Regression modeling of rodent locomotion data.

We present the case for a model-based approach to the statistical analysis of rodent locomotion data for which previous research has established that the time course of behavioral change follows a specified parametric form. Inadequacies of statistical methods commonly applied to such data are described. Regression models can reveal time-dependent aspects of response that might escape more elementary analyses, as we illustrate with experimental data regarding the locomotion response to amphetamine. Inter-individual variation in response profiles is incorporated through the use of random coefficients (i.e. mixed effects models). These models provide an effective means of quantifying the extent of individual differences in behavioral response to stimulant drugs not attributable to treatment. The need to adjust for multiple comparisons when searching for regions of significant difference is discussed. We present specific models that we have found useful for characterizing responses to low doses of amphetamine or cocaine (the log-logistic random effects model) and higher or repetitive doses of amphetamine or cocaine (the polynomial random effects model). Code for performing these analyses with standard statistical software is provided.