Random sampling will be no better at improving safety than the police who post radar traps once every 6 months on a given stretch of road. The other 363 days a year, people will break the speed limit in that area with impunity. Want to stop it, put up a speed camera that gets everyone. Same would apply to analysis of CVRs, but as has been already pointed out correctly, to do so would be prohibitively complex and expensive with virtually no provable benefit to be had. Unlike flight data, it requires human interpretation, and not just some of the time, but all of the time. Ask a CVR team from your favourite accident investigation authority how long it takes to perform analysis of just one accident CVR, and they already have some idea of what it is they are looking for.
The other thing about random sampling is that while you may only look at a portion of your data, you're responsible for 100% of it. While you're not likely to ever get to 100% of your data, even if you wanted to, it's going to look very bad if one of the ones you "missed" in your 1 in 50 program ends up being later shown to have been a precursor to something serious.
If cockpit discipline is a serious issue for the industry, there are much better ways to assure compliance through reinforcement in training and checking; with frequent published reminders; and by having professionals "self checking" each other. The way forward is to build a culture of compliance, not by using a stick and a carrot.
Last edited by J.O.; 26th Oct 2010 at 11:21.