Spatial Analysis of Gunshot Reports on Twitter in Mexico City

Round 1

Reviewer 1 Report

I recommend updating Key Words to include words that are not already in the title, this will expand potential search hits/matches.

Larger issues to address:
If you are measuring pre- and during COVID crime reports (via Twitter), why not examine March through March since that more accurately reflects when behaviors were changing and municipal protocols changing? Otherwise, your time frames encompass the December holidays of 2018 and 2019, which were not very different from one year to the other. Or, even a smaller time period if some places in your study area began opening up earlier than March 2021, though this will decrease your sample size so if you do this you would need to test for validity of sample size.

Did you consider/test for change of behaviors/reporting via Twitter due to people being at locations (home) that they would not normally be at during certain times of day? And how would you propose to account for the increased use of social media in general (contrary to your decreased number of “gunshot” tweets) during lockdown? For example, analyzing not just the number of shootings posted on Twitter but analyzing the number as a ratio of all Twitter posts in that location? Or some other normalization.

There was a slight increase in total number of tweets, especially in December 2019 through May 2020, but an overall decline in “gunshot” references (141 to 101) and this is only briefly addressed in reference to data for the NNR. This may warrant further examination as mentioned above.

Specifics, line items:
Good distinction and reference to domestic violence on p. 5.

“Gunshots during the Pandemic” is not a section just on gunshots but rather on crime studies during the pandemic. I suggest renaming this section to be more reflective of the content. Alternatively, limit the content of the section to studies strictly on gunshots.

Grammar on p. 6 highlighted.

p. 7 “Georeferenced information” implies geographic coordinates. Remove one of these.

Table 1 still contains the Spanish “a” instead of “to” please update. Also, if row 1 goes until 6/02/2019, how does row 2 start on the same date? Subsequent rows cannot be inclusive of the dates before them so row 2 should begin on 7/02/2019, etc.

p. 9 “referring” typo

p. 10 "zonal effect" of MAUP, not "zoning"

p. 10 administrative boundaries would alter the results as compared to a continuous surface analysis – correct this “typo”

p. 13 QGis should be QGIS and ArcGis is ArcGIS

p. 13 You have Octubre in Spanish instead of the English October in the Table column titles.

p. 17 should refer to “Kernel Estimation analysis undertaken reveal…” and not “maps undertaken…”

p. 17 revise sentence before caption for second map to be more accurate

p. 18 You do not conduct spatial analysis by municipality to avoid MAUP (as stated earlier in the manuscript) however I think there should be some analysis done based on the 16 municipalities. You can do clustering by municipality before and during COVID as well as total number of tweets in the municipalities over the two time periods, or whatever time periods you ultimately choose. This may shed light on the spatial distribution of Twitter activity overall and not just the criminology patterns the may or may not underlie the “gunshot” tweets.

pp. 18 “29 and 18 tweets respectively” while not a typo this is confusing because your next sentence states that 22 tweets (average) as second highest number remains constant. Itzapalapa has a higher occurrence, so this goes against what you previously state – clarify why you rank Cuauhtmeco as higher. Is this due to location of this municipality and the Kernel Estimation results? If so, overlaying these municipal boundaries on the Kernel Estimation outputs would be very useful.

p. 19 I recommend that Graph 2 be changed into a map to visualize the distribution. The current graph is useful to show that all (but one) municipalities had a decrease in number but to show where the higher occurrences were would be better done through 2 maps side by side, or one map showing change (-) or (+) by value.

p. 20 re-read/re-write highlighted sentence??

p. 21 you state “…explained by the location of the tweets…and the underlying criminological processes in these areas…” Are you simply referring to the overall decrease in other crimes that are reported? If so, that is a bit different than “underlying criminological processes” which you do not actually address throughout the manuscript. Revise how you have this written.

p. 21 Again, if you are going to name the municipalities in reference to the Kernel estimation then you should at minimum include/overlay the boundaries on the Kernel estimation surface for reference. Otherwise, discussion points such as this are meaningless.

pp. 21-22 you make good discussion points regarding informal economies and different income. You just need to ensure that the spatial analyses that you conduct are not limited and therefore steered to a certain result.

p. 23 or simply more people being around in Cuauhtemoc during the pandemic to post on Twitter. How do you rule this out? Do the overall number of tweets in Cuauhtemoc also increase in the same time period or not?

pp. 23-24 annual festivities are not a criminological phenomenon

Overall, a worthwhile topic and good start to the spatial analysis. I recommend a deeper examination of how Twitter patterns in general may have changed during the same time periods, as well as consideration to changing the time period of analysis given when SAH orders were put in place.

