Posted by Larry Hoover on January 28, 2009, at 8:32:33
In reply to Re: Stats...ooops, posted by linkadge on January 27, 2009, at 10:59:37
There's another way you can visualize what I'm trying to describe. Again, it starts with the data in this slide:
http://medicine.plosjournals.org/perlserv/?request=slideshow&type=figure&doi=10.1371/journal.pmed.0050045&id=96831&ct=1You can plot your own histogram (bar graph) from this data set.
If the two comparator groups, drug and placebo, are not different, then a histogram of their difference in mean scores should yield that classic bell curve, with the peak of the curve at or near the zero line, which is called the normal distribution. If you create this histogram, the x axis should have the range of -10 to + 10, and the y-axis should the number of trials at each unit interval. Estimates are fine. You don't need to be perfect.
If you do that, you get nothing at all like the classic bell curve. Instead, you get something more or less like half of a bell curve. The data are said to exhibit homoskedacity, which is to say they are skewed or biased (-skedacity part) in one direction (the homo- part). In all my years of studying stats and such, I doubt I've ever seen such a classic example of this type of distribution.
Also, in a normal distribution, only 2.5% of the data points should lie so far above the zero difference point that they are considered to be significantly different from the rest. In this example, we have 43% (i.e. 100% minus 57%) which are significantly above the zero difference line.
And yet, with this overwhelming evidence of drug superiority, critics use sophistry to declare things like "80% of the response of drugs is seen in the placebo group". But that's due to confounding variables, uncontrolled independent variables, in the clinical trial methodology. If the critics were intellectually honest about the findings, they would be criticizing clinical trial methodology, not the drugs.
In my perhaps not so humble opinion.
Lar
poster:Larry Hoover
thread:876214
URL: http://www.dr-bob.org/babble/20090104/msgs/876719.html