Predicting and Plotting Crime in Seattle(racketracer.wordpress.com)
racketracer.wordpress.com
Predicting and Plotting Crime in Seattle
https://racketracer.wordpress.com/2015/03/02/predicting-and-plotting-crime-in-seattle/
3 comments
The maps lack normalization by population density. But even that might be misleading, for example industrial areas having low population densities but nonetheless a lot of people in the area during working hours. And while population density might be relevant for some crimes it might be irrelevant for others. Where not to park your car is probably better normalized by the number of parking spaces or even the number of cars present. So don't try to read to much out of maps or charts not specifically crafted to answer the question you are interested in.
I agree, but they are nice visualizations nevertheless. Relevant xkcd: http://xkcd.com/1138/
I don't know Seatle, but it looks like the major highways / avenues are highlighted in these crime maps, perhaps they don't have too much population or parking lots but they have a lot of traffic. (I don't know Seatle, perhaps they do have a lot of population and parking lots.)
I don't know Seatle, but it looks like the major highways / avenues are highlighted in these crime maps, perhaps they don't have too much population or parking lots but they have a lot of traffic. (I don't know Seatle, perhaps they do have a lot of population and parking lots.)
The hotspots north of the band of water (canals) are definitely lower-income concentration areas; a couple of them highways with seedy hotels (prostitution/drugs).
The main blob is "city center". Some low income areas in there don't look very visible at this scale, but also "where everyone is out on streets" (work in daytime, nightlife at night).
The main blob is "city center". Some low income areas in there don't look very visible at this scale, but also "where everyone is out on streets" (work in daytime, nightlife at night).
I think that a really cool feature, that I am not sure if it has been implemented or not, is a date package that can do a lot of splicing. If I wanted to know the exact day of the week, sunset time or dusk time, or isolate specific hours without tons of tedious work in Pandas. It would probably help out with normalizing population density such as downtown when not in the working hours but maybe on weekends.
All of the highlighted streets have a lot of businesses with a lot of parking and a lot of traffic through the area.
This is super tricky. One project I worked on looked at pedestrian deaths at a regional scale. We were lucky that the regional government had daytime population estimates for the CBD. Otherwise downtowns tend to look like pedestrian killing fields. Even with those daytime estimates, you're stuck treating those areas as "special," and thus limiting yourself in terms of the kinds of comparisons you can defensibly make.
You don't always need to supply a rate to get a useful visualization either. In the case of pedestrian fatalities, you might still want to invest heavily in safety improvements where absolute numbers of fatalities are high, but rates are low.
On the visualization side, I've found that maps are not always the best strategy for spatial data.
You don't always need to supply a rate to get a useful visualization either. In the case of pedestrian fatalities, you might still want to invest heavily in safety improvements where absolute numbers of fatalities are high, but rates are low.
On the visualization side, I've found that maps are not always the best strategy for spatial data.
I think you should better read works like this one: http://arxiv.org/abs/1409.2983