Crime rates at the neighborhood level: American, Canadian neighborhoods
Introduction
The data I'm going to present in this post
In this post, I'm going to show quite a bit of data.
(1). Comparing violent crime at the neighborhood level in several Canadian cities (for example: how does the worst neighborhood in Calgary compare to the worst neighborhood in Saskatoon?)
(2). Comparing violent crime at the area / small city sized level in several Canadian cities (for example: how does the North End, Winnipeg compare to Thompson, Manitoba?)
(3). Comparing violent crime at the neighborhood level in Canadians cities and American cities (for example: how does the worst neighborhood in Buffalo compare to the worse neighborhood in Saskatoon?)
(4). Comparing the central business districts / downtowns in each of the Canadian and American cities I have data for (for example: how does the downtown in Edmonton compare to the downtown in Minneapolis?)
(5). City-wide crime rates juxtaposed with their downtown violence score and their most violent neighborhood (do city-wide crime rates predict anything about the downtown's violence or the worst neighborhood?)
(6). What the crime rates in Canadian cities would look like if you boundary them like American cities (removing the suburbs from the city limits).
(7). A list of high crime small towns in Canada (under 10,000 people) with their violent crime statistics which can be compared at the neighborhood level.
What made me want to do this analysis
A month ago, I did an analysis of crimes rates at the neighborhood level. I really just wanted to see what city-wide crime rates actually correlate with. Long story short, I found that they mostly correlate with the percentage of the population that lives in safe or moderate areas. You can read that analysis here if interested.
I took a look at five cities, and I took a look at the crime in each neighborhood. One of these cities was Canadian, and that city was Winnipeg, and although at the city-wide level, Winnipeg had the fourth out of fifth highest "violence score", when we looked at its downtown, it was the worst, statistically speaking, by double than the next downtown. And when we looked at the worst neighborhoods, the worst neighborhoods were in Winnipeg. And its top two neighborhoods were quite a bit worse than the worst neighborhoods in the other four cities I looked at, which were Pittsburgh, Minneapolis, Portland and Buffalo.
So, that got me interested to see what other Canadian cities had in regards to crime at the neighborhood level. So, I looked for as many cities as I could, and I found data for seven cities in Canada. But for two of them (Vancouver and Toronto), the neighborhood sizes are just too big to be useful. That being said, I do make a fairer metric to compare the other five cities to these two cities: amalgamating several smaller neighborhoods into larger areas to make the populations comparable. That way, we can get sort of an idea of how those Toronto and Vancouver areas would look like if they were broken up into smaller neighborhoods.
The five cities I found good data for were funnily enough, the five biggest cities in the three prairie provinces: Calgary, Edmonton, Regina, Saskatoon and Winnipeg. I knew I'd only find data for like 25% of the cities I searched for, I just didn't expect them to all be in the same region. My American data is a bit more geographically separated!
The violence score: what is it?
Before we get into looking at these statistics, let me explain what the violence score is ("Vio." in the tables). It is a metric that weights and sums the violent crimes in terms of their frequency. Frequency is used as a proxy for severity here; it's like the free market for crime severity.
It's necessary to make a metric like this for a few reason.
Often times we see data that just has a total violent crime per capita metric. It might say "3,000 violent crimes per 100,000 residents" or something like that. This isn't a good metric because violent crime includes everything from homicides to uttering threats or harassment. Obviously, homicide is more severe than uttering threats or a simple assault like slapping somebody in the face. In these generalized metrics, a homicide is worth the same as an uttered threat. No good.
But we also need the violence score because often times people just tend to jump to homicide and just look at homicide to determine how dangerous or violent an area is. But because homicide is so infrequent, it's not really the best marker. Now, of course homicides matter, but just looking at homicide on its own is going to mislead you. If we only look at homicides: a couple of extra homicides in an area can make it look way worse than it is. Likewise, a couple less homicides in a bad area than normal will make it look better than it actually is. Let me give some examples.
Lakewood U.C. has a higher homicide rate than Mayfair, Riversdale and Downtown in Regina. I don't live in Regina but the rates of every other violent crime indicate that Lakewood U.C. is a pretty nice area while the latter three are rougher spots.
Dovercourt in Edmonton happens to have a pretty high homicide rate but its other violent crime rates are very low and indicate it is a safe area. As a consequence it has a higher homicide rate than neighborhoods like McCauley and Spruce Avenue, which again, have much higher violent crime rates in general.
Southland Park and Tuxedo have a higher homicide rate than Central Park in Winnipeg, but the other violent crime statistics indicate that Central Park is a very rough area, while Southland Park and Tuxedo are safe areas.
For homicide data to be accurate at the neighborhood level, you need a lot of years of data. For the other more common violent crimes, three years of data is great.
Apart from that, you as an individual are so unlikely to be a victim of a homicide, especially at random. What you're more likely to be a victim of is a robbery or an assault, and so these crimes also need to be taken into account when we look at how violent or dangerous an area is.
Calculating the violence score
You can skip this section if you do not care about how the violence score, assault score and sexual violence score are calculated. It's probably very boring for most people.