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 2 Report

The article addresses important issues, but requires major changes to be published. The title is journalistic, not scientific. There is no bacground in the abstract regarding the discussed issues - it only contains a presentation of the research itself.

The introduction contains elements that should be included in the following sections, such as the presentation of the research method or partial results. Incorrect layout of part of the article, not compliant with MDPI standards. 

The article is not formatted according to the MDPI requirements. Poor quality of graphics. 

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 3 Report

The authors explored the reports of gunshots in Mexico City collected on Twitter before and during Covid-19, using spatial analysis techniques. In sum, I liked the article because is well framed and has interesting content. However, the paper needs improvement to be published. My overall recommendation is to accept after major revision. I tried to help with some straightforward comments below.

Major comments

1) It is not clear how the authors selected the 242 tweets after removing the colloquial sentences. Did the authors read all the tweets? Please, explain it better in the Data section.

2) Why the pre-Covid period is from October 2018 to October 2019? Covid's first case in Mexico dates back to February 2020. I suggest the authors adjust the sample period.

3) In my opinion, the main problem of the paper is that tweets are not capturing the real dynamics of gunshots. Is it possible to estimate how many Twitter users are in Mexico City? Or is it possible to map the location of all tweets and correlate it with the population? I hypothesize that this correlation is higher with the wealthier population. Another demonstration of this flaw could be the measure of the 2-years official counts of deaths by shooting in comparison with the 242 unofficial reports on Twitter. The analysis is telling us just a part of the story.

Minor comments   

Page 4 – The last paragraph of the introduction should be moved to the discussion section.

Table 2 – October.

Page 14 – Graph or Figure?

Please use used the MDPI journal template for your submission.

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 4 Report

Decision: Reconsider after major revisions

This paper presents an exploratory analysis on patterns of gunshots in Mexico city, using Twitter data as the source for the analysis, and comparing two periods: before and during the Covid-19 pandemic. The manuscript in generally well written and has a nice flow; the methodology, while not novel, is adequate, and I certainly appreciate seeing a quantitative spatial analysis work in a Latin American country, a region that in my experience is still relatively underserved of studies of that type (despite high crime rates in many of its countries).

However, I do have some issues with this study and the manuscript as of now. First and foremost, I believe the contributions are a bit limited. While the issue of how the pandemic influenced the geography of gunshots in Mexico city has not been explored, the analysis done was a rather simple and more could be done. I have a few suggestions that I believe will add more depth to the analysis; these are detailed in my comments on the Methodology section.

Second, I think the theoretical background should be developed a bit more and some caveats considered. The authors state that Twitter is a more reliable source than official data for gunshots (that one being very restricted), but Twitter data is not devoid of its own issues of reliability. More detail on this is given later in this document (particularly in my comments for the Introduction section). Similarly, I believe more context should be given about the situation in Mexico and Mexico city, particularly in how the pandemic was/is being experienced and tackled, as well as any context on gunshots (even if only before the pandemic). A general description of Mexico city should also be provided, especially since specific municipalities are mention in the Discussion.

Thirdly, the results could be presented a bit better, particularly maps 1 and 2 (I give more specifics suggestion in my comments for the Results section). In general, I think this paper would benefit from having a bit more maps (to either illustrate results or give context).

In the remainder of this document, I give more specific details on what I think should be addressed.

General formatting

This is minor, but I think using Mexico city throughout the text instead of CDMX would be better. It is a better-known term, and Mexico city is not that long of a term.

