Content uploaded by Helena Matute
Author content
All content in this area was uploaded by Helena Matute
Content may be subject to copyright.
A preview of the PDF is not available
Causal and predictive-value judgments, but not predictions, are based on cue-outcome contingency
Abstract and Figures
In three experiments, we show that people respond differently when they make predictions as opposed to when they are asked to estimate the causal or the predictive value of cues: Their response to each of those three questions is based on different sets of information. More specifically, we show that prediction judgments depend on the probability of the outcome given the cue, whereas causal and predictive-value judgments depend on the cue-outcome contingency. Although these results might seem problematic for most associative models in their present form, they can be explained by explicitly assuming the existence of postacquisition processes that modulate participants' responses in a flexible way.
Figures - uploaded by Helena Matute
Author content
All figure content in this area was uploaded by Helena Matute
Content may be subject to copyright.
... Most causal learning experiments exploit a basic principle of causality: causes and effects (outcomes) correlate with each other, unless a third factor masks this relationship. Since causality cannot be directly observed (Hume, 1748), people use this simple principle and rely on a proxy measure, the contingency between the cause and the outcome, to estimate causality (Allan, 1980;Wasserman et al., 1996;Vadillo et al., 2005;Blanco et al., 2010). In a simple situation with only one binary cause and one binary outcome, the contingency can be computed by means of the p index (Allan, 1980). ...
... Across the three experiments, we found that the departure from contingency was detected in effectiveness judgments, formulated as a causal question, but not in the conditional probability judgments. This is in line with recent studies on the causal illusion (Chow et al., 2019) and coincides with previous claims that, generally, causal estimations are more prone to bias than are other types of judgments, such as predictions (Vadillo et al., 2005). This also has theoretical implications: some authors have proposed that biases in causal learning are the result of processes that appear in the moment of emitting the judgment, rather than in the encoding phase (Allan et al., 2008). ...
Causal illusions have been postulated as cognitive mediators of pseudoscientific beliefs, which, in turn, might lead to the use of pseudomedicines. However, while the laboratory tasks aimed to explore causal illusions typically present participants with information regarding the consequences of administering a fictitious treatment versus not administering any treatment, real-life decisions frequently involve choosing between several alternative treatments. In order to mimic these realistic conditions, participants in two experiments received information regarding the rate of recovery when each of two different fictitious remedies were administered. The fictitious remedy that was more frequently administered was given higher effectiveness ratings than the low-frequency one, independent of the absence or presence of information about the spontaneous recovery rate. Crucially, we also introduced a novel dependent variable that involved imagining new occasions in which the ailment was present and asking participants to decide which treatment they would opt for. The inclusion of information about the base rate of recovery significantly influenced participants’ choices. These results imply that the mere prevalence of popular treatments might make them seem particularly effective. It also suggests that effectiveness ratings should be interpreted with caution as they might not accurately reflect real treatment choices. Materials and datasets are available at the Open Science Framework [https://osf.io/fctjs/].
... Most causal learning experiments exploit a basic principle of causality: causes and effects (outcomes) correlate with each other, unless a third factor masks this relationship. Since causality cannot be directly observed (Hume, 1748), people use this simple principle and rely on a proxy measure, the contingency between the cause and the outcome, to estimate causality (Allan, 1980;Wasserman et al., 1996;Vadillo et al., 2005;Blanco et al., 2010). In a simple situation with only one binary cause and one binary outcome, the contingency can be computed by means of the p index (Allan, 1980). ...
... Across the three experiments, we found that the departure from contingency was detected in effectiveness judgments, formulated as a causal question, but not in the conditional probability judgments. This is in line with recent studies on the causal illusion (Chow et al., 2019) and coincides with previous claims that, generally, causal estimations are more prone to bias than are other types of judgments, such as predictions (Vadillo et al., 2005). This also has theoretical implications: some authors have proposed that biases in causal learning are the result of processes that appear in the moment of emitting the judgment, rather than in the encoding phase (Allan et al., 2008). ...