The main crimes
Each crime is relative to the most frequent severe violent crime, which is aggravated assault. The rest of the crimes get a point value that is determined by how much more frequent aggravated assaults occur than it. Aggravated assaults are also called "major assault" in Canada, which includes assault 2, assault 3 and attempted murder. I will be using these two terms interchangeably throughout the post.
Although this particular analysis is anchored to Canadian neighborhoods, I first conceptualized this when looking at neighborhood statistics in mostly American cities. The formula was designed so that both Canadian and American neighborhoods, and presumably in any other country, could be compared. I say this to explain why I'm using national data from the USA to get the bulk of figures for calculating the violence score.
The American national crime frequencies for severe violent crimes in 2020 are as follows: homicide (6.5 per 100,000 people), robbery (73.9), aggravated assault (279.7). So for example, a robbery is worth 3.8 points because aggravated assaults are 3.8 times more frequent than robberies. A homicide is worth 43 points for the same reason. And aggravated assaults are worth 1 point.
To get the worth for sexual assaults, we'll need to look at Canada's statistics because the FBI does not provide sexual assault frequencies at the national level, but StatCan does. The American aggravated assault rate is 3.2 times higher than Canadian sexual assault rate, so sexual assaults are worth 3.2 points.
Dealing with different crime categories in different cities: the assault score and the sexual violence score.
It is preferable when a city delineates between degrees of assaults and types of sexual offences. However, many did not have data with these specific categories. For example, for Saskatoon, I had the following categories: homicide, robbery, sexual offences, total assaults. But for Winnipeg, its data was a lot more precise, so I was able to make the following categories: homicide, robbery, aggravated assault, common assault (simple assault / assault 1), sexual assault. Among all the cities I found data for, there were a lot of different combinations for categories. In total, there were eight categories I grabbed data for.
The main problems I ran into were: cities that grouped major assaults and minor assaults into one "total assaults" category; cities that only had rapes and not all sexual assaults; and cities that grouped all sexual offences together which includes offences that are not sexual assaults.
I originally presented the comparative data with every category, so many cities had a bunch of squiggly lines on the columns that they did not have data for. I didn't like that the data was so cluttered. Presenting them all together like this was overwhelming and unintuitive and confusing for comparison. For example, how can you really compare a neighborhood and its sexual violence if one stat is "rape" and the other is "sexual offence"? Or "aggravated assault" vs. "total assault"?
So, I made two indexes: the sexual violence score and the assault score. They take the incomplete or generalized data and turn them into a score. The sexual violence score predicts the sexual assault rate from rape or sexual offences, and the assault score predicts the major assault rate from total assault. So cities that had data for major assaults and sexual assaults did not need these indexes to be calculated.
The formula for these indexes are relative to each city. Different cities have different ratios for certain offence groupings. For example, 19.1% of the assaults in Pittsburgh were aggravated, while this figure is 33.7% for Portland. So to calculate the most accurate score, I had to find what percentage of these offence groupings were what per city.
Calculating the sexual violence score
So when we look at city-wide data from 2021 for Cincinnati, we see that 75.3% of the sexual assaults were rapes. So to predict the sexual assault rate, we multiply the rape rate by 1.3 (100 divided by 75.3), and that gives us the sexual violence score.
Following the same process, we find these figures for the following cities:
* St. Louis: 72.1%. So multiply by 1.4
* Minneapolis: 44.5%. So multiply by 2.2
* Pittsburgh: 39.5%. So multiply by 2.5
For Washington, D.C. I couldn't find the statistic. So I had to estimate it another way. To do this, I looked at the percentages in cities in the same metro area: Montgomery and Fairfax. Montgomery was at 37.4% and Fairfax was at 41.8%. Pretty close in the ratios, so I'll go in the middle and say 39.6% for D.C., which is multiply by 2.5.
Now we need to get the sexual violence score from sexual offences. Luckily, Canada has some pretty excellent and easy to find data on this for every single municipality. So, I was able to get into every single city here and find the rates in each city for the same 3 years of the data (so, 2016 to 2018 for Edmonton; 2019 to 2021 for Calgary, etc.). For example, in Regina 74.1% of the sex offences were sexual assaults. So, to get the sexual violence score, I multiplied the sexual offence rate by 0.741. This figure is 0.875 for Vancouver and 0.759 for Saskatoon.
Calculating the assault score
To get the assault score from total assaults, I did the same process as I did to get the sexual violence score from sex offences. In Calgary, 33.6% of assaults were major assaults, so the total assault rate is multiplied by 0.336 to get the assault score. This figure is 0.421 for Regina, 0.409 for Saskatoon, 0.361 for Edmonton, 0.399 for Vancouver and 0.254 for Toronto.
