THE INFLUENCE OF KNOWLEDGE ABOUT CAUSAL MECHANISMS ON COMPOUND PROCESSING

：芭芭拉

THE INFLUENCE OF KNOWLEDGE ABOUT CAUSAL MECHANISMS ON COMPOUND PROCESSING
Rocio Garcia-Retamero. The Psychological Record. Gambier: Spring 2007. Vol. 57, Iss. 2; pg. 295, 12 pgs

Abstract (Summary)
Empirical evidence has shown that several factors influence whether a compound is represented as several independent components or as a configuration. However, most of the previous research focused on data-driven factors (e.g., modality of the stimuli presented in the experimental task). In one experiment, I analyzed the influence of people's knowledge about causal mechanisms on compound processing in a causality judgment task. Specifically, via the experimental instructions participants' knowledge about the causal mechanism through which several potential causes could bring about an effect was manipulated. In the configural causal model condition, the potential causes acted through the same causal mechanism; in the elemental causal model condition, these causes acted through different causal mechanisms; in the neutral causal model condition, the causal mechanisms were not specified. Results showed that participants spontaneously and when they were induced a configural causal model via instructions processed compounds as configurations. In contrast, they processed compounds as several independent components when they were induced via instructions an elemental causal model. I interpret these results as being in line with the predictions of current learning models about compound processing.

：芭芭拉

Full Text (5173  words)

Copyright The Psychological Record Spring 2007

[Headnote]
Empirical evidence has shown that several factors influence whether a compound is represented as several independent components or as a configuration. However, most of the previous research focused on data-driven factors (e.g., modality of the stimuli presented in the experimental task). In one experiment, I analyzed the influence of people's knowledge about causal mechanisms on compound processing in a causality judgment task. Specifically, via the experimental instructions participants' knowledge about the causal mechanism through which several potential causes could bring about an effect was manipulated. In the configural causal model condition, the potential causes acted through the same causal mechanism; in the elemental causal model condition, these causes acted through different causal mechanisms; in the neutral causal model condition, the causal mechanisms were not specified. Results showed that participants spontaneously and when they were induced a configural causal model via instructions processed compounds as configurations. In contrast, they processed compounds as several independent components when they were induced via instructions an elemental causal model. I interpret these results as being in line with the predictions of current learning models about compound processing.


The causal texture of the environment is rather complex. In fact, a single cause is often perceived as contributing toward producing an effect and yet as being insufficient to produce it on its own (Novick & Cheng, 2004). For example, low resistance to infection in the absence of a flu virus is not by itself sufficient to cause one to have the flu; neither, typically, is the presence of a flu virus per se. These two factors together, however, often do cause one to come down - with the flu.

Many researchers have become increasingly interested in studying how compound stimuli composed of several components are processed and represented. Interestingly, research on compound processing has generally focused on conditioning in animals (e.g., Deisig, Lachnit, Giu**, & Hellstern, 2001; Kehoe & Graham, 1988; Pearce & George, 2002; see Pearce, 2002 for a review). In humans, this issue has been studied in predictive learning (Shanks, Charles, Darby, & Azmi, 1998; Williams & Braker, 1999; Young, Wasserman, Johnson, & Jones, 2000), categorization (Kruschke, 1992), multiple cue probabilistic learning (Edgell & Roe, 1995; Kruschke & Johansen, 1999), decision mak-ing (GarciaRetamero & Hoffrage, 2007; Garcia-Retamero, Hoffrage, Dieckmann, & Ramos, in press), and Pavlovian conditioning (Lachnit & Kimmel, 1993; Lachnit, Kinder, & Reinhard, 2002). Psychobiological models of learning and memory very often refer to this issue as well (e.g., Fanselow, 1999; Rudy & Sutherland, 1995; Sutherland & Rudy, 1-9-8-9). However, only a few studies have focused on compound processing when the components are several candidate causes of an effect (see Novick & Cheng, 2004).

