Print

---

I am working for re-submitting of a grant. The focus of my problem is how to use Bayesian approaches in computing power in a logistic regression model. The detailed reviewer's comment to my pervious submission is as follows:

‘One additional problem with the power estimates is that the hypotheses of interest appear to employ order constraints, i.e. are often of the form A is greater than B and both are greater than C. Order restricted inference is notoriously hard in a frequentist setting, and its use invalidates most of the traditional approaches in calculating power. The investigators might be better served by switching to Bayesian inference instead, where the posterior probability of the hypothesis A>B>C can be more meaningfully evaluated given some flat prior. Packages such as WinBUGS could be used.’

I am new to WinBUGS and have no any knowledge of BUGS language. Due the short time and the urgency (July 1 deadline), I appeal to you, the WinBUGS experts, for helps. I'll greatly appreciate it if you can provide a sample WinBUGS program for computing power for a logistic regression model. The assumptions for the power calculation are:

1) model: logit(Y) = alpha + beta1*X1 + beta2*X2 + beta3*X3 + beta4*X4, where Y, X1--X4 are dichotomous variables (0 or 1);

2) available sample size N=1153
3) exp(beta1) = 2.5 (odds ratio), exp(beta2) = 1.7, exp(beta3) = 1.6 and exp(beta4) = 1.8;
4) the proportions: 48% for X1 = 1, 8% for X2 = 1, 11% for X3 = 1 and 9% for X4 = 1. The proportion for Y = 1 is 33%.

I just downloaded the WinBUGS and had installed it to my PC. I am learning how to run a program and view the results.
Thank you very much in advance for your help.

Hong Xian