0
$\begingroup$

I was reading the book I came across the conditional probability. The statement is:

Imagine a certain disease that affects 1 in every 10,000 people. And imagine that there is a test for this disease that gives the correct result (“diseased” if you have the disease, “nondiseased” if you don’t) 99% of the time.

T = Test result is positive. D = Person is diseased.

So, P(T|D) = 0.99 and P(D) = 0.0001. The author calculated P(D|T) by Bayes theorem.

P(D|T) = P(T|D) P(D) / [P(T|D)P(D) + P(T|~D)P(~D)]

For that, he calculated different probabilities.i.e.

  • P(~D)= 1- P(D) = 0.9999
  • P(T|~D) = 1 - P(T|D) = 0.01

I could understand intuitively that P(~T|D) = 1 - P(T|D). But I am unable to understand how P(T|~D) = 1 - P(T|D) is valid? Can somebody explain mathematically?

P.S. The snapshot of the book snippet is: Probabilty

$\endgroup$
  • $\begingroup$ ... It isn't true in general; in this case the "gives the correct result (...) 99% of the time" sentence makes two distinct things the same, which is why the result holds in this particular case $\endgroup$ – Glen_b -Reinstate Monica Jun 4 '17 at 11:04
5
$\begingroup$

In general, P(T|~D) ≠ 1 - P(T|D). To see why, suppose (1) the test results were completely independent of the disease and (2) the test is positive 25% of the time.

(1) implies that $$P(T|\lnot D) = P(T|D) = P(T).$$ If the test results are completely independent of the disease, then it doesn't matter what the disease state is. Then the equation you're asking about becomes $$P(T|\lnot D) =^? 1 - P(T|D)$$ $$P(T) =^? 1 - P(T)$$ The second version is true if, and only if, P(T) = .5. In particular, filling in the value from (2), $$.25 ≠ 1 - .25 = .75.$$

It looks like the book section you were reading is here. I don't see where the author is assuming that P(T|~D) = 1 - P(T|D).

Maybe you're getting confused by the way the author moves between two versions of the total probability. One version is $$ P(F) = P(F,E) + P(F,\lnot E).$$ The second is $$ P(F) = P(F|E)P(E) + P(F|\lnot E)P(\lnot E).$$

The first version expresses the probability of F in terms of two conjunctions: F-and-E, F-and-not-E. The second version expresses the probability of F in terms of two conditional probabilities: F-given-E and F-given-not-E. The definition of conditional probability makes them equivalent. But it's important to note that the conditional version includes the marginal (unconditional) probabilities of E and ~E.

Another potential point of confusion is that the author does most of the calculation in non-percentage numbers (.99 rather than 99%), then reports the final value as a percentage (.98% rather than .0098).

$\endgroup$
  • $\begingroup$ I inferred P(T|~D)= 1 - P(T|D) because the author has calculated P(T|~D)=0.01 and P(T|D)=0.99. So, I became confused that how he calculated P(T|~D)=0.01. $\endgroup$ – chuckskull Jun 4 '17 at 11:57
  • 2
    $\begingroup$ The author assumes that the test "gives the correct result (“diseased” if you have the disease, “nondiseased” if you don’t) 99% of the time." That is, P(T|D) = .99 and P(~T|~D) = .99. So the equation they use is P(T|~D) = 1 - P(~T|~D). $\endgroup$ – Dan Hicks Jun 4 '17 at 12:19

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.