Associative learning is perhaps the field in which the most controversial debate about the nature of the representation of compounds has taken place (see Pearce, 2002; Wagner, 2003, for reviews). In general, associative learning models characterize causal learning as reflecting the development of associative connections between the candidate causes and the effect. When it comes to compound processing, we can differentiate between the elemental and the configurai theoretical approach. In the former, a compound is treated as a merely divisible entity composed of different components (e.g., Mackintosh, 1975; McLaren & Mackintosh, 2000, 2002; Pearce & Hall, 1980; Rescorla & Wagner, 1972; Wagner & Brandon, 2001). At the other extreme, the components are assumed to interact with each other to create a new stimulus: a configuration (e.g., Kruschke 1992; Kruschke & Johansen, 1999; Pearce, 1994).

A common starting point for the elemental associative models is the well-known theory of Rescorla and Wagner (1972). This model is elemental because it assumes that a compound is decomposed such that each of its components acquires a different connection to the effect (Wagner & Rescorla, 1972; Rescorla, 1972). The strength of this associative connection is modified according to the following equation:

Δ V^sub A^ = α . β . (λ - Σ^sup n^^sub k=l^ V^sub AK^ (1)

where the change in the weight of the connection linking cause A to the effect e (i.e., ΔV^sub A^) in a given trial is a function of the discrepancy between λ and the sum of the weight of all the causes, A to k, presented on that trial (i.e., Σ^sup V^^sub Ak^). The λ parameter is the maximum level of activation that the effect e could achieve. The discrepancy between λ and Σ^sup V^^sub Ak^ is multiplied by α and β, which represent the associability of A and e, respectively.

Empirical evidence also shows that a configuration itself can be directly associated with the effect (Young et al., 2000; see Pearce, 2002 for a review). One formal configurai theory that currently holds promise is Pearce's (1994) model (see also Pearce, 2002). This theory assumes that the strength of the connection between a configuration, AB, and an effect, e, is proportional to the discrepancy between λ and the weight of that configuration, V^sub AB^, (see Equation 1). The weight of the configuration is computed according to the following equation:

V^sub AB^ = E^sub AB^ + ∑ S^sub AB,i^. E^sub i^ (2)

where E^sub AB^ is the associative strength of the configuration AB; E^sub i^ is the associative strength of the i similar configurations that the individual has learned about in the environment; and S^sub AB,i^ is the similarity, judged by the proportion of shared components, between the configuration AB and these / configurations. Therefore, Equation 2 assumes that any configuration can influence the effect directly by its own associative strength (E^sub AB^). At the same time, other familiar configurations (e.g., AC) may exert an indirect influence on the effect to the extent that they are similar to AB.

Both the elemental and the configurai theoretical approach have received partial empirical support. The general conclusion is that people process compounds elementally in certain environments and configurally in others. For example, many authors have suggested that the modality of the stimuli presented in the experimental task may determine whether a compound evokes an elemental or a configurai representation. Specifically, stimuli from different modalities seem to favor an elemental representation whereas stimuli from the same modality seem to favor a configurai representation (see Kehoe, Home, Home, & Macrae, 1994; Myers, Vogel, Shin, & Wagner, 2001; Pearce & George, 2002; Wagner, 2003). Another factor of influence is the specific demands of the task such as previous experience with problems that can be solved either elementally or configurally (Williams & Braker, 1999; Williams, Sagness, &McPhee, 1994).

Note that most of the previous research that analyzed the factors that influence whether a compound is represented elementally or configurally mainly focused on data-driven factors. Only a few studies considered theory-driven factors such as participants' previous beliefs or intuitions. For instance, in a series of three decision-mak-ing experiments, Garcia-Retamero and Hoffrage (2006) showed that compound processing is strongly influenced by participants' knowledge about causal mechanism (see also Garcia-Retamero, Hoffrage, & Dieckmann, in press; Hoffrage, Garcia-Retamero, & Czienskowski, 2005). In the experimental task, participants had to make cue-based inferences on which of two alternatives had a higher criterion value. That is, they had to predict which of two patients had a higher body temperature. To make such decisions, participants could use information about three cues (i.e., whether the patients had ingested any of three different substances; see Bröder, 2000; Newell & Shanks, 2003 for a similar experimental procedure). In two experiments, two cues were amalgamated into a highly predictive compound by applying the XOR the AND logical rule, respectively.1 In the third experiment, there was no highly predictive compound.

