Questions about the Recognition Heuristic

 

rh6002This is the second post based on Pohl’s paper, “On the Use of Recognition in Inferential Decision Making.”  Pohl looks at what have come to be seen as weaknesses of the recognition heuristic.

Recognition as a memory-based process
While acknowledging that recognition should generally be treated as a continuous variable, Goldstein and Gigerenzer  focused on the outcome of this recognition process, which is either “recognized” or “not recognized” with only a small and negligible gray zone of uncertainty in between. Accordingly, the quality of these subjective recognition judgments, that is, whether
they were true or not or with what confidence, was originally not considered. Meanwhile, some researchers have asked whether and how the recognition process itself possibly affects subsequent inferences.

Another question concerns what it actually means that an object is recognized. Recognition is no doubt helpful in many situations. For example, it helps when someone meets people on the street to know whom to greet (because they are recognized as neighbors) and whom not (because they are not recognized, suggesting that they are strangers). However, even in this simple situation, it is not recognition itself that is helpful, but rather the information associated with it. Maybe the recognized passerby is someone severely disliked or known for other reasons (because he or she is a famous actor or local politician). In these cases, recognition alone wouldn’t suffice to tell what to do. One needs to remember who these persons are, that is, one needs to retrieve further information about them from memory. In other words, it is the combination of recognition and further knowledge that drives behavior in many everyday situations.  This argument could exemplify why some researchers may feel uneasy that there should be cases in which one’s inferences are based on recognition alone. The same argument applies to the classical city-size task, in which cities are not only recognized, but are recognized for being a state’s capital, being located at the coast, being a tourist site, or hosting a big automobile company. All this knowledge is intertwined with recognition and is probably retrieved in an instant. If that were true, the postulated “search memory” and “stop searching memory” assumptions of the RH possibly need to be changed to inhibitory working-memory processes, trying to prevent any of the already retrieved information beyond recognition to enter the decision making process.  In sum, it  appears to be useful to look more closely into the memory processes that lead to the recognition (or rejection) of an object.

Recognition heuristic as a cognitive process model
Some of the heuristics in the adaptive toolbox proved quite successful in predicting people’s behavior. Yet, the next and in Pohl’s view highly important question is whether and how these heuristics can be translated into cognitive process models, describing how people actually proceed when making an inferential decision. The question then is of how to derive adequate predictions from the RH, for example, for reaction times (RT). Pohl thinks that the step-wise procedures described in the RH (as well as in other heuristics) should basically allow one to derive such predictions. In a similar vein, Pachur and Hertwig claimed that “recognition is first on the mental stage and ready to enter inferential processes when other probabilistic cues still await retrieval.” Using recognition should therefore be rather fast, while searching for further information will need additional time. Supporting evidence for a stepwise TTB process resulting in increasing reaction times the more cues had to be searched was provided by Bröder and Gaissmaier. Pachur and Hertwig  found that inferences in line with the RH were slower when additional inconsistent information was present. They also reported that under time pressure inferences more often followed the RH.

 Reasons for not using the RH
One argument that could be used to explain evidence that is contradictory to the RH is to assume that people decide in each case whether the RH would be the best strategy to apply. If not, they use some other strategy. Pachur and Hertwig  stated that “people appear to decide case by case whether they will obey the recognition heuristic. Moreover, these decisions are not made arbitrarily, but demonstrate some ability to discriminate between cases in which the recognition heuristic would have yielded correct judgments and cases in which the  recognition heuristic would have led astray.” They also assumed that the RH is typically chosen as the default strategy in recognition cases (i.e., whenever one object is recognized and the other not), but that it can be “suspended” for a number of reasons and thus not applied to the current case. The reasons for suspending the RH include (1) availability of probabilistic cues with
larger validities than the recognition validity; (2) source knowledge (i.e., knowing that an object is recognized for other reasons than its criterion value; e.g., Chernobyl is recognized by most people, but not because of its size, but because of the nuclear accident in 1986); and (3) conclusive criterion knowledge. In sum, the decision process is more complicated than previously assumed. Besides, proposing that, before the RH is applied, memory is searched in order to check whether anything better can be found or whether something speaks against using the RH, appears tantamount to saying that recognition is used as a cue whenever nothing better is available. But then the RH is no longer a shortcut, intentionally ignoring other, potentially useful information.

