Bayesian Structural Time Series and Geographically Weighted Logistic Regression Modelling Impacts of COVID-19 Lockdowns on the Spatiotemporal Patterns of London’s Crimes

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This article selected five types of crimes and used the spatiotemporal data of crime spots. Taking London’s 982 MSOAs as basic spatial units, it conducted a detailed analysis of the impact of COVID-19 lockdowns on the spatiotemporal patterns of crimes in London, elaborating on the spatiotemporal changes of various types of crimes during the three lockdown phases. It quantitatively analyzed the socioeconomic factors influencing these changes and obtained some valuable results and conclusions. There are some areas for improvement:

1. The article mentioned that the restriction levels during the three lockdown phases were different, and they were incorporated into the BSTS model to improve the model’s prediction accuracy. It is recommended that the authors explain how the restriction levels were quantified.

2. To improve the BSTS model fit, the article introduced factors such as holidays and temperature as covariates into the model and reached the conclusion stated in line 567. I would kindly request that the authors provide a comparison chart in order to help readers understand how the inclusion of holidays and temperature indeed improved the model fit.

3. In the logistic regression analysis in section 5.3.2, the same factor showed huge differences in the odds ratio among the three phases. In addition to the descriptions of these differences, I would hope that the authors could analyze their underlying real-world significance.

4. Lines 469--470 mentioned a significant decline in violent crimes after the lifting of restrictions, which seemed to contradict Figure 7(e).

5. It appears that the purpose of figure 9 was to show the correlation between spatial units and their adjacent units, but the authors analyzed this as if they were using a hotspot spatial analysis map. Would it be more appropriate for figure 9 to apply a hotspot spatial analysis map directly, since the two types of maps can convey different content and have different calculation principles? It is only when all correlations are positive (i.e., all “high-high” or “low-low”) that the two types of maps share a high similarity. It is suggested that the authors rephrase this analysis.

Author Response

1. The article mentioned that the restriction levels during the three lockdown phases were different, and they were incorporated into the BSTS model to improve the model’s prediction accuracy. It is recommended that the authors explain how the restriction levels were quantified.

 

Thank you for the great suggestion, and the unclear expression had been revised accordingly. In lines 272-276 of the article (section 4.2.4), we modified the quantification for varied restriction levels as below:

“Given that the data on restriction levels are organized on weekly basis, the restriction level for each week has been calculated by counting the quantity of varied restriction measures in place during the target week. For example, if a week experienced three types of restrictions, namely school closure, pub shutdown, and shop closure, then its restriction level is valued as 3”.

 

2. To improve the BSTS model fit, the article introduced factors such as holidays and temperature as covariates into the model and reached the conclusion stated in line 567. I would kindly request that the authors provide a comparison chart in order to help readers understand how the inclusion of holidays and temperature indeed improved the model fit.

 

It would be an innovative idea to test out comparing the model performances with/without the inclusions of temperature and holiday factors in the BSTS model. But we may hold on to add such practices at current stage, and would be more than happy to try it out in our ongoing project on London’s stop & search exploration (another paper producing) for following reasons:

• the factors were selected to represent and capture the crime data’s seasonality - temperature (e.g., literature [13]) and temporal emerging events related to crime opportunities – holiday (e.g., literature [18]), in the equal status of other selected socio-economic variables. If we compare the results inclusive/exclusive of these two covariates, then future readers may wonder how about other scenarios for other factors? It will then add more contents and discussions under varied scenarios, instead shift the readers’ attention from the main outcomes we’d like to impress them with.
• Practically, it seems impossible for us to complete the request within the journal’s required 7 days’ returning period (especially during Christmas holiday when the University IT supports are away until early January). Because it is in fact to redo all data processing and model tests but omitting the 2 factors; the maps and figures would need to be compared as well for compactness. In addition, the previous modelling running took longer than using the University server. But this is a good idea and we would like to adopt such comparisons inclusive and exclusive of selected factors in the ongoing project (on stop & search).

 

3. In the logistic regression analysis in section 5.3.2, the same factor showed huge differences in the odds ratio among the three phases. In addition to the descriptions of these differences, I would hope that the authors could analyze their underlying real-world significance.

