Strategy Selection — Single or Multiple?

spannerindexThis post tries to do a little tying together on a familiar subject. I look at a couple of papers that provide more perspective than typical research papers provide. First is the preliminary dissertation of Anke Söllner. She provides some educated synthesis which my posts need, but rarely get. Two of her papers which are also part of her dissertation are discussed in the posts Automatic Decision Making and Tool Box or Swiss Army Knife? I also look at a planned special issue of the Journal of Behavioral Decision Making to address “Strategy Selection: A Theoretical and Methodological Challenge.”

Söllner’s work is concerned with the question:  which framework–multiple strategy or single strategy– describes multi-attribute decision making best? In multi-attribute decision making we have to choose among two or more options. Cues can be consulted and each cue has some validity in reference to the decision criterion. If the criterion is an objective one (e.g., the quantity of oil), the task is referred to as probabilistic inference, whereas a subjective criterion (e.g., preference for a day trip) characterizes a preferential choice task. The multiple strategy framework is most notably the adaptive toolbox that includes fast and frugal heuristics as individual strategies. Single strategy frameworks assume that instead of selecting one from several distinct decision strategies, decision makers employ the same uniform decision making mechanism in every situation. The single strategy frameworks include the evidence accumulation model and the connectionist parallel constraint satisfaction model.

Söllner summarizes her findings as follows:

  1.  Automatic (compensatory) information integration (in line with the connectionist parallel satisfaction model) can be observed for multi-attribute decisions from given information, when information search is reduced to a minimum.
  2.  Decision makers do not ignore additional information, but seem to automatically integrate it as they vary their behavior (choice outcomes, information search, confidence judgments) in accordance to its content. This finding supports the idea of a single uniform mechanism as proposed by the connectionist network framework and the evidence accumulation framework. This conclusion decides that the integration of information does not seem to be as costly as assumed by the multiple strategy framework.
  3.  Stopping behavior is dependent on the level of evidence given and inter-individually diverse – much more than predicted by the stopping rules incorporated in the multiple strategy framework. This finding lends further support to the adequacy of the evidence accumulation framework to describe multi-attribute decision making.

So Söllner clearly comes down on the side of the single strategy framework. She does note that much of the multiple strategy framework can be helpful to the theoretical development of single strategy frameworks. I have certainly been convinced by the single strategy framework, especially the connectionist parallel satisfaction model. At the same  time I have missed many of the fine details.

Söllner provides synthesis for me on single strategy models. She states that connectionist models assume that the decision problem is represented in a neural network that captures all decision-relevant information. Activation spreads in parallel through the network until a stable state of maximized consistency is reached and the option with the highest positive activation is chosen.  According to Söllner, the process of information search is not formally modeled within PCS – the model representing the connectionist network framework of multi-attribute decision making. The parallel constraint satisfaction model starts with the assumption that there is only one all-purpose rule for integration and choice, but that there are different rules for search, generation, and change of information. In other words, they assume that individuals use different strategies for input formation but that they apply a general, multifunctional mechanism for output formation. Another prominent class of single strategy models that includes the process of information search is called evidence accumulation models. These models assume a sequential sampling process that terminates as soon as the accumulated evidence passes an evidence threshold.

Söllner points out that the underlying idea of adaptation to environmental characteristics, however, is not universal to the multiple strategy framework, but can also be part of connectionist network models and evidence accumulation models  that can adjust their parameters adaptively. As has been argued before, the question how these parameters are adjusted constitutes a theoretical challenge for the single strategy frameworks just as the strategy selection problem continues to challenge the multiple strategy framework. This is the question of weights.

In the request for papers for the planned special issue  of the Journal of Behavioral Decision Making to address “Strategy Selection: A Theoretical and Methodological Challenge,” the debate between single and multiple strategy frameworks is well characterized in the description of the editors:

While Andreas Glöckner can be considered a proponent of single-strategy approaches, and connectionist as well as evidence-accumulation models in particular, Julian Marewski is one of multi-strategy frameworks, and known for adopting an approach to modeling strategy selection that relies on cognitive architectures and production-systems (i.e., ACT-R). For instance, while Glöckner can be considered a critic of the fast-and-frugal heuristics and cost-benefit frameworks, Marewski’s research aligns with these approaches. At the same time, both Glöckner and Marewski stand for a group of strategy selection researchers that, while using experimental methods, place emphasis on investigating strategy selection via computer simulations. Arndt Bröder, in turn, takes a theoretically more agnostic position, neither siding with multi- nor with single-strategy approaches. He represents those strategy selection researchers that rely heavily on sophisticated experimental methods to investigate the behavioral contingencies that any strategy selection model should be able to explain. Finally, in contrast to Bröder and Glöckner, who strictly control for confounds in their research, Marewski belongs to those cognitive scientists who stress the importance of complementing controlled designs with those that model naturally-occurring memory, perceptual, and other confounds, all of which are exploited by the cognitive system in the real-world.

Söllner is as likely as anyone to have been assigned the job of writing that description of the editors since Arndt Bröder is her principal advisor . No matter who wrote it, it was likely a difficult job, but it says a lot that is difficult to determine just by reading papers.  Glockner (see post Parallel Constraint Satisfaction Theory and many others) and Marewski (see post ACT-R and Cognitive Niches) are clearly rivals as can be seen from commentaries they have written on each others work.