Good work. This is truly much better than any of the similar services I've seen in the past. I especially like that it's an easy lookup for what looks to be the complete array of crimes in all areas; and, the heatmap is helpful. What is it normalized based on?
Hi, thanks so much! The heatmaps are normalized on a per county basis, and red is a 97-98 percentile block in terms of crime density.
We're not doing per capita normalization because we haven't had that data on a per block basis, but definitely looking in to ways to start to do more granular normalization.
I'm confused, then. It seems like all of King County, in Washington (where Seattle, Redmond, Bellevue are located) is all green. If the heat-map is normalized by county, shouldn't even a place like King County have some red spots?
Could you normalize at the census tract level? Better yet, assign average occupants, shoppers, cars parked, etc. to the locations at which the crime occurred.
A bigger issue is variance in less dense areas. It looks like what you're doing today is simple counts whereas for areas with lower crime rates, more history might provide a more stable crime rate. It might also be useful to weight against a demographic prediction of crime or at least average the crime rate over the period that the demographic prediction is stable.
> A bigger issue is variance in less dense areas. It looks like what you're doing today is simple counts whereas for areas with lower crime rates, more history might provide a more stable crime rate. It might also be useful to weight against a demographic prediction of crime or at least average the crime rate over the period that the demographic prediction is stable.
The purpose of this tool seems to be to decide where to live. Since the demographics of cities change over the years, what is the advantage of looking at more-past demographics?
Second, demographic instability is a very real attribute of neighborhoods, and it may be correlated with higher crime. I imagine you're trying to reduce that visibility. Why? Again, the goal is to give a current view of the crime in the area, not what the crime will look like when things calm down.
In all, I think it's good how it is, as it represents exactly what I would be looking for when I examine crime/local facilities/housing prices.
If the real rate of crime is varying, that's something of interest to the user. The problem is that in a neighborhood with 1000 people, how am I to distinguish between a rate of 1 in 10,000 and 10 in 10,000? In a short time span, that's impossible to do.
That said, you're right that one needs to make sure that real changes aren't smoothed. That's why considering an average over a period of demographic stability likely makes sense.
This is useless for my area 22202 Arlington VA. You basically mark all areas with highrises as red, and areas with single track housing as green. Sorry, if there is a row of 22 story building on one block and 3 blocks away is single family dwellings of course there is going to be more crimes where there is 30x the number of people.
Fair enough. Then make the influence of each crime inversely proportional to the number of people there. All of that said, don't go to the smooth spots of the function that aren't current and declare "this is the crime rate of this place". That was my point
We're partnering with SpotCrime.com and CrimeReports.com for their crime data, who both work with police agencies to syndicate their data. In most cases our crimes are updated on a daily or weekly basis.
Unfortunately, New York doesn't publish their crime data in a granular way. I think much of SpotCrime's data is scraped from various sources including newspapers, and we're showing what we get from them. This is a bummer for us, as it's probably not complete, but hopefully New York will publish their data soon!
Ummm, according to the San Francisco heat map, Hunter's Point looks like a very safe area while China town seems way more dangerous. I don't know how accurate this is.
In terms of % chance of a crime happening near you, that's totally accurate. We're working on another companion heatmap that shows the percentage of violent crimes (then, Hunter's Point pops out quite a bit)
Here's a preview of the violent crime map - it's colored based on the percentage of violent crimes in a given area, for areas with > 50th percentile crime for the county.
23 comments
[ 2.7 ms ] story [ 67.1 ms ] threadWe're not doing per capita normalization because we haven't had that data on a per block basis, but definitely looking in to ways to start to do more granular normalization.
Perhaps I don't understand ~how~ you normalized.
That fix has been made, but as I understand it our current heatmap tiles will still be in the cache until next week.
A bigger issue is variance in less dense areas. It looks like what you're doing today is simple counts whereas for areas with lower crime rates, more history might provide a more stable crime rate. It might also be useful to weight against a demographic prediction of crime or at least average the crime rate over the period that the demographic prediction is stable.
The purpose of this tool seems to be to decide where to live. Since the demographics of cities change over the years, what is the advantage of looking at more-past demographics?
Second, demographic instability is a very real attribute of neighborhoods, and it may be correlated with higher crime. I imagine you're trying to reduce that visibility. Why? Again, the goal is to give a current view of the crime in the area, not what the crime will look like when things calm down.
In all, I think it's good how it is, as it represents exactly what I would be looking for when I examine crime/local facilities/housing prices.
That said, you're right that one needs to make sure that real changes aren't smoothed. That's why considering an average over a period of demographic stability likely makes sense.
http://www.trulia.com/local/#crimes/washington-dc
PS: http://en.wikipedia.org/wiki/File:ArlingtonTODimage3.jpg High density, mixed use development is often concentrated within 1/4 to 1/2 mile from the County's Metrorail rapid transit stations, such as in Rosslyn, Courthouse, and Clarendon (shown in red from upper left to lower right). http://en.wikipedia.org/wiki/Arlington_County,_Virginia
Unfortunately, New York doesn't publish their crime data in a granular way. I think much of SpotCrime's data is scraped from various sources including newspapers, and we're showing what we get from them. This is a bummer for us, as it's probably not complete, but hopefully New York will publish their data soon!
As you can see, Hunter's Point definitely sticks out in this map. http://dl.dropbox.com/u/338114/Screenshots/02l.png