Statistical Significance

(Mike Glasbrener) #1

I really enjoy Russ Roberts Econtalk podcast. He has quite a diverse guest population on a wide variety of topics.

Anyhow this week was on statistical significance and they touch a bit on it’s application to health care studies. I wish anyone generating, reviewing, publishing and publicizing studies understood this so study results and their significance could be digested prior to implementation of crappy public policy. I also wonder how many results are tossed due to confirmation biases.
Link to this week’s page.

(Bunny) #2

I agree, there should be a standard set of protocols like hard empirical cross validation and then contrasted to another level to compare related subsets to search for any stragglers then and only then this or that can be said (universal consensus within the respectable medical and scientific community) about what the real deal is for “application!”

(Mike Glasbrener) #3

One point was that very statistally significant results can be obtained from smaller data sets. Researchers often conclude that if it’s so significant from a small data set then the results from a larger data set will be even more significant and valid. However, the reverse is true. The smaller the data set the more significant the result and the higher the likelyhood the result conclusion is wrong.

Again, there were multiple tie ins to the medical research field which ties back into Taubes and Feldman’s Assertion that medical researchers aren’t great scientists…

(Bunny) #4

Time being a significant factor!