GGDMLE: a computer program which finds maximum likelihood estimates for the generalized log gamma distribution.

A FORTRAN program is described which finds maximum likelihood estimates for the generalized log gamma model for survival time data. The model includes regression variables which are assumed to be linearly related to log survival time and handles censored data. The shape parameter is treated as fixed in the optimization procedure (a modification of Powell's hybrid method), but the program can find maximum likelihood estimates (MLE) for several different values of the shape parameter in one run. The other parameters can be optionally fixed so special cases of the generalized log gamma (GLG) model, such as the log exponential, extreme value and normal models, can be studied. A subroutine is included which calculates initial parameter estimates for a given shape parameter value which can be used to start the optimization procedure. After some background mathematics, a description of how to use the program is given, including input/output features, a description of the subroutines and an explanation of the flow of control. An application, using data from Princess Margaret Hospital, Toronto, is presented to illustrate the program's use.