5 ms·

Predicting and Plotting Crime in Seattle

35 points by racketracer 12y ago

seanhandley 12y ago
"Plotting crime" - great word choice :D
- racketracer 12y ago
  Pun completely intended
danbruc 12y ago
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.
- gus_massa 12y ago
  I agree, but they are nice visualizations nevertheless. Relevant xkcd: http://xkcd.com/1138/ 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.)
  - jschwartzi 12y ago
    All of the highlighted streets have a lot of businesses with a lot of parking and a lot of traffic through the area.
  - squidfood 12y ago
    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).
    - racketracer 12y ago
      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.
- peatmoss 12y ago
  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.
marcodena 12y ago
I think you should better read works like this one: http://arxiv.org/abs/1409.2983 http://arxiv.org/abs/1409.2983

35 points by racketracer 12y ago

seanhandley 12y ago

"Plotting crime" - great word choice :D

racketracer 12y ago
Pun completely intended

danbruc 12y ago

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.

gus_massa 12y ago
I agree, but they are nice visualizations nevertheless. Relevant xkcd: http://xkcd.com/1138/ 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.)
- jschwartzi 12y ago
  All of the highlighted streets have a lot of businesses with a lot of parking and a lot of traffic through the area.
- squidfood 12y ago
  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).
  - racketracer 12y ago
    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.
peatmoss 12y ago
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.

marcodena 12y ago

I think you should better read works like this one: http://arxiv.org/abs/1409.2983 http://arxiv.org/abs/1409.2983