mudiff.mblmodwoc.equalvar uses a mixed Bayesian/likelihood approach to
determine conservative sample sizes, in the sense that the desired posterior credible interval coverage and length
for the difference between two normal means are guaranteed over a given proportion of data sets that can arise according to the prior information, when variances are equal.
The desired total length of the posterior credible interval for the difference between the two unknown means
First prior parameter of the Gamma density for the common precision (reciprocal of the variance)
Second prior parameter of the Gamma density for the common precision (reciprocal of the variance)
The desired fixed coverage probability of the posterior credible interval (e.g., 0.95)
The probability that the length of the posterior credible interval of fixed coverage probability level will be at most len
The number of points simulated from the preposterior distribution of the data. For each point, the length of the highest posterior density interval of fixed coverage probability level is estimated, in order to approximate the (100*worst.level)%-percentile of the posterior credible interval length. Usually 50000 is sufficient, but one can increase this number at the expense of program running time.
The Maximum number of Consecutive Steps allowed in the same direction in the march towards the optimal sample size, before the result for the next upper/lower bound is cross-checked. In our experience, mcs = 3 is a good choice.
Assume that a sample from each of two populations will be
collected in order to estimate the difference between two independent normal means.
Assume that the precisions of the two normal sampling distributions are
unknown but equal, with prior information in the form of a Gamma(alpha,
mudiff.mblmodwoc.equalvar returns the required sample sizes to attain the desired length len
for the posterior credible interval of fixed coverage probability level
for the difference between the two unknown means.
The Modified Worst Outcome Criterion used is conservative, in the sense that the posterior credible interval
length len is guaranteed over the worst.level proportion of all
possible data sets that can arise according to the prior information, for a fixed coverage probability level.
This function uses a Mixed Bayesian/Likelihood (MBL) approach. MBL approaches use the prior information to derive the predictive distribution of the data, but uses only the likelihood function for final inferences. This approach is intended to satisfy investigators who recognize that prior information is important for planning purposes but prefer to base final inferences only on the data.
The required sample sizes (n1, n2) for each group given the inputs to the function.
The sample sizes are calculated via Monte Carlo simulations, and therefore may vary from one call to the next.
It is also correct to state that the coverage probability of the posterior credible interval of fixed length len will be at least level with probability worst.level with the sample sizes returned.
Lawrence Joseph email@example.com and Patrick Belisle
Joseph L, Belisle P.
Bayesian sample size determination for Normal means and differences between Normal means
The Statistician 1997;46(2):209-226.
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.