Swiss Army Knife or Adaptive Tool Box?

giant-swiss-army-knife-3This post is based on the paper: “Single-process versus multiple-strategy models of decision making: Evidence from an information intrusion paradigm,” written by Anke Söllner,  Arndt Bröder,  Andreas Glöckner, and Tilmann Betsch. It appeared in Acta Psychologica in January 2014. It is a well done overview of multi-attribute decision models (Multi-attribute decision making deals with preferential choice e.g., “Which dessert do you like better?” and probabilistic inferences e.g., “Which dessert contains more calories?”),  along with clever experiments.  I am confused that it is single process vs multiple strategy. I would think that it would be process vs. process or strategy vs strategy.

This appears to me to be another polite skirmish in the continuing battle between fast and frugal heuristics and compensatory connectionist models. Do we change strategies or adjust decision thresholds or weights? However, the researchers have moved back to broader frameworks to get a different way to study and attack.  This paper has an interesting group of authors. Sollner and Broder wrote a paper last year that looked at similar issues, but focused on the importance of looking separately at how information is acquired and how information is integrated.  Glockner and Betsch are prime proponents of parallel constraint satisfaction theory– a single process model that apparently is weak on information acquisition. I will expect a Gigerenzer  or Marewski counter move soon for the fast and frugal heuristics side. I should note that there seems to be much respect between those with differing views, and the idea that probably everyone is a little bit wrong and a little bit right seems to pervade.

Multiple Strategy Models (MSMs) propose that decision makers have several  decision strategies or heuristics at their disposal, for example, the “adaptive toolbox,”  and choose adaptively between them. The selected decision strategy determines the sequence of information search (search rule), the amount of information searched (stopping rule), and how information is integrated (decision rule). The “take-the-best” heuristic (TTB) assumes a cue-wise information search along a cue validity hierarchy—from the cue with the highest validity to the cue with the lowest validity (TTB’s search rule). Information search terminates as soon as a cue discriminates between the considered options and favors only one of them (TTB’s stopping rule). The decision maker chooses the option favored by the discriminating cue (TTB’s decision rule). Thus, if the stopping rule is satisfied before all cues have been investigated, TTB uses only a subset of available and applicable information.

The Single Process Models (SPMs) comprise the second framework for multi-attribute decision making. Here, the decision maker employs one single decision making mechanism. Two classes of the SPMs are connectionist models  and evidence accumulation models. (I should note that evidence accumulation models are new to me. They seem to be TTB married to a rational model.) Connectionist models assume that decisions are formed by parallel consideration of all available decision-relevant information in a neural network representing the decision problem. Activation spreads in the network until a stable state is reached and consistency is maximized. The option with the highest positive activation is chosen. The connectionist models focus on the process of information integration, given a set of information. Evidence accumulation models assume a sequential sampling process that terminates as soon as one option surpasses a certain threshold of preference or confidence. Whenever this happens, a choice is made in favor of this option. Evidence accumulation models do not focus exclusively on information integration, but often also model the process of information search—either in a probabilistic or a deterministic way. Although the SPMs avoid the strategy selection problem by assuming only one single mechanism that is applied to all multi-attribute decisions, one might argue that they merely replace this issue with a different problem: How do decision makers adjust the proposed uniform mechanism? According to Sollner et al the theoretical advantage of the SPMs over the MSMs lies in the fact that the MSMs often do not confine the set of decision strategies. Hence, new behavioral phenomena may be captured by extending the toolbox with more sophisticated strategies. The downside of SPMs is that they currently do not provide strict predictions for search or the selection of decision boundaries. I should note that I thought that parallel constraint satisfaction relied on the deliberative system for information acquisition. For Glockner and Betsch et al to point out these weaknesses is convincing.

To distinguish between the two frameworks, the researchers approach rests on basic assumptions of the two frameworks. MSMs comprise decision strategies of different degrees of complexity. A key feature of the less complex strategies is that they concentrate on a specific part of the available information only, ignoring the remaining strategy-irrelevant part. One example for such a strategy is the TTB heuristic– “take the best, ignore the rest.” SPMs, in contrast, do not share this notion of valid, but potentially irrelevant information. Instead, they would predict that any applicable piece of information readily available to the decision maker is fed into the single decision making mechanism and therefore influences the decision maker’s behavior. Importantly, SPMs can, of course, ignore information by giving a weight of zero to it.  Hence, if people have learned or decided to employ a specific strategy as assumed in the MSM framework, their behavior should all things being equal, not be influenced by information that is irrelevant to execute this strategy. If, however, valid information is automatically evaluated as is assumed in evidence accumulation or connectionist models, the decision maker’s behavior should be influenced by this information.

The authors approached this question by introducing an information intrusion paradigm that builds on these basic assumptions of the competing frameworks.  In each trial one or two pieces of  information showed up for free—they were not actively purchased by the participant and no information costs were attached to them. These information intrusions happened as soon as the first cue value information was intentionally acquired by the participant. (For details, please refer to the paper.)

The  results of the experiments show that information search and choice behavior followed SPMs’ predictions—the strategy irrelevant information intrusions were not ignored as their content influenced the decision makers’ behavior in the direction predicted by the SPMs. To summarize, the results were in line with the SPMs’ assumption that any applicable information will be fed into an assumed single decision making mechanism. From the view of evidence accumulation models, the results can be interpreted as follows: In the learning phase the evidence threshold is lowered until any discriminating cue reliably causes the overstepping of the threshold. In the test phase, TTB incompatible information intrusions automatically will be fed into the mechanism and can therefore cause an undershooting of the threshold.  Furthermore, in these cases they will not blindly follow TTB’s decision rule, but integrate the searched (and intruded) information in their decision.

For the connectionist model SPMs, decision makers can learn to calibrate their information search to the environment in the learning phase. Glöckner and Betsch  further propose that “deliberate constructions” help to form and adjust the network. As this process is not fully specified, it can only be assumed that incompatible information intrusions might lead to such a low level of consistency (achieved after the automatic maximization process) that deliberate information search is initiated. When feeding incompatible information into the network, the activation of the TTB option decreases and the alternatives’ activation increases. Thus, the
difference in activation is reduced, leading to lower confidence judgments when choosing the TTB-option. If the activation of an alternative option exceeds the activation of the TTB-option, this alternative will be chosen and confidence judgments will mirror the absolute difference in activation between the TTB-option and the chosen alternative.

Both frameworks are metaphors that try to describe human behavior, and as you learn more, the metaphors look worse. Sometimes, competing metaphors are so flexible that they are able to account for any empirical finding and it becomes therefore impossible to empirically distinguish between them.  According to Sollner et al, the more elegant explanation for the results presented here is given, however, by assuming a single uniform mechanism for decision making.

Söllner, A., Bröder, A., Glöckner, A., & Betsch, T.(2014). “Single-process versus multiple-strategy models of decision making: Evidence from an information intrusion paradigm.” Acta Psychologica 146 (2014) 84–96.

One thought on “Swiss Army Knife or Adaptive Tool Box?

  1. Pingback: Revisiting Swiss Army Knife or Adaptive Tool Box - Judgment and Decision Making

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.