A TED talk by epidemiologist Ben Goldacre, who hasn’t been taking his Ritalin lately. He’s quite entertaining, and during the second half of the talk, he also gets quite serious. Have a look.

He talks about a lot of bad ways to present science, to misinform. One of them is through design flaws. This happens a lot in trials, often without any malicious intent. And then, some of the time, the design is truly rigged. The flaws will be made less conspicuous, hopefully left undiscovered. See the talk for some examples.

His last main point for this talk is on publication bias, the bias you find when research showing the “wrong” results has been withheld and never been admitted to the research pool. You do several trials. Then, after you see the results, you cherry pick the ones that are to be published. The others are left buried in a dark corner somewhere. Publication bias can be uncovered by putting publications on a specific treatment into a funnel plot and see if it looks symmetrical, or if one side is missing.

Above is a nice and symmetrical funnel plot showing no publication bias.

This last figure is a plot demonstrating publication bias – in actual studies on publication bias(!)

This entry was posted in Statistics. Bookmark the permalink.

Leave a Reply

Your email address will not be published. Required fields are marked *