Buffalo was the only American city to not have an aggravated assault rate provided. So to get Buffalo's aggravated assault percentage, I had to find a different source: the FBI. Buffalo has the highest rate of all cities for aggravated assault percentage at 60.7% (in 2021). So the multiplier for Buffalo's assault score is .607
One last thing is that Calgary's neighborhood-level data for assaults was even less complete than other cities because it excluded domestic assaults. Other cities' data did not exclude domestic assaults. I made sure of this by checking the city-wide rates on StatCan and comparing them with the city-wide rates summed up from the neighborhood data. Calgary's assault rate summed up from the neighborhood data is 297.9. However, it's StatCan rate from 2019 to 2021 is 630.1. So to adjust for this, its assault score will be multiplied by 2.12.
Division points
This last point is very self-explanatory. Most cities had data for all four main categories: homicide, robbery, sexual violence score, assault score. But a few cities were missing one of these categories. So, each category gets a "division point".
To calculate the violence score: the rate per 100,000 of each category is multiplied by its worth (homicide by 43, for example), and then the four categories are summed together. Finally, the sum will be divided by the number of categories involved in the sum. The data for Pittsburgh has all four categories, so its sum gets divided by four. But Calgary only has data for three categories, so its sum gets divided by three.
Example calculation: The neighborhood Beacon Heights in Edmonton has a homicide rate of 46.3, robbery of 393.2, sexual violence score of 150.4, and an assault score of 530.4. So it is calculated like this: [(46.3 x 43) + (393.2 x 3.8) + (150.4 x 3.2) + (530.4 x 1)] / 4, which gets us a violence score of 1123.8.
Datasets
Data #1: Neighborhoods in Canada
So now that that's out of the way, let's see the data in the Canadian five cities. The table below shows every neighborhood that has a violence score over 700. Any row that is red, that's the worst neighborhood in that city. For example, we see in the chart here, we have a red row at rating #1, "Point Douglas" in Winnipeg, and at rating #9, we have another red one, which is "Pleasant Hill" in Saskatoon. These are red rows and are the worst neighborhoods in the respective cities.
By the way, Pleasant Hill, definitely the most ironic name for a neighborhood that I came across looking at this. The most dangerous neighborhood in Saskatoon... "Pleasant". This is why proper nouns are capitalized I guess!
Every neighborhood that has a yellow row, that's a downtown or a commercial district. Most of the commercial districts are downtowns, but also in cities like Saskatoon, Calgary and Edmonton, we find there's a lot of neighborhoods that are like town centres for suburban areas, and so they're mostly stores; that's basically the suburban version of a downtown... so they're business districts, just not "central" ones. So, for business districts, I halved the rates / indexes because they attract more people and there's more commercial buildings when compared to a typical neighborhood. This makes them comparable in a more accurate way.
You can see the full list of neighborhoods in each city in the following links. By the way, I only bothered to do the commercial district halving on neighborhoods with high crimes rates, so you'll see certain commercial districts with their full rates: Calgary, Edmonton, Regina, Saskatoon, Winnipeg
Data #2: Areas and small cities in Canada
Now, the next two cities that I found stats for were Toronto and Vancouver. But again, the problem with these cities was that the populations of their neighborhoods were way too big for a useful crime resolution. The average population of neighborhoods in Toronto and Vancouver are: 22,116 and 26,439. The figures for Winnipeg (3,943), Regina (7,401), Saskatoon (4,133), Calgary (6,572) and Edmonton (3,629) are much more indicative because they have a higher resolution.
One delineation I'm making is that I'm calling a "neighborhood" the sort of things that typically have under 10,000 people. Not all neighborhoods in the five "good data" cities have under 10,000 people. Daniel McIntyre in Winnipeg has over 10,000; Rundle has more than 10,000 in Calgary. But for the most part, these neighborhoods have less than 10,000 people. Whereas the "areas" are these things that typically have 10,000 to 50,000 people, so they're a collective of neighborhoods. And so only five cities have data by neighborhood but all seven have data by area.
To make insightful or fair comparisons, I had to compare apples to apples. I couldn't compare neighborhoods of under 10,000 people to neighbourhoods of 10,000 to 50,000 people. There are ten neighborhoods in a 25,000 resident area that have 2,500 people! So, I went through the other five Canadian cities and combined areas to be over 10,000 people, usually somewhere between 10,000 and 30,000 residents, because that's what were typically seen in Vancouver and Toronto. So this can give us an idea of the area. For example, Moss Park in Toronto has a violence score of 970.0, so perhaps its neighborhood breakdown is similar to that of Downtown in Calgary (983.6) or South Circle in Saskatoon (949.2).
I also included some of the smaller (between 10,000 and 50,000 residents) notorious cities as well for comparison. When we look up, "most dangerous city in Canada", apart from seeing larger cities like Winnipeg and Thunder Bay, we tend see smaller cities like Thompson, North Battleford and Prince Albert come up. So, I wanted to see how they compared to similarly populated areas within larger cities.
On the table below, you can view the violent crime data for the top 50 areas.
You can click any of these links to view other relate tables to area statistics.
