Criminal chart
Mar 08, 2007
The Times found a sharp surge in violent crimes.
Uh-oh.
The legend for the columns is missing.
The maximum murder rate of about 45 per 100,000 in the top chart is depicted by a column 9x as tall as that showing the minimum rate of about 60 per 100,000 of aggravated assaults in the bottom chart.
Sorting by murder rate does disservice to the bottom chart, rendering it essentially unreadable.
Reference: "Violent Crime in Cities Shows Sharp Surge", New York Times, March 9 2007.
The bars would presumably be 2004, 2005, and 2006 from left to right, making this effectively a small multiple. Definitely a case where a "panel" or "trellis" line graph format would have been less cluttered with ink.
Also, the shading from light to dark of the three years privileges 2006 in the fight for eyeball time; more ink has been lavished on 2006 compared to the equivalent number in pale 2004.
Posted by: derek | Mar 09, 2007 at 03:53 AM
Perhaps a scatter graph, with assaults on one axis and murders on the other might be worth trying. The direction of time might be represented by symbols for 2004 and 2006, and a dog-legged line connecting them via 2005. The lines would need enough room for clear labels identifying each city, if they all have to share the same graph space.
Posted by: derek | Mar 09, 2007 at 03:59 AM
About the sorting issue: this could be (very loosely...) inspired on Jacques Bertin's reorderable matrix. This is a very useful technique to find clusters of records (in this case cities) with similar, or dissimilar, patterns. I tried this in http://bizviz.jorgecamoes.com/uma-matriz-ordenavel/en/ with EU countries.
Since we have a small number of cases, I would test something like a cross between a line chart for the time series and a scatterplot.
Posted by: Jorge Camoes | Mar 09, 2007 at 06:19 AM
can we not afford guns in Detroit?
Posted by: Jeremy Kandah | Mar 09, 2007 at 03:47 PM
A scatter plot with time series is not always the best option, but in this case it could work, because we have a small number of cases and a general trend. I had to try it, so take a look (I think this is what Derek had in mind, also):
http://bizviz.jorgecamoes.com/exemplo-de-grafico-de-dispersao-com-series-temporais/en/
(sorry about the automatic translation).
Posted by: Jorge Camoes | Mar 10, 2007 at 07:53 PM