Patients' beliefs about the effectiveness of their treatments are key to the success of any intervention. However, since these beliefs are usually formed by sequentially accumulating evidence in the form of the covariation between the treatment use and the symptoms, it is not always easy to detect when a treatment is actually working. In Experiments 1 and 2, we presented participants with a contingency learning task in which a fictitious treatment was actually effective to reduce the symptoms of fictitious patients. However, the base-rate of the symptoms was manipulated so that, for half of participants, the symptoms were very frequent before the treatment, whereas for the rest of participants, the symptoms were less frequently observed. Although the treatment was equally effective in all cases according to the objective contingency between the treatment and healings, the participants' beliefs on the effectiveness of the treatment were influenced by the base-rate of the symptoms, so that those who observed frequent symptoms before the treatment tended to produce lower judgments of effectiveness. Experiment 3 showed that participants were probably basing their judgments on an estimate of effectiveness relative to the symptom base-rate, rather than on contingency in absolute terms. Data and materials are publicly available at the Open Science Framework: https://osf.io/emzbj/
... This measure was included to determine whether average predictions made at test would still show reliable illusory causation effect-that is, participants predict greater performance for students given Kalavatic teaching than those given ordinary teaching despite overall base rate of performance being identical for cue-present and cueabsent trials-and whether differences in prediction are greater for High-OD relative to Low-OD participants, indicative of an outcome density effect. Based on findings from Chow et al. (2019), outcome density effects are most reliably produced in causal judgements, with little evidence of the effect in prediction judgements (see also Vadillo, Miller, & Matute, 2005 on the dissociation between causal and prediction judgements). Thus, we did not anticipate an interaction between cue type and outcome condition (indicative of an outcome density effect) on average prediction ratings, however, we expected participants to show reliable illusory causation by predicting greater performance with Kalavatic teaching than ordinary teaching. ...
... Some researchers have argued that this disassociation between prediction ratings and causal ratings indicates the involvement of a different system or psychological process when participants are making predictions rather than causal judgements (Allan, Siegel, & Tangen, 2005;Waldmann, 2001). However, given that this is a single dissociation where the effect is reliable on one measure and inconsistent on the other, it may simply come down to the sensitivity of each individual measure to subtle biases (De Houwer, Vandorpe, & Beckers, 2007;Vadillo et al., 2005;Vadillo, Musca, Blanco, & Matute, 2011). Ultimately, the extent to which aggregate predictions or causal ratings predict actual classroom behaviour is an empirical question requiring further investigation. ...
Teachers sometimes believe in the efficacy of instructional practices that have little empirical support. These beliefs have proven difficult to efface despite strong challenges to their evidentiary basis. Teachers typically develop causal beliefs about the efficacy of instructional practices by inferring their effect on students' academic performance. Here, we evaluate whether causal inferences about instructional practices are susceptible to an outcome density effect using a contingency learning task. In a series of six experiments, participants were ostensibly presented with students' assessment outcomes, some of whom had supposedly received teaching via a novel technique and some of whom supposedly received ordinary instruction. The distributions of the assessment outcomes was manipulated to either have frequent positive outcomes (high outcome density condition) or infrequent positive outcomes (low outcome density condition). For both continuous and categorical assessment outcomes, participants in the high outcome density condition rated the novel instructional technique as effective, despite the fact that it either had no effect or had a negative effect on outcomes, while the participants in the low outcome density condition did not. These results suggest that when base rates of performance are high, participants may be particularly susceptible to drawing inaccurate inferences about the efficacy of instructional practices.
... associative strength). This allows causal models to make a distinction between the different indexes people use to describe how events are related (Vadillo, Miller & Matute, 2005). A possible solution would be to integrate the distinction between learning and performance used in causal models into associative learning models. ...
... A possible solution would be to integrate the distinction between learning and performance used in causal models into associative learning models. Vadillo, Miller and Matute (2005) suggested that this could be achieved by assuming that a cue → outcome association is not directly mapped onto a response. Instead, an individual might use information from associations in the process of organising a response; therefore, when people are trying to derive a causal relationship then, and only then, do they change their attention. ...
