Measures derived from the thermodynamic formalism for dynamical systems are applied to human behavior to determine the degree of interaction within sequences of choices. Sequences of 500 binary choices generated by 22 human subjects are analyzed using autocorrelation and mutual information functions, as well as the fluctuation spectrum of local dynamical entropies or local spatial scaling exponents. The main results are as follows: (1) choices generated by the subjects are interdependent on short, midrange, and large scales relative to the 500 choices generated in the paradigm; (2) human behavior is characterized by large fluctuations of dynamical and geometrical choice characteristics; and (3) these geometric and dynamical characteristics can be related to a few competing principles.