@article {311, title = {Metapopulation models: An empirical test of model assumptions and evaluation methods}, journal = {Ecology}, volume = {86}, number = {11}, year = {2005}, note = {Nov Metapopulation models: An empirical test of model assumptions and evaluation methods}, pages = {3088-3098}, abstract = {Patch occupancy models provide a simple phenomenological approach to evaluating ecological questions on a metapopulation scale. In this study, I use and modify a patch occupancy model to evaluate the effects of synchronous extinctions correlated with flooding and patch-size-dependent migration on the regional dynamics of a neotropical beetle, Cephaloleia fenestrata. Various methods have been used to evaluate patch occupancy models. Previous authors most commonly have evaluated patch occupancy models by logistic regression of incidence functions. Likelihoods produced from regression methods, however, neglect autocorrelation in spatial occupancy patterns even though spatial autocorrelation is common in ecological data. In this study, I used a Monte Carlo method of model evaluation, which accounts for spatial autocorrelation. Results suggest that patches undergo synchronous extinctions correlated with flooding and this affects regional dynamics of C. fenestrata. Immigration probability positively correlated with patch size, and emigration probability negatively correlated with patch size, also affecting C. fenestrata regional dynamics. The patch occupancy pattern was positively spatially autocorrelated at only two of the sites, but was nearly significant at another. The logistic regression method was a reasonable alternative to the Monte Carlo method for model evaluation. Other model evaluation methods (fits to model development data, proportion of occupied patches, and turnover rates) were inconsistent (at best) quantitative measures of a model{\textquoteright}s fit to independent data. When used to select among competing models, however, the fits to model development data were reasonably good indicators of fits to independent data. Given the increased complexity involved in using the Monte Carlo method, the simpler logistic regression method may be a preferable alternative, especially when spatial autocorrelation is minimal.}, author = {Johnson, D. M.} }