In these experiments, participants' causal mental models were manipulated through instructions. In the configurai causal model, cues acted through the same causal mechanism. In the elemental causal model, they acted through different causal mechanisms. In the neutral causal model, the causal mechanism was not specified. In these experiments, knowledge about causal mechanisms described what could have mediated or intervened between the cause and the effect, thereby enabling the cause to bring about the effect (Ahn, Kalish, Medin, & Gelman, 1995; Bullock, Gelman, & Baillargeon, 1982; Koslowski & Okagaki, 1986). Interestingly, a high percentage of participants processed the highly predictive compound as a configuration when they had a configurai causal model.

However, to the best of my knowledge, there is no research that has analyzed how much influence people's knowledge about causal mechanisms has when they make a judgment about the extent to which a compound integrating several potential causes brings about an effect. This is the main aim of this experiment.

：芭芭拉

☆Experiment

Participants received a causality judgment task in which information about four potential causes and one effect was shown in a summary format. A simultaneous presentation procedure such as this has been used frequently in recent studies on causal processing (e.g., Buehner, Cheng, & Clifford, 2003; Wu & Cheng, 1999). Specifically, participants received a sample of 40 hypothetical clinical patients who were symbolized by pictures of faces on the computer screen. They were all presented simultaneously. These patients could have ingested four different substances (the candidate causes, namely, A, B, C, and D) and they could suffer a disease (the effect).

In the experiment, B appeared either alone or in compound with A, C, or D (i.e., B, AB, CB, DB). Each of these stimulus presentations was equally likely. When B appeared alone, the effect was shown with a likelihood of .2. When it appeared in compound, this likelihood was .8. After analyzing the sample of patients, participants were asked to estimate to what extent each of the stimulus presentations (i.e., B, AB, CB, and DB) caused the effect. Participants were also asked about a new stimulus combination: the compound ACD.

Participants' knowledge about the mechanisms through which the potential causes acted was manipulated between groups via the experimental instructions. Particularly, I differentiated a configural, an elemental, and a neutral causal model. In the configurai causal model condition, the instructions emphasized that all the potential causes acted through the same causal mechanism in bringing about the effect. These instructions would then induce participants to process the compounds AB, CB, and DB as configurations. In the elemental causal model condition, the instructions emphasized that A, C, and D acted through the same causal mechanism to bring about an effect, but B acted through a different causal mechanism. Consequently, these instructions would induce participants to process the compounds AB, CB, and DB as several independent components. In the third condition, the neutral causal model, participants did not receive any information about the possible causal mechanisms through which the potential causes acted. Therefore, this condition would show how participants process compounds spontaneously.

I hypothesized that if participants process compounds as several independent components, as the Rescorla and Wagner model predicts, they would learn that A, C, and D are reliable causes of the effect, but B is not. Participants' causality judgments of the new compound, ACD, would then be the sum of their judgments of its components. Consequently, participants' causality judgment of ACD would be higher than those of AB, CB, or DB. In contrast, if compounds are processed as configurations, as the Pearce model predicts, participants would learn that AB, CB, and DB are reliable causes of the effect, but B is not. Participants' causality judgments of ACD would then be a function of the similarity between this compound and the other familiar stimuli. Note that the compound ACD has one component out of three in common with AB, CB, and DB. Therefore, participants' causality judgments of ACD would be equal to those of AB, CB, or DB.

