Analyzing the Dynamics of Affective Dyadic Interactions Using Patterns of Intra- and Interindividual Variability

Abstract

There are many compelling accounts of the ways in which the emotions of 1 member of a romantic relationship should influence and be influenced by the partner. However, there are relatively few methodological tools available for representing the alleged complexity of dyad level emotional experiences. In this article, we present an algorithm for examining such affective dynamics based on patterns of variability. The algorithm identifies periods of stability based on length of time and amplitude of emotional fluctuations. The patterns of variability and stability are quantified at the individual and dyadic level, and the approach is illustrated using data of the daily emotional experiences of individuals in romantic couples. With this technique, we examine the fluctuations of the emotions for each person and inspect the overlap fluctuations between both individuals in the dyad. The individual and dyadic indices of variability are then used to predict the status of the dyads (i.e., together, apart) 1 year later.

Notes

¹Matlab code for the algorithm, as well as detailed information for its implementation, is available on the first author's website: http://psychology.ucdavis.edu/labs/ferrer/pubs/

²The equation for estimating reliability of within-person change is R_C = , where σ² _person*day represents the variance due to the person × day interaction, σ² _Error represents expected error variation, and m represents the number of items in the scale. Estimates for all generalizability coefficients computed from variance components estimates are available upon request.

³We conducted analysis removing the constraint of the individual distances between their respective centers having to be less than the maximum dispersion parameter. The results were similar to those with the constraint. Moreover, removing this condition did not result in loss of information, as the lower bound of the distance is the same and the upper bound represents how far from each other the two individuals' fixations are.

⁴To assess the performance of our RC identification algorithm, we conducted a Monte Carlo study. Details and results of the simulation study are in the Appendix.

⁵We thank the Multivariate Behavioral Research editor for suggesting these alternative metrics.

^aNo within-series variability, thus SD estimates are not applicable. The simulated error standard deviations were 0.20 for Realistic, 0.10 for Single RC, and 2.00 for No RC. N_R = 1,001 replications. RC = relative constancy.