 

Thank you for your suggestion! Exploring the reasons behind the change in odds ratios over time can provide a great perspective. We had taken the suggestions and added three paragraphs (in red) in Section 5.3.2, specifically in

• Lines 793-808: “During London’s first lockdown, a higher male ratio seemingly reduced burglary risks, possibly due to restrictions curtailing male offenders’ activities. However, such effect diminished in subsequent lockdowns, owing to potential adaptations to lockdowns. In the second lockdown, a higher proportion of teenagers significantly increased burglary risks, coinciding with its being the only period with schools open. It also experienced increased burglary risks in areas with higher population density and long-term unemployment, potentially due to financial strains and relaxed restrictions. Additionally, increased drug offences during this period might interactively influence the rises in burglaries. Simultaneously, the economic impacts during the first and second lockdowns was direct and profound, with the Coronavirus Job Retention Scheme (CJRS), also known as the ’furlough’ scheme, helping to bridge the support gaps. By the time of the third lockdown, enhancements to the ’furlough’ scheme had helped to relief unemployment pressures, reducing the likelihood of potential offenders resorting to burglaries. Instead, during the third lockdown, a higher proportion of elderly population surprisingly decreased burglary risks, who were thought to adhere more strictly to lockdown measures, enhancing household occupancy and thus prevent burglary”.
• Lines 822-829: “The increases of robbery crimes during the initial lockdown phases appeared to be significantly correlated with the rises in long-term unemployment, a trend likely exacerbated by the economic pressures resulting from the lockdowns. However, with the continuous refinement of the 'furlough' scheme, this program started to see significantly ease of the financial strain on the unemployed population in the third lockdown, thereby reducing the propensity for robbery. Moreover, it is noteworthy that during the second lockdown, the escalation in drug offences also increased the risks of robbery, suggesting that the uptick in acquisitive crimes may be attributed to the rises in drug offences under the relaxed lockdown measures.”
• Lines 840-854: “Initially, it was observed that the population aged over 60 exhibited a reduced risk of theft crimes, specifically during the second lockdown period. Such reduction could be attributed to their continuing adherence to lockdown regulations and lesser frequencies leaving home despite the relaxation of restrictions. As a consequence, it diminished their likelihoods of becoming victims of thefts. Secondly, individuals without long-term unemployment showed a decreased risk of thefts during the second and third lockdowns, as opposed to the first lockdown. This variation may be due to the less severe economic pressures comparing with the first lockdown. During later lockdowns, those who had been employed in the preceding 12 months were likely to have a more robust financial status, making them less prone to engage in thefts. Lastly, an intriguing phenomenon was noted during the second lockdown in that, areas with higher drug offences inversely correlated with a lower risk of increased thefts. This trend could be associated with the reallocation of police resources, especially following the surge in drug offences during the first lockdown, and further led to a heightened focus on such crimes, possibly at the expense of addressing theft-related activities”.

 to elucidate the factors driving these changes in odds ratios over time.

 

4. Lines 469--470 mentioned a significant decline in violent crimes after the lifting of restrictions, which seemed to contradict Figure 7(e).

 

Thank you for your suggestion, and we apologize for the errorous statement. Violent crime did not show a significant decrease, and we have corrected this in our presentation.

 

5. It appears that the purpose of figure 9 was to show the correlation between spatial units and their adjacent units, but the authors analyzed this as if they were using a hotspot spatial analysis map. Would it be more appropriate for figure 9 to apply a hotspot spatial analysis map directly, since the two types of maps can convey different content and have different calculation principles? It is only when all correlations are positive (i.e., all “high-high” or “low-low”) that the two types of maps share a high similarity. It is suggested that the authors rephrase this analysis.

 

Thank you for your suggestion. We did intend to use regions’ spatial associations to their “high-high” cluster or hot spots, and “low-low” cluster or cold spots (Anselin, 1995; GeoDa). To make the statements clearer, we had rephrased to clarify our intent (Lines 546-552) as below.

“Red areas exhibit a "high-high" clustering, or so-called “hot spots”, indicating that both the target areas themselves and their surrounding neighbourhoods have high crime rates. This pattern suggests regions with frequent criminal activity, possibly due to specific socio-economic or environmental factors. Conversely, blue areas show a "low-low" clustering, or so-called “cold spots”, meaning that both these regions and their adjacent areas have lower crime rates. This may indicate better community management, a higher sense of safety, or other crime deterrent factors in place”.

It is understood the density-based point pattern hotspots analysis, but considering this dataset’s polygon nature, and the widely utilised LISA cluster mapping in crime research to identify “hot spots” and “cold spots”, we hope the current revision could be accepted.

 

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

All comments are written in a pdf file.

 

 

Comments for author File: Comments.pdf

Author Response

1. Routine activity and strain criminological theories are assumed valid, yet there is substantial criticism of them. In effect this research can be seen as a test of the validity of routine activity and strain theories rather than the application of them

 

Thank you for your suggestion and we had made the updates accordingly. It is designed to test out theories’ applicability and suitability in London, some paragraphs had been updated in Discussion section as below:

 

Lines 705-713: “This phenomenon effectively validates the applicability of Routine Activity Theory, which postulates the occurrence of a crime to be a converge of three elements in space and time: a motivated offender, a suitable target, and the absence of a capable guardian. Lockdown measures in London had significantly altered the routine activities of potential offenders and targets, disrupting possibility of crucial convergence. Consequently, as individuals adapted to social distancing measures, crime patterns were anticipated to change into a more stable state; it then reflected the dynamic nature of criminal activity in response to changes in societal patterns. The results consolidated routine activity theory’s ability explaining crime rate fluctuations during lockdowns.”

