The biggest story to come out of Chicago racing this summer has been the number of fatal breakdowns that have occurred at Arlington Park - 19 so far this season (DRF story  Rick Telander).  Del Mar has also had a rash of breakdowns (bring on the Polycrap). Any or all of the following five reasons may explain why this is happening and why it is such big news.

#1 Course superintendents, track maintenance crews and groundskeepers have suddenly become incompetent idiots.

#2 Trainers are using illegal and legal drugs that are pushing horses beyond their physical limits and masking their own perceptions of pain.

#3 Breeders are producing unsound horses from lightly raced precocious sires with weak constitutions. If Ghostzapper managed only four starts per campaign and eleven races total, should we expect his off-spring to excel racing every two weeks? (Was he unsound or just well managed?)

#4 Barbaro's breakdown and the subsequent media attention have put a microscope on racetrack injuries and fatalities.

#5 Bad Luck.

I would guess that #4 and #5 are the most likely reasons behind the breakdowns and the associated media blitz. Arlington had more breakdowns in 2004 (24) but the Barbaro story is so fresh in everyone's minds that any tragedy at the track is getting publicity. The effects of #2 and #3 would not be felt in such a short timeframe, though these are serious long run issues that probably have already contributed to a gradual increase in breakdowns (assuming there really has been an increase in breakdowns - I need to see some more numbers). I have a hard time accepting #1 given the advancements in track maintenance. However, the fact that most of the attention is on just one track does make you wonder.  

I would love to build a dataset that could be used to study the factors that determine breakdowns (training patterns, breeding, surface condition, age, # previous starts, days since last race, etc). I am not quite sure how to do this outside of mining results charts. The DRF sells charts in a text file for $800 per year. There are 13 full years of downloadable charts, so the data would cost $10,400. Anyone think of a better way?