But to reiterate, these large populations aren't that indicative. At a certain point, a population becomes too big to be insightful about crime and violence because it loses specificity and resolution. Many different smaller parts of the larger area get blended and averaged into one. This is problematic because in reality, some of the smaller parts are worse than others and some are better than others and we don't really see how big the difference is or how good the good parts are and how bad the bad parts are. We can think about it like this: a neighborhood with a population of 2,000 people is a crystal clear photo, while an area of 20,000 is a somewhat pixelated photo, a city of 500,000 is a pixelated piece of garbage photo taken with a 2000's flip phone zoomed in 200%, and a country of several million is a photo literally taken with a potato.
For example, if we look at the Notre Dame area in Winnipeg, the neighborhood Centennial (violence score: 3,457.6) is much worse than the neighborhood Weston (831.1). Or in the Regina area North Central, the neighborhood North Central (2534.0) is much worse than the neighborhood Eastview (648.2). So how would Moss Park breakdown in Toronto if it was 8 neighborhoods with about 3,000 residents a piece? I don't know, but surely it would have large disparities like we see in Notre Dame and North Central.
Even when we look at the North End in Winnipeg, the Lord Selkirk Park (5,702.9) part is much worse than the Dufferin (2,943.8) part, even though they're both really bad, there is still a significant difference between them.
Anyway, I just put this "areas" table for curiosity's sake (the big popular cities of Canada) and context's sake (comparing the notorious smaller cities). Vancouver and Toronto seem to be the most desirable places in Canada that people like to talk about, and that people want to move to. To be frank, if the cities that only had area sized "neighborhoods" were Hamilton and Kitchener, I probably would not have bothered with the areas. But I'm glad I made the areas because it allowed for fairer comparisons with the smaller notorious cities.
Data #3: Neighborhoods in the USA
I already had data for some American cities, so I wanted to compare the Canadian neighborhoods to the American ones. When I read around Canadian subreddits, people often times talk about the danger in Canada not being as bad as it is in the States. For example, I looked up things like "crime" and "danger" in the Winnipeg subreddit, and a common notion was, "it's bad for Canada, but it's probably average for the United States". So the question here is: do the worst neighborhoods in Canadian cities compare to the worst ones in American cities? Do the central business districts compare?
So here we go: I have some statistics for a good variance of cities: like St. Louis, Washington D.C., Buffalo, Cincinnati, Denver, Minneapolis, Pittsburgh and Portland. So quite a bit of cities here.
The table below shows all of the neighborhoods in the five Canadian cities and these eight American cities that have a violence score over 1,800.
You can click any of these links to view other relate tables to area statistics.
* The entire table of neighborhoods in each of the American cities: Buffalo, Cincinnati, Denver, Minneapolis, Pittsburgh, Portland, St. Louis, Washington D.C.
So this chart is quite interesting. St. Louis has the worst neighborhoods, but Winnipeg is second. In the top 20 neighborhoods, 16 are in St. Louis and 4 in Winnipeg,
Winnipeg's worst neighborhood has a higher violence score than Minneapolis and Washington D.C.'s worst neighborhood. Its worst neighborhood is closer to St. Louis's worst neighborhood than to Minneapolis's
Saskatoon's worst neighborhood is roughly on par with (within 250 violence score) Cincinnati and Pittsburgh's worst neighborhood.
Regina's worst neighborhood is roughly on par with Buffalo's worst neighborhood.
Edmonton's worst neighborhood is roughly on par with Denver's worst neighborhood.
So, we see a lot of diversity amongst these cities. But secondly, the worst neighborhoods in Canadian cities are not "better" than the worst neighborhoods in American cities. It's more dependent upon the individual cities. We don't really see any national correlations with violence scores at the neighborhood level. Canadian neighborhoods fit and mix right into the same spectrum that American neighborhoods are on. We have to take this on a city-by-city basis.
#3.1: The low resolution obscurity bias
It's always important to point out certain flaws or biases in the data and its limitations because they are going to influence our instincts with regards to interpretations and analysis.
One important limitation of this dataset goes back to population size / resolution. It's what I call the "low resolution obscurity" bias. It's similar to the concept I talked about while making the "areas vs. neighborhoods" distinction, but its application is in the context of high crime neighborhoods with high populations. For example, when we look at a city like Regina, its worst neighbourhood has almost 10,000 people. Whereas St. Louis's worst three neighbourhoods have less than 2,500 people each. In fact, we have to scroll all the way down to the 36th worst neighborhood in St. Louis, Dutchtown, to find a neighborhood that has more people than Regina's worst neighbourhood.
If we take a look at St. Louis's neighborhoods on map, we find that there are neighborhoods right by each other that have drastically different violence scores. Do we find a safe neighborhood right by a horrible one? No, not really. But we do find things like: super horrible neighborhoods right by less horrible neighborhoods, or bad neighborhoods by moderate neighborhoods. For example, the neighborhood Academy (violence score: 3,556.8; pop.: 2,355) sits right between Fountain Park (7,105.7; 1,075) and Hamilton Heights (6,660.4; 2,187). Sure, Academy is still a horrible area, but the difference between Academy and these other two neighborhoods that surround it is pretty drastic. If Fountain Park and Academy together happened to be considered as one neighborhood, then it'd have a violence score of 4,669.1, which severely obscures Fountain Park's badness.
