Factors Influencing Water Point Functionality in Liberia: A Regression and Bayesian Network Analysis

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The article presents a well structured and convincing argument on the factors associated with Afridev handpump functionality in Liberia using systems thinking (Bayesian network analysis) methodology. This article will be of interest to WASH practitioners, academics and policy makers alike on how these methods provide deeper insights into water infrastructure improvement and service delivery implementation. 

As it stands the paper approach and narrative is strong, with a well laid out methodology and clear study limitations taken into account. It contributes to the growing body of knowledge and understanding of predictors for functionality of Afridev handpumps. While there are other categories that could be included in the survey and overall results, the authors have made it clear the data collection limits in survey collection and data quality. The findings are valuable to the methods and large dataset. There are some areas that can be improved throughout the paper to help narrative and impact in communicating the overall results. 

Abstract

Line 7-8: This line could be misleading if taken in silo. It is true that the results show no statistical significance but misses key findings in the abstract that service delivery is important - as discussed in section 4.4 as the ‘management entity is associated with at least twice the odds of a water point being functional’. This is a key message. It is briefly covered in line 12-13 but it could be stronger.

 

Section 1

Line 29-33: Very important statistics although lack of access to basic drinking water is the lowest statistic (25% compared to 70% basic sanitation and over 95% lack of hygiene services). The link between access to basic drinking water, and using water resources that are distant, reliable and contaminated could be clearer. This would set up early why the study chose to focus on the water supply dimension of WASH, as the stats could be read that sanitation and hygiene infrastructure is a more pressing need to address.

Line 39-40: An important point to make about the impact of climate change but reads as though it was an add-on. Suggest editing along the lines of “Challenges in ensuring reliable functionality may also be exacerbated due the increasing impact of the climate emergency on precipitation patterns and aquifers”.

Line 46-49: Is this referring to work using large quantitative datasets in Liberia? Or more generally in SSA? Large statistical datasets have been used before to inform predictors for functionality - refer to work in Malawi (https://www.mdpi.com/2073-4441/11/3/494 and https://pubs.rsc.org/en/content/articlelanding/2020/ew/d0ew00283f)  and Liberia, Sierra Leone, and Uganda (https://pubs.acs.org/doi/10.1021/es402086n) and Ghana for Bayesian methods (https://doi.org/10.1002/2014WR016770). Can this line be expanded to cover how this study furthers the work in large data models and systems thinking in the WASH context?

 

Section 2.2.1

Line 83: what is defined as a ‘toolkit’? Are special tools required? This is where the study could reference the simplicity of maintaining the Afridev handpump.

Section 4

The results in this section are well discussed and present a strong argument for professionalised service provision for water infrastructure at the local level. However, this link could be clearer in section 4.3 and 4.4. In the results section and this discussion it would have been interesting to see the linkages between formal institutional management structure and the findings that a fee structure is not statistically significant. This is touched on in the transition between sections but could be stronger and support research across SSA over the last decade of the impact of informal community based management compared to professionalised management (e.g. Chowns 2015, Hutchings 2015 and 2018, Moriarty 2013, Hope 2015, Truslove 2020 and 2021, van de Broek 2015, Carter 2016, Murray 2024). Section 4.4 could go stronger on this point. 

Section 4.6

Line 290: It would be good to expand on what the risks are from using data from 2017. Have there been any policy or substantial changes to water implementation in Liberia since then? This study presents important findings and this line could go stronger on how these are still relevant.

Overall, this was a really interesting read! Well done all.

Author Response

The article presents a well structured and convincing argument on the factors associated with Afridev handpump functionality in Liberia using systems thinking (Bayesian network analysis) methodology. This article will be of interest to WASH practitioners, academics and policy makers alike on how these methods provide deeper insights into water infrastructure improvement and service delivery implementation. 

 

As it stands the paper approach and narrative is strong, with a well laid out methodology and clear study limitations taken into account. It contributes to the growing body of knowledge and understanding of predictors for functionality of Afridev handpumps. While there are other categories that could be included in the survey and overall results, the authors have made it clear the data collection limits in survey collection and data quality. The findings are valuable to the methods and large dataset. There are some areas that can be improved throughout the paper to help narrative and impact in communicating the overall results. 

 

Abstract

Line 7-8: This line could be misleading if taken in silo. It is true that the results show no statistical significance but misses key findings in the abstract that service delivery is important - as discussed in section 4.4 as the ‘management entity is associated with at least twice the odds of a water point being functional’. This is a key message. It is briefly covered in line 12-13 but it could be stronger.

 

Response: Thank you. We revised the abstract to clarify that, although the payment structure was not statistically significant, the management entity was strongly associated with functionality. We now explicitly state that service delivery factors, particularly management, are critical. This also reinforces the novelty of our scenario-based simulation approach, which quantifies the impact of service delivery structures on functional outcomes.

 

Section 1

Line 29-33: Very important statistics although lack of access to basic drinking water is the lowest statistic (25% compared to 70% basic sanitation and over 95% lack of hygiene services). The link between access to basic drinking water, and using water resources that are distant, reliable and contaminated could be clearer. This would set up early why the study chose to focus on the water supply dimension of WASH, as the stats could be read that sanitation and hygiene infrastructure is a more pressing need to address.

