In a study performed
a few years ago [1], the following problem was proposed to 18 expert
consultants for AIDS patients:
Helen has tested positive for AIDS.
How likely do you think she is really an AIDS patient? What would you advise
her?
The input
data are:
1. The probability of having AIDS, when
one belongs to a population without special risk, is 1 in 10,000.
2. The sensitivity of the AIDS test is
99.9%. In other words, the probability of a false negative is 0.1%.
3. The specificity of the AIDS test is
99.99%. In other words, the probability of a false positive is 0.01%.
The result
of the study was as follows:
·
The 18 experts agreed
that the probability that Helen is an AIDS patient is greater than 90%. Most
thought that the probability is greater than 99%. Some even claimed that is
greater than 99.9%.
·
All experts said that they
would advise Helen to inform her family, make everybody test for AIDS, and start taking medication.
1. As the probability of having AIDS in
this population is 1 in 10,000, there will probably be a person with AIDS and
9,999 AIDS-free:
2. Since the probability of a false
negative is 0.1%, it is almost certain that the person with AIDS will test
positive.
3. Since the probability of a false
positive is 0.01% (1 in 10,000), when testing the 9,999 individuals who do not
have AIDS, one of them will test positive without having AIDS (the probability
of this is 0.9999).
4. Summarizing: when testing those
10,000 people for AIDS, two will test positive. One is an AIDS patient, the
other is not.
5. Therefore, the probability that Helen
is really an AIDS patient is 50% (the same as getting heads when flipping a
coin).
6. The advice the experts should have
given Helen is: repeat the test in another laboratory.
Are the experts
really experts? In whose hands are we?
[1] E.Kurz-Milcke,
G.Gigerenzer, L.Martignon, “Transparency in risk communication: graphical and
analog tools,” in Strategies for risk
communication: evolution, evidence, experience, Annals of the New York
Academy of Sciences, vol.1128, 2008.
The same post in Spanish
Thematic Thread on Statistics: Next
Thematic Thread on Linguistics and Medicine: Next
Thematic Thread on Statistics: Next
Thematic Thread on Linguistics and Medicine: Next
Manuel Alfonseca
No comments:
Post a Comment