Another example is that Walnut Park East (5,390.6) and Mark Twain (2,833.7) are bordering. Again, both bad neighborhoods, but if they were conjoined into one neighborhood, the badness or Walnut Park East would be obscured, because collectively they have a violence score of 4005.9.
The importance of this observation is that this is probably the same reality in all of the other highly populated high crime neighborhoods: the badness in some parts is obscured by the fact that they are conjoined with neighboring "less bad" parts.
To demonstrate this in another way, I made a "worst contiguous 10,000" for each city. Basically, I combined several neighboring neighborhoods to equal around 10,000 people (or in some cases, a sole neighborhood with a higher population). The worst contiguous 10,000 is the neighborhood collective with the highest violence score in the city. Here we can see that, although Regina's worst neighborhood was ranked 8th out of the 13 cities, its worst contiguous 10,000 is 4th.
The not so high population resolutions of neighborhoods like North Central obscure how bad they are in some parts. So, if we compare North Central, Regina (violence score: 2,534.0) to McKinley, Minneapolis (2,366.1), they have similar violence scores. But North Central has three times the population. So, someone might look at these two places juxtaposed and think that they're even. But, are they actually even? Probably not. There are probably areas of North Central that are worse than McKinley, and also parts of North Central that are better than McKinley.
This same bias affects Calgary's three worst neighborhoods. Downtown Commercial Core (pop: 8,683), Rundle (11,699) and Marlborough (9,162). They only have violence scores in the mid-900s to the low 1000s, but if they were divided into 5 neighborhoods a piece, surely some would be way over 1,000. So although Denver and Portland's worst neighborhoods have much higher violence scores than Calgary's worst neighborhoods, when we look at the worst contigious 10,000s, Calgary is nearly on par with Portland and Denver.
This bias also effects Denver. One of its worst neighborhoods, Five Points, has a violence score of 1,211.4 and a population of 19,104. What if this neighborhood was divided into ten different neighborhoods? We've already seen how varying population areas of around 15,000 to 25,000 can be in the Canadian Areas section.
And we can ask this same question for a ton of other neighborhoods in the cities. What we would expect if Pleasant Hill, Saskatoon (violence score: 2,820.8), which has almost 5,000 people, was divided into two neighborhoods? It's likely that one would have a violence score over 3,000, while the other would dip below the current score. So, when we look at North Central, Regina, what if it was divided into seven neighborhoods? How would it look? And we can even make the same comments on some Winnipeg neighborhoods like William Whyte. What if it was divided into three neighborhoods? What if Daniel McIntyre was divided into eight neighborhoods? How would they look?
What if we divided Broadway-Fillmore in Buffalo into five neighborhoods? Or Genesee-Moselle into two or three neighborhoods? If Buffalo had these two neighborhoods cut up into multiple neighborhoods, instead of having five neighborhoods with a violence score over 1,500, we might instead see nine neighborhoods with a violence score over 1,500. We might even see one that has a score over 3,000, depending on the distribution of crime in Broadway-Fillmore and Genesee-Moselle.
If all of these splits happened, it would give us the same degree of population resolution we see in St. Louis's worst neighborhoods for the most part: mostly between 1,000 and 3,000 people.
It's without question that St. Louis does have the worst neighborhoods in terms of violence scores, and especially in terms of homicide rates. No matter how you slice the pie, the homicide rates in St. Louis neighborhoods are going to be the highest. But, because of the specificity and resolution, it looks like St. Louis has a higher number of bad neighborhoods when compared to other cities than it really does. For example, only 34% of neighborhoods in St. Louis with a violence score over 1,000 have a population over 4,000 people. This figure is 57% for Edmonton, 63% for Denver, 70% for Buffalo and 88% for Minneapolis. What if Ventura Village was chopped into three? What if Hawthrone was chopped into two? What if Folwell was chopped into two or three?
#3.2: New hypothesis - weapon frequency and the aggravated assault to homicide ratio
Here are three more tables that show the highest neighborhoods in three violence stats.
St. Louis neighborhoods have homicide rates that we don't see in other cities. Winnipeg neighborhoods have robbery rates that we don't see in other cities. And Winnipeg and St. Louis neighborhoods have major assault rates we don't see in other cities.
This comparison between Winnipeg and St. Louis neighborhoods generated a hypothesis for me: is the reason why St. Louis neighborhoods have such high homicides rates, and generally speaking why the USA has higher homicide rates than Canada, not because it is more violent, rather because of the tool (weapon) used to fulfil the violence?
For example, of the neighborhoods in Winnipeg with an aggravated assault rate between 3000.0 and 3300.0, the average homicide rate is 81.8. In St. Louis, the average homicide rate in neighborhoods with such aggravated assault rates is 212.6, 2.5 times higher. Likewise, we can see similar trends if we look at the neighborhoods with aggravated assault rates between 600 and 900 in some of these cities: Buffalo (22.2 homicide rate), Denver (15.4), Portland (15.1), Edmonton (8.8) and Saskatoon (7.3).