 

Response: Thanks - to clarify the message, and to respond to other reviewer comments, we have dropped the references to sanitation and hygiene and focus our message on water only. We hope this clears up the confusion. 

 

 

 

Line 39-40: An important point to make about the impact of climate change but reads as though it was an add-on. Suggest editing along the lines of “Challenges in ensuring reliable functionality may also be exacerbated due the increasing impact of the climate emergency on precipitation patterns and aquifers”.

 

Response: Thank you for this suggestion, we modified it to “Challenges in ensuring reliable functionality may also be exacerbated due to the increasing impact of the climate on precipitation patterns and aquifers.”

 

Line 46-49: Is this referring to work using large quantitative datasets in Liberia? Or more generally in SSA? Large statistical datasets have been used before to inform predictors for functionality - refer to work in Malawi (https://www.mdpi.com/2073-4441/11/3/494 and https://pubs.rsc.org/en/content/articlelanding/2020/ew/d0ew00283f)  and Liberia, Sierra Leone, and Uganda (https://pubs.acs.org/doi/10.1021/es402086n) and Ghana for Bayesian methods (https://doi.org/10.1002/2014WR016770). Can this line be expanded to cover how this study furthers the work in large data models and systems thinking in the WASH context?

 

Response: Thank you for this suggestion, we expanded this section to include references to prior studies in Malawi, Ghana, and Liberia, and clarified how our study contributes to these findings by applying Bayesian methods to a large national dataset in Liberia. This is the first study to integrate both multivariable logistic regression and Bayesian Network modeling to assess rural water point functionality in Liberia, offering both statistical rigor and systems-level insight. We think this study is novel due to the dataset size (10k+ observations), no previous publications from this dataset, and we examine the mechanisms through which financial management leads to sustainability.

 

 

 

Section 2.2.1

Line 83: what is defined as a ‘toolkit’? Are special tools required? This is where the study could reference the simplicity of maintaining the Afridev handpump.

 

Response:Thank you for pointing this out. We clarified that the toolkit refers to a basic set of tools required for routine maintenance of Afridev handpumps, which are designed for ease of repair. We also clarified in our limitations section that we do not have access to the original survey instrument, which may have defined the toolkit more specifically.

 

Section 4

The results in this section are well discussed and present a strong argument for professionalised service provision for water infrastructure at the local level. However, this link could be clearer in section 4.3 and 4.4. In the results section and this discussion it would have been interesting to see the linkages between formal institutional management structure and the findings that a fee structure is not statistically significant. This is touched on in the transition between sections but could be stronger and support research across SSA over the last decade of the impact of informal community based management compared to professionalised management (e.g. Chowns 2015, Hutchings 2015 and 2018, Moriarty 2013, Hope 2015, Truslove 2020 and 2021, van de Broek 2015, Carter 2016, Murray 2024). Section 4.4 could go stronger on this point. 

 

Response: Thank you for this comment. We expanded Sections 4.3 and 4.4 to discuss the implications of professionalized service provision and referenced relevant literature (e.g., Chowns 2015, Truslove 2020). This discussion is enhanced by our use of scenario-based simulations, which provide a novel, policy-relevant lens for evaluating management structures. This is an important finding as the rural water supply sector debates the merits of community-based management versus more professionalized service provision models. 

 

Section 4.6

Line 290: It would be good to expand on what the risks are from using data from 2017. Have there been any policy or substantial changes to water implementation in Liberia since then? This study presents important findings and this line could go stronger on how these are still relevant.

 

Response: 

Thank you for this comment. We agree that the age of the dataset is an important consideration. In response, we have expanded the limitations section to further clarify that, although the data are from 2017, the underlying service delivery structures and institutional capacity challenges in Liberia have remained largely consistent in recent years. This supports the continued relevance of the findings for current WASH planning and policy.

 

Overall, this was a really interesting read! Well done all.

 

Thank you, we appreciate you taking the time to review.

Reviewer 2 Report

Comments and Suggestions for Authors

General Comments

This manuscript presents an empirical analysis of factors influencing water point functionality in Liberia using a large national dataset of over 11,000 Afridev handpumps. By applying multivariable logistic regression and Bayesian Network modeling, the study assesses associations between functionality and variables such as maintenance structure, repair time, mechanic availability, and payment status. The inclusion of scenario-based simulations offers a practical dimension to the analysis, making it policy-relevant and timely.

The manuscript offers a well-structured, policy-relevant analysis using a large national dataset and robust methods, including logistic regression and Bayesian Networks. It is clearly presented and practically valuable. To strengthen its scholarly contribution, the paper should more clearly articulate its novelty, better integrate related literature, clarify model assumptions and limitations, separate interpretation from results, and refine policy recommendations to reflect the study’s scope. Reducing redundancy between the Results and Discussion will also improve clarity and analytical depth.

