Measurement of Potential Victims of Burglary at the Mesoscale: Comparison of Census, Phone Users, and Social Media Data

Round 1

Reviewer 1 Report

Thank you for revising your manuscript. My concerns have been properly addressed.
Now I understand what an "urban village" is in your context. I hope readers do not misunderstand that it is completely different from the "Urban village" in UK that is related to New Urbanism.

Author Response

Point 1: Thank you for revising your manuscript. My concerns have been properly addressed. Now I understand what an "urban village" is in your context. I hope readers do not misunderstand that it is completely different from the "Urban village" in UK that is related to New Urbanism.


 Response 1: Thank you for pointing it out. In order to prevent readers from misunderstanding the meaning of "urban village", we have added the following to the manuscript:

“It is completely different from the ‘Urban village’ in the UK, an idealised model for sustainable development and planning.” (Page 2, Lines 119-120)

The new draft uses the “track changes” function in Microsoft Word (recommended by the journal), and the line number are generated in the simple markup/no markup format.

Reviewer 2 Report

I wish to thank the authors for their changes and detailed responses to my points. I had misinterpreted a couple of elements of the paper but now they are much clearer.

I also wish to comment that the new discussion in lines 318-347 considerably strengthens the paper.

I have some minor points

1)

In response to my comment R3 you state:

"This is actually key to our research question, which is to develop a measure of residential population as a surrogate measure of burglary targets."

I would add this in the introduction and expand on sentence in line 72-73.

Replace "to find the optimal measure for potential burglary victims" with

"and aims to identify the most appropriate measure of residential population as a surrogate measure of potential burglary targets."

2) Lines 288-290.

This is only a possible explanation for your finding. You are making assumptions here and offering a plausible explanation.

Please rephrase as currently this reads as factual and there is no data to support this.

Replace "After getting of the bus, the thief..."

With (or similar)

One possible explanation here is that offenders arrive at a locality by bus, and then move away from the stop to a high density residential areas to find suitable targets, thus bus stops are negatively related to burglary. An alternative explanation is bus stops offer a measure of guardianship that increases the risk of committing a burglary.

 3) Lines 300

This show that thieves

This is one study findings so try to avoiding as proof.

This suggests that burglars prefer residential areas rather than commercial districts, likely due to a greater number of  suitable and more rewarding targets.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

The authors have done a pretty good job of revising the manuscript to address my previous critiques. 

The manuscript needs a good proofread by a native speaker -- there are some misspellings that a spell checker will not pick up. Another small writing issue is that some words seem randomly capitalised. A good proofreader will sort these problems quickly.

A few additional comments about the revised manuscript:

1) Is there any explanation for why the number of burglary cases dropped so much between 2017 and 2019? Were there different police enforcement or patrol actions? The count is more than 50% lower in the last year of the study period. It may be that whatever process is leading to the reduction is not captured in your model and this may be why it only explains a modest percentage of the variation (26% with the best fit model).

2) I wonder if you could develop a way to fuse the absolute and relative population counts. Both may be important influences on burglary counts. The former in terms of how much opportunity there is to have something to steal, the latter in terms of presence in the area (i.e., neighbours) that might observe burglaries (another important source of guardians). Perhaps this could be suggested as an avenue for future research.

3) It would be helpful to note what 'community-based variables' you are referring to on line 369.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.

Round 1

Reviewer 1 Report

I appreciate the opportunity to review the manuscript titled “Measurement of Potential Victims of Burglary on A Micro Scale.” This study aims to find the optimal measure for potential burglary victims comparing four measures; census population, census households, cell phone users, and THR. As part of the review, I have raised a few questions and made some comments that would be helpful in improving the manuscript.

1) The discussion is not convincing due to the lack of information on burglaries in the study area. What time and what day did burglaries occur frequently? Such information may improve the analyses using cell phone and THR data. For example, if there were many burglaries during the daytime, the ratio of daytime to nighttime population would be a promising variable.

2) [L.291] Is there any evidence that thieves use buses? This should also be discussed based on information about burglaries in the study area. Natural surveillance by bus users may deter crimes.