 

Lines 957-962: “Another explanation for the abnormal increase in drug offences could be that the police resources freed up due to the reduction in acquisitive crimes during the lockdown, were then redirected to combat with drug offences. Hence, although the total number of crimes had increased, the actual number of cases had not changed. Correspondingly, the reduction in drug offences observed in some areas during Lockdown 3 might owe to drug dealers and consumers’ adapting to the restrictions in developing new anti-surveillance strategies, which led to a decrease in recorded numbers of drug offences.”

 

2. Unclear whether research data is about crime instance counts or crime rates (e.g. crime instances/10,000 residents). If the latter, then analyses relating to density may need revising? Ditto analyses in which daily population in a location differs from resident population (e.g. central areas of London.

 

Yes, that’s absolutely right, these details are crucial and should be explained more clearly. In this submission, we used the count of crimes for analysis. Among the modelling factors, population density is measured by the number of residents per square kilometer within the census tract, LSOA in London. We have provided further explanations for these points in the Data section, specifically in 4.2.2 and 4.2.5 (Lines 247-248, 283). Thank you for your feedback!

 

Daily population as suggested e.g., central London, will be a challenging topic considering the large volume of commutes in weekdays pre- or post- pandemic, but “hollow city” during the pandemic, which will add up more complexity in data mismatching issue, if be taken for this project. In light of such, we would still use Census data for consistency and also the comparability among all London LSOAs.

 

 

3. Police resources are limited and insufficient to address and record all crime (by a significant amount!). Where acquisitive crime rates fall, there are more resources available to address other issues such as expressive crime and thus recorded incidents will rise with no increase in actual crime.  Hence there are alternative explanations of increases in expressive crimes and drug crimes. For example and alternative explanation of reduced drug crimes is that drug traders and users developed new anti-surveillance methods and new ways to avoid the increased attention of stop and search policing.

 

Your insights are very meaningful, and the shift in police resources as well as changes in criminal counter-surveillance strategies are indeed likely to be factors contributing to the variation in drug crime numbers during the isolation period. We have included this information at the end of section 5.2.3 (Lines 736-743, also could be found from the response to comment Point 3).

 

4. In fact, the data and analyses in several places offer the potential for alternative criminological explanations including about the viability of the use of routine activity and strain theories of crime.

 

Thank you for your suggestion. Some of the analyses indeed appear to provide empirical support for these criminological theories based on real-world situations. We have highlighted this perspective in the discussion. In the first paragraph of section 5.2.3 (Lines 705-713).

In the first paragraph of section 5.3.4 (Line 957-962). Also could be found from response to comment Point 1.

 

5. English language is good. Some errors include extra words, missing words and misused words. Suggest word by word edit by professional.

 

Thank you for your suggestions; we have made some efforts to correct grammar errors and typos as much as possible.

Author Response File: Author Response.docx

Reviewer 3 Report

Comments and Suggestions for Authors

I enjoyed reading the paper and the data and analyses are useful to the field as a whole. The paper is well worth publishing and will be of benefit broadly. I commend the authors.

Regardless, in such a rich complicated paper there are some questions unanswered.

Routine activity and strain criminological theories are assumed valid, yet there is substantial criticism of them. In effect this research can be seen as a test of the validity of routine activity and strain theories rather than the applicaiton of them

Unclear whether research data is about crime instance counts or crime rates (e.g. crime instances/10,000 residents). If the latter, then analyses relating to density may need revising? Ditto analyses in which daily population in a location differs from resident population (e.g. central areas of London.

Police resources are limited and insufficient to address and record all crime (by a significant amount!). Where acquisitive crime rates fall, there are more resources available to address other issues such as expressive crime and thus recorded incidents will rise with no increase in actual crime.  Hence there are alternative explanations of increases in expressive crimes and drug crimes. Fro example and alternative explanation of reduced drug crimes is that drug traders and users developed new anti-surveillance methods and new ways to avoid the increased attention of stop and search policing.

In fact, the data and analyses in several places offer the potential for alternative criminological explanations including about the viability of the use of routine activity and strain theories of crime.

Comments on the Quality of English Language

English language is good. Some errors include extra words, missing words and misused words. Suggest word by word edit by professional.

Author Response

Below I list minor ambiguities that should be corrected and revised.