For every 17.7 major assaults in St. Louis, there is a homicide. This figure is 102.6 for Saskatoon, 51.3 for Winnipeg, 30.3 for Denver and 13.6 for Cincinnati. I'm going to investigate this for a later post, but my hypothesis is that the higher percentage of aggravated assaults that use a gun result in more aggravated assaults that end up in homicides. Some cities had data that gave the weapon used for each crime incidence, so I will look at the major assault to homicide ratios at the neighborhood level and see if weapon used correlates with the major assault to homicide ratios, and generally speaking, the homicide rates.
My hypothesis therefore predicts that guns are used as a weapon 2.5 times more frequently in those St. Louis neighborhoods than those Winnipeg neighborhoods, and that guns are used about 2 times more frequently in those Portland neighborhoods than those Saskatoon neighborhoods. Guns are more lethal than knives, blunt objects and fists; that is not ground-breaking information.
I don't actually anticipate that it would be 2.5 times more. I say this because between 2019 and 2021, 87.7% of major assaults in Winnipeg were stabbings, while between 2018 and 2020, 68.4% of major assaults in St. Louis were shootings. A more sophisticated model would predict something like "lethality", taking into account how lethal the weapons used are.
I will investigate and do some analysis on this soon, but I'm not going to digress on this any longer.
Data #4: Downtown, central business districts
The next thing I looked at was the downtowns. To measure crime and violence in the downtown areas, I had to come up with a different metric. One might naturally use the residential population of the downtown as the "population" part of the "crime per capita" equation, but this is an imprecise method because downtowns have a lot more "temporary" residents (those who work, shop, do nightlife, go to museums there, etc.) than "permanent" residents (those who actually live there). So, I figure the best population metric to use was the urban area population of that particular city. Note that I am not using the "city's" population. A city's population can be very misleadingness because "city" is a political boundary, while "urban area" (also called agglomerations) is a geographic boundary. And we're studying crime here from geographic perspective, not a political one. I talk more about the misleadingness of city populations in a later section here called "suburban flight".
The urban area is a pretty good metric to measure the population of people who do and could visit that city's center on a habitual or even daily basis. It's basically all of the cities, towns and suburbs that are contiguous with the city. For example, politically, somebody who lives in Fairfax, Ohio does not live in Cincinnati. But geographically speaking, they do, and thus it is apart of the Cincinnati urban area.
The last thing to mention here is that, I only took the crime from the central business parts of downtown. For example, Minneapolis has a couple neighborhoods that are considered to be in downtown, Loring Park and Elliot Park, but they are residential areas so the crime from those neighborhoods is not included.
This was really surprising for me to see at first, but it does make sense. One of the things I noticed when doing my first analysis was that Canadian cities tend to include a lot more suburban areas into their city limits while American ones don't, which as a consequence reduces the Candian cities' overall city-wide crimes rates. So if our base notion for thinking one downtown is more violent than another's is the city-wide crime rates, well this would help to explain why that reasoning is destined to fail. I'll demonstrate the consequences of this in an upcoming section where I take off the suburbs of Canadian cities and give them more American-like boundaries.
But generally speaking, the downtown's don't even correlate with how bad the worst neighborhoods in the city are. Who would've predicted that Calgary's downtown area is slightly more violent than St. Louis's downtown area?
Data #5: City-wide crime rates, a geographically useless metric
This section will take a look at the city-wide crime rates and analyze them in a varying of ways. This is an extension on my previous post exploring the utility of city-wide crime rates. This section further demonstrates why city-wide crime rates have practically no utility from a geographic standpoint.
#5.1: Juxtaposed with downtown and violence score of worst neighborhood
The table below is simple. It juxtaposes the city-wide violence score with the downtown violence score and the violence score of the most dangerous neighborhood in that city.
As you can see, the city-wide violence score does not correlate simultaneously with the downtown's violence score and the worst neighborhood's violence score. One basic way analyze this is to look at each city and rank it by the three different violence scores. For example, Buffalo as a city has the 5th highest violence score, but has the 9th highest worst neighborhood and the 12th highest downtown. Denver has the 7th highest city-wide violence score, but the 11th highest worst neighborhood and the 8th highest downtown. Minneapolis has the 2nd highest city-wide violence score, the 3rd highest worst neighborhood and 7th highest downtown. Saskatoon has the 10th highest city-wide violence score, the 7th highest worst neighborhood and the 2nd highest downtown.
#5.2: City-wide correlation models
We can look at the correlation between the city-wide rates and the other two rates more complexly as well. This metric looks at how well the city-wide violence score predicts the violence scores of the city's downtown and worst neighborhood.
Of the 13 cities here, the average city-wide violence score is 600.3. The average downtown violence score is 1,387.1. The average worst neighborhood violence score is 3,082.1.
So, if city-wide crime rates correlated with the violence in the downtown area and violence in the worst neighborhoods, the city-wide violence score should predict that the downtown has a violence score that is 2.31 times higher than the city-wide score, and that its worst neighborhood is 5.13 times higher. For example, if a city had a violence score of 300.3, if it correlated, it would have a downtown violence score of 693.69 and a worst neighborhood with a violence score of 1,540.5. Let's put this into practice.