Bearing this in mind, I hypothesized that participants would process a compound as several independent components in the elemental causal model condition. In this condition, participants' causality judgments of ACD would then be higher than those of AB, CB, or DB. In contrast, participants would process a compound as a configuration in the configurai causal model condition. Therefore, in this condition, participants' causality judgments of ACD and AB, CB, and DB would be similar. Previous research shows that people spontaneously process compounds as configurations when they have no knowledge about causal mechanism (see Shanks, Charles, et al., 1998; Shanks, Darby, & Charles, 1998; Williams & Braker, 1999). In line with this previous research, I hypothesized that participants would also process a compound as a configuration in the neutral causal model condition. Therefore, participants' causality judgments of ACD and AB, CB, and DB would also be similar in this condition.

Method

Participants

Ninety-six undergraduate students took part in the experiment. Sixty-four were female and 32 were male. They had a mean age of 24 years (range 18-30). They were randomly assigned to one of three equally sized groups based on the type of causal mental model that was induced. Participants received course credits for their participation in the experiment.

Stimuli

The experiment was programmed using Visual Basic 6.0. A causality judgment task was used. Specifically, participants were presented with a sample of 40 clinical patients, represented through pictures of faces displayed in five rows of eight pictures each. These pictures were randomly assigned to the different positions on the computer screen. The patients could have ingested four different substances, called "Red," "Blue," "Green," or "Yellow." When a patient had ingested one of these substances a square of its color was presented below the picture. Moreover, the patients could suffer from "Rifastan" syndrome. If so, a picture of an oval face would be shown. Otherwise, a round face would be shown. Participants were told that the substances act through a hormonal, a blood, or a neural mechanism. The assignment of the substances to these different causal mechanisms was counterbalanced across participants. Which substance appeared alone (B) was also counterbalanced across participants.

Participants' knowledge about the mechanisms through which the causes acted was manipulated between groups with three levels, namely, the configurai, the elemental, and the neutral causal model. In the configurai causal model condition, participants were told: "A crucial point is the mechanism of action of these substances. Please consider carefully how the substances act: They act via the same mechanism, that is, they act via (either the hormonal system, the bloodstream, or the nervous system)." In the elemental causal model condition, participants were told: "A crucial point is the mechanism of action of these substances. Please consider carefully how the substances act: The Yellow substance acts via the hormonal system, and the Red, Blue, and Green substances act via the bloodstream (with color and the causal mechanism counterbalanced across participants)." Finally, in the neutral causal model condition, participants were not given information about causal mechanism. Additionally, participants in all conditions were told that the test for "Rifastan" syndrome was conducted at a point in time when the substances, if taken, would have already shown an effect.

Procedure

At the start of the experiment, participants were given full written instructions on the computer screen, in which they were informed about the experimental procedure. They were then presented a sample of 40 patients that they had to analyze at their own pace. After that, participants clicked a button to advance. For each of the potential (combination of) causes, a rating scale was shown on the computer screen. Participants had to estimate the extent to which they thought the corresponding substance(s) caused/prevented the disease. Specifically, they were asked, "Bearing in mind the sample of clinical cases with which you were presented, to what extent is ingesting substance(s) X the cause of Rifastan syndrome?" where X was B, AB, CB, DB, or ACD. Near to the rating scale a square of the color(s) of the corresponding substance(s) was shown. The order in which the rating scales for the different stimuli presentations were presented was counterbalanced across participants. The rating scales were scroll bars that ranged from -100 to 100. A positive rating meant that the substance caused the disease, whereas a negative rating meant that the substance prevented the disease. Zero meant that the substance had no effect on the disease. Participants were able to move the scroll, which initially was at zero, using the mouse. They could modify their causality judgments until they pressed "accept."

Results

Figure 1 displays the average causality judgments for B, AB, CB, DB, and ACD in the elemental, neutral, and configurai causal model conditions. An analysis of variance (ANOVA) with factors Causal Mental Model × Stimuli showed a significant interaction between these two factors, F(B, 372) = 5.78, p < .001. In line with the experimental hypotheses, post hoc analyses using Tukey's HSD (honest significant difference) test showed that, in the elemental causal model condition, participants' causality judgments of the compound ACD were higher than those of AB, CB, and DB, p = .032, p = .044, and p = .03, respectively. Furthermore, these judgments did not differ in the neutral causal model condition, p > .90. Finally, in the configurai causal model condition, participants' causality judgments of the compound ACD were lower than those of AB, CB, or DB, p = .03, p = .048, and p = .046, respectively.

