Review Reports

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This paper addresses a highly relevant and timely topic in the field of modern urban logistics and adopts a rigorous Stated Preference (SP) modeling framework. While the study is methodologically sound and clearly written, I would like to recommend the following three points for revision and improvement:

1. Integration with Observed Data Context

While the use of Stated Preference (SP) data is appropriate for exploratory behavioural research, the manuscript should provide a more comprehensive discussion of potential divergences between stated intentions and actual driving behaviour. To strengthen the practical validity of the findings, the authors are encouraged to discuss how the identified SP patterns relate to, or could be validated using, existing Naturalistic Driving Data (NDD) or accident statistics specific to last-mile delivery operations — even if such data were not collected for this study. This would enhance the credibility and policy relevance of the work.

1. In-Depth Model Interpretation and Practical Relevance

The discussion of the Binary Logit model results should be expanded substantially. Beyond presenting parameter signs and statistical significance, the manuscript should interpret the magnitude of coefficients and elasticities to clarify the relative importance of different attributes. A clear and direct linkage should be established between these key attributes and the corresponding design of training programs or policy interventions, thereby enhancing the practical implications of the modeling outcomes.

1. Cross-Country Heterogeneity Analysis

The study includes last-mile delivery professionals from five distinct European countries (Croatia, Cyprus, Greece, Italy, and Slovenia). Given the well-documented variations in traffic culture, infrastructure, and delivery regulations across these contexts, the authors should explicitly conduct—or at least discuss the feasibility of—a segmented analysis or introduce interaction terms to examine whether behavioural patterns and training needs significantly differ between countries. This addition would greatly improve the generalizability and contextual robustness of the findings.

 

Author Response

See the Rebuttal attached

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The manuscript is fairly well written and focuses, on my perspective, on a timely and significant topic, which is related to understand last-mile delivery professionals perceptions about the activity. 

In what follows, I will describe some concerns I found misleading or not rigorous enough, while revising this research study.

The authors claim right in the Abstract that the "study aimed to identify the training needs of last-mile delivery workers", which I think is not adequate, not even the correct focus.

The methodology proposed is focused on Discrete Choice Modelling and used Stated Preference surveys to conduct the research.

Data is gathered from different countries that, unfortunately does not represent the potential unobserved heterogeneity that lies in different countries of Europe.

The results are a major concern since the sample data is completely imbalanced, and the results discussion must be focused on relative differences and not in absolute values, such as the frequency that is used. Slovenia, with 151 respondents, accounts for over 45% of the total sample... This would reflect bias in the obtained results and other issues related to the uneven sample distribution, is that Slovenian responses will dominate, and may affect the Binary Logit Model findings. The heavy weighting toward one region limits the external validity.

Besides, once the authors learn that the samples are not balanced, they should already incorporate statistical techniques that could be relevant to provide more rigorous results, and potentially scalable. 

In terms of methodological framework, it is not appropriate the econometric model specification and it must be improved. A Mixed Logit Model would be better for analysing data.

In the design of the survey, the authors did not consider other relevant factors, such as workload, weather conditions...and these are know to be very important for this type of professionals.

Although overall the work has an adequate structure, it should be noted that the section describing the Methodology, in particular the part relating to the development of the Logit Model, is presented before the Data Overview (Experimental Design), which should not be the case, because it is necessary to have a general understanding of the data to be worked with before deciding which statistical analysis is most appropriate for obtaining meaningful results. In this case, I would suggest reformulating these parts of the Methodology.

The manuscript requires major improvements before it can be considered for potential publication in the journal.

 

 

 

Author Response

See the Rebuttal attached

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

This study conducted a stated preference survey to identify the driver behavior patterns for delivery riders. The topic is interesting but several questions should be addressed before acceptance. 
1. The commonly used discrete choice models with logit family function include binary logit model, nested logit model, multinomina logit model, random parameter logit  (RPL) model and other advanced logit models, such as RPL model with heterogeneity in mean/heterogeneity in means and variance, etc., However, models pertaining to RPL related model was not include in the methodology section, highlighting the need to include these content in this manuscript. Some articles are useful for reference.

(1) Applied Choice Analysis, David A. Hensher, John M. Rose, and William H. Greene. Cambridge University Press, June 2015, https://doi.org/10.1017/CBO9781316136232

(2) Statistical and Econometric Methods for Transportation Data Analysis, BySimon Washington, Matthew G. Karlaftis, Fred Mannering, Panagiotis Anastasopoulos.https://doi.org/10.1201/9780429244018.

2. The binary logit model was applied in this study, which could not address the unobserved heterogeneity accross observation, ignoring this heterogeneity may lead to biased estimated results. Though, models, such as the random parameter binary logit model was not applied, the limitation of ignoring this heterogeneity should be discussed at the end of the manuscript. Some articles are useful for reference.

(1) Unobserved heterogeneity and the statistical analysis of highway accident data

(2) Investigating the severity of expressway crash based on the random parameter logit model accounting for unobserved heterogeneity

(3) The statistical analysis of highway crash-injury severities: A review and assessment of methodological alternatives

3. The equation of the McFadden's pseudo R-square shold be included.

4. Three factors were considered in the stated preference survey, namely the driver behavior (legal and illegal), trip time (5, 10, 15, 20), and salary type (flat salary and performance-based salary). It should be 2*4*2=16 delevery scenarios. However, only five delivery scenarios were included. Please explain why.

5. Figures' quality in the manuscript are suggested to be improved with Python or R. 

Author Response

See the Rebuttal attached

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

all comments are addressed.

Reviewer 2 Report

Comments and Suggestions for Authors

The authors revised and improved the manuscript, taking into consideration my previous concerns.