A public database of workplace injuries

The U.S. Department of Labor has plans to create a publicly available database of workplace safety incidents for large corporations.  The idea is that adding this kind of public visibility for injuries and illnesses will encourage companies to develop more effective safety programs and will raise the public visibility for the 3 million workers a year who are hurt or sickened.  If companies do a really good job, they can use this as a benefit by promoting their safety record in a more publicly credible way. 

Companies have legitimate concerns because data is often misused by advocates and by people who don’t understand statistical validity and reliability.  They might compare the safety records of companies in different industries, with different workforces, and come to erroneous conclusions. There are a slew of information processing heuristics that I have blogged about at length that would lead company advocates to underappreciate the problems and safety advocates to overappreciate the problems.

The trick then comes down to how the database user interface is designed.  With data from 500,000 companies available, accurate and fair comparisons will be available in most cases.  The interface should be designed to highlight these rather than to allow full flexibility to surf and mine the data to get whatever the user wants.  This is different from the approach taken for many database projects, but may be a better way to go in this case, both politically and in recognition of human nature. 

One way of achieving this would be to use a derived industry-specific metric.  Weighted and adjusted rates are common in public health, economics, and many other domains.  For example, a company with a 1.3 for injuries could indicate that they have 30% more injuries or 0.3 standard deviations more or whatever, compared to a carefully matched set of other companies.  Once the political minefield of developing the weighting and adjusting statistics has been waded through, it might be reasonable to presume that the rates will be used accurately or if not could easily be disputed. 

Of course, organizations will still try to game the system.  That is also human nature.  There will be underreporting incentives targeted directly at employees.  There will be coding schemes that minimize the apparent severity of injuries that are reported.  But if this is done across the board, it could be that the weighting and adjusting process will cancel these all out and we can compare companies within an industry, industries across the economy, and individual companies over time.  This benefit is too much to give up on.