The modeling of a behavioral stream here begins with the set of all T-patterns detected in a real-time behavior record coded in terms of the real-time occurrences of many different types of behavioral events. The T-pattern type and an algorithm (in the Theme software) for its detection have been described elsewhere [1,2,3,4]. T-patterns, which are recursively defined (as patterns of patterns), are essentially repeated chains of events characterized by fixed event order and significantly similar time distances between the consecutive parts of the chain over its repeated occurrences. While each detected T-pattern may capture structural aspects of particular temporal segments of the behavioral stream, the kind of model described in this paper combines the information contained in different patterns to describe the whole stream as alternating and/or temporally nested performances of sub-sets of the detected T-patterns.

A special algorithm first considers mutually exclusive patterns. It establishes all possible sub-sets of detected patterns where none of their respective occurrences overlap in time and selects the set (A), which has the maximum combined duration of its patterns, i.e., covers the largest proportion of the observation period (typically at least tens or hundreds of millions of sets need to be considered); see Figure 1. Nested patterns are considered next. Per definition, a T-pattern X is nested in T-pattern Y, if X is not a sub-pattern of Y but occurrences of X only take place during occurrences of Y (partially nested if at least one occurrence of X does). Patterns that are nested in any one of the patterns in A are then identified and added to the model, which thus gives a more complete description of the whole behavioral stream. Models of this kind are presented describing different types of human interaction such as children's collaborative problem solving and discussions between adults.

Figure 1. The lower part of this figure illustrates how a behavioral stream (here 320 s of children's dyadic object play) can be described in terms of alternating occurrences of (here three, marked 1, 2 and 3) different non-overlapping T-patterns. Considering all possible combinations of the T-patterns that the Theme program detected within this behavioral stream, it automatically identified this particular sub-set as the one covering the greatest percentage of the observation time (total duration optimization). The upper part of this figure shows in more detail only one of the three patterns; marked 2. For an explanation of how to read this kind of pattern diagrams see Magnusson (1996a, 1998, 2000). It turns out that still other independent patterns may be nested within (i.e. occur completely within) the patterns of such optimized pattern sets.

References

1. Magnusson, M.S. (1996a). Hidden real-time patterns in intra- and inter-individual behavior: description and detection. European Journal of Psychological Assessment, 12, 112-123.
2. Magnusson, M.S. (1996b). T-patterns, Theme and The Observer. In: Measuring Behavior '96. Proc. Int. Workshop on Methods and Techniques in Behavioral Research (16-18 October 1996, Utrecht, The Netherlands), 69. Electronic publication at www.noldus.com/events/mb96/abstracts/magnusson.htm.
3. Magnusson, M.S. (1998). Real-time pattern detection versus standard sequential and time series analysis. In: Measuring Behavior '98. Proc. 2^nd Int. Conf. on Methods and Techniques in Behavioral Research (18-21 August 1998, Groningen, The Netherlands), 211-213. Electronic publication at www.noldus.com/events/mb98/abstracts/magnusson.htm
4. Magnusson, M.S. (2000). Discovering hidden time patterns in behavior: T-patterns and their detection. Behavior Research Methods, Instruments & Computers, 32, 93-110.

© 2000 Noldus Information Technology b.v.