Parsing a new-ish CPSC report

Update: I just wrote, and lost, an entire post. Blast! Sigh. I'll try to replicate it as best I can.

Last post, I promised regressions and more data-digging. Forgive me, but I won't do that today! Instead, I'll write about a new report from the Consumer Product Safety Commission (CPSC) on playground-related injuries from 2001-2008. The report mostly uses data from an internal CPSC database, in addition to the NEISS data (which we saw last time), and death certificate data. First, kudos to the CPSC for putting out a revised report. There are so many playground injuries each year (over 200,000!) that it's important people - parents, educators, regulators - don't grow complacent. I found the report on the Maryland Materials website, link here.

Still, the report isn't perfect. My main gripe with it is that the CPSC database is not a nationally representative sample; it's a collection of reports culled from newspapers, direct complaints, letters from lawyers, etc. There were 2,691 incidents reported to the CPSC from 2001-2008... meanwhile, there were about 60,000 incidents collected in the NEISS database over the same period! That means that one can extrapolate the NEISS data with a reasonable degree of confidence and make claims based on that data on the state of playground safety in the country as a whole, but one can't do the same with the CPSC data.

A random and interesting foible in the CPSC data is that Wisconsin day care centers comprise 35% of all reports (as opposed to restaurants in New Mexico, homeowners in Vermont, schools in California, and so forth) because all product complaints from Wisconsin are automatically sent to the CPSC database. Bizarre!

There are some striking discrepancies in the CPSC data and NEISS data. For instance, 28% of all accidents in the CPSC data are reported as occurring at home (and that number rises to 44% of you exclude the day care center-heavy Wisconsin data), compared to about 17% in the NEISS data over the same period. Moreover, 4% of all injuries in the CPSC data (without Wisconsin) occur at school (7% with the Wisconsin data), whereas the NEISS data tells us that about 25% of all playground-related accidents occur at school! We'll assume that the NEISS data are "correct", since they are statistically representative (assuming that all sample hospitals submit accurate and timely injury reports... *gulp*). So, why are home accidents over-reported and school accidents under-reported in the CPSC data?

There seem to be three possible reasons. First, the two datasets represent different things. The CPSC data cover for reported accidents, whether hospital-worthy or not, and the NEISS data are strictly confined to hospitalizations. So, maybe parents are just over-anxious, reporting every little incident that happens, and schools are more lackadaisical. Second, it could just be statistical noise - i.e., there are so few observations that it's just coincidental that so many accidents happened at home. Third, and most cynically, and therefore most likely, it could be the case that parents report home accidents in the hopes of building a lawsuit against the equipment manufacturer, while schools DON'T report because they want to open themselves to liability in that particular direction (since the equipment manufacturer could blame the school for poor upkeep, or improper installation, etc.).

Final thoughts: one must be careful when drawing conclusions from data! Even with large samples like NEISS, there are always possible reasons for why the data are skewed in a certain way.