SI
SI
discoversearch

We've detected that you're using an ad content blocking browser plug-in or feature. Ads provide a critical source of revenue to the continued operation of Silicon Investor.  We ask that you disable ad blocking while on Silicon Investor in the best interests of our community.  If you are not using an ad blocker but are still receiving this message, make sure your browser's tracking protection is set to the 'standard' level.
Biotech / Medical : XOMA. Bull or Bear? -- Ignore unavailable to you. Want to Upgrade?


To: aknahow who wrote (9142)3/12/1999 9:10:00 AM
From: Tharos  Read Replies (5) | Respond to of 17367
 
George,
Are we maybe looking at a test construction problem? Because the FDA relies heavily on statistical evidence, the data would have to be statistically stable so data could be assumed to represent the population, and the mean and std deviation could be determined.

You would want to construct your test small enough so that it only contains the number of samples necessary to meet the relative population representation requirements, but large enough not to suffer from any data skewing or anomalies. With the desire for p = 0.005 the calculated sample size would necessarily be critical, and without reaching that number you would not be able to statistically verify that your sample did indeed represent the population. If the tests were constructed with a perceived higher death rate than is currently being experienced, then the data could easily appear skewed or non significant if the death rate in the untreated group had been halved by improvements in classical treatment. If my belief is true, then I think someone more qualified in statistics would be able to produce a couple of numeric examples.

If the test sample statistic was determined not on total samples, but on expected results extrapolated from then knowns, then it is easy to expect changes to the formula would have a significant impact. I think that is what we are seeing in the test and why it is being continued.