Saturday, August 8, 2020

reading review - seeking wisdom, riff offs (encore)

OK, back on the stage for one last riff - I hope this clears up the last thought I shared on Thursday:

11c) Let’s say someone ID’s one of two colors with 80% success. If the color exists in 10% of the population, the other color is misidentified in 18% of total IDs. So how sure can we be it's the minority based on an identification? Around 31%.

This asks something like - was the car you saw blue or green? - and sets a condition that there is a 20% chance of identification error (I say blue when it was green). Further, if green is seen in only 10% of the total of all cars added together, then we know 2% of green answers are incorrect; green is incorrectly called blue 18% of the time.

If you write it out, it looks like this:

  • Blue car is correct answer - 72%
  • Blue car is incorrect answer - 2%
  • Green car is correct answer - 8%
  • Green car is incorrect answer - 18%

So if you add up the percentages and take the proportion, you get the odds of each answer being correct:

  • Probability blue is correct: 72/74 = 97.3%
  • Probability blue is incorrect: 2/74 = 2.7%
  • Probability green is correct: 8/26 = 30.8%
  • Probability green is incorrect: 18/26 = 69.2%

In other words, if someone says they saw the less frequently seen car, it only takes a small identification error rate before the answer becomes more than 50% likely to be incorrect.