Specific Comments and Suggestions for Authors:

1. Clarify Novel Contribution (Lines 222–229):
   Clearly articulate how this study extends previous research beyond describing associations. Highlight what is methodologically, empirically, or regionally novel—particularly with respect to the Bayesian Network application.
2. Improve Literature Integration (Lines 230–284):
   Strengthen the Discussion by comparing your findings to those in similar contexts (e.g., Nigeria, Tanzania). For example, claims about WASH committee effectiveness (Lines 259-264), repair timelines (Lines 242–245) and section 4.5 (Lines 271-287) would benefit from references to previous studies.
3. Clarify Model Assumptions and Limitations (Lines 288–306):
   The Discussion would benefit from a clearer explanation of the assumptions made in both the logistic and Bayesian models—particularly the assumptions behind imputation and the “most ideal” and “least ideal” scenario simulations.
4. Clarify Data Scope and Sampling Frame (Lines 294–297):
   Reaffirm that Afridev pump analysis covers the majority of national data and that results are not generalizable to all water point types in Liberia or SSA.
5. Clarify and Reframe Policy Implications (Lines 307–324):
   Recommendations are relevant and practical but should be more clearly framed as implications drawn from the study’s findings. Where possible, directly link each recommendation to specific results, and avoid prescriptive language by using conditional phrasing (e.g., “may benefit from” or “could consider”). This will strengthen alignment between evidence and policy advice.
6. Tighten Language and Reduce Redundancy (Throughout):
   Consider streamlining the Discussion (e.g., Lines 153–162 vs. Lines 230–239) and improving phrasing for clarity and flow (e.g., Lines 253–258) by reducing repetition of specific statistical results already detailed in the Results section. For instance, rather than restating odds ratios and percentages related to mechanic/toolkit availability, you could focus on interpreting what these findings mean for water point sustainability and local capacity building. This would allow the Discussion to emphasize implications, draw connections to existing literature, and provide deeper insight—thereby enhancing the manuscript’s analytical value.

Author Response

This manuscript presents an empirical analysis of factors influencing water point functionality in Liberia using a large national dataset of over 11,000 Afridev handpumps. By applying multivariable logistic regression and Bayesian Network modeling, the study assesses associations between functionality and variables such as maintenance structure, repair time, mechanic availability, and payment status. The inclusion of scenario-based simulations offers a practical dimension to the analysis, making it policy-relevant and timely.

The manuscript offers a well-structured, policy-relevant analysis using a large national dataset and robust methods, including logistic regression and Bayesian Networks. It is clearly presented and practically valuable. To strengthen its scholarly contribution, the paper should more clearly articulate its novelty, better integrate related literature, clarify model assumptions and limitations, separate interpretation from results, and refine policy recommendations to reflect the study’s scope. Reducing redundancy between the Results and Discussion will also improve clarity and analytical depth.

Specific Comments and Suggestions for Authors:

1. Clarify Novel Contribution (Lines 222–229):
   Clearly articulate how this study extends previous research beyond describing associations. Highlight what is methodologically, empirically, or regionally novel—particularly with respect to the Bayesian Network application.

 

Response: Thank you for this comment- we revised the introduction to emphasize novelty. We’ve addressed the novelty element elsewhere, but briefly - it’s a large dataset (10,000+ observations) of which there are few national datasets out there like this; its the first time someone has published on these data; and there are some novel interactions between variables. We hope our revisions are helpful.

 

 

1. Improve Literature Integration (Lines 230–284):
   Strengthen the Discussion by comparing your findings to those in similar contexts (e.g., Nigeria, Tanzania). For example, claims about WASH committee effectiveness (Lines 259-264), repair timelines (Lines 242–245) and section 4.5 (Lines 271-287) would benefit from references to previous studies.

 

Response: Thank you for this comment,  we added references to other key studies on these topic.

 

1. Clarify Model Assumptions and Limitations (Lines 288–306):
   The Discussion would benefit from a clearer explanation of the assumptions made in both the logistic and Bayesian models—particularly the assumptions behind imputation and the “most ideal” and “least ideal” scenario simulations.

Response: Thank you for this comment, we have modified in our methodology section that we followed BN best practices when constructing the model. The scenarios are derived from the modifiable variable with the highest odds ratio from the multivariate logistic regressions. 

1. Clarify Data Scope and Sampling Frame (Lines 294–297):
   Reaffirm that Afridev pump analysis covers the majority of national data and that results are not generalizable to all water point types in Liberia or SSA.

 

Response: Thank you for this comment, We clarified that the analysis is limited to Afridev handpumps, which represent 90% of the dataset, and results may not generalize to other pump types.

 

1. Clarify and Reframe Policy Implications (Lines 307–324):
   Recommendations are relevant and practical but should be more clearly framed as implications drawn from the study’s findings. Where possible, directly link each recommendation to specific results, and avoid prescriptive language by using conditional phrasing (e.g., “may benefit from” or “could consider”). This will strengthen alignment between evidence and policy advice.

 

Response: Thank you for this comment, We revised the conclusion to use conditional language (e.g., “may benefit from”) and directly linked recommendations to specific findings.

 

1. Tighten Language and Reduce Redundancy (Throughout):
   Consider streamlining the Discussion (e.g., Lines 153–162 vs. Lines 230–239) and improving phrasing for clarity and flow (e.g., Lines 253–258) by reducing repetition of specific statistical results already detailed in the Results section. For instance, rather than restating odds ratios and percentages related to mechanic/toolkit availability, you could focus on interpreting what these findings mean for water point sustainability and local capacity building. This would allow the Discussion to emphasize implications, draw connections to existing literature, and provide deeper insight—thereby enhancing the manuscript’s analytical value.