Also, typically this journal has a template to be followed, so I was a bit surprised that the manuscript does not follow the template. It is not up to me to decide on this, but I believe you may need to reformat the manuscript to fit the template.

Introduction and “gunshots during the pandemic” section

I think these section are well written and with a nice flow. I do have some specific comments and suggested changes, which are mentioned below:

• Defining gunshots

For clarity, you should explicitly define what is considered as “gunshots”, since it could be ambiguous. Is that any event that involve a gunshot? Or is that the mentions of such event? For instance, if there was a bar fight that ends with a X shots being fired, with Y people reporting this event on Twitter, would the assumption that one gunshot event happened, or were there X events (or Y)? Are domestic violence episodes, bar fights, robberies, etc all considered as possible gunshot episodes (provided that there were shots fired, and that people actually reported them on Twitter). Or are you considering more specific events (gang fights, shooting sprees, etc)?

• “By March, with the exception of essential services (trash collecting, medical services, electricity, water, public security, etc.), most companies, industries, private and public institutions were closed, and a large proportion of the population were quarantined or subject to serious restrictions on mobility (Wilder- Smith and Freedman, 2020:1). “

This has actually varied a lot across countries or even regions and cities: some were much harsher than others in their quarantining measures. This has also varied across time, with countries sometimes relaxing or ramping up their distancing protocols etc. You should rewrite this sentence to account for that nuance. In addition, the citation Wilder-Smith and Freedman, 2020 is not listed in your references towards the end.

 

• “In fact, there is a growing body of literature on crime / Covid-19, that studies the effect of lockdown on crime and violence”

I would rewrite this sentence as “In fact, there is a growing body of literature that studies the effects of Covid-19 related lockdowns on crime and violence”

• “Out of distrust or fear, the population generally do not seek to involve themselves directly by calling the police, but rather, comment on shootings on Twitter with other social network users, and thereby maintain the confidentiality of their personal details. In Mexico City, Twitter has thus become a more precise source of information (or, at least, an alternative source) regarding this issue than incident calls to the police or 911.”

Do you have sources to cite? While it does sound plausible, there can other reasons for people not reporting (e.g. bureaucracy, time availability), so this type of statement should be backed by empirical information, or at least presented as a conjecture. Same for the issue of Twitter being a more precise source of information. What are the grounds for such statement? Have there been any victimization surveys that can be used to compare the quality of Twitter data with official data? Twitter data can also be faulty (since you need to process the text, there can be ambiguity, duplicates etc), and there can be bias in who is more likely to tweet. If twitter data for gunshots is abundant compared to official data, leading to more data points, this may be an indicator of quality, but even then the issue of bias (of who is tweeting) will still be present. This all boils down to the problem of the  “dark figure of crime”, and so you may want to cite that as well.

• “One of the most interesting findings of this analysis is that during lockdown, shootings appear to have moved from one municipality to a neighboring one. The model suggests that the municipality of Cuauhtemoc had the highest concentration of shootings and experienced a 50% increase in Twitter reports. In contrast, the neighboring municipality of Miguel Hidalgo registered a reduction in the average frequency with a pronounced fall in reports in the second period of 72.4%. This may be due to changes in crime patterns in the city.”

Are these municipalities in the Mexico city metropolitan area? If so, you should mention that (if not, you should provide more qualifiers to where these cities are located, such as state and country). The reader is probably aware the study is in Mexico city (due to the title and abstract), but may be confused by the sudden mention of these other municipalities (that often constitute their own town or city).

Study area

This study would benefit from a study area section or subsection. Authors may not be familiar with CMDX and its municipalities (that you do mention later on!), so you should briefly describe it and show a map or two illustrating the location of CMDX, its municipalities and other useful information (otherwise, the maps 1 and 2 seem pretty abstract!)

Also, while the Covid-19 pandemic is affecting the whole globe, it had different timings in different countries and even different cities. When did the pandemic really started taking effect in Mexico, and specifically, in Mexico City? How was the quarantining/social-distancing done in CDMX? Were there different policies/levels of quarantining at different moments in the time period considered?