3) [L.249] Your explanation for the abnormal performance of TRH data from 0:00 to 1:00 is not convincing. So why did it perform well at 23:00? Why did it perform worse at 3:00 than 2:00? Is it possible to generalize this result in other areas? Considering the instability of AIC of TRH, why not consider splitting time periods into less than 15?

4) Please add what limitations there are on taking measures against burglaries due to the relative (not absolute) value of TRH data.

Minor comments

Subsection 3.1

- Serious differentiation exists in what?

- Please add the relationship between 15 communities and three urban villages.

- Please add what is urban village.

Reviewer 2 Report

1) The NB (Negative Binomial) model used is not defined in paper, and the reference [33] is hard to access. It seems that authors use some NB parametrization differ from traditional one, and use dependent variable Y=k*g, k~Poisson(lambda), g~Gamma(1/alpha,aplha). I have never met such parameterization

2) The quality of the models fit (deviance explained) is not given in the text, so it's hard to understand the total covariates influence

3) It seems from Figure 2 that Zero Inflated/Zero Altered Poisson can be use also to compare with NB.

Typos:
135: "The grids with an area less than 1/2 of the complete grids are cut to obtain 192 effective samples". "Cell" should be used instead of "grid"?

182: "Gamama(1/alpha,alpa)" - Gamma?

Reviewer 3 Report

This paper attempts to develop a measure of residential ambient population to support an understanding of the occurrence of residential burglary.

However, I am not sure if the theories used are fully understood, nor if they can be applied in the way the author(s) are currently seeking to do. I below explain my theoretical problems with the paper. I would also suggest the writing could be improved in several sections as it is difficult to follow. However, the comments below relate to theoretical and methodological issues I have with the paper.

The paper needs to reconsider the likely impact of the ambient population on residential burglary and both (i) the justification for this analysis; (ii) the expected/anticipated results based on theory, and (iii) the interpretation of the findings in light of this.

1) There has been an extensive micro level analysis of burglary internationally, indeed spatially it is one of the crimes we know most about given its spatial certainty. In the Western world it is one of better reported crimes due to insurance requirements. However, the timing of burglary is problematic given it often occurs when a person is not in their property – see work of Ratcliffe on Aoristic analysis to support this.

2) Micro level theories of burglary draw heavily on the concepts of repeat victimisation and near repeat victimisation which you do not discuss. These have emerged from routine activity, rational choice theory and crime pattern theory. Therefore, at the micro level - the unit of analysis is frequently the individual property rather than a census area of grid. I suggest your need to consider each of these theories as they concurrently influence opportunities for burglary – I am not sure you can only draw on rational choice theory given you are working with spatial grids.

3) When aggregated units (grids are used) a burglary rate is often included – to account for the number of houses in each grid. Thus you may identify a burglary rate per 100 households for example to standardise across grids. I suspect that the findings on page 8 lines 273-275 are reflective of increasing household densities.

4) The ambient population has been developed to examine population exposure to crime - at a particular time and place. In other words how many people are present in an area that could be exposed to a crime, such as theft from person or violence.

5) However, for burglary as you correctly state the target is not the person it is the property of the household (or vehicle if someone breaks into a house to steal a car). More importantly most burglary occurs when a person is not in their home – so you would expect burglary to be higher when the ambient population is low as this is reflective of lower levels of guardianship (and vice versa). You suggest on line 138 that “Besides, more potential victims attract more burglars” but in fact this is not true as what a burglar is seeking is good opportunities for burglar – one of these is people not being at home.

6) I think the dependent variable in your NBR analysis uses the count of burglaries per grid – it is not clear what size your grid is. However, when ambient population is high, burglary is likely to be low (increased guardianship).

7) By including all burglaries without a time stamp - you cannot considering time of burglary relative to time of ambient population measure. Therefore  you are unable to distinguish if burglaries happen when people are at home or away from their home. There is limited evidence of the effectiveness of CCTV for reducing burglary as it is easy to mask/hide a face from the camera.

8) Your hotspot analysis is rather basic and seems to be based on a count of burglaries within a grid, and an assignment of what is a low/high value. This is a rather unsophisticated and subjective identification of hot spot areas and ignores a wealth of literature and techniques in this area.

You need to reconsider the theoretical framework for this paper, the underlying assumption and the methodology you are using. Your dependent variable (burglary) includes daytime and nighttime burglary. It does not make sense therefore to use ambient population as an independent variable as you mix two very different conditions for burglary in your dependent variable. 

