Performs likelihood-ratio tests on nested models. Currently, one method was implemented
for beta-binomial models (betabin
) or negative-binomial models (negbin
).
# S4 method for glimML
anova(object, …)
Fitted model of class “glimML”.
Further models to be tested or arguments passed to the print
function.
An object of formal class “anova.glimML” with 3 slots:
A vector of character strings with each component giving the name of the models and the formulas for the fixed and random effects.
A data frame containing the results. Row names correspond to the models.
logL | numeric | maximized log-likelihood |
k | numeric | number of parameters in the model |
AIC | numeric | Akaike information criterion for the model |
AICc | numeric | Corrected Akaike information criterion for the model |
BIC | numeric | Bayesian information criterion the model |
Resid. dev. | numeric | Residual deviance |
Resid. Df | numeric | df of the residuals |
Test | character | Nested models which are tested |
Deviance | numeric | Deviance difference between the 2 models |
Df | numeric | df associated with deviance difference |
A character chain indicating the kind of fitted model: “BB” for beta-binomial, or “NB” for negative-binomial model.
The comparison between 2 or more models will only be valid if they are fitted to the same data set.
The anova
method for models of formal class “glimML” needs at least 2 nested models of the
same type (either beta-binomial or negative-binomial models: they cannot be mixed). The quantity of interest is
the deviance difference between the compared models: it is a log-likelihood ratio statistic. Under the null
hypothesis that 2 nested models fit the data equally well, the deviance difference has an approximate
\(\chi^2\) distribution with degrees of freedom = the difference in the number of parameters between
the compared models (Mc Cullagh and Nelder, 1989).
McCullagh, P., Nelder, J.A., 1989. Generalized linear models. London, Chapman & Hall, 511 p. See Appendix C. Likelihood ratio statistics, p. 476-478.
# NOT RUN {
data(orob2)
# likelihood ratio test for the effect of root
fm1 <- betabin(cbind(y, n - y) ~ seed, ~ 1, data = orob2)
fm2 <- betabin(cbind(y, n - y) ~ seed + root, ~ 1, data = orob2)
anova(fm1, fm2)
# }
Run the code above in your browser using DataLab