Watching the local news in St Louis is like receiving reports from a war zone. A woman is shot in her car. A mother of five is shot and killed. A family of four is robbed at gunpoint.
St Louis is one of just a couple of cities where there has been a real (as opposed to imagined by scandal-monging media) spike in violent crime. I wouldn’t attribute it to any specific events because there is always a spike in crime in St Louis. The authorities are making heroic efforts to make the city more habitable but the progress is slow and uneven.
Clarissa's Blog 
Current statistics don’t support “imaginary.” St. Louis was mentioned in this article, but wasn’t considered a particular hot spot. Baltimore this year has more murders than NYC, despite being substantially smaller. http://www.usatoday.com/story/news/2015/07/09/us-cities-homicide-surge-2015/29879091/
The pattern is inconsistent. My belief is that cities with better social service programs are having fewer problems, but that’s hypothesis at this point.
LikeLike
Nate Silver says the spike in violence is limited to just a couple of places. And St Louis is among the worst. His analysis tends to be very careful.
http://fivethirtyeight.com/features/scare-headlines-exaggerated-the-u-s-crime-wave/
LikeLike
“Lies, Damned Lies and Statistics” was the title of a book from which I taught many years ago. Their effort to gather data is admirable; the statistical analysis isn’t. From what I can gather from the article, they used something along the lines of a t test to determine the likelihood of getting the number of 2015 homicides as a simple matter of random variation.
What’s wrong with that?
(1) Applying sampling statistics to non-sample data. We’re dealing with actual counts of what happened. There’s no sampling error involved.
(2) We know there is bias in reporting of statistics. The t test assumes accurate reporting without bias. It cannot deal with bias. The test in fact assumes a normal distribution around the mean and that would be highly unusual for population data. The test also assumes a common underlying rate for all cities, which is simply not true. Without that assumption, there’s nothing to calculate.
(3) The t test ignores context. What makes the data especially interesting is that rates were in decline prior to this year in most of these places.
A better approach would be to build a time series model separately for each city for which data is available. There’s no evidence they did that. Such a model would include demographic and ecnomic data and would be much more interesting than what they did. However, it might not show what they want.
Now a simple test as to whether a community has a problem is this:
(1) Assume there is an equal probability that crime will increase or decrease in a community each month. For simplicity, we’ll ignore the no change alternative.
(2) Thus there’s a 50% chance of crime increasing in any give month. The odds of an increase in 2 consecutive months would be 25% (.5 x .5) and in 3 consecutive months 12.5% (.5 x .5 x .5), etc. By the time you get to five months, you’re seeing a result that’s rather improbable — and you had better be thinking about why its happening.
Right now, I suspect there are real problems in a number of cities. At best, my position favors the Scottish verdict of “Not Proven.” I don’t believe the data support use of the term imaginary.
LikeLike
The reason it is being denied in the 5-10 urban core’s where it is happening is highly political… I think you can piece together what agenda might be on the line..
LikeLike