Discussion

In sum, results in the experiment show that participants' knowledge about causal mechanism influenced whether they processed compounds as configurations or as several independent components. Specifically, in line with the predictions of the elemental model of Rescorla and Wagner (1972), participants assigned higher causality judgments to ACD than to AB, CB, or DB when they were primed with an elemental causal model. That is, they processed the compound ACD as several independent components when these components were said to act through different causal mechanisms in bringing about an effect. However, in line with the predictions of Pearce's (1994) configurai model, participants assigned similar causality judgments to ACD as to AB, CB, and DB when they were in the neutral causal model condition. Consequently, participants processed compounds as a configuration spontaneously, that is, when the experimental instructions did not suggest a possible underlying causal mechanism.

Finally, when instructions suggested that the component causes acted through a common causal mechanism in bringing about the effect, thereby prompting a configurai causal model, participants' causality judgments of ACD were lower than those of AB, CB, or DB. This result is partially in disagreement with the experimental hypothesis because a similar causality judgment in this condition was predicted. However, it is in line with findings of previous research that has challenged Pearce's (1994) configurai model (e.g., Kinder & Lachnit, 2003; Shanks, Darby, & Charles, 1998; Williams & Braker, 1999). Specifically, Pearce's model has been criticized for not allowing for flexibility in the degree of generalization between configurations. That is, the model assumes that the generalization between configurations is a function of the degree of similarity between them. Similarity is defined as a the proportion of shared components. Yet the actual amount of generalization between configurations has been shown to be lower than the predictions of the model, both in predictive learning (Shanks, Darby, & Charles, 1998; Williams & Braker, 1999) and in Pavlovian conditioning (Kinder & Lachnit, 2003). This also holds for the results of the experiment I presented above.

The human capacity for assessing relationships between events has long been a topic of interest in the area of judgment research (see Alloy & Tabachnik, 1984, for a review). One interesting effect consistent with my results is that participants' judgments are strongly influenced by their prior knowledge (Alba & Hasher, 1983). For instance, in categorization judgments, Wattenmaker, Dewey, Murphy, and Medin (1986) found that providing participants with a hint that encouraged the additive integration of features greatly facilitated learning of linearly separable categories compared to nonlinearly separable categories. Note that the optimal response in a nonlinearly separable category does not depend on individual features but rather on the relation between them. Therefore, to be able to deal with such a category individuals have to be sensitive to the configurai property. When, in contrast, a hint induced encoding in a manner compatible with nonlinearly separable categories, then these categories were easier to learn than linearly separable ones (see also Waldmann, Holyoak, & Fratianne, 1995). Results in my experiment are also in line with the research in causal attribution by Ahn and Bailenson (1996). Specifically, these authors presented participants with event scenarios and provided different explanations for these events. They showed that when two events had the same underlying mechanism, participants preferred conjunctive explanations. In contrast, they preferred independent explanations when the events conflicted at the mechanism level, or even when only information about covariation between them was given without mak-ing clear that there was an underlying causal mechanism. These findings demonstrate that people are able to map their previous causal knowledge to a learning task in a way that influences compound processing.

In this study, information about the potential causes and the effect was presented using a summary format. Previous studies focusing on causal processing also used a procedure in which information was presented simultaneously (e.g., Buehner et al., 2003; Wu & Cheng, 1999). From an associative learning prospective, it could be assumed that participants in the task randomly sampled and explored the provided information and updated the associate weights of the potential causes accordingly (see an example of this in Kruschke & Johansen, 1999). A question requiring future research, however, is whether the results reported here can be generalized to a procedure in which information is presented in a trial-by-trial format. In fact, previous research shows that procedural details exert a crucial influence on participants' responses (Allan, 1993; Matute, Arcediano, & Miller, 1996; Yamauchi, Love, & Markman, 2002). Another open question is whether similar results could be found using stimuli of different modalities. As I mentioned above, stimuli from different modalities seem to favor an elemental representation whereas stimuli from the same modality seem to favor a configurai representation (see Kehoe et al., 1994; Myers et al., 2001; Pearce & George, 2002; Wagner, 2003). The present findings seem to be a promising starting point for such future research.

