Naked Science Forum
General Science => General Science => Topic started by: thedoc on 13/12/2011 17:44:33
-
How do you analyse data without having a presupposition of what you're actually looking for? Because that's important too. You're actually biasing the search system.
Asked by Evil Eye Monster via Twitter
Visit the webpage for the podcast in which this question is answered. (http://www.thenakedscientists.com/HTML/podcasts/show/2204/)
-
We answered this question on the show...
Steven - Actually, different people can look at the same data set and try and find different information from it. So, it’s not necessarily one approach. Different people will try and tease out different bits from the same data source.
-
One can certainly crunch data to look for something that is significant. But, there is often a fallacy in doing it that way. They say that a good statistician can prove anything!!
If one is analyzing the data without an a-priori hypothesis, then one should at least find a way to split the data. Perhaps doing a small pilot study, then repeating with a second larger and more rigorous study to see if the new hypothesis holds up.
Generally a researcher should have some expectations of what to expect, then do some kind of a blinded study so that their expectations don't interfere with the data collection. The biggest risk, of course, is to have a study that is designed to prove the hypothesis without thoroughly considering alternative explanations.
Some of the cohort studies like the Framingham Heart Study (http://en.wikipedia.org/wiki/Framingham_Heart_Study) were designed precisely to look for obscure causes of heart disease. But, as I mentioned, one must be careful to ascertain that the data is generally applicable, and not just an artifact of the one particular study.