Experimental design: Problems in understanding the dynamical behavior-environment system.
Variability is data—track the client’s recent history to explain replication failures.
01Research in Context
What this study did
Iivanainen (1998) wrote a theory paper. It says we should stop treating ups and downs in data as noise.
Instead, we should see them as clues about the client’s past learning history.
The paper calls this idea “dynamical systems thinking.”
What they found
Old-school A-B-A designs miss the real story. They treat each data point as separate.
The author says behavior is more like weather: today’s response depends on yesterday’s.
So a failed replication may just mean the client’s “behavioral state” changed.
How this fits with other research
Falligant et al. (2020) backs this up. They show that longer phases and baseline type change false-positive risk. That is variability carrying information, exactly what Iivanainen (1998) wants us to track.
Carey et al. (2014) give a warning. Sampling only the first few trials hides true mastery curves. Their graphs look cleaner but lie. This is the same “noise” mistake Iivanainen (1998) attacks.
Arango et al. (2023) echo the humility theme. Both papers tell analysts: your lens is limited—check your assumptions, whether cultural or statistical.
Why it matters
Next time you see bounce in a client’s graph, pause. Add one quick probe: ask parents what changed at home, check sleep, meds, or new teachers. Write that “state” note on the graph. Over a few cases you will spot hidden variables that explain the bounce and save you from pointless program changes.
Want CEUs on This Topic?
The ABA Clubhouse has 60+ free CEUs — live every Wednesday. Ethics, supervision & clinical topics.
Join Free →Add a “state” box on your graph sheet: jot sleep, meds, or life events each session.
02At a glance
03Original abstract
In this paper, I attempt to describe the implications of dynamical approaches to science for research in the experimental study of behavior. I discuss the differences between classical and dynamical science, and focus on how dynamical science might see replication differently from classical science. Focusing on replication specifically, I present some problems that the classical approach has in dealing with dynamics and multiple causation. I ask about the status and meaning of "error" variance, and whether it may be a potent source of information. I show how a dynamical approach can handle the sort of control by past events that is hard for classical science to understand. These concerns require, I believe, an approach to variability that is quite different from the one most researchers currently employ. I suggest that some of these problems can be overcome by a notion of "behavioral state," which is a distillation of an organism's history.
The Behavior analyst, 1998 · doi:10.1007/BF03391965