 

Response: Thank you for this comment, we have restructured the discussion section to reduce redundancy and improved the clarity and flow while offering meaningful insight based on the result section. 

 

 

Reviewer 3 Report

Comments and Suggestions for Authors

This manuscript investigates the factors associated with rural water point functionality in Liberia, using a large national dataset and applying both logistic regression and Bayesian network analysis. The topic is important and policy-relevant, particularly in the context of efforts to strengthen rural WASH services across Sub-Saharan Africa. The structure of the paper is generally clear, and the methods are appropriate. Below are my specific comments and suggestions for improvement:

1. It would help to explain more clearly why specific variables were selected for analysis. While the variables seem reasonable, it’s not always obvious how they were chosen or what prior work supports their inclusion. A stronger link to existing literature or theoretical frameworks would make the analysis more grounded.
2. There are inconsistencies in the reported results. For example, the abstract mentions a 92.5% functionality rate in the most ideal scenario, but the main text refers to 77.7%. Please double-check and make sure all figures are aligned throughout the manuscript.
3. The interpretation of the Bayesian network findings seems a bit overstated. The Bayes Factor of 1.71 suggests only weak evidence in favor of the most ideal scenario, yet the paper describes this as a significant improvement. I would recommend tempering the language in this section to better match the strength of the evidence.
4. The result that WASH committees are associated with lower functionality than other managing groups is unexpected. This goes against findings from other countries, and it would be useful to explore possible reasons—whether related to capacity, oversight, or other contextual issues.
5. The handling of missing data should be explained in more detail. While the method used is noted, readers would benefit from knowing how much data were missing for key variables and whether the assumption of data being missing completely at random is reasonable in this context.
6. The policy recommendations in the conclusion could be made more specific. For example, rather than recommending better management or training in general terms, it would be more helpful to suggest concrete steps or programs that could be implemented based on the findings.
7. The statistical model diagnostics could be presented more fully. The AUC value is reported, but the results of the Hosmer-Lemeshow test or other fit statistics would help to assess the reliability of the logistic regression model.
8. The Bayesian network diagrams are quite dense and difficult to read. Simplifying them or supplementing the figures with a summary table showing key probability changes under different scenarios would make the results easier to interpret.
9. Since the data were collected in 2017, it would be helpful to include a short note on how conditions in Liberia may have changed since then, and whether the findings are still likely to reflect current challenges in water point functionality.
10. The study finds no significant relationship between payment mechanisms and water point functionality, which contrasts with studies from other regions. It would be helpful to discuss why this may be the case in the Liberian context, and what this implies for designing sustainable financing models at the community level.

Author Response

This manuscript investigates the factors associated with rural water point functionality in Liberia, using a large national dataset and applying both logistic regression and Bayesian network analysis. The topic is important and policy-relevant, particularly in the context of efforts to strengthen rural WASH services across Sub-Saharan Africa. The structure of the paper is generally clear, and the methods are appropriate. Below are my specific comments and suggestions for improvement:

 

1. It would help to explain more clearly why specific variables were selected for analysis. While the variables seem reasonable, it’s not always obvious how they were chosen or what prior work supports their inclusion. A stronger link to existing literature or theoretical frameworks would make the analysis more grounded.

 

Response: Thank you for this comment, part of the reason for the selection of variables was based on what was in the dataset that also had hypothesized or theoretical relationships with functionality, though may have not been empirically demonstrated. Based on comments from another reviewer we have added additional literature and linked our study to what has been previously done. 

 

1. There are inconsistencies in the reported results. For example, the abstract mentions a 92.5% functionality rate in the most ideal scenario, but the main text refers to 77.7%. Please double-check and make sure all figures are aligned throughout the manuscript.

 

Response: Thank you for your comment, we have confirmed that the abstract, tables, and text match

 

1. The interpretation of the Bayesian network findings seems a bit overstated. The Bayes Factor of 1.71 suggests only weak evidence in favor of the most ideal scenario, yet the paper describes this as a significant improvement. I would recommend tempering the language in this section to better match the strength of the evidence.

 

Response:Thank you for this comment, we have removed using Bayes factor as an evaluation to strength of association and used posterior odds ratio instead. We hope this change makes our result more initiative to understand. 

1. The result that WASH committees are associated with lower functionality than other managing groups is unexpected. This goes against findings from other countries, and it would be useful to explore possible reasons—whether related to capacity, oversight, or other contextual issues.

 

Response: Thank you for this comment.  We have expanded the discussion to include an exploration of why this might be, including recent work on the professionalization of handpump service delivery which intends to address some of the limitations of community-based management

 

1. The handling of missing data should be explained in more detail. While the method used is noted, readers would benefit from knowing how much data were missing for key variables and whether the assumption of data being missing completely at random is reasonable in this context.

 

Response: Thank you for this comment. The amount of missing data for each variable has been specified in the footnote of table 3. We modified the manuscript to include more details of justification for our MCAR assumptions.

 

1. The policy recommendations in the conclusion could be made more specific. For example, rather than recommending better management or training in general terms, it would be more helpful to suggest concrete steps or programs that could be implemented based on the findings.

