Feeding decisions by Steller's jays (Cyanocitta stelleri): The utility of a logistic regression model for analyses of where, what, and with whom to eat

It is widely recognized that animal behavior is simultaneously affected by many variables. Both the study of interactions between these variables unde r naturalistic conditions and the proper statistical analysis of data deriv ed from such studies remain particular problems for ethologists. In the pre sent study we investigated choices by Steller's jays (Cyanocitta stelleri ) selecting between two feeding locations under a variety of conditions. A m ultifactor logistic regression analysis of our data showed that the jays' b ehavior was simultaneously affected by several variables, including proximi ty of the feeding site to cover, food preferences, and the presence of cons pecifics and other animals. We found that (1) jays strongly preferred an un occupied feeder over one occupied by another jay or a sympatric mammal with the effect of squirrels being much greater than that of other jays, (2) co ntrary to our expectations, in the absence of a reason to prefer the other feeder, the jays generally selected the feeder that was further from neares t cover, and (3) the presence of sunflower seeds on one feeder but not on t he other provided a reason to prefer the feeder offering sunflower seeds. T he logistic regression analysis provided a more complete and integrated mod el of the birds' behavior than more commonly used univariate methods. Our a pproach and results are also applicable to studies of other animals. While univariate analyses are useful in some instances, multifactor procedures re veal more details about the interactions of single factors and future exper imental studies can take advantage of this additional knowledge.