direction for future work.