 

Response:  Thank you for this comment, We have clarified one such policy improvement is the training of mechanics to conduct repairs and equipping them with necessary tools. We have also clarified that by effective management we mean a management structure capable of collecting funds to pay such an adequately-trained and -equipped mechanic.

 

1. The statistical model diagnostics could be presented more fully. The AUC value is reported, but the results of the Hosmer-Lemeshow test or other fit statistics would help to assess the reliability of the logistic regression model.

 

Response: Thank you for this comment. In large datasets like ours, traditional goodness-of-fit tests such as the Hosmer-Lemeshow test often flag models as misfitting due to their sensitivity to sample size. To address this, we have included additional model evaluation techniques, such as a calibration plot in our updated manuscript. 

 

1. The Bayesian network diagrams are quite dense and difficult to read. Simplifying them or supplementing the figures with a summary table showing key probability changes under different scenarios would make the results easier to interpret.

 

Response: Thank you for this suggestion, We have replaced these figures with simplified diagrams

 

1. Since the data were collected in 2017, it would be helpful to include a short note on how conditions in Liberia may have changed since then, and whether the findings are still likely to reflect current challenges in water point functionality.

Thank you for this comment. We agree that the age of the dataset is an important consideration. In response, we have expanded the limitations section (beginning at line 290) to clarify that, although the data are from 2017, the underlying service delivery structures and institutional capacity challenges in Liberia have remained largely consistent in recent years. This supports the continued relevance of the findings for current WASH planning and policy.

 

1. The study finds no significant relationship between payment mechanisms and water point functionality, which contrasts with studies from other regions. It would be helpful to discuss why this may be the case in the Liberian context, and what this implies for designing sustainable financing models at the community level.

 

Response:Thank you for your comment,  we have expanded the conclusion section to highlight this finding of our work, stating that payment mechanisms or revenue generation are not sufficient to improve functionality. Rather, using those funds to pursue adequate technical management (including a mechanic with necessary tools) improves functionality. We hope this editing is helpful. 

Reviewer 4 Report

Comments and Suggestions for Authors

Dear authors,

 

The manuscript focuses solely on a statistical analysis of hand pump data in Liberia, using an already published dataset. Unfortunately, the novelty of the study is not clearly articulated, and the contribution to the current body of knowledge remains unclear.

The analysis is limited to two statistical methods: linear regression and Bayesian network analysis. While these are valid techniques, the paper does not incorporate more advanced or modern methodologies, such as machine learning techniques, which are increasingly standard in contemporary research for improving predictive accuracy and insight extraction.

Furthermore, there is no critical or visual exploration of the dataset that illustrates the interrelationships among the predictors. Such analysis would have helped to clarify and support the findings.

Overall, the manuscript lacks the depth and innovation required for consideration as a technical or scientific paper. Rather, it represents a basic application of statistical tools to an existing dataset. The work could be considered a component or preliminary layer within a more comprehensive framework—for instance, in the context of Digital Twin Models, where survey data and statistical analyses serve as just one layer in a much more complex and practically applicable system.

For these reasons, I regret to recommend that the manuscript is not suitable for publication in its current form.

Author Response

Dear reviewer,

Thank you for taking time to review our manuscript, we respectfully disagree with several of the points raised in this review.

First, the suggestion that our methods lack sophistication because they include "only" logistic regression and Bayesian networks overlooks the fact that both are widely recognized as machine learning techniques more specifically, supervised and unsupervised learning models used in a broad range of applied scientific research. Logistic regression, in particular, has long been a foundational tool in both statistics and ML, and Bayesian networks represent a structured probabilistic model that offers interpretable insights, especially valuable in policy-relevant research. Ironically, if the concern is that these are “basic” techniques, one wonders why they are simultaneously excluded from being considered “modern” or “advanced.”

Second, the expectation that novel scientific research must necessarily invent new methods or employ the flashiest available tools (such as unspecified “machine learning techniques”) misunderstands the nature of applied research. Our manuscript does not aim to develop methodology; it applies established and interpretable techniques to a novel setting, with real-world relevance. Much policy-focused empirical work—especially in global development and infrastructure—relies precisely on “existing datasets” and “basic” statistical techniques to inform decision-making. If these approaches are too pedestrian, one might also question the bulk of evidence-based policymaking research published in journals like Water Resources Research or World Development.

Third, while we understand the reviewer’s enthusiasm for Digital Twin Models, this paper does not purport to build such a system. Nor should every study be required to do so in order to merit publication. The expectation that all researchers adopt the reviewer’s specific modeling preference suggests a rather narrow view of acceptable contributions within this domain.

Rregarding novelty: this is the first academic paper to our knowledge using this dataset. In response to other reviewer comments, we’ve strengthened our articulation of the study’s novelty and situated it clearly within the existing literature. Our visualizations (DAGs, regression tables, and a geospatial map) are standard for this type of work and meet or exceed what is typically found in comparable peer-reviewed studies of rural water point functionality.

 

Reviewer 5 Report

Comments and Suggestions for Authors

The paper presents a robust analysis of factors influencing hand pump functionality in Liberia, combining logistic regression and Bayesian network (BN) models. The study is well-structured, methodologically sound, and addresses an important gap in rural water supply sustainability. However, several areas require clarification, refinement, and additional context to enhance the manuscript’s impact and readability.

