Comments by "George Reynolds" (@karhukivi) on "Ловушка Байеса" video.

  1. The test is 99% sensitive, so if 100,000 people who have the disease are tested, then 1% or 1000 people won't test positive. However two independent tests change that scenario as the ones who don't test positive will not be the same for both tests.
    2
  2. Easiest to comprehend. We all know about medical screening and the fear of the result coming back positive. Even medical people get it wrong and make the wrong diagnosis. Coin flipping, selecting drawers in quiz programs, eye-witness accounts and marketing studies are also used as examples, but the medical ones seem to give the best explanations.
    1
  3. OK here's an example: if 80% of people with measles have spots, does a person with spots have an 80% chance of having measles? What about spots not caused by measles? Or the 20% of people who actually have measles but don't have spots? The problem is inverse probability and Bayes Rule can be proved by using sets theory. Google on it, as there are plenty of explanations and proofs for all levels of ability.
    1
  4. If you have gout, you might be given a uric acid test. Well, the probability of a positive given gout = 50% so the test is actually worthless under Bayesian statistics. However, your doctor will probably still do it to reinforce their opinion. If the doc thinks it is gout, then a positive uric acid will confirm it. If the test comes back negative, the doc will still think you have gout! A survey some years ago showed that medical people got this wrong 80% of the time.
    1