The poster implies we should all worry because half of the studies say it's a health risk...
But by that same logic none of us should worry because half of the studies say there is no damage.
This logic doesn't hold up, because that's not how NHST (null hypothesis statistical testing) works (assuming that's what is used in these studies to determine an effect).
When you do NHST, you do a statistical to determine the probability that you could have the results by chance, assuming that the effect you are testing for does not exist (this is called "the null hypothesis"). If the probability is below a certain threshold (typical ones used are.05 and.01), then we say that we "reject the null hypothesis" and the results are "statistically significant", which basically means that they were unlikely to have occurred by chance if the effect was not actually there.
If the probability is above the threshold, all we can say is that we "failed to reject the null hypothesis". We do *not* say that the we accept the null hypothesis. It means we don't have enough confidence in the results to reject the possibility that the difference was due to chance. The problem may be that we did not use enough subjects, so we had insufficient power. ("Power" is the probability of detecting an effect, given that the effect exists). If your power is less than 50%, then it is clearly wrong to conclude anything from a result of "no effect found". You're better off flipping a coin than doing the study!
NHST has been criticized in psychology because the outcome depends upon a combination of effect size and number of subjects (sample size). The more subjects you have, the more likely you are to detect an effect. Very large sample sizes can detect very small effects, which may not really mean much but are still "statistically significant". (e.g. if something increases the risk of cancer by.00000001%, we probably don't care). More emphasis is now being placed on measuring effect sizes.
(Note that I have no idea if these studies use NHST, or even whether most medical studies do, but this is what happens in experimental psych).
It's happening anyway, 50% of engineering/computer sci PhDs are earned by foreign born students
Yes, but how many of them end up staying? More than a few, I'd wager.
Slashdot Mirror
The poster implies we should all worry because half of the studies say it's a health risk...
But by that same logic none of us should worry because half of the studies say there is no damage.
This logic doesn't hold up, because that's not how NHST (null hypothesis statistical testing) works (assuming that's what is used in these studies to determine an effect).
When you do NHST, you do a statistical to determine the probability that you could have the results by chance, assuming that the effect you are testing for does not exist (this is called "the null hypothesis"). If the probability is below a certain threshold (typical ones used are .05 and .01), then we say that we "reject the null hypothesis" and the results are "statistically significant", which basically means that they were unlikely to have occurred by chance if the effect was not actually there.
If the probability is above the threshold, all we can say is that we "failed to reject the null hypothesis". We do *not* say that the we accept the null hypothesis. It means we don't have enough confidence in the results to reject the possibility that the difference was due to chance. The problem may be that we did not use enough subjects, so we had insufficient power. ("Power" is the probability of detecting an effect, given that the effect exists). If your power is less than 50%, then it is clearly wrong to conclude anything from a result of "no effect found". You're better off flipping a coin than doing the study!
NHST has been criticized in psychology because the outcome depends upon a combination of effect size and number of subjects (sample size). The more subjects you have, the more likely you are to detect an effect. Very large sample sizes can detect very small effects, which may not really mean much but are still "statistically significant". (e.g. if something increases the risk of cancer by .00000001%, we probably don't care). More emphasis is now being placed on measuring effect sizes.
(Note that I have no idea if these studies use NHST, or even whether most medical studies do, but this is what happens in experimental psych).
It's happening anyway, 50% of engineering/computer sci PhDs are earned by foreign born students Yes, but how many of them end up staying? More than a few, I'd wager.
You mean, a New York Times reporter like Judith Miller whose looking at jail time if she doesn't reveal the name of a source?