Slashdot Mirror
How the Next US Nuclear Accident Might Happen
Lasrick writes: Anthropologist Hugh Gusterson analyzes safety at US nuclear facilities and finds a disaster waiting to happen due to an over-reliance on automated security technology and private contractors cutting corners to increase profits. Gusterson follows on the work of Eric Schlosser, Frank Munger, and Dan Zak in warning us of the serious problems at US nuclear facilities, both in the energy industry and in the nuclear security complex.
Except he really doesn't do anything. There doesn't appear to be any study, only the subjective (i.e., qualitative) third party claims, which doesn't mean that they are wrong, just that he didn't do anything himself. He does however launch an attack at quantitative methodology, which isn't a surprise, given that his article approach is a defence of his own field, at heart: If you can measure it, it is by default open to quantitative assessment.
This applies to scales (hello psychometrics) which are almost never measured without error (heh, look: Error in variables and latent measurement models!), open ended responses (latent direlecht allocation models and similar) and multiple measurements from different sources (back to reliability and latent measurement models). He is right in principal, and makes the point in the article that having poor test security and design (where the testees' employers have access to, or even provide the examinations and assessments themselves) is wrong, and that systems that provide too many false positives are ignored.
The correct approach to the final system would be a layered system, in which sensitivity increases with depth. As for assessments: no shit, don't let people grade themselves. Ever. And impose penalties and randomly conducted tests by third parties. If you want to hire this out: make it so that whoever succeeds get a bonus. Make the two sides compete. This only defines why QA is of vital importance.