1.The abstract lacks specific quantitative takeaways (e.g., "33% to 92.5% functionality" is mentioned but not contextualized).

2.The SDG 6 discussion is generic. Liberia’s specific challenges (e.g., 25% lack basic water access) are buried in later sections.

3.The data cleaning process (e.g., hot deck imputation) is described technically but lacks justification for choosing MCAR over MAR (Missing at Random).

4.The BN model’s "hypothesized causal relationships" are not explicitly tied to literature or expert input.

5.Table 3’s descriptive statistics are clear, but Table 4’s odds ratios (ORs) could better highlight key findings.

6.The BN’s "most ideal" scenario (77.7% functionality) contrasts with the logistic regression’s "92.5%" claim in the Abstract.

7.The WASH committee paradox (lower functionality vs. literature) is noted but not theorized.  Propose explanations (e.g., "Liberian committees may lack training or resources compared to Nigerian counterparts [28]").

8.The BN’s "weak evidence" (BF=1.71) undermines the strong logistic regression results. Discuss why BN and regression diverge (e.g., "BNs capture nonlinear interactions missed by regression").

9.The cross-sectional design limitation is acknowledged but could be sharper. Emphasize causality gaps (e.g., "Cannot determine if management improves functionality or functional systems attract better management").

10. Missing data assumptions (MCAR) need validation. Suggest sensitivity analyses (e.g., "Future work could test MAR-based imputation").

11. Define "AUC," "ROC" at first use.

12.Label axes in Figures 2–3 (e.g., "Probability of functionality (%)").

13.Simplify dense passages (e.g., "Bayes factor quantifies evidence strength" vs. current phrasing).

14. Policy Implications: Expand recommendations (e.g., "Prioritize toolkit distribution to mechanics in Q4 distance areas").

15. Consider adding a map of Liberia’s water point distribution.

16.Ensure all citations match the journal’s format (e.g., [12] lacks a volume number).

Author Response

The paper presents a robust analysis of factors influencing hand pump functionality in Liberia, combining logistic regression and Bayesian network (BN) models. The study is well-structured, methodologically sound, and addresses an important gap in rural water supply sustainability. However, several areas require clarification, refinement, and additional context to enhance the manuscript’s impact and readability.

 

1.The abstract lacks specific quantitative takeaways (e.g., "33% to 92.5% functionality" is mentioned but not contextualized).

 

Response: Thank you for your comment. We have added specific quantitative results to the abstract.

 

2.The SDG 6 discussion is generic. Liberia’s specific challenges (e.g., 25% lack basic water access) are buried in later sections.

 

Response: Thank you for this comment, we have simplified the intro and we now hope Liberia’s challenges come across more clearly.

 

3.The data cleaning process (e.g., hot deck imputation) is described technically but lacks justification for choosing MCAR over MAR (Missing at Random).

 

Response: Thank you for this comment, we have elaborated on justification on MCAR over MAR. 

 

4.The BN model’s "hypothesized causal relationships" are not explicitly tied to literature or expert input.

 

Response: Thanks for pointing this out, we should have clarified further in the methods. We developed these based on BN best practices, where we used expert input and available literature to set up the model. 

 

5.Table 3’s descriptive statistics are clear, but Table 4’s odds ratios (ORs) could better highlight key findings.

 

Response: Thank you for this comment, we have modified our logistic regression tables to highlight key findings. 

 

6.The BN’s "most ideal" scenario (77.7% functionality) contrasts with the logistic regression’s "92.5%" claim in the Abstract.

 

Response: Thank you for this comment, we have corrected this mistake.

 

7.The WASH committee paradox (lower functionality vs. literature) is noted but not theorized.  Propose explanations (e.g., "Liberian committees may lack training or resources compared to Nigerian counterparts [28]").

 

Response: Thank you for this comment, we have elaborated on this paradox in our discussion section.

 

8.The BN’s "weak evidence" (BF=1.71) undermines the strong logistic regression results. Discuss why BN and regression diverge (e.g., "BNs capture nonlinear interactions missed by regression").

 

Response: Thank you for this comment, we have removed using BF as a strength of association and used Posterior odds ratio instead, we hope this will make our results more initiative to interpret and understand. 

 

9.The cross-sectional design limitation is acknowledged but could be sharper. Emphasize causality gaps (e.g., "Cannot determine if management improves functionality or functional systems attract better management").

 

Response: Thank you for pointing out, we have revised our manuscript to elaborate on the causality and limitations regarding cross-sectional design. 

 

 

10. Missing data assumptions (MCAR) need validation. Suggest sensitivity analyses (e.g., "Future work could test MAR-based imputation").

 

Response: Thanks you for this comment, we have elaborated our justification of using MCAR in our updated manuscript. 

 

11. Define "AUC," "ROC" at first use.

 

Response:Thank you for your comment. We have now defined "AUC" (Area Under the Curve) and "ROC" (Receiver Operating Characteristic) at their first mention. We have also added a glossary of abbreviations at the end of the manuscript. While these are well-known statistical terms, we agree that defining them improves accessibility for a broader audience.

12.Label axes in Figures 2–3 (e.g., "Probability of functionality (%)").

 

Response: Thank you for this comment, we have simplified our figures to improve clarity.