Reviewer 4 Report

This manuscript presents an analysis of which of four data sources used to measure potential burglary victims performed most effectively in a model trying to explain burglary incidence. From my knowledge of crime theory, the authors’ choice of rational choice theory is an appropriate theoretical framework for predicting/explaining burglary. The analysis seems largely well carried out and negative binomial regression is an appropriate method for this type of dependent variable. There are some additional details about your methods/findings that could be beneficial to report. The main weakness of the present paper is that the the discussion lacks some depth. I recommend a moderate revision to improve the contribution of the paper.

Major points

1. I think it would be very helpful to include a reference map that shows the major geographical features of the study area. You talk about some of them in your discussion of the spatial pattern, but as a reader who is not familiar with this particular Chinese city, it would help me to follow that description much better if you had a reference map that accompanied Figure 2, showing, for example, the locations of the villages & their spatial extent, the major roads, etc.
2. p. 7 makes reference to VIFs, but I don’t see anywhere in the manuscript that reports the VIFs comprehensively.
3. Figure 3: I think it’s easy to miss the fact that the axes are scaled quite differently across the three graphs. I can appreciate that you want to make them fit nicely together and show the values clearly, but there needs to way to draw the reader’s attention to the fact that the values are in fact quite different across the three graphs. Adding a y-axis break like the ones shown here might help: https://en.m.wikipedia.org/wiki/File:Y-axis_break.svg
4. It seems an omission not to report the pseudo-R2 or deviance explained value for each model. Although the AIC helps to understand which model fits best, it doesn’t really tell us about the overall explanatory power of the model - how well the model actually explains the burglaries. You also make reference to community based variables that were not included in the models - this may contribute to lower pseudo-R2 / deviance explained values perhaps. It would also be good to see some discussion of your residuals. Are they also spatially clustered?
5. The discussion needs more depth. It seemed a bit cursory to me and I think could better situate the findings within the literature. There are also additional things about your results and data that can be discussed. For example, it seems that household density is more important than population density, given that the IRR is higher for the household counts than the population counts. Why might that be? Are multi-person households more likely to have things worth stealing? I wonder also if one reason that the TRH data outperforms the census data is that the census data are much older than the TRH data, so there could be more error from changes in population patterns over time. I know that Chinese cities can change very quickly and the TRH data are contemporaneous with the burglary data, while the census data are ~6 years older. It would be interesting to repeat this analysis with contemporaneous Chinese census data. I guess this would be possible soon? I believe the census was in 2020 but am not sure how long before the data become available.
6. In the conclusion, I think you have misordered the ranking of the performance: According to the AIC, the census households outperformed the census population model.

Minor points

1. Overall: The English language expression is pretty good though there are occasional odd translations and a few repeated errors. One is researches: the plural of research is research. A find/replace will easily correct that. Microscopic should read microscope, and discussion of ‘levels’ should probably be replaced with ‘scales’. APPs should be written apps.
2. p. 1: the US actually has more frequent demographic estimates available through the American Community Survey which has replaced the long-form decennial census.
3. p. 3: table 2: it might be good to explain what the n represents. This is explained later in the paper (192 grid cells), but at that point where the table comes it has not yet been explained. (n = 192 grid cells)
4. p. 4, line 138: …procedure to automatically settle the case point… replace with procedure to geocode the case point. It would also be good to report how many surveillance cameras were removed when you cleaned the surveillance camera dataset.
5. p. 5, line 164: ‘current people stream’ do you mean active users? as in the number of people currently using the WeChat app?
6. p. 6 line 180: ‘so discrete that variance greater than the mean value significantly’. rephrase: …is a count variable, distributed in a small number of grid cells such that the variance is much greater than the mean number of burglaries in each cell.
7. p. 6 line 207: you talk about decent buildings. Do you mean well-constructed buildings? I’m not sure what decent means in this context.
8. p. 7, line 235: I think this should refer to Figure 3, not Figure 2.
9. p. 8, line 285: ‘intestinal trails’ - this is not really what you mean, I am sure. But I’m also not quite sure what you’re trying to say here so cannot suggest an alternative.