In a generalized linear model the variance of an observation yi with mean μi is
Var(yi) = φ V(μi)
where φ is the dispersion parameter. With a Normal distribution, for example, V(μi)=1 and φ=σ2, while for Poisson and binomial distributions φ=1. However, if there is heterogeneity in the data (that is, the residual deviance is noticeably larger than its expected value i.e. the residual number of degrees of freedom), you can use the DISPERSION option of MODEL to request some other value.