Over the last forty years, experimental support for different models of associative learning has come from a range of phenomena. Support for the Rescorla-Wagner (1972) model comes from blocking and overshadowing experiments; however, this model is unable to explain the findings of latent inhibition experiments. The Mackintosh (1975) model, on the other hand, is able to accommodate the findings from blocking, overshadowing and latent inhibition experiments, as well as discrimination learning, relative validity, learned irrelevance, intra-/extra-dimensional shift (IDS/EDS) and learned predictiveness experiments. The model proposed by Pearce and Hall (1980) is also able to explain the findings of blocking, overshadowing and latent inhibition experiments, but in addition to this it is also able to accommodate the effects of partial reinforcement and negative transfer. In an attempt to unify the theories into a single model that is able to explain all the aforementioned phenomena, Le Pelley (2004) proposed a hybrid model of associative learning, but it was not easily able to incorporate the effects of learned value. Alternatively, Esber and Haselgrove (2011) proposed a model that reconciles the influence of predictiveness and uncertainty into a single mechanism for attentional allocation, and this model was better able to explain the experimental findings of learned value. Theories of associative learning claim that a cue’s predictive validity determines the amount of attention it attracts and to what extent it is subsequently learned about (e.g. Mackintosh, 1975; Pearce & Hall, 1980). In Chapter 2, using eye-tracking methodology during a learned predictiveness task, several measures of overt attention were recorded and compared on trials where the predictive contingency was certain or less certain. Findings revealed that, at a within-trial level, good predictors of an outcome attracted more attention compared to irrelevant cues. Although, at a between-trial level, uncertain trials attracted more attention compared to certain trials. These findings provide support for the conflicting attentional modulation predictions made by the Mackintosh (1975) and Pearce-Hall (1980) models. Consequently, these findings can only be fully explained by appealing to a model of associative learning that incorporates both the principles of predictiveness and uncertainty (e.g. Le Pelley, 2004; Esber & Haselgrove, 2011). Prior to eye-tracking becoming more widely available as a measure of overt visual attention, stimulus associability was used as an indirect measure of attention since it is assumed that the speed at which a stimulus is learned about reflects the amount of attention it attracts. This is demonstrated in the IDS/EDS task which consistently finds that IDS are easier than EDS because in the IDS condition the higher associability of the predictive dimension in Stage 1 facilitates learning when generalised into Stage 2. Until now, eye gaze during an IDS/EDS task has not been investigated to determine whether the effect results from a shift in overt attention from Stage 1 into Stage 2. Chapter 3 revealed that participants acquired an attentional bias towards predictive cues in Stage 1 which transferred into Stage 2; however, in the EDS condition this bias was maintained only very briefly. Eye-tracking during learned predictiveness tasks using adult participants has revealed that cues which are good predictors of an outcome attract more overt visual attention than cues which are irrelevant. However, thus far, little research has investigated whether good predictors of reinforcement and non-reinforcement show a comparable effect. Moreover, it is currently unclear whether children and non-human animals demonstrate the learned predictiveness effect. Chapter 4 employed the same design and stimuli to examine eye gaze towards cues during a simple learned predictiveness task (AX+, AY+, BX-, BY-) in adults, children and an orangutan. Results revealed that all participants demonstrated the learned predictiveness effect, directing more attention towards cues that were good predictors of the outcome compared with cues that were irrelevant. However, for adult humans this effect was only present on reinforced trials and questionnaire data suggested they had only learned about one of the predictive contingencies. Contemporary discussions of associative learning have emphasised the importance of a cue’s predictive relevance in determining learned variations in attention. However, most theoretical accounts of the effect do not capture the notion of prediction – only associative strength, or relative associative strength (e.g. Mackintosh, 1975). In Chapter 5, letters were established as congruent or incongruent cues of other letters presented simultaneously or serially with a target cue. Results revealed no difference in the amount of attention directed towards congruent and incongruent cues if stimuli were presented simultaneously or serially when participants were required to respond to the identity of the target cue. However, an attentional bias towards congruent cues compared to incongruent cues was found when cues were presented serially, if participants were permitted to predict the identity of the target before its onset.
... On each trial, just after showing whether the patient took the treatment or not, the participants answered whether they thought the patient would overcome the crisis or not, by giving a yes/no response. This question is frequently included in causal illusion experiments because it helps participants to stay focused on the task (Vadillo et al., 2005). The occurrences of the outcome (i.e., the healings of the crises) did not depend on the participants' responses, but were pre-programmed to occur in random order. ...
Previous research suggests that people may develop stronger causal illusions when the existence of a causal relationship is consistent with their prior beliefs. In the present study, we hypothesized that prior pseudoscientific beliefs will influence judgments about the effectiveness of both alternative medicine and scientific medicine. Participants ( N = 98) were exposed to an adaptation of the standard causal illusion task in which they had to judge whether two fictitious treatments, one described as conventional medicine and the other as alternative medicine, could heal the crises caused by two different syndromes. Since both treatments were completely ineffective, those believing that any of the two medicines worked were exhibiting a causal illusion. Participants also responded to the Pseudoscience Endorsement Scale (PES) and some questions about trust in alternative therapies that were taken from the Survey on the Social Perception of Science and Technology conducted by FECYT. The results replicated the causal illusion effect and extended them by revealing an interaction between the prior pseudoscientific beliefs and the scientific/pseudoscientific status of the fictitious treatment. Individuals reporting stronger pseudoscientific beliefs were more vulnerable