In sum, this work tried to build a bridge between compound processing and other areas of cognitive psychology, especially work on causal processing. I have explored the impact of knowledge about causal mechanisms on a very specific problem, namely, the representation of compounds either as several components or as a configuration. I hope there will be future attempts to elucidate more factors that could have a crucial influence on compound processing.

[Footnote]
1In the AND experiment, an alternative for which both cues forming the compound were present had a higher criterion value than an alternative for which only one or neither cue was present. In the XOR experiment, an alternative for which one, and only one, of the cues that formed the compound was present had a higher criterion value than an alternative for which either both or neither cue was present.

：芭芭拉

[Reference]
References
AHN, W., & BAlLENSON, J. (1996). Causal attribution as a search for underlying mechanisms: An explanation of the conjunction fallacy and the discounting principle. Cognitive Psychology, 31, 82-123.
AHN, W., KALISH, C. W., MEDIN, D. L, & GELMAN, S. A. (1995). The role of covariation versus mechanism information in causal attribution. Cognition, 54, 299-352.
ALBA, J. W., & HASHER, L. (1983). Is memory schematic? Psychological Bulletin, 93, 203-231.
ALLAN, L. G. (1993). Human contingency judgments: Rule based or associative?. Psychological Bulletin, 114, 435-448.
ALLOY, L. B., & TABACHNIK, N. (1984). Assessment of covariation by humans and animals: The joint influence of prior expectations and current situational information. Psychological Review, 91, 112-149.
BR&Ouml;DER, A. (2000). Assessing the empirical validity of the "Take-The-Best" heuristic as a model of human probabilistic inference. Journal of Experimental Psychology: Learning, Memory, and Cognition, 26, 1332-1346.
BUEHNER, M. J., CHENG, P. W., & CLIFFORD, D. (2003). From covariation to causation: A test of the assumption of causal power. Journal of Experimental Psychology: Learning, Memory, and Cognition, 29, 1119-1140.
BULLOCK, M., GELMAN, R., & BAILLARGEON, R. (1982). The development of causal reasoning. In W. J. Friedman (Ed.), The development psychology of time (pp. 209-255). New York: Academic Press.
DEISIG, N., LACHNIT, H., GIU**, M., & HELLSTERN, F. (2001). Configurai olfactory learning in honeybees: Negative and positive patterning discrimination. Learning and Memory, 8, 70-78.
EDGELL, S. E., & ROE, R. M. (1995). Dimensional information facilitates the utilization of configurai information: A test of the Castellan-Edgell and the Gluck-Bower models. Journal of Experimental Psychology: Learning, Memory and Cognition, 21, 1495-1508.
FANSELOW, M. S. (1999). Learning theory and neuropsychology: Configuring their disparate elements in the hippocampus. Journal of Experimental Psychology: Animal Behaviour Processes, 25, 275-283.
GARCIA-RETAMERO, R., & HOFFRAGE, U. (2006). How causal knowledge simplifies decision mak-ing. Minds & Machines, 16, 365-380.
GARCIA-RETAMERO, R., HOFFRAGE, U., & DIECKMANN, A. (in press). When one cue is not enough: Combining fast and frugal heuristics with compound cue processing. The Quarterly Journal of Experimenal Psychology.
GARCIA-RETAMERO, R., HOFFRAGE, U., DIECKMANN, A., & RAMOS, M. (in press). Compound cue processing whthin the fast and frugal heuristics approach in non-linearly separable environments. Learning & Motivation.