Minneapolis has a violence score at the city-wide level of 871.2. This predicts a downtown violence score of 2,012.5. However, its actual downtown violence score is 1,225.6. So, we take the predicted score and minus the actual score to see how close or far off the prediction is. The actual rate is 786.9 points different from the predicted rate. The model overshot: the city-wide rate predicted something more dangerous than what actually exists
Saskatoon has a violence score at the city-wide level of 335.9. This predicts its worst neighborhood has a violence score of 1,723.0. However, its actual worst neighborhood has a violence score of 2,820.8. The actual rate is -1,097.8 points different from the predicted rate. The model undershot: the city-wide rate predicted something safer than what actually exists.
Below is the table for this data. The higher the city is on the table, the more the city-wide rates undershot. The lower the city is on the table, the more the city-wide rates overshot.
Looking at the entire table of stats, we can really see that city-wide rates do not predict these two metrics at all. St. Louis and Washington D.C. have huge overshoots, while Winnipeg and Saskatoon have huge undershoots. The city-wide rates severely overestimate how bad the downtown and worst neighborhoods are in St. Louis and Washington D.C., and they severely underestimate how bad the downtown and worst neighborhoods are in Winnipeg and Saskatoon. The only city where the model did a half-decent prediction was Denver. The rest of the cities are full of huge overshoots and undershoots
No model can be made to predict this stuff based off of city-wide crime rates because city-wide crime rates are a geographically useless metric.
#5.3: Crime inequality score
In my previous post, I created a metric called the "crime inequality score". You can read more about it there. The main idea is that it contrasts the difference between the worst areas of the city (most violence 5%) and the best areas of the city (safest 50%), similar to wealth equality metrics. The lower the crime equality score is, the smaller the difference between the best and worst areas. Or in other words, the more "equal" the crime is throughout the city. A higher score means it's more unequal.
Canadian cities tend to have worse crime equality scores than American cities. The biggest reason for this in my opinion is because Canadian cities include a lot more suburbs into their city-wide boundaries than American cities do. In the next subsection, we will explore this very phenomenon.
#5.4: Suburban flight! Taking off the 'burbs
As previously mentioned, Canadian cities tend to include more suburbs into their boundaries than American cities do. To see this claim substantiated, look at a map. Just kidding... In my first analysis I compared the urban area population to the city population of many cities and checked what percentage of the urban area population the city population was. You can find this on my previous post.
But seriously, look on a map! Minneapolis, St. Louis and Buffalo on a map are literally 100% gridded urban areas. Calgary is mostly suburban loopy roads. You can find these loopy roads near Minneapolis, but not inside its boundaries (for example, there are suburbs in the neighboring city Edina).
So in this section, I removed all of the suburbs from the Canadian cities (and the American ones that had them) to see what their city-wide statistics would look like if they were American cities. My classification for if something was suburban or not was simply the road layout. If it was gridded, it was urban. If it had the "street hierarchy" (lots of cul-de-sacs, lots of those half-loop roads that only have two entrances) road lay-out, it was suburban. Sometimes this was a bit harder to do in cities like Pittsburgh due to the hilly nature of the city, so in some cases I actually looked on Google Street View to make a choice.
The fact that Canadian cities tend to include suburbs into their boundaries and American cities don't has huge consequences on the city-wide crime rates. And because people typically have been looking at city-wide crime rates to gauge safety and the like, Canadian city-wide crime rates have given people a statistical basis to think that Canadian cities are categorically safer than American ones.
I'm not saying that Canadian cities are going to be topping the charts now, but they will be at the levels of some of the other cities in the table. For example, before we did the de-suburbification, Portland, Denver and Pittsburgh were all at least a level higher than Saskatoon, Regina and Edmonton. But post-de-suburb-ification, these three Canadian cities are now at least a level higher than those three American cities.
You can view this table in higher resolution here. You can also view each neighborhood I classified as "suburban" here.
Note: for crime with an unknown address, I distributed it like this: in Regina there were 217 assaults with unknown locations. 72% of the known assaults happened in the urban area, so 72% of the unknown assaults went into the urban totals and 28% went into the suburban totals. I did this process for every single stat in all cities that had crime with an unmentioned location.
You can also see how small Canadian cities actually are. For example, Canton, Ohio has a city population of 70,872, but its urban area population is 279,2451. That makes sense when you see that Saskatoon's urban city population is only 83,170, but its urban area population is over 280,000.
Providence, Rhode Island, has a city population of 190,934 but an urban area population of 1,190,956, so it's similar to Calgary in this sense. But even when you look on a map, many urban gridded areas are not apart of Providence's boundaries, so Providence's true "urban population" is going to be larger than 190,934.
We can even see that, although Winnipeg does have the worst neighborhoods in Canada, it is the least suburban of the five Canadian cities here; this is one of the reason why Winnipeg's city-wide crime rates are much higher than the other Canadian cities. This is an example of the suburban fluff effect being less extreme in Winnipeg than say, Regina (see my previous post for more talk about "suburban fluff") .