 

13.Simplify dense passages (e.g., "Bayes factor quantifies evidence strength" vs. current phrasing).

 

Response: Thank you for this comment, we have made bulk editing throughout the manuscript to simplify dense passages. 

 

14. Policy Implications: Expand recommendations (e.g., "Prioritize toolkit distribution to mechanics in Q4 distance areas").

 

Response: Thank you for this comment, we have expanded our policy recommendations in the study outcomes section. 

 

15. Consider adding a map of Liberia’s water point distribution.

 

Response: Thank you, a map has been included and listed a figure 1.

 

16. Ensure all citations match the journal’s format (e.g., [12] lacks a volume number).

 

Response: Thank you for pointing this out, we have double checked our citation.

Round 2

Reviewer 3 Report

Comments and Suggestions for Authors

Thank you for the opportunity to review this manuscript. The topic is highly relevant and the authors have done a solid job presenting their analysis. The revised version shows clear improvements in structure, clarity, and depth of discussion. Overall, it’s a well-executed piece of work with meaningful implications for rural water service delivery. If possible, it would be helpful to retain the track changes version alongside the clean one in future revisions.

Author Response

Thank you for the opportunity to review this manuscript. The topic is highly relevant and the authors have done a solid job presenting their analysis. The revised version shows clear improvements in structure, clarity, and depth of discussion. Overall, it’s a well-executed piece of work with meaningful implications for rural water service delivery. If possible, it would be helpful to retain the track changes version alongside the clean one in future revisions.

Response: 

Thank you for taking the time to review our manuscript, your input is highly valued and appreciated. In future submissions we will be mindful to submit a version that highlights changes. 

Reviewer 4 Report

Comments and Suggestions for Authors

The authors have not adequately addressed the reviewer’s suggestions in the revised manuscript. Consequently, the manuscript should be rejected, as it does not present any significant novelty or contribution to the field. The analysis provided does not meet the standards required for publication in a prestigious journal.

Author Response

The authors have not adequately addressed the reviewer’s suggestions in the revised manuscript. Consequently, the manuscript should be rejected, as it does not present any significant novelty or contribution to the field. The analysis provided does not meet the standards required for publication in a prestigious journal.

Response:

 

Thank you for taking our time to review our manuscript. Please see the previous review comments and the comments in this round of comments. If these are inadequate, specifics would be helpful. 

Reviewer 5 Report

Comments and Suggestions for Authors

This manuscript presents an analysis of 11,065 Afridev handpumps in Liberia, using both multivariable logistic regression and Bayesian network models to explore determinants of water point functionality. The study has clear strengths, including a large and nationally representative dataset, a combination of complementary analytical methods, and scenario simulations that provide actionable policy insights. The paper is generally well-organized, the statistical analyses are sound, and the discussion connects the findings to relevant literature. The conclusions are policy-relevant and could inform sustainable rural water supply management in Liberia and similar contexts.
However, improvements are needed to enhance clarity, methodological transparency, and presentation quality. Specifically, sections could benefit from more concise writing, clearer justification of certain methodological choices, improved data visualization, and deeper reflection on the limitations and generalizability of the findings. I recommend minor revision before acceptance.

1. Abstract:

1). Simplify sentence structures and reduce methodological detail.

2). Emphasize key numerical results and policy implications.

3). Consider adding methodological keywords such as Bayesian Network and logistic regression for better indexing.

2. Introduction

1). Update literature review with more recent studies (2022–2025) specific to Liberia or comparable contexts.

2). Clearly articulate the study’s novelty compared to prior work (e.g., larger dataset, combined modeling, scenario simulation).

3. Materials and Methods

1). Expand on the rationale for using the hot-deck imputation with Approximate Bayesian Bootstrap (ABB), including parameter settings and any sensitivity checks.

2). Clarify the definition of “toolkit” earlier in the methods section, given its importance as a predictor variable.

3). Provide more details on how the Bayesian network structure was determined (e.g., expert panel size, variable discretization thresholds).

4. Results

1). Improve table readability (e.g., highlight statistically significant results, group categories for clarity).

2). Reduce redundancy between text and tables; focus the narrative on the most policy-relevant findings.

3). Ensure that figure captions, especially for scenario simulations, are self-contained and clearly explain the scenario assumptions and probability values.

5.  Discussion

1). Expand on cross-country comparability by integrating examples from Malawi, Ghana, Uganda, etc., to assess the transferability of findings.

2).Separate technical recommendations from management recommendations for clarity.

3). Clearly distinguish between statistical significance and policy relevance, particularly for non-significant predictors like payment structure.

6. Limitations

1). Further discuss sample representativeness and potential differences between Afridev and other pump types.

2). Provide quantitative evidence supporting the MCAR assumption (e.g., missingness tests).

7. Language and Formatting

1). Reduce sentence length and eliminate repetitive statements for conciseness.

2). Ensure that in-text citations and reference formatting strictly follow Sustainability guidelines.

3). Check figure and table numbering for consistency with the main text.