1. Line 27: Near references 14 and 15 listed in line 28, author Yang is not cited. I suggest that the authors check the correctness of what is written.

 

Thank you for your suggestion. In this context, we should cite the study by Yang et al. This is because there were issues with the content of references [15] and [19], which have now been corrected, and reference [15] now corresponds to the research by Yang et al.

 

2. Line 37: No reference is given for the work of Paramasivan and colleagues.

 

Thank you for your suggestion. The citation for Paramasivan et al.’s work is located in Line 40. We hope that this citation is in compliance with the standards.

 

3. Lines 38, 150 and 455: No reference citation is given for Yang et al.

Lines 42-43, 191-194, 293-294: The reference [15] does not correspond to the work of Yang et al.

 

Thank you for your suggestions. Since reference [15] has been corrected to Yang et al.’s study, the contents in these sections have all been properly cited upon revisions.

 

4. Line 43: No reference to the paper of Campedelli et al. is given.

 

Thank you for your suggestion; we have added a citation to Campedelli et al.’s study here.

 

5. Line 49: I suggest giving an example of the work when "most research" is mentioned.

 

Thank you for your suggestion; we have cited some representative studies that did not quantify the impact of COVID-19.

 

6. Line 137: The sentence starts with "Another theory proposed by Agnew,...", and references are given at the end of the sentence [12, 13]. I suggest that the authors check that reference [12] is correctly cited.

 

Thank you for your suggestion; indeed, we should not have cited reference [12] here, and it had been deleted accordingly.

 

7. Figure 2: I suggest that the figure is supplemented by a description of the coordinate axes, the units of measurement and the name of the map projection.

 

Thank you for your suggestion; we have provided explanations in the figure for the coordinate axes, measurement units, and map projection name.

 

8. Lines 251 and 256: I suggest that the authors check that the year 2013 is written correctly in reference [26].

 

Thank you for your suggestion; the publication date of the reference has been corrected to 2023.

 

9. Figure 3: In the description of the horizontal axis, "month" is given, and it can be seen from the figure that the years are given and there is no insight into all months. If necessary, it is better to enlarge the images to make the information easier for the reader to understand. Adding vertical and horizontal guidelines within the diagram would certainly improve readability. The number of holidays is always a whole number, so I see no reason to indicate 1.5, 2.5, etc. on the vertical axis of Figure 3b.

 

Thank you for your suggestion; we have added gridlines to the images to enhance readability. Additionally, we have changed the y-axis of Figure 3b to integers and updated the horizontal axis scales of both figures to years.

 

(a) Temperature	(b) Holiday count

 

10. Line 283: Unnecessary "the" can be deleted.

 

Thank you for your suggestion; we have made the correction.

 

11. Line 285: Please check the clarity of the sentence and whether it is necessary to put a comma in the part "...haven't, to assess...".

 

Thank you for your suggestion; we have restructured the sentence to make the content neater in this context.

 

12. Figures 4, 10, 11 and 12: I suggest that the name of the axis is given on the coordinate axes.

 

Thank you for your suggestion. Due to Figure 4 being referenced from other papers, we may not be able to modify its axes. However, Figures 10-12 have already been revised upon suggestion.

 

13. Line 405: The subsection describing Logistic Regression is missing.

 

Thank you for your suggestion; we have added a subsection describing logistic regression in 4.3.5.

 

14. Figures 5 and 6: For ease of interpretation, I suggest adding vertical axes for each month indicated.

 

Thank you for your suggestion. To enhance readability, we have changed the time series to cover the period from 2019 to 2023 and added axis labels for each month. This will make it easier to reference important time points mentioned in the text.

 

15. Line 464: I suggest specifying which Figure 7 should be considered and deleting the unnecessary brackets

 

Thank you for your suggestion; we have clarified which figure should be referenced and removed unnecessary parentheses.

 

16. Figure 7: The values given on the coordinate axes are too small to read

 

Thank you for your suggestion; we have increased the font size of the axes and added axis labels.

(a) All Crime	(b) Burglary
(c) Drug Offences	(d) Robbery
(e) Violence Against the Person	(f) Theft

 

17. Figure 9: The light blue color (low high) is not visible in the figure. I suggest enlarging the displayed figures slightly to make the displayed content clearer.

 

Thank you for your suggestion; we have adjusted the color bar scale to make the differentiation between colors more distinct.

 

18. I suggest to merge the sub-chapters "Discussion" into one main chapter "Discussion", which comes before the chapter "Conclusion".

 

Thank you for the insightful suggestion! We had restructured the discussion sections by merging previously Section 5.1.3, 5.2.3 and 5.3.3 into Section 6 Discussion, followed by Section 7 Conclusions.

Author Response File: Author Response.docx