Also, if there is any literature on shootings in CDMX pre-Covid, that would be beneficial to mention too.

Data section

The description of the data is overall good, but you should display a map with the raw dataset (i.e., the outline of CDMX with the point cloud for the ~3k tweets considered)

“The data were obtained from reports of shootings posted on Twitter between 6 October 2018 and 6 October 2020,”

Why this specific time frame? Would it have made sense to look at a longer time frame pre-pandemic, for instance? (you would have more data-points)

“Of the total universe of 9,819,255,653 tweets analyzed, 3,588 contained the word “gunshots” and had enough information to be georeferenced”

It seems that the assumption here is that a tweet mentioning gunshots refers to an individual event (“a shooting”). However, it could be that people are just talking about gunshots: being afraid of gunshots, or seeing news about gunshots, or discussing violence as a problem, as is very common on Twitter. Also, people that are more afraid of violence on a subjective basis may be more likely to tweet about their fears – however, fear of crime is not necessarily a good proxy to actual crime (see Alkimin et al., 2013 for instance). You need to address these theoretical assumption and limitations

Alkimim, A., Clarke, K. C., & Oliveira, F. S. (2013). Fear, crime, and space: The case of Viçosa, Brazil. Applied Geography, 42, 124-132.

Methodology section

The methodology is adequate but rather limited, as well as the findings. I recommend adding the following bits of analyses to make the contributions of this paper more substantial:

• Time analysis pre-covid and during covid. What times of the day are gunshots more commonly reported? How about months, days of the week? Does it vary when comparing the two periods considered?
• Does the spatial concentration of gunshots follow the Law of Crime Concentrations? In my experience, the Law of Crime concentration has typically been tested by using crime counts aggregated per street segment (or some other unit), but I see no problem in using crime densities (from KDE). You can calculate what is the (smallest) percentage of pixels that can account for 50% (or 25% etc) of the gunshots, and compare that to what is expected by the Law of Crime Concentrations (it is an interesting result, whether it confirms the law or not).
• Analyzing qualitative change in gunshots: what are the other words that most appears in the tweets analyzed (after filtering out preposition, articles, pronouns etc)? Is there any change between the two periods considered? For instance, there might be more gunshots associated with the terms bar and street (adequately translated into Spanish, of course) in pre-covid times, and more gunshots associated with the terms house and apartment during covid times.

A few more things that should be addressed:

For KDE, what was the bandwidth used and why? If you used the standard one for ArcGIS, mention which method is actually used by ArcGIS (I believe it is Silverman’s criterion, but it could have changed).

I believe the equations for KDE and for Ripley-K don’t need to be shown here, since they are fairly known within the GIS & Spatial Analysis community. Therefore, it suffices to just mention that you used this and that and cite it accordingly (which you already do).

Results section

The presentation of the results should be improved. I suggest putting Maps 1 and 2, side-by-side, using the same intensity scale (to save space, you can include the scale bar just once, or have it on the top or bottom). This will facilitate comparing the two distributions. There is no need to have both the isocurves and the color in the map, so I would just keep one (probably the color). Having both renders the map harder to read in the end, since one obscures the other. I suggest that you include the borders would CDMX and its municipalities overlaying the raster, since you do mention the spatial variation of gunshots in relation to the municipalities. You might want to consider different color schemes to improve visualization of the raster with the overlaying borders. Finally, while not mandatory, including a scale bar and a north arrow will also improve the maps.

Discussion section

Overall, the discussion section looks good to me. I expect that, with the additions suggested in the methodology, you will also edit the discussion to mention these other results. You may also want to bring up and discuss the uncertainties related to Twitter data that I mentioned in my comments for the Introduction. Additionally, since you mention specific regions and peaks throughout the discussion, I suggest adding at least a map in which you clearly mark which regions/peaks you are referring (e.g. you can circle them on a map).

