Determining Synchrony Between Behavioral Time Series: An Application of Surrogate Data Generation for Establishing Falsifiable Null-Hypotheses

Synchrony between interacting systems is an important area of nonlinear dynamics in physical systems. Recently psychological researchers from multiple areas of psychology have become interested in nonverbal synchrony (i.e., coordinated motion between two individuals engaged in dyadic information exchange such as communication or dance) as a predictor and outcome of psychological processes. An important step in studying nonverbal synchrony is systematically and validly differentiating synchronous systems from non-synchronous systems. However, many current methods of testing and quantifying nonverbal synchrony will show some level of observed synchrony even when research participants have not interacted with one another. In this article we demonstrate the use of surrogate data generation methodology as a means of testing new null-hypotheses for synchrony between bivariate time series such as those derived from modern motion tracking methods. Hypotheses generated by surrogate data generation methods are more nuanced and meaningful than hypotheses from standard null-hypothesis testing. We review four surrogate data generation methods for testing for significant nonverbal synchrony within a windowed cross-correlation (WCC) framework. We also interpret the null-hypotheses generated by these surrogate data generation methods with respect to nonverbal synchrony as a specific use of surrogate data generation, which can then be generalized for hypothesis testing of other psychological time series. Translational Abstract Recently psychological researchers from multiple areas of psychology have become interested in nonverbal synchrony (i.e., coordinated motion between two individuals engaged in dyadic information exchange such as communication or dance) as a predictor and outcome of psychological processes. Researchers interested in quantifying nonverbal synchrony from bivariate time series face a challenge when attempting to distinguish between synchrony quantifications obtained from synchronous systems and synchrony quantifications obtained from nonsynchronous systems. In this article, we propose and review four methods of surrogate data generation for distinguishing between synchronous and nonsynchronous systems. Surrogate data generation is a method commonly used in physics to distinguish meaningful order within a data set from randomness by generating new null-hypotheses. To understand these new null-hypotheses obtained from the proposed surrogate data generation methods in the context of nonverbal synchrony, we performed surrogate data analyses on two simulated data sets (sine waves and random noise) and two real data sets obtained from dyadic conversations. We also break down each surrogate generation method in detail and clearly state each newly generated null-hypothesis in the framework of nonverbal synchrony. Through simulation, we show that presented surrogate data generation methods appear to be well-suited for testing nuanced hypotheses regarding nonverbal synchrony.