Analyzing sequential categorical data on dyadic interaction: A comment on Gottman.

Abstract

Recent theoretical developments emphasize that social interactions are dynamic and reciprocal, and this has led to widespread use of time-series data on behavior in two-person systems. In principle, such data allow one to separate the influences of two actors on each other, Statistical methods currently being used, however, are deficient in several respects. In this article, we show that a statistic proposed by Sackett and later "proved" by Gottman is incorrect. We also show that the failure to control for autodependence can produce misleading results. Finally, we introduce new procedures that are based on both traditional and more recently developed methods for the analysis of contingency tables. Using these procedures, we show how to test for dependency, how to measure dependency, how to test for differences in dependency across subgroups, and how to test for dominance in reciprocal behavior sequences.

The influence of prior knowledge on viewing and interpreting graphics with macroscopic and molecular representations

“…Each expected transition frequency was calculated by multiplying the sum of the observed transition frequencies in the criterion row with the sum of the observed transition frequencies in the target column, and then dividing by the total number of twoevent sequences in the table. Finally, each observed transition frequency was compared to the expected transition frequency for the same two-event sequence by means of an adjusted residual z score (Allison & Liker, 1982;Bakeman & Gottman, 1997). Following the z score distribution for the normal curve for one-tailed tests of significance, a z score of 1.65 or higher is at a significance level of p < .05, a z score of 2.33 or higher is at a significance level of p < .01, and a z score of 3.11 or higher is at a significance level of p < .001 (Pagano, 1990).…”

Marital interaction and satisfaction: A longitudinal view.

“…Within each couple, Bakeman's ELAG4 program was used to compute the Allison and Liker (1982) z score of sequential connection. The z score measures the direction and gain in prediction of the consequent code's occurrence given knowledge that the antecedent code has occurred.…”

Detecting group differences in sequential association using sampled permutations: Log odds, kappa, and phi compared

“…To investigate bias, we generated empirical sampling distributions for phi, transformed kappa, and Yule's Q. The distributions were based on 10,000 sequences, each 101 events long (so that :c Allison &Liker,1982, andFaraone &Dorfman, 1987; for additional examples of situations more complex than those considered here, also see Budescu, 1984, andIacobucci &Wasserman, 1988). In such cases, Dumas (1986), citing Tavare and Altham (1983), argues that estimates of cross influence need to be corrected for autocorrelation within each stream, yet not all experts apply such corrections (e.g., Wickens, 1993), and in any event, the proposed correction is nil if autocorrelation is absent.…”