This post looks at the paper “Do people learn option or strategy routines in multi-attribute decisions? The answer depends on subtle factors” authored by Arndt Bröder, Andreas Glöckner, Tilmann Betsch, Daniela Link, and Florence Ettlin (Acta Psychologica 143 (2013) 200–209). The researchers note that in their classic book on Einstellung effects, Luchins and Luchins (1959) demonstrated the robustness of maladaptive routinization in problem solving strategies. A specific strategy that had been successful in several trials was still used after changes in the environment that rendered simpler solutions available. Routinization even prevented many participants from finding simple solutions to new problems in which the routinized strategy could not be used. Hence, routinization may be beneficial in a stable task environment, but it may become detrimental in a changing world. It has been demonstrated that even experts fall prey to Einstellung effects, although their magnitude reduces for top-experts.
A strategy is defined as a set of processes operating on available information to reach a decision (e.g., steps of information search, evaluation, stopping rule). Strategies can be applied to any given set of information, and hence, strategies are more abstract than options. For example, the take-the-best heuristic (TTB) for choice follows the search rule to inspect attributes (cues) in the order of their importance (validity) until one is found which differentiates between options. If the first cue differentiates, search is stopped, and a choice is made. Otherwise, more cues have to be searched. A strategy can typically be represented as an abstract set of operations, often including if–then rules. Given that both levels of routinization (option or strategy) have been demonstrated in the literature, the important question emerges how feedback in a choice experiment is spontaneously interpreted if both kinds of routinization are possible.
In the experiments, the subjects tried to pick the winners of card games with four card players in each game. Each card player was an option for the purposes of the experiments. The subjects were given the predictions of four so-called experts who also predicted the winner of each card game. To over simplify, if you picked a card player that was an option routine, and if you picked an expert or based the decision on experts, that was a strategy routine. Over 110 trials, routines were rewarded and then suddenly things were changed so that a new routine was needed.
In typical decision tasks, participants choose between options that are described by attributes or cues. Often, these attributes are in conflict (e.g., price vs. quality), and different strategies can be used to resolve this conflict, that is to map the multi-dimensional decision situation onto a one-dimensional judgment or choice. A predominant view in decision research assumes that people possess a repertoire of decision strategies from which they choose adaptively or contingent on task and environmental demands. For example, search costs or payoff structures of the environment, time pressure or memory retrieval costs have been shown to impact on people’s decisions, compatible with the view that they choose appropriate strategies How these strategies are selected, however, remains a matter of controversy. Glockner and Betsch question the predominant view, but that is not the issue here (Intuition in J/DM).
SSL (strategy selection learning) theory assumes that the learning process is driven by reinforcement, that is, the outcome information for each choice . The authors conjecture that one important moderator may be how the feedback is interpreted by the decision maker. For example, in a recent study Pachur and Olsson showed that different types of feedback in a multiple cue probability learning task could lead to different internal representations (and presumably learning processes) of decision-relevant knowledge.
One major tenet of cognitively oriented decision research over the last three decades emphasizes the constructive nature and adaptivity of decision making. People construct preferences and select strategies “on the fly” contingent on task demands and decision goals. One example of this constructive nature is the notorious framing effect: Depending on the option descriptions, preferences may reverse. The original explanation of Prospect Theory explains this by a different problem representation (gain or loss domain), and alternative models emphasize the role of attention that may be directed to different aspects of the decision problem. Due to the constructive nature of decision making, there has been a growing interest of research in underlying cognitive processes like memory and attention in recent years.
For example, Armel et al. varied the amount of attention given to decision options by manipulating their presentation time, and they found an accentuation of pre-existing preferences in the choice probabilities. Ashby et al. showed in eye movement analyses that buyers and sellers focused on different attributes of a product, thus producing the asymmetric pricing known as the “endowment effect” . Platzer and Bröder manipulated the visual salience of decision options’ attributes, thereby affecting their accessibility in memory. Salient attribute information had a higher likelihood to be considered in choices.
In the reported experiments, the researchers investigated whether people tend to learn options or strategies in repeated decision tasks. The success rates for a cue-based TTB strategy and one of several options were equated in a learning phase. Unbeknownst to participants, either the option or the TTB success rate dropped after 110 trials. Two conclusions can be reached from what happened. First, in all conditions both kinds of routines are developed by some of the participants which should be taken into account in theorizing, but the experimental conditions in all cases clearly favored one mode of routinization. Second, the question originally posed concerning the preponderance of one kind of routines cannot be answered in general.
Whether participants established the routine of a single-cue option or another cue-based strategy in the experiments depended on the graphical display of the feedback. The authors note that It has long been acknowledged in associative learning research that attentional processes play a central role. These models extended the well-known Rescorla–Wagner theory of associative learning essentially by making assumptions about the dynamic change in its cue salience parameters. The authors conjecture that a classical salience mechanism as postulated in the Rescorla–Wagner model would suffice to explain the results.
Viewed from a perspective of parallel constraint satisfaction networks, attention effects could be modeled on the level of individual cues. Rather than assuming that whole strategies are the units of selection, connection weights between options and cues are modified by feedback across learning trials in PCS. Hence, cues eventually gain different weights for determining decisions. Conceptually, it would be possible to make the modification of weights dependent on attention processes that focus on different cues in a certain experimental situation.
The authors conclude that their experiments showed the dramatic effect of a seemingly subtle detail demonstrated and that this should caution decision researchers to generalize results across different contexts. The experiments strike me as clever, but bring up in my mind the tangled bank of Egon Brunswik. I need to work harder to understand this better.