Graphing to Tell a Story: St. Louis homicides
I recently discussed the availability and presentation of St. Louis’ regional crime data. For the city of St. Louis, the Post Dispatch’s site STLtoday.com has the best graphic presentation of homicide data. An interactive bar chart that presents 4 years’ data. While it is colorful and does provide individual monthly data when the mouse hovers over bar - it is not at all clear what you can learn by looking at this graph. It is not easy to compare changes month by month, to look at trends across a year, or to see underlying patterns in monthly rates.
Rather than simply complain I thought I would try my hand at making the graph understandable. 
First I tried a simple line graph. But the overlapping lines made it difficult to see individual patterns. There was no reference to what one might expect on average. So it did not improve the utility of the graph at all.
Next I tried each year paired with the average of all years. This allowed a variety of comparisons. 1) By comparing each year with the average it is easy to see when months have more or less than would be expected. So we can see that 2008 has been pretty much consistently higher each month. 2) It also allows us to see the pattern across the months - you are less likely to get murdered in January and February than any of the other months. 3) Additionally, it also allows us to predict that there will be a spike in rates in November of 2008 when that data becomes available.
Now following the suggestions of graphics visionary Edward Tufte I decided to see if we could further simplify the presentation - conveying the same information with less ink. I believe that the final graph does this by removing the extraneous lines and the vertical axis information. This has the advantage of focusing the viewers attention on the main points - examining the pattern of homicides across the year and differences between the years. It looses the interactive features of the original but it does deal with the fact that the original graph really did not provide any more information to the viewer than a simple table of numbers.
Are there other ways to graphically ring information out of this data? Let me know what you come up with.
NOTE: This entry is cross posted on both my personal and professional blogs.