4 responses to “Bayesian Experimental Design Part III”

  1. ben dwamena

    Have you tried your hands on Stata’s bayesmh command yet?

  2. Alex Sutton

    John, Just discovered your blog generally – v.impressive! Agree with the sentiment and approach to sample size determination described over the 3 blog parts here. I believe this is not appreciated widely enough and hopefully your writing will help in this matter.

    Economists have their own take on this, of course, often termed value of information analysis. Because their outcomes are utilities which they ultimately equate to money the fully Bayes approach is quite natural

    One important aspect in all this is knowing the degree of impact a trial will have. Will the results only influence the clinician doing the trial, or all in the hospital where he/she works; or perhaps it impacts nationally, or even internationally? I think you have to assume something about this before you can measure total utility change a trial may cause. Then, if the new treatment is adopted, how long will the new treatment be used for until a superior treatment comes along? – That also needs estimating to estimate total utility impact. But even if you could estimate all this, the funder of the trial may not be “interested” in utilities gained outside their jurisdiction. So the perspective taken by the analysis can influence the calculations significantly.

    What I would like to know is in what proportions are the following the reasons why more sophisticated sample size methods are not used: a) people don’t know about them/don’t know how to carry them out; b) people don’t have the time for the extra work required; or c) given the difficulty in estimating the necessary quantities, the approach is perceived too difficult to carry out reliably?

    My guess is that a) is the current main driver, but wonder if c) would be a further stumbling block once a) is addressed?

Leave a Reply

Network-wide options by YD - Freelance Wordpress Developer