A test of fit for lattice distributions
Thomas W. Epps · Communication in Statistics- Theory and Methods · 1995
The goodness of fit of a model for count data f 0(x;θ)x ∊ {0,1,2,…}, may be judged by measuring the discrepancy between the probability generating functions of model and data--p 0)(t;θ) =; Σt x f 0(x;θ) vs. p n(t) = Σt x f n(x). A formal test based on the mean difference between them, is widely applicable, has better power than the Pearson chi-squared test in-distinguishing common lattice models, and compares favorably with newer, specialized tests. Plots of p n − p 0 give a visual Interpretation of the results and the nature of departures from H 0.