Author Response

This manuscript presents an analysis of 11,065 Afridev handpumps in Liberia, using both multivariable logistic regression and Bayesian network models to explore determinants of water point functionality. The study has clear strengths, including a large and nationally representative dataset, a combination of complementary analytical methods, and scenario simulations that provide actionable policy insights. The paper is generally well-organized, the statistical analyses are sound, and the discussion connects the findings to relevant literature. The conclusions are policy-relevant and could inform sustainable rural water supply management in Liberia and similar contexts.
However, improvements are needed to enhance clarity, methodological transparency, and presentation quality. Specifically, sections could benefit from more concise writing, clearer justification of certain methodological choices, improved data visualization, and deeper reflection on the limitations and generalizability of the findings. I recommend minor revision before acceptance.

1. Abstract:

1). Simplify sentence structures and reduce methodological detail.

Response: Thank you for this comment, we have simplified sentence structures and reduced methodological detail.

2). Emphasize key numerical results and policy implications.

Response: Thank you for this comment, we have  further revised our abstract to emphasize key numerical results and policy implications.

3). Consider adding methodological keywords such as Bayesian Network and logistic regression for better indexing.

Response: Thank you for this comment, we have added Bayesian Network into the keywords.

2. Introduction

1). Update literature review with more recent studies (2022–2025) specific to Liberia or comparable contexts.

Response: Thanks. In the previous round, we updated the manuscript with other studies from comparable contexts, however, there are no studies recently conducted from Liberia on this topic, hence why we did this study which helps to demonstrate its novelty.

2). Clearly articulate the study’s novelty compared to prior work (e.g., larger dataset, combined modeling, scenario simulation).

Response: Thank you for this comment, we have stated in our introduction that “Unlike earlier studies, we integrate both modeling techniques to assess management and repair dynamics at scale, and simulate how combinations of service delivery factors influence functionality outcomes. This approach helps clarify the mechanisms through which governance and operational capacity affect rural water service sustainability.”

3. Materials and Methods

1). Expand on the rationale for using the hot-deck imputation with Approximate Bayesian Bootstrap (ABB), including parameter settings and any sensitivity checks.

Response: Thank you for this comment, we have expanded the section to clarify the rationale behind this methodology and included further details regarding parameter settings.

2). Clarify the definition of “toolkit” earlier in the methods section, given its importance as a predictor variable.

Response: Thank you for this comment, we have clearly defined the definition of “toolkit” to the best of our ability very early on (please see section 2.2 “ The toolkit likely refers to a basic set of tools needed for routine maintenance of Afridev handpumps, which are designed to be easily repaired with minimal equipment.”

3). Provide more details on how the Bayesian network structure was determined (e.g., expert panel size, variable discretization thresholds).

Response: We used expert knowledge and literature review to determine the Bayesian network structure.

4. Results

1). Improve table readability (e.g., highlight statistically significant results, group categories for clarity).

Response: Thank you for this comment, we have made significant changes to our tables to highlight statistically significant results and group categories for clarity

2). Reduce redundancy between text and tables; focus the narrative on the most policy-relevant findings.

Response: Thank you for this comment, we have reduced redundancy and have revised our section to be focused on policy driven finding. However, most policy-relevant findings can be found in the discussion section. 

3). Ensure that figure captions, especially for scenario simulations, are self-contained and clearly explain the scenario assumptions and probability values.

Response: Thank you for this comment, we have revised the figure title to make them self-contained.

5. Discussion

1). Expand on cross-country comparability by integrating examples from Malawi, Ghana, Uganda, etc., to assess the transferability of findings.

Response: While this study is focused on Liberia and cross-sectional, given the size of the dataset, there is potential transferability of findings to other country contexts using similar water point types. However, differences in environmental, policy conditions, and other factors warrant careful interpretation across contexts.

2).Separate technical recommendations from management recommendations for clarity.

Response: Thank you for this comment, we have separated technical recommendations from management recommendations

3). Clearly distinguish between statistical significance and policy relevance, particularly for non-significant predictors like payment structure.

Response: Thank you for this comment, we have specified that although Fee structure was not statistically significantly associated with water point functionality in this study, it may be an important financial management factor related to other rural water supply or community outcomes. However, the situation also reflects a wider shift in the rural water supply sector, which increasingly recognizes that financial sustainability cannot depend solely on informal, volunteer-led systems.

6. Limitations

1). Further discuss sample representativeness and potential differences between Afridev and other pump types.

Response: 

Thank you for this comment, we have updated the limitations section to include maintenance and repairability across other pump types and generalizability concerns. 

 

2). Provide quantitative evidence supporting the MCAR assumption (e.g., missingness tests).

Response: Thank you for this comment, in large survey data like ours with more than 11,000 entries and more than 50 variables, the quantitative missing tests often has its power limited and it can reject MCAR in large samples for trivial differences. We have specified that this assumption was informed by visual inspection of missingness patterns and missing data matrices. However, further work is needed to fully assess the validity of this assumption and understand the potential influence of missingness on our findings. 

7. Language and Formatting

1). Reduce sentence length and eliminate repetitive statements for conciseness.

Response: Thank you for this comment, we have made bulk editing of our manuscript to eliminate repetitive statements and long sentences. 

2). Ensure that in-text citations and reference formatting strictly follow Sustainability guidelines.

Response: Thank you for this comment, we have double checked out in-text citations and references that are in compliance with journal guidelines. 

3). Check figure and table numbering for consistency with the main text

Response: Thank you for this comment, we have ensured our figure and table numbering match.