HOFFRAGE, U., GARCIA-RETAMERO, R., & CZIENSKOWSKI, U. (2005). The robustness of the Take The Best Configurai heuristic in linearly and nonlinearly separable environments. In B. G. Bara, L. Barsalou, & M. Bucciarelli (Eds.), Proceedings of the XXVII Annual Conference of the Cognitive Science Society (pp. 971-976). Mahwah, NJ: Lawrence Erlbaum Associates.
KEHOE, E. J., & GRAHAM, P. (1988). Summation and configuration: Stimulus compounding and negative patterning in the rabbit Journal of Experimental Psychology: Animal Behavior Processes, 14, 320-333.
KEHOE, E. J., HORNE, A. J., HORNE, P. S., & MACRAE, M. (1994). Summation and configuration between and within sensory modalities in classical conditioning of the rabbit. Animal Learning & Behavior, 22, 19-26.
KINDER, A., & LACHNIT, H. (2003). Similarity and discrimination in human Pavlovian conditioning. Psychophysiology, 40, 226-234.
KOSLOWSKI, B., & OKAGAKI, L. (1986). Non-Humean indices of causation in problem-solving situations: Causal mechanism, analogous effects, and the status of rival alternative accounts. Child Development, 576, 1100-1108.
KRUSCHKE, J. K. (1992). ALCOVE: An exemplar-based connectionist model of category learning. Psychological Review, 99, 22-44.
KRUSCHKE, J. K., & JOHANSEN, M. (1999). A model of probabilistic category learning. Journal of Experimental Psychology: Learning, Memory and Cognition, 25, 1083-1119.
LACHNIT, H., & KIMMEL, H. D. (1993). Positive and negative patterning in human classical skin conductance response conditioning. Animal Learning & Behavior, 21,314-326.
LACHNIT, H., KINDER, A., & REINHARD, G. (2002). Are rules applied in Pavlovian electrodermal conditioning with humans general or outcome specific? Psychophysiology, 39, 380-387.
MACKINTOSH, N. J. (1975). A theory of attention: Variations in the associability of stimuli with reinforcement. Psychological Review, 82, 276-298.
MATUTE, H., ARCEDIANO, F., & MILLER, R. R. (1996). Test question modulates cue competition between causes and between effects. Journal of Experimental Psychology: Learning, Memory and Cognition, 22, 182-196.
MCLAREN, I. P. L., & MACKINTOSH, N. J. (2000). An elemental model of associative learning: I. Latent inhibition and perceptual learning. Animal Learning & Behavior, 28, 211-246.
MCLAREN, I. P. L1 & MACKINTOSH, N. J. (2002). Associative learning and elemental representation: II. Generalization and discrimination. Animal Learning & Behavior, 30, 177-200.
MYERS, K. M., VOGEL1 E. H., SHIN, J., & WAGNER, A. R. (2001). A comparison of the Rescorla-Wagner and Pearce models in a negative patterning and a summation problem. Animal Learning and Behavior, 29, 36-45.
NEWELL, B. R., & SHANKS, D. R. (2003). Take The Best or look at the rest? Factors influencing "one-reason" decision mak-ing. Journal of Experimental Psychology: Learning, Memory, and Cognition, 29, 53-65.
NOVICK, L. R., & CHENG, P. W. (2004). Assessing interactive causal influence. Psychological Review, 111, 455-485.
PEARCE, J. M. (1994). Similarity and discrimination: A selective review and a connectionist model. Psychological Review, 101, 587-607.
PEARCE, J. M. (2002). Evaluation and development of a connectionist theory of configurai learning. Animal Learning & Behavior, 30, 73-95.
PEARCE, J. M., & GEORGE, D. N. (2002). The effects of using stimuli from three different dimensions on autoshaping with a complex negative patterning discrimination. The Quarterly Journal of Experimental Psychology, 55B, 349-364.
PEARCE, J. M., & HALL, G. (1980). A model for Pavlovian learning: Variations in the effectiveness of conditioned but not unconditioned stimuli. Psychological Review, 87, 532-552.
