5 ms·

Just for reference: https://www.neighborhoodscout.com/blog/highest-murder-rate-cities-2019 https://www.neighborhoodscout.com/blog/highest-murder-rate-c...

rsj_hn 6y ago
That is a poor reference. Basically a lesson in how to not analyze data. What it is doing is taking murders per capita of small areas (10,000 people) and comparing them to big cities with over a million people, thus it is dominated by random noise in the small areas. Do that over the whole nation and you create a noise collector rather than a "reference". This is why the list is led by small populations that have experienced an anomalous number of murders in the last year, rather than larger cities that tend to have more stable murders per capita. A second weakness is the inconsistency in city boundaries. Although many of these neighborhoods are indeed violent, it's not really fair to compare E St. Louis with LA or Chicago. An apples to apples comparison would be to compare E St. Louis with the worst neighborhoods of LA or Chicago, because E St. Louis is a small (and particularly blighted) portion of the overall St Louis MSA. To fix this, what the author should do is normalize the data before performing the division to get a per capita rate. There are two options: * fix the geography and collect data over several years, say a decade for the town with 10,000 people to at least get 100,000 as a minimum sample size. * look at the MSA rather than what are effectively neighborhoods in a given MSA. The second approach is much better because it gives you recency of data which also solves the problem of changing populations, and gives you more of an apple to apple comparison. But that means that rural areas will simply drop out. That's OK, if you want a general rural vs urban comparsion, you can do that by bucketing lots of rural locales into larger ones and then comparing.

rsj_hn 6y ago

That is a poor reference. Basically a lesson in how to not analyze data. What it is doing is taking murders per capita of small areas (10,000 people) and comparing them to big cities with over a million people, thus it is dominated by random noise in the small areas. Do that over the whole nation and you create a noise collector rather than a "reference". This is why the list is led by small populations that have experienced an anomalous number of murders in the last year, rather than larger cities that tend to have more stable murders per capita. A second weakness is the inconsistency in city boundaries. Although many of these neighborhoods are indeed violent, it's not really fair to compare E St. Louis with LA or Chicago. An apples to apples comparison would be to compare E St. Louis with the worst neighborhoods of LA or Chicago, because E St. Louis is a small (and particularly blighted) portion of the overall St Louis MSA. To fix this, what the author should do is normalize the data before performing the division to get a per capita rate. There are two options: * fix the geography and collect data over several years, say a decade for the town with 10,000 people to at least get 100,000 as a minimum sample size. * look at the MSA rather than what are effectively neighborhoods in a given MSA. The second approach is much better because it gives you recency of data which also solves the problem of changing populations, and gives you more of an apple to apple comparison. But that means that rural areas will simply drop out. That's OK, if you want a general rural vs urban comparsion, you can do that by bucketing lots of rural locales into larger ones and then comparing.