Cranial Size and Intelligence Link

[Mandrake]

We have to differ on this, r=0.5 is not a high correlation. It's borderline, at least in the field of biology, and I think biologists tend to be more generous than those in other sciences. Further, if you look back through the threads here, you will see that there are some who are claiming r=0.2 is a high correlation, and that is the background for my comments.

Do you consider medicine to be a biological science? When a drug company seeks approval for a new drug, it must submit various studies to the government for approval. If a drug is found to correlate at r=.2 to a side effect that causes permanent disability or death, will the drug be approved? At what value of r would the license be granted? The same question can be asked with respect to drugs that might be given to pregnant women. If r=.2 for deformed babies or death to the fetus, would the drug be allowed for use with pregnant women?

If you were considering an elective operation and knew that the correlation between the procedure and a debilitating outcome was .2, would you have the operation? If not, what value of r would you consider to be acceptable?

 

[humanino]

This is a ridiculous argument ! There are some sides effects, always. If the drug compagnies had the same requirements as in physics, there would be none.

Of course, there would also be no drug passing so high requirements. This is only due to market considerations. If they had long enough time (say over decades) they would be able to produce such medicines.

 

[Moonbear]

Do you consider medicine to be a biological science? When a drug company seeks approval for a new drug, it must submit various studies to the government for approval. If a drug is found to correlate at r=.2 to a side effect that causes permanent disability or death, will the drug be approved? At what value of r would the license be granted? The same question can be asked with respect to drugs that might be given to pregnant women. If r=.2 for deformed babies or death to the fetus, would the drug be allowed for use with pregnant women?

If you were considering an elective operation and knew that the correlation between the procedure and a debilitating outcome was .2, would you have the operation? If not, what value of r would you consider to be acceptable?

In the examples you cite, the data wouldn't be analyzed that way. A Pearson's correlation (r) is the wrong statistic to use. So, if a company seeking drug approval handed a report to FDA that included such a statistic, no, the drug would not be approved because FDA would tell them their analysis was flawed. Drug trials would not include correlations or even post-hoc analyses; those are very frowned upon by FDA. Severe side effects that required people drop out of the study or led to serious health problems would either be reported as a proportion of the subjects reporting the side effects or using a survival analysis. Using a Pearson's correlation requires comparing two variables with a normal distribution and similar standard deviations. The closer r is to 0, the more scattering there is of the values from a linear relationship.

 

[Mandrake]

This is a ridiculous argument ! There are some sides effects, always. If the drug compagnies had the same requirements as in physics, there would be none.

I didn't make any argument. I asked a few questions. We all know that drugs have side effects and are sometimes licensed with side effects that can be serious. The question is whether a .2 correlation is insignificant with respect to biology. Your answer is apparently a resounding YES. But the question I raised is at what correlation is the risk of a serious complication accepted as small enough to allow. I seriously doubt that a drug with life threatening side effects in the r = .2 range would be licensed. If that is true, we can conclude that this example of biology recognizes small correlations as very important.

 

[Mandrake]

In the examples you cite, the data wouldn't be analyzed that way. A Pearson's correlation (r) is the wrong statistic to use.

If you can produce a scatter diagram with the data, you can determine a correlation coefficient. I am not claiming anything about the licensing prodedures of the FDA. The simple question is whether a drug would be licensed if it had that kind of correlation. I asked a similar question with respect to whether or not readers here would undergo elective surgery, if they believed that the correlation between that operation and severe impairment could be represented by a correlation coefficient of .2.

 

[humanino]

Your answer is apparently a resounding YES.

Nope. Never said that. I said, with regards to lethality, a widespread product cannot take such a high risk. 1% of consumer dying would be far above acceptable level for instance. However, 1% of likelihood for a scientific result, does that look acceptable to you ?

 

[Moonbear]

If you can produce a scatter diagram with the data, you can determine a correlation coefficient. I am not claiming anything about the licensing prodedures of the FDA. The simple question is whether a drug would be licensed if it had that kind of correlation. I asked a similar question with respect to whether or not readers here would undergo elective surgery, if they believed that the correlation between that operation and severe impairment could be represented by a correlation coefficient of .2.

My point is you can't produce a scatter diagram with the type of data you suggested. Instead, you'd have something along the lines of drug vs placebo or new drug vs currently approved drug, and for each, you'd have incidence of reported side effect as your dependent variable. So you're just creating an argument with no validity because it doesn't fit with the way real data are collected or analyzed. If you're thinking of this differently than that, please demonstrate the type of data you're envisioning would be collected for such an analysis...feel free to make up a fictitious data set if it helps illustrate the point.

 

[Mandrake]

My point is you can't produce a scatter diagram with the type of data you suggested.

Yes, I agree. I understand your point. Correlations relate variables. I picked poor examples, since I was suggesting a binary outcome, such as survival or death. In order for a correlation to be used with respect to a drug or a medical procedure, there must be two variables. In the context of my comments, one of the variables would have to be an outcome. When the outcome is a parameter that is measured over a range, it is unlikely that a life and death situation exists. I presume outcomes that would apply to these categories may be things such as vision acuity, lung capacity, blood pressure, IQ, hearing acuity, physical strength, etc. All of these are measurable over a range.