RESCORLA, R. A. (1972). "Configurai" conditioning in discrete-trial bar pressing. Journal of Comparative and Physiological Psychology, 79, 307-317.
RESCORLA, R. A., & WAGNER, A. R. (1972). A theory of Pavlovian conditioning: Variations in the effectiveness of reinforcement and nonreinforcement. In A. H. Black & W. F. Prokasy (Eds.), Classical conditioning II: Current theory and research (pp. 64-99). New York: Appleton-Century-Crofts.
RUDY, J. W., & SUTHERLAND, R. J. (1995). Configurai association theory and the hippocampal formation: An appraisal and reconfiguration. Hippocampus, 5, 375-389.
SHANKS, D. R., CHARLES, D., DARBY, R., & AZMI, A. (1998). Configurai processes in human associative learning. Journal of Experimental Psychology: Learning, Memory and Cognition, 24, 1353-1378.
SHANKS, D. R., DARBY, R. J., & CHARLES, D. (1998). Resistance to interference in human associative learning: Evidence of configurai processing. Journal of Experimental Psychology: Animal Behavior Processes, 24, 136-150.
SUTHERLAND, R. J., & RUDY, J. W. (1-9-8-9). Configurai association theory: The role of the hippocampal formation in learning, memory, and amnesia. Psychobiology, 17, 129-144.
WAGNER, A. R. (2003). Context-sensitive elemental theory. The Quarterly Journal of Experimental Psychology, 56B, 7-29.
WAGNER, A. R., & BRANDON, S. E. (2001). A componential theory of Pavlovian conditioning. In R. R. Mower & S. B. Klein (Eds.), Handbook of contemporary learning theories (pp. 23-64). Mahwah, NJ: Lawrence Erlbaum Associates.
WAGNER, A. R., & RESCORLA, R. A. (1972). Inhibition in Pavlovian conditioning: Application of a theory. In R. A. Boakes & M. S. Halliday (Eds.), Inhibition and learning (pp. 301-336). London: Academic Press.
WALDMANN, M. R., HOLYOAK, K. J., & FRATIANNE, A. (1995). Causal models and the acquisition of category structure. Journal of Experimental Psychology: General, 124, 181-206.
WATTENMAKER, W. D., DEWEY, G. I., MURPHY, T. D., & MEDIN, D. M. (1986). Linear separability and concept learning: Context, relational properties, and concept naturalness. Cognitive Psychology, 18, 158-194.
WILLIAMS, D. A., & BRAKER, D. S. (1999). Influence of past experience on the coding of compound stimuli. Journal of Experimental Psychology: Animal Behavior Processes, 25, 461-474.
WILLIAMS, D. A., SAGNESS, K. E., & MCPHEE, J. E. (1994). Configurai and elemental strategies in predictive learning. Journal of Experimental Psychology: Learning, Memory, and Cognition, 21, 129-142.
WU, M., & CHENG, P. W. (1999). Why causation need not follow from statistical association: Boundary conditions for the evaluation of generative and preventive causal power. Psychological Science, 10, 92-97.
YAMAUCHI, T., LOVE, B. C., & MARKMAN1 A. B. (2002). Learning nonlinearly separable categories by inference and classification. Journal of Experimental Psychology: Learning, Memory and Cognition, 28, 585-593.
YOUNG, M. E., WASSERMAN, E. A., JOHNSON, J. L, & JONES, F. L (2000). Positive and negative patterning in human causal learning. The Quarterly Journal of Experimental Psychology, 53B, 121-138.

[Author Affiliation]
ROCIO GARCIA-RETAMERO
Max Planck Institute for Human Development, Berlin

[Author Affiliation]
I thank Ulrich Hoffrage and AnJa Dieckmann for their helpful discussion of the results and Anita Todd for editing the manuscript.
Correspondence concerning this article should be addressed to Rocio Garcia-Retamero, Departmento de Psicologia Experimental y Fisiologia del Compartamiento, Facultad de Psicologia, Universidad de Granada, Campus Universitario de la Caztuja s/n, 18071 Granada (Spain). (E-mail: rretamer@ugr.es).