Also, some specific points to be addressed:

• “The increase in the dispersion of violence associated with reports of gunshots, thus moved from the west towards the center of the city, and can be explained by the location of tweets on the administrative borders of the city’s municipalities and the underlying criminological processes in these areas (Graph 2).”

This is part is a bit vague. It is not clear what you mean by “moved from west towards the center of the city and can be explain by the location of the tweets on the administrative borders..”. It is trivial that the density map is explained by the tweets; do you mean that it is explained by tweets being more specifically located at the borders of the municipalities? Is there any reason why they would be concentrated there? Or are you referring to some edge effect related to these borders? Also, what specific underlying criminological processes are being referred at the end of this part?

• “The neighboring municipalities of Miguel Hidalgo and Cuauhtemoc registered first order changes in intensity, detected by the Kernel estimation. Reports in the municipality of Miguel Hidalgo decreased by 72.41% during the Covid-19 lockdown, while the Cuauhtemoc registered an increase of 50%.

One possible explanation for this phenomenon, based on the spatial proximity of the municipalities and the routine activities approach, is the differing financial capacity of potential victims of crimes committed with firearms to comply with lockdown measures. The municipality of Cuauhtemoc, which includes the historical center of the city, has many street vendors and informal businesses, who, as a result of the severe decline in their income, struggled to respect the health lockdown (that is, they continued with their activities despite the restrictions). “

No changes recommended here but notice how this part requires a Study area description earlier on to familiarize the reader with these locations.

• “This study also identifies peaks in violence by gunshots in the city, which do not appear to be related to either the criminological surroundings or to activity linked to the illicit drug market. In the municipality of Iztapalapa, 42.5% of tweets reported parties by neighbors and religious celebrations as the origin of gunshots during the pandemic.”

It is nuclear to me the source for this inference. What are the peaks? It might be worth adding a map that marks this and any other peaks that are specifically mentioned (like potential peaks in Cuauhtemoc and Miguel Hidalgo). How do you know that 42.5% of tweets about gunshots are related to parties and religious celebrations? Was that done by directly inspecting the data? You should mention more explicitly that this is the source, otherwise you should cite any external source.

• “Second, it identifies that reports of gunshots may be related to three different criminological phenomena: a decrease in commercial activity and criminal opportunities”

Notice that there was no actual systematic study done about this, and it seems to be mostly conjectures based of qualitative comparisons that are developed in the Discussion (instead of being described in the Methodology and Results). You should make this conjectural nature more explicit by adding something like “Further studies are required to more systematically assess these relations”.

 

 

Author Response

Please see the attachment

Author Response File: Author Response.docx

Round 2

Reviewer 3 Report

Based on the authors' improvement on the draft, I accepted it in the present form.

Author Response

We thank reviewer 3 for his comments. They were all very helpful in improving the paper.

Reviewer 4 Report

Dear authors,

I believe the manuscript is now fit for publication, and I'm glad to see that you addressed (in one way or another) all the points raised in my review. The following are just some small comments and suggestions that you may find useful. Since one of the comments involves maybe adding a graph, I've marked my decision as "accept after minor revision".

• Time analysis: interesting to see that no significant change in temporal patterns happened. I'm wondering if it wouldn't be interesting for the reader to still include a graph showing these patterns, for instance, one graph showing the variation per day of the week for the pre-Covid (on the left), and during-Covid (on the right), and then do the same of month, and time of the day. Not mandatory, but worth considering (maybe add as support material).
• Law of crime concentration: I think you bring a fair case of why not include those results in the paper (methodological "controversies", scope). Nevertheless, it was interesting to see the results for KDE, and how they don't quite fit was is expected from the Law of Crime Concentrations. Which to me is to an issue, but an indicator that the law might need to be tweaked and better defined. You might want to consider approaching these results as future work (no action is needed for the current paper, that's just a comment I thought was worth mentioning)

 

Author Response

Three graphs were added in the appendix of the paper (1.- Pre-covid -blue- and Covid-19 -red- shot reports in Mexico City. 2.- Day of the week and 3.- Time of day).

We thank reviewer 4 for his comments. All of them were very useful to improve the article.

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.