« The paradox of screening | Main | Projecting iPad sales »


Feed You can follow this conversation by subscribing to the comment feed for this post.

Cris Luengo

A test that gives two thirds false positives is not a very accurate test, and is not very useful at all. I cannot believe they would tell these 69% of people they have cancer. Are you sure these numbers are correct?

This particular test doesn't help your argument much, because it's worthless no matter what the a priori probabilities are. There are plenty of examples for screenings that produce more false positives than true positives, and that actually use accurate tests.


Cris, note that the 69% is a worst-case scenario as I'm assuming that every male gets screened but there is no way around it because the specificity is so low.


You also have to figure in what you expect people who get positives to do. In this particular case, there are a number of men with prostate cancer for whom the treatment is no treatment, because the cancer is so slow-moving and the men are so old.

So if you take these men out of the pool, then what happens?

Pierre-Hugues Carmichael

I was really surprised by that very low specificity. It would be interesting to see if that's the norm for all screening tests or just the sad story of the PSA. I also like John's comment about what happens after the screening results are revealed.

But again, do doctors rely simply on the screening test to give a diagnostic ? I personally wouldn't give much credence to a screening result until I passed a more official (though potentially more expensive and/or invasive) diagnostic test.

On a lighter note, I find it hilarious that PSA stands for Prostate-Specific Antigen, while the specificity of the test is really very poor.

Jon Peltier

"56% healthy people who test positive"

If you flip a coin, you'll only have 50% false positives.


John and PHC: Anyone testing positive in a screening test should go for a confirmatory test, as you suggested. Whatever that test is, it also has its own false positive rate. In addition, there must be a reason why that second test is not used as a screening test: as PHC pointed out, it's probably because it is very expensive, or very invasive.

Jon: I said the same thing in my reply to PHC. A specificity of 33% means 2/3 of the time, a healthy person will test positive. If you flip a coin, only 1/2 would test positive. But this is not quite right because in reality, we don't know who's healthy and who's not.


Why not continue with an application of Bayes rule?

Pr( c | + test ) = Pr( + test | c )Pr( c )/[Pr( + test | c )Pr( c ) + Pr( + test | ~c )Pr( ~c )]
= .8*.16/[.8*.16 + .67*.84] = .185

Which is to say, the probability of having cancer given that you get a positive test result is about 1/5.


Noahpoah: Thanks for computing the positive predictive value (PPV).

It shows that the PSA test is mildly useful, i.e. better than random. Without the test, 16% of the population have cancer ("the prior"). Given the positive test result, 18.5% of those have cancer.



I hadn't thought of it that way. I was thinking in terms of how the bad specificity of the test and relative rarity of the disease make a positive test result fairly useless, maybe even damaging, given the stress that thinking you have cancer produces, the cost and possible complications of further, possibly more invasive tests, etc...


Noahpoah: No doubt your conclusion is the right one, and I agree with it too. I brought the other interpretation to show that whoever adopted this test wasn't completely nuts. The test is better than random selection but not much better.

Verify your Comment

Previewing your Comment

This is only a preview. Your comment has not yet been posted.

Your comment could not be posted. Error type:
Your comment has been posted. Post another comment

The letters and numbers you entered did not match the image. Please try again.

As a final step before posting your comment, enter the letters and numbers you see in the image below. This prevents automated programs from posting comments.

Having trouble reading this image? View an alternate.


Post a comment

Your Information

(Name is required. Email address will not be displayed with the comment.)

Marketing and advertising analytics expert. Author and Speaker. Currently at Columbia. See my full bio.

Spring 2015 Courses (New York)

Jan 26: Business Analytics & Data Visualization (14 weeks) Info

Feb 23: Statistics for Management (10 weeks) Info

Mar 28: Careers in Business Analytics & Data Science (one-day seminar) Register

Apr 7: The Art of Data Visualization Workshop (6 weeks) Register

Next Events

Sep: 28 Data Visualization New York Meetup, New York, NY

Oct: 5 Andrew Gelman’s Statistical Communications class, Columbia University

Oct: 13 AQR ProSeminar, NYU Sociology

Oct: 22 Leading Business Change Through Analytics, Columbia Business School

Oct: 30 Ray Vella’s Designing Infographics class, NYU

Past Events

See here

Junk Charts Blog

Link to junkcharts

Graphics design by Amanda Lee


  • only in Big Data