The RH as a non-compensatory strategy
Choosing the recognized object is not identical with basing one’s decision on recognition alone.
In natural environments, recognition and knowledge are most likely confounded, that is, the probability of recognition increases with the cities’ size and so does cue knowledge that speaks for the cities’ largeness. Pohl, for example, reported differences in adherence rates, depending on (a) whether participants merely recognized the object’s name or knew more about it, and
(b) whether the recognized object was actually the correct choice or not. When participants knew more about the recognized object or when it represented the correct choice, they chose it consistently and significantly more often.  Note that these observations could be based on two cases: (1) The recognized object was chosen for another reason than recognition alone; and (2) the unrecognized object was chosen despite non-recognition. The latter could possibly represent compensatory inferences, where the decision based on recognition was overturned by another cue (that spoke for the unrecognized or against the recognized object), thus suggesting that something different than the non-compensatory RH was used in these cases. But they could also result from yet some other non-compensatory mechanism not considering recognition at all.

Evidence for a Less-is-more effect (LIME)                                                                                              The LIME entails that the overall inferential accuracy of a person who recognizes only about half of the objects in a domain could be higher than that of a person who recognizes all objects. The assumed reason for this at first glance surprising effect is that a person with full recognition can never use the more valid recognition cue, because all objects are recognized and so recognition does not discriminate. Instead, this person has to rely on her (in this case) less valid knowledge. However, a person with fewer recognized objects can utilize the highly valid recognition cue more often and thus will be more often correct.  The empirical evidence for a LIME, however, remains scarce with the reported effects mostly being of minor size. Perhaps it is difficult to find real domains that exactly possess those conditions that theoretically foster a LIME.

The RH as part of the toolbox
In typical experiments using paired comparisons (e.g., with city names), participants answer a series of such comparisons and infer which of the two objects in each pair is the larger one. For example, in a set of 20 objects and with all possible pairwise combinations, participants work through some 190 trials. Given that not all objects are recognized or not all are unrecognized,
there will be different types of pairs, or “cases”, depending on how many of the objects in a pair are recognized: (1) Recognition cases, in which one object known and the other not. These cases represent the central ones for studying the RH. (2) Guessing cases, consisting of two unknown objects, such that persons have nothing left but to guess (or to infer probabilistic cues from the names of the objects, e.g., to which country a city might belong, thus allowing inferences about its size). Recognition is not helpful here, because none of the objects is recognized. (3) Knowledge cases, consisting of pairs in which both objects are known. Again, recognition is of no help since both are recognized. In this case, other knowledge has to be assessed to reach a decision.  All this makes clear that the studied paired comparisons are more complicated than each of the “fast and frugal” heuristics when viewed as a single strategy suggests.

Inherent in this description is also one of the main problems of the toolbox approach:
How does one know when to take which heuristic?  Goldstein and Gigerenzer  suggested that the knowledge which strategy should be applied in which situation might be either (a) genetically coded, (b) socially or culturally transmitted, or (c) learned individually. While these mechanisms seem plausible, they also remain somewhat vague yet, so that the strategy-selection problem is certainly an area in which more research is needed.

Pohl, Rudiger F, (2011). On the use of recognition in inferential decision making:  An overview of the debate. Judgment and Decision Making, Vol. 6, No. 5, July 2011, pp. 423–438.

Bröder, A., & Gaissmaier,W. (2007). Sequential processing of cues in memory-based multiattribute decisions. Psychonomic Bulletin & Review, 14, 895–900.

Goldstein, D. G., & Gigerenzer, G. (2002). Models of ecological rationality: The recognition heuristic. Psychological Review, 109, 75–90.

Pachur, T., & Hertwig, R. (2006). On the psychology of the recognition heuristic: Retrieval primacy as a key determinant of its use. Journal of Experimental Psychology: Learning, Memory, and Cognition, 32, 983–1002.

 

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.