Development and application of a genetic algorithm-informational modeling approach to exploatory statistical modeling of lizard-habitat relationships

Anolis carolinensis, an arboreal lizard common to the southeasternUnited States, has been studied often in lab settings, but infrequently in its natural habitats with respect to the ecology of this species. The current study conducted exploratory statistical modeling of associations between 18 habitat features and the occurrence of A. carolinensis in study plots at the northern distributional limits of this species in eastern Tennessee.Statistical hypothesis-testing procedures and stepwise computer algorithms are commonly used by ecologists to analyze observational (nonexperimental)multivariate data, such as the data analyzed in this study.However, such procedures and algorithms are frequently, but inappropriately, used to find the single supposedly \u27best statistical models and/or support interpretations of the importance or causal nature of variables in the model. Thus, such analyses provide only a narrow scientific view of the multivariate data and the many potentially useful models.The present study developed a genetic algorithm-information modeling (GAIM) approach to a) reduce certain computational and statistical limitations imposed by stepwise algorithms and hypothesis testing procedures, respectively, and b) conduct a wider exploration of any observational multivariate data set. The GAIM approach utilizes a genetic algorithm, which bases its searching power on biological and evolutionary concepts, and the informational approach to statistics, which bases its ability to rank and evaluate models on statistical likelihood and information theory. It is suggested that researchers use an approach that provides a wider view of the data (e.g., finds many models that fit the datawell instead of just one or a few models), such as the GAIM approach, to more fully explore observational multivariate data. The set of well-fitting models obtained from a GAIM analysis can then be used to propose combinations of variables or factors that could be investigated by experiments in order to test causal hypotheses and/or produce predictivemodels.One hundred sixty-six plots were placed in four different habitats along the Little Tennessee River where A. carolinensis occurs. Plots were surveyed for the presence/absence of this species in summer and winter seasons and habitat variables, both in and adjacent to the plots, were measured. Logistic regression modeling using the GAIM approach was conducted separately on the summer and winter data sets. For the summer data, the most frequent variables in the final set of GA models were(including the intercept): distance to potential overwintering rock, summercanopy categorization, distance to habitat edge, herb/shrub/vine cover,swimmer sunlight index, ambient temperature, and standardized distance along the habitat edge from the west boundary of habitat.For the winter data, the most frequent variables in the final set of models were (including the intercept): ambient temperature, presence of live overstory evergreen tree trunks, presence of overwintering rock,standardized distance along the habitat edge from the west boundary of habitat, distance to potential overwintering rock, and canopy cover categorization. In each data set, the variables which most frequently occurred in the final model set were also the ones which most frequently assessed statistically significant parameter estimates.The summer models suggest that further research on A. carolinensis might focus on a) sunlight and thermal factors and b) habitat features related to certain spatial scales beyond the skimmer home range scale.Future research might also examine responses of this species to winter habitat features such as a) shelter and potential basking sites, b) sunlight availability and temperature, and c) spatial features beyond the typical winter home range size. Methods using experimental control, or at least partial control, over field variables are needed to determine the specific responses of this species to key habitat features and the causal mechanisms underlying those responses. In addition, more studies are needed whichtake approaches based on biophysical and physiological ecology, especially if they can be linked to reproductive output, population ecology, and habitat use on local and regional scales