This Title All WIREs
How to cite this WIREs title:
WIREs Cogn Sci
Impact Factor: 3.175

Strategy selection: An introduction to the modeling challenge

Full article on Wiley Online Library:   HTML PDF

Can't access this content? Tell your librarian.

Modeling the mechanisms that determine how humans and other agents choose among different behavioral and cognitive processes—be they strategies, routines, actions, or operators—represents a paramount theoretical stumbling block across disciplines, ranging from the cognitive and decision sciences to economics, biology, and machine learning. By using the cognitive and decision sciences as a case study, we provide an introduction to what is also known as the strategy selection problem. First, we explain why many researchers assume humans and other animals to come equipped with a repertoire of behavioral and cognitive processes. Second, we expose three descriptive, predictive, and prescriptive challenges that are common to all disciplines which aim to model the choice among these processes. Third, we give an overview of different approaches to strategy selection. These include cost‐benefit, ecological, learning, memory, unified, connectionist, sequential sampling, and maximization approaches. We conclude by pointing to opportunities for future research and by stressing that the selection problem is far from being resolved. WIREs Cogn Sci 2014, 5:39–59. doi: 10.1002/wcs.1265 This article is categorized under: Psychology > Reasoning and Decision Making
Knowledge‐ and accessibility‐based strategies. Knowledge‐based strategies use attributes of alternatives as cues to make inferences about the alternatives. For instance, to infer which of two cell phone brands, Sumsan or Sinostar, is of better quality, a person could rely on an integration strategy. Following the tallying heuristic a person would consider one or more cues, such as the phones' price or their reliability. An alternative's value on a cue is coded as positive (e.g., high price, high reliability), as negative (e.g., low price, low reliability), or as unknown. For each phone, the person adds up the number of positive values and infers that the phone with the larger sum is better. Weighted‐additive integration strategies additionally assign weights to the various cues and compute the weighted sum of positive and negative cue values for each phone. Lexicographic heuristics, in turn, search through cues sequentially, basing a decision, for instance, on the first cue on which the two alternatives differ (e.g., where one phone has a negative value and the other a positive). To illustrate this, take‐the‐best considers cues in the order of how likely they are to help a person make accurate inferences. An example of an accessibility‐based strategy is the fluency heuristic. This heuristic simply opts for the alternative whose name (e.g., Sumsan) is perceived as having been more quickly retrieved. Another accessibility‐based strategy is the recognition heuristic, which chooses recognized brands over those that have never been heard of before.
[ Normal View | Magnified View ]
Parallel constraint satisfaction network (PCS) model. The PCS model assumes that alternatives and their attributes are represented as nodes in a simple connectionist network. These nodes are connected via inhibitory or excitatory links, which represent logical relations between the alternatives and their attributes. The strength of the relations between the nodes is captured by weights which can be estimated as free parameters. What is called the general validity node serves to activate the network. Adapted from Glöckner and Betsch. Copyright 2008 Society for judgement and decision making.
[ Normal View | Magnified View ]
Schematic representation of cognitive niches. P (Applicability) is the probability of a person being able to execute a given strategy when deciding between two alternatives. The fluency heuristic's applicability is the probability of a person perceiving a difference in retrieval times between the alternatives; the applicability of knowledge‐based strategies is the probability of a person retrieving the knowledge such a strategy (e.g., tallying) requires to make a decision. The applicability of the strategies is a function of a person's exposure to the alternatives in the environment. Specifically, alternatives that occur very frequently in the environment are likely to be encountered more often than those that occur less frequently. Encounters with an alternative are often also associated with the acquisition of knowledge about the alternative. This correlation between encountering an alternative and acquiring knowledge about it makes it likely that people can retrieve knowledge about alternatives they encounter more frequently than about those they encounter less often. Correspondingly, knowledge‐based strategies are most likely applicable when one or both alternatives occur very frequently in the environment. Roughly the reverse holds for the fluency heuristic, because retrieval times tend to be easier to tell apart when little knowledge is available which, in turn, is likely the case when one or both alternatives occur less frequently in the environment.
[ Normal View | Magnified View ]
When will the recognition heuristic help a person make accurate inferences? An unknown criterion (e.g., the likely future sales of brands) is reflected by an environmental mediator (e.g., the press). The mediator makes it more likely for a person to encounter alternatives (e.g., brand names) with larger criterion values than those with smaller ones (e.g., the press mentions popular brands more frequently). As a result, the person will be more likely to recognize alternatives with larger criterion values than those with smaller ones, and ultimately, recognition can be relied upon to accurately infer the criterion (e.g., to infer which brand will generate more sales). The relations between the criterion, the mediator, and recognition can be measured in terms of three environmental correlations, labeled surrogate, ecological, and recognition, respectively. Figure adapted from Goldstein and Gigerenzer. Copyright 2002.
[ Normal View | Magnified View ]
Factors in strategy selection. The adaptive‐decision‐maker framework has set out to explicate a number of factors that have been found to shape the selection of strategies. Besides the accuracy, effort, and time involved in using a strategy, these factors also include, for instance, so‐called reference point, framing, and task effects. To illustrate just one such effect, complex tasks, as characterized by a large number of alternatives to choose from, multiple attributes, or time pressure, will lead people to adopt simplifying lexicographic heuristics.
[ Normal View | Magnified View ]
Production systems. Production systems are computer‐based theories typically used to model some form of (artificial) intelligence. They can be composed of, for example, models of declarative memory or other components of cognition, and a series of production rules that interact with these modules. Productions are if–then statements; the if‐part of the rule states a condition that must be fulfilled in order for an action to be carried out, which is specified in the then‐part of the rule. For instance, when implemented in terms of production rules, the recognition and fluency heuristics (Figure ) can be thought of as matching the names of two alternatives, say two brands, Sumsan and Sinostar, against the contents of declarative memory.
[ Normal View | Magnified View ]
Testing models of strategies and strategy selection—two sides of the same medal. Much descriptive research in psychology, economics, and biology shares a methodological problem: without adequate strategy selection theories, it is difficult to test heuristics and other models of behavior in experiments.
[ Normal View | Magnified View ]
Overfitting. Schematic illustration of how two models fit past observations (circles) and how they predict new observations (triangles). As can be seen from the deviation between the lines and the data points, the complex Model 1 (dashed line) overfits the past observations. Model 1 takes into account many variables, xi, including potentially irrelevant ones. This model is not as accurate in predicting the new observations as the simple Model 2 (solid line), which only relies on one relevant variable. Figure adapted from Pitt and colleagues. Copyright 2002.
[ Normal View | Magnified View ]

Browse by Topic

Psychology > Reasoning and Decision Making

Access to this WIREs title is by subscription only.

Recommend to Your
Librarian Now!

The latest WIREs articles in your inbox

Sign Up for Article Alerts