Wednesday, February 02, 2005
The Will Rogers Effect
This ironic effect pleases me because it is not intuitive and is humorously named. While it may be tedious from a technical perspective, I believe the reader can understand it if I am lucid. I shall try...
It is found in many settings where one is trying to detect the presence of a condition in order to predict the outcome of the condition with higher likelihood. If one could remember all the cards already dealt in a black jack game he could predict the likelihood that the next card to be drawn would make him a winner more accurately. This would be a more skilled player than one who simply treats each hand as an entity unto itself.
For tedium sake, I will use an example from clinical science. In this example let's imagine that we have a condition of interest (such as cancer, Alzeheimer's, TB, etc.), an outcome (such as death, profound dementia, consumption, etc.), an old test and a new ("better") test to detect the presence of the disease.
For simplicity sake let's assume that the amount of the disease present is proportional to the likelihood of a "bad" outcome. This is generally true to a point and makes sense i.e. if you only have a "little" cancer, Alzeheimer's, TB you will likely live longer before you die, become demented or are consumed; if you have "a lot" of disease you have a higher expectation that the disease will "get you" soon.
Now let's talk about the diagnostic tests for a moment. The "old" test might be Digital Rectal Exam for prostate cancer, or psychological testing for Alzeheimer's or presence of blood in sputum compared to PSA or the new test on the horizon for a specific protein related to Alzeheimer's or a test for the presence of the TB bacterium. I've tried to illustrate this below:
The new test has moved people who were "not sick" yet had "substantial" disease into an area where they are "now sick," but with less disease than those who have always been "very sick."
The effect that using the new test has on predicting the outcome is the ironical part. The sick group now has a better outcome predicted than before because they have some new less sick folks included in their group. In a like manner the not sick group also has a better outcome than before because they have fewer "nearly sick" folks in their group.
Now to Will Rogers...who is remembered as having said something like..."When the Oakies left the state for the California Gold Rush they raised the average IQ in both states."
Did any of us do that when we left WV?
It is found in many settings where one is trying to detect the presence of a condition in order to predict the outcome of the condition with higher likelihood. If one could remember all the cards already dealt in a black jack game he could predict the likelihood that the next card to be drawn would make him a winner more accurately. This would be a more skilled player than one who simply treats each hand as an entity unto itself.
For tedium sake, I will use an example from clinical science. In this example let's imagine that we have a condition of interest (such as cancer, Alzeheimer's, TB, etc.), an outcome (such as death, profound dementia, consumption, etc.), an old test and a new ("better") test to detect the presence of the disease.
For simplicity sake let's assume that the amount of the disease present is proportional to the likelihood of a "bad" outcome. This is generally true to a point and makes sense i.e. if you only have a "little" cancer, Alzeheimer's, TB you will likely live longer before you die, become demented or are consumed; if you have "a lot" of disease you have a higher expectation that the disease will "get you" soon.
Now let's talk about the diagnostic tests for a moment. The "old" test might be Digital Rectal Exam for prostate cancer, or psychological testing for Alzeheimer's or presence of blood in sputum compared to PSA or the new test on the horizon for a specific protein related to Alzeheimer's or a test for the presence of the TB bacterium. I've tried to illustrate this below:
The new test has moved people who were "not sick" yet had "substantial" disease into an area where they are "now sick," but with less disease than those who have always been "very sick."
The effect that using the new test has on predicting the outcome is the ironical part. The sick group now has a better outcome predicted than before because they have some new less sick folks included in their group. In a like manner the not sick group also has a better outcome than before because they have fewer "nearly sick" folks in their group.
Now to Will Rogers...who is remembered as having said something like..."When the Oakies left the state for the California Gold Rush they raised the average IQ in both states."
Did any of us do that when we left WV?
Comments:
Post a Comment
<< Home
<< Home
