I got in to an interesting discussion today over at sciencebasedmedicine.org about medical errors and the huge variety of numbers that are out there.
It reminded me of a project I started to work on a few years ago at work, where I attempted to correlate low staffing levels to an increased error rate.
When I finished the graph, I discovered that the trend was completely in the wrong direction....low staffing levels actually appeared to be correlated with fewer errors. When I did some fishing around, it became obvious what had happened. The error reporting system is all self reporting by staff. When they don't have enough staff to cover all the work, they also don't have time to fill out error reports, and thus many small to mid sized errors went undocumented.* When there was more staff available, people were more compliant in their error reporting.
This is yet another risk of self reported data.
I will admit, it still amuses me that somewhere deep in my documents file, I still have a spreadsheet that suggests that the fastest way to get rid of errors would be eliminate most of the staff. Correlation does not equal causation, and graphs are only as good as the source data.
*Hospital policy dictates that we should record "near miss" errors.....situations where nothing actually went wrong, but almost did. The theory is that we need a record of these issues so we can address things before they actually cause a real problem. However, if no one documents them at the time, they are nearly impossible to find later.