#5.5: Crime ribbons
I did another series of crime ribbons. I updated the way I made these crime ribbons from the last post. This post, I had a different color for every increment of 100, so there was more specificity. Every neighborhood was rounded to the nearest 100. So for example, violence scores of 50 and under got the color for "0"; 50.1 to 149.9 got the color for "100", and so on.
There are still 100 bars per city: 1 bar per 1 percent (of the population that lives in neighborhoods with the denoted violence scores). The main utility of the crime ribbon is that it's a nice visualization to show us what is responsible for a city's crime rate. You can read more about this on my original blog post where I introduced this graphic.
The crime ribbons in the picture below are sorted by city for its violence score. The most left, St. Louis, has the highest violence score, and the most right, Calgary, has the lowest violence score.
So what we can see in the five most left cities is that there's really a lot of yellow space. And that's because there is a lot of moderate areas. There's also not much blue, there's not much super quiet safe areas. The consequence of this is that these five cities don't have much low crime areas to lower their city-wide rates down.
When we look at Denver and Portland, which were some of the cities whose worst neighborhoods had some of the lowest violence scores, we can see that there's barely any bright blue in them; there's a lot of that "light green" or turquoise color. Whereas if we compare this to Winnipeg, which has a higher violence score than these two cities, there's a ton of blue space in Winnipeg. But another huge different is that Winnipeg's low end (low on the crime ribbon; the city's worst areas) is very red and orange. So, although these three cities have very similar violence scores at the city level, Winnipeg is getting a lot of more score from the very bad areas, while Denver and Portland have a lot less score being watered down from a lot of blue areas.
We can see in a couple other of these Canadian cities like Saskatoon and Regina, there low ends are slightly more intense, more red-orangey, than the low ends of Washington D.C., Cincinnati, Buffalo, Denver, Pittsburgh and Portland. The reason why these two cities at the city-wide level have such a low score is because there's so much blue space in their top ends. Recall the suburban fluff effect: this helps to explain why there is so much more blue space in the Canadian cities.
The crime ribbons do a nice job visualizing why city-wide crime rates don't really tell us much about the worst areas of the city. We mostly see a correlation with the blue and green space with the violence score. The low ends don't correlate at all. If we just focus our vision on the low ends of the cities, there's not really a correlation when we shift our eyes left to right. But there is a decent correlation with the blue space when do the same. Winnipeg and Pittsburgh kind of mess this correlation up because their low ends are much worse than Denver and Portland's low ends.
#5.6: Safe zone skew
I wrote more about this in my original post, but since I have more data, I'm looking at the safe zone skew phenomenon again.
The correlation is still very strong between the percentage of the population that live in safe and moderate areas and the city-wide rates.
The point of this is to show that the city-wide rates don't correlate to the severity of violence in the worst parts of a city. They just correlate with how many areas are not that violent. It doesn't say anything about how violent the violent parts are.
With data like this, I'm trying to show in a visual way that city-wide crime rates aren't of much utility. Certainly, we might be able to find some utility in them. If you have a city with St. Louis's crime rates, it's highly probable that there's going to be some pretty bad neighborhoods there. But it is possible that a city with a violence score of 1,600 has every neighborhood between a 1,000 and 2,500 violence score; that could be how a city with violence score of 1,600 is built. So even when we look at a city with such a high violence score, it might not even be the case that the you find mega horrendous neighborhoods there. In the case with St. Louis, it is true, but it's not inherently going to be true.
Data #6: Small towns
The last stats I compiled was small towns or population centres with neighborhood sized populations with high crime rates in Canada. When we look at them, we find a lot of really interesting phenomena. I haven't actually looked at how crime is in small towns versus in cities. And here I'm just looking at small towns with high crime rates, I did not look at every single small town that there is in Canada. I theoretically could do that. And maybe one day I'll investigate this phenomenon a bit more, but it really seems that small towns do not experience robberies that much. I've never really considered this or thought about it, but maybe robberies are only frequent in cities.
When we look at the data here, the sexual assault is bonkers! And so are the major assaults. I mean we have sexual assault rates in many of these towns over 1,000. Keep in mind, the highest one we had in a city neighborhood, in all of the cities we looked at including the American ones, was Point Douglas in Winnipeg with a rate of 786.7. There are 24 towns in Canada with a sexual violence score higher than that. The highest one is 4555.6!
And then the same thing, we see with the major assault rates. The highest we saw with all of the neighborhoods, was again, the Point Douglas neighborhood with a rate of 5479.6. There are 13 towns with an aggravated assault higher than that! The highest one is 8995.7!
Now, I only looked at high crime ones here, so don't get any ideas here that all small towns or all reservations have crime rates like these. I mean even some of them I looked at here don't have high crime rates, like Wikwemikong and Peguis. I had a method of finding the most violent towns, and some of them that weren't very violent were caught by this method. Basically, I looked at the violent crime severity index given by StatCan, and some of these towns had one abnormal year with a high score but the rest of the years were low.
Comments
Post a Comment