06.03.2017 Views

Mathematics for Computer Science

e9ck2Ar

e9ck2Ar

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

“mcs” — 2017/3/3 — 11:21 — page 771 — #779<br />

18.9. Probability versus Confidence 771<br />

seems to suggest that it’s worth betting with high odds that you have TB, because<br />

it makes sense to bet against something unlikely happening—like the test being<br />

wrong. But having TB actually turns out to be a lot less likely than the test being<br />

wrong. So the either-or of Corollary 18.9.1 is really an either-or between something<br />

happening that is extremely unlikely—having TB—and something that is<br />

only very unlikely—the diagnosis being wrong. You’re better off betting against<br />

the extremely unlikely event, that is, it is better to bet the diagnosis is wrong.<br />

So some knowledge of the probability of having TB is needed in order to figure<br />

out how seriously to take a positive diagnosis, even when the diagnosis is given with<br />

what seems like a high level of confidence. We can see exactly how the frequency<br />

of TB in a population influences the importance of a positive diagnosis by actually<br />

calculating the probability that someone who tests positive has TB. That is, we want<br />

to calculate Pr TB j pos , which we do next.<br />

18.9.2 Updating the Odds<br />

Bayesian Updating<br />

A standard way to convert the test probabilities into outcome probabilities is to use<br />

Bayes Theorem (18.2). It will be helpful to rephrase Bayes Theorem in terms of<br />

“odds” instead of probabilities.<br />

If H is an event, we define the odds of H to be<br />

Odds.H / WWD PrŒH <br />

PrŒH D<br />

PrŒH <br />

1 PrŒH :<br />

For example, if H is the event of rolling a four using a fair, six-sided die, then<br />

PrŒroll four D 1=6; so<br />

Odds.roll four/ D 1=6<br />

5=6 D 1 5 :<br />

A gambler would say the odds of rolling a four were “one to five,” or equivalently,<br />

“five to one against” rolling a four.<br />

Odds are just another way to talk about probabilities. For example, saying the<br />

odds that a horse will win a race are “three to one” means that the horse will win<br />

with probability 1=4. In general,<br />

PrŒH D Odds.H /<br />

1 C Odds.H / :<br />

Now suppose an event E offers some evidence about H . We now want to find<br />

the conditional probability of H given E. We can just as well find the odds of H

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!