The presentation of the data is extremely confusing, especially with the lack of titles and axis names. Does each bin between each month represent a year from 2001 to 2013? On some charts, like narcotics and criminal sexual assaults, you switch to weekly bins with no indication at all.
Well, I'm not sure what the point of this was. My team at the Chicago Tribune has been working on visualizing and explaining Chicago crime data for quite some time. Here's one of our biggest projects, which pulls in (I think) the same data Socrata used, as well as our own tracking of shootings, homicides and suburban crime data: http://crime.chicagotribune.com
Hopefully one of my coworkers will join in here who's more knowledgeable than me, but remember crime data, such as what Socrata is displaying, is just a snapshot of crimes that have been reported. It could mean those crimes were just more enforced at that time, or more people decided to report crimes, etc. It is not a crime victimization survey.
I checked the data myself and I believe it's a coding error on his end. I'm not seeing a spike in December or January for Criminal Sexual Assaults. (it's worth noting that using a month function on a null date usually results in a 1 being returned (January))
It's always really important to understand what you're analyzing when dealing with crime data. In this case this is when crimes are reported or when arrests are made. That's not the same thing as when the crime occurs. So the spikes for burglary in the morning and at lunch hour are more likely because that's when people notice the crime has occurred and call the cops, not when someone is actually breaking in. And the prostitution trends are probably when the cops are choosing to be out and about arresting, which might not at all be correlated with when the most illegal activity is actually happening.
It's interesting that the author thinks the spikes on the first of the month are bad data, but that the huge spike in sexual assault on Jan 1 of each year is valid data. I wouldn't doubt that there are slightly more sexual assaults on new year's eve, but on the order of 10x more than any other day of the year? That doesn't seem right. I'd guess there are more nuances about how sexual assaults are reported or logged that are at play here.
Socrata and Statwing are both very useful services, but without some domain knowledge and a good understanding of how the data was collected, even the best tools won't get you valid insights.
I did (some of) this with chicagocrime.org in 2005... :-)
...and I've since learned the error of my ways. It's just too misleading to do these types of overly simplistic data reports. The data set is flawed in many ways, including:
* Data model. The Chicago crime data only has a single date/time field. For many crimes, such as break-ins, the victim/reporter isn't able to pinpoint an exact time; they might just provide a time range. That doesn't jive well with the data model.
* Data mistakes. I dealt with public record databases extensively from 2005-2012 (Washington Post, chicagocrime.org, EveryBlock). Government data sets (like any data sets!) contain mistakes, which are compounded when you do aggregate queries.
All in all, obviously this post is harmless link bait at face value, but more thought should be given to these issues. Open govt. data is a good thing, but it's healthy to be skeptical.
9 comments
[ 4.2 ms ] story [ 25.5 ms ] threadI don't know how to say it without coming off overly negative, but those graphs read like textbook examples of what not to do.
Thankfully, there's no shortage of good, free resources for learning these things [1].
1: http://datasciencemasters.org/
How we built it: http://blog.apps.chicagotribune.com/2013/02/28/the-chicago-c...
More about the suburban crime data: http://blog.apps.chicagotribune.com/2014/01/31/displaying-cr...
Hopefully one of my coworkers will join in here who's more knowledgeable than me, but remember crime data, such as what Socrata is displaying, is just a snapshot of crimes that have been reported. It could mean those crimes were just more enforced at that time, or more people decided to report crimes, etc. It is not a crime victimization survey.
Here's what I'm seeing with the data: https://www.icloud.com/iw/#numbers/BAJdMdQ8oUn9yYC-REmBpZlzD...
It's interesting that the author thinks the spikes on the first of the month are bad data, but that the huge spike in sexual assault on Jan 1 of each year is valid data. I wouldn't doubt that there are slightly more sexual assaults on new year's eve, but on the order of 10x more than any other day of the year? That doesn't seem right. I'd guess there are more nuances about how sexual assaults are reported or logged that are at play here.
...and I've since learned the error of my ways. It's just too misleading to do these types of overly simplistic data reports. The data set is flawed in many ways, including:
* Data model. The Chicago crime data only has a single date/time field. For many crimes, such as break-ins, the victim/reporter isn't able to pinpoint an exact time; they might just provide a time range. That doesn't jive well with the data model.
* Data mistakes. I dealt with public record databases extensively from 2005-2012 (Washington Post, chicagocrime.org, EveryBlock). Government data sets (like any data sets!) contain mistakes, which are compounded when you do aggregate queries.
* Systematic police department effort to reduce crime numbers through data trickery. See the amazing recent piece by Chicago Magazine: http://www.chicagomag.com/Chicago-Magazine/June-2014/Chicago...
All in all, obviously this post is harmless link bait at face value, but more thought should be given to these issues. Open govt. data is a good thing, but it's healthy to be skeptical.
(Note chicagocrime.org is no longer around, as I redirected it to my other project, EveryBlock, several years ago: http://www.holovaty.com/writing/chicagocrime.org-tribute/)