Simulation studies have shown that when data were generated with no group covariate effects between populations (the null hypothesis), the Likelihood Ratio Test consistently rejected the null hypothesis far too often (Smith 1991). The Analysis of Deviance (ANODEV) approach, however, performed according to its expected nominal distribution (Smith 1991). For this reason, SURPH.2 provides an Analysis of Deviance procedure for testing models with group covariate effects.
The theory behind Analysis of Deviance and a complete description of the models can be found in the SURPH.1 Manual; the following is simply a description of the procedure for using the Analysis of Deviance with SURPH.2.