Re: JDK version popularity
On Mar 24, 9:35 am, Patricia Shanahan <p...@acm.org> wrote:
Lew wrote:
Lew wrote:
Umpteen people agreeing on a guess doesn't make it correct.
Nathan wrote:
Do you agree that two guesses are better than one though?
Of course not. Without facts to at least hint at the truth, two
ignorant guesses are twice as ignorant as one. Is that better?
It all depends on the probability of correctness of the individual guesse=
s.
Consider a yes/no question. If each individual guess is indeed based on
absolutely no facts, so that its probability of being correct is exactly
0.5, a million guesses are no more useful than one guess.
However, I strongly suspect that most people who respond to questions in
this newsgroup have a better than 0.5 chance of being right.
If each individual guess has a probability of correctness that is at
least 0.5+e for some positive number e, then you can achieve any
required probability of correctness by taking a majority vote among
enough guesses.
See "weak learners" and "boosting" for application of this principle to
more complicated cases.
I said, "without facts to at least hint at the truth", which would
indicate a probability of less than 0.5 of being right. I also
disagree with your assessment that people in this newsgroup have a
better than 0.5 chance of being right. I'd be surprised if they have
even a 0.5 chance. They would need *some* evidence for the
probability to improve. Anecdotal evidence is not very helpful, as
memories are inaccurate and individual samples sizes are too small. I
would assess random guesses at the percentage of Java version
installations, *absent evidence*, to be essentially worthless even
among the fine minds present here.
That said, my comment was directed at the question of whether two
guesses were better than one. Even if the probability that guesses
are anywhere close is some tiny epsilon greater than 0.5, only two
guesses still does not yield a discernible improvement. Now if we
could get a large sample size where people here *accurately* report
how many current installations of their knowledge have which version
of Java, then we can get a reasonable estimate. Again, absent any
facts with at least a hint of truth, we are not serving the OP by
pulling percentages out of our posteriors.
What is the prejudice against facts and evidence, anyway?
--
Lew