Compares two or more fitted models. The single-hit exponential is the common
nested sub-model, so each higher-parameter model is tested against it by the
chi-squared deviance difference on the extra degrees of freedom. preferred
is the lowest-AIC model among those that significantly improve on the
exponential, or the exponential itself when none do. Models with equal
parameter counts (approximate vs exact beta-Poisson) are not nested in each
other and are separated only by AIC/BIC. Because the exponential is a limiting
case rather than a regular nested model, the chi-squared comparison is an
approximation. With exactly two models this reduces to the original
simpler-vs-fuller deviance test.
Value
A tibble with one row per model: fit metrics, a converged flag,
AIC/BIC and their deltas, the nested chi-squared test of each model against
the exponential baseline (NA on the baseline row), logical preference
flags, a stable selection code, and a human-readable conclusion. A
warning is issued when any compared model did not converge, since a poorly
converged richer model can make a simpler model appear preferred by default.