Many municipalities have begun providing frequently-updated data on city activities as part of an effort to increase transparency in government. My hope in this project is to leverage these data to provide a better idea of how people are responding (and reducing social contact) in real time to the COVID-19 pandemic. Data used in this project encompass four broad categories: crime, automobile crashes, first responder calls and air pollution. When possible, I have aggregated these data into broad categories (i.e. property crime, violent crime, crime victims by race, etc) so comparisons may be made. Data for each city is rescraped approximately daily. I currently have data for the following cities, but am continuing to update whenever I have free time:
Crime: Austin, Baltimore, Baton Rouge, Boston, Chatanooga, Chicago, Cincinnati, Detroit, Dallas, Denver, DC, Gainesville, Kansas City, Little Rock, Los Angeles, Louisville, Memphis, Mesa, Montgomery County (MD), Naperville, Nashville, Omaha, Phoenix, Raleigh, Riverside, Rochester, Rockford, Sacramento,Saint Paul, San Diego, San Francisco, San Jose, Santa Ana, Seattle, Sonoma County, St. Petersburg, Tacoma, Tucson, Virginia Beach, Wichita
Auto Accidents: Baltimore, Baton Rouge, Boston, Chicago, Cincinnati, Denver, Fort Worth, Los Angeles, Milwaukee, Montgomery County, NYC, Nashville, Riverside, Sacramento, San Diego,San Jose, Santa Ana, Scottsdale, Seattle, St. Petersburg, Tucson, Virginia Beach, Wichita
First Responder/Fire Dept: Baton Rouge, Cincinnati, Mesa, Milwaukee, San Diego, San Francisco, Scottsdale, Seattle, Virginia Beach, Wichita
In each plot, I show when two events occurred: orders to shut down the restaurants and bars (red line) and shelter-in-place (black line).
For the crime data, I aggregate data into various categories. The breakdowns are as follows:
For the crash data, I aggregate data into two categories:
For the Fire 911 data, I aggregate data into three categories:
Data Last Updated:
Caveats:
1) I have done almost no quality control on these data. This is meant as a quick and dirty project to look across as many cities as possible and see how behavior is changing. At a future point, I would like to go back and clean these data up, but that is not what is presented here. A particular concern is that the most recent day (or days) of data may not be fully updated, thus one should be especially careful interpreting these points.
2) Particularly with the crime data, cities include very different events and so comparing all crimes are not all comparable across cities. For instance, some cities include only UCR Part 1 crimes, some include all crimes and some include police interactions many of which are not crimes at all. Comparing across crime category types is more valid as these are always taken from a standard set of crimes (Violent: Robbery, Murder, Rape and Assault/Battery), (Property: Theft/Larceny, Burglary excluding Motor Vehicle Theft), (Motor Vehicle Theft: MV Theft), but even in these categories there is variation in reporting across cities. I have indicated roughly what each city includes in its crime data on each cities' page.
3) For the air pollution data---values are not adjusted for weather conditions. Large changes can occur because of weather patterns even independent of any change in emissions.
4) Make sure to check where cities' data end. Cities update at different frequencies and so some cities are more up-to-date than others.
5) I have no skills at web design, so this website is very ugly
6) If you see something in one of the graphs that looks obviously wrong, let me know and I can try to fix it.
Data and Code
Here is a link to my (again ugly, uncommented) Stata code, which pulls data from various cities' open data portals---all plots can be recreated directly from the code and it should be moderately easy to tweak to create other cuts of the data (of course because each city documents their crime differently, this requires digging at least partially into each cities' data) . Feel free to borrow for any projects that might be useful. Alternatively, here is a Stata dataset of all cities' crime collapsed to the daily level broken out by the categories used in the charts.
Findings (as of 4/7)
1) There is a large reduction in reported crime as social distancing measures are put in place. The largest decreases appear to be in large metros on the East Cost, Midwest and West Coast. Smaller metros and metros in the South and Interior West exhibit much smaller decreases.
2) This reduction is driven almost entirely by property crimes and other lower-level offenses like drug crimes (which fall to nearly 0 in some cities). Property crime declines much more in neighborhoods with low average commute times (which I use as a proxy for a job destination neighborhood).
3) Changes in violent crime are much smaller and typically occur only after a shelter-in-place order---many (or even most) cities see no decrease at all. Violent crime falls much more in neighborhoods with higher levels of median income---there are no decreases or even increases in low-income neighborhoods.
4) There do not not appear to be large increases in domestic violence (for cities in which data are available), which is reassuring. However, an important caveat to this is that willingness to report may have fallen considerably because access to shelters has presumably decreased. In some cities, there does seem to be an increase in violent crimes committed in residences (and a commiserate decline in violent crimes committed outside the home) in some cities , but it is unclear what types of crimes are driving this change.
5) Crime declines are largely driven by crimes committed during the workday (8AM-7PM) and crimes committed outside the home. Crimes committed inside residences see little change.
6) As expected, there is a very large reduction in auto accidents after social distancing measures are put in place (accidents are about 33% of their pre-virus levels). However, there is a much smaller reduction in serious accidents. Additionally, accidents fall in many cities which see small (or) no reduction in crime.
7) There appears to be no increase in EMS calls with the exception of Baton Rouge (and maybe Milwaukee), which beginning 3/22 saw an enormous increase in serious EMS calls. If anything, in most cities EMS calls decline after social distancing measures come into place, perhaps suggesting that exposure to everyday activities leads to higher degrees of medical distress.
8) There are no evident declines in PM 2.5 pollution. There is some evidence for declines in Ozone, and these appear to be the largest in the regions that anecdotally engaged in social distancing earlier (New England, Mid-Atlantic, Upper Midwest and Pacific Coast). However, in both cases there is a high degree of day-to-day variation and I do not control for weather, making it difficult to disentangle anything but large changes.