Difficulty of changing topologies: lambda

The parameter lambda determines the difficulty of changing topologies. There are two options in the program. In the run settings menu you are asked if you want to update lambda.

Update lambda: NO

The parameter lambda is kept fixed. You are prompted for this fixed value in the model options menu.

Update lambda: YES

The parameter lambda is sampled with Gibbs sampling from the posterior distribution, which has the form of a beta distribution. You are again prompted for a value of lambda in the model options menu. This value, however, is not the value of lambda itself, but the mean of its prior distribution. The beta distribution depends on two parameters, alpha and beta. These parameters are related to the mean according to mean = alpha/(alpha+beta). In the program, the value of beta is kept fixed: beta=2. Specifying the mean just determines both parameters of the prior distribution. Note, however, that as the mean increases, the variance of the prior distribution decreases. The following figure shows plots of the prior distribution for various values of the mean, as shown at the top of each subfigure.

Beta priors


Last modified: July 2001

Back to the beginning of this documentation.