Staying with my off the wall bad examples, here is a old math problem:
Three people go to stay at a motel, and the man at the desk charges them $30.00 for a room. They split the cost ten dollars each. Later the manager tells the desk man that he overcharged them, that the actual cost should have been $25.00. The manager gives the bellboy $5.00 and tells him to give it to them. The bellboy, however, decides to cheat the 3 customers and pockets $2.00, giving each of the them only one dollar.
Now each has paid $9.00 to stay in the room and 3 x $9.00 = $27.00. The bellboy has pocketed $2.00. $27.00 + $2.00 = $29.00 - so where is the missing $1.00?
I know where the missing $1.00 is and I also know the next flip of the coin in 50%.
But looking at data that does not influance the outcome or looking at the wrong data to come to a conclusion (my two bad examples) will not give you a accurate conclusion on the odds of another large problem on QA.
Now for this coin of mine that comes up heads everytime......the coin is for sale who wants to start the bidding..??
You better bid fast because I am going to walk outside now and my chances of getting hit by lightning is a lot higher than I thought......
Phystutordotcom: My best suggestion is that the MTBF for QA (with no code changes) should be a bell curve with a median of 8 years. But that's not sufficient to calculate a probability of failure within N years because we don't know the variance. A single data point doesn't give us any basis to even guess the variance. Sure, you could guess at 4 years because it's half way between zero and 8, and do the calculation (integral of the Gaussian function) -- but what's the point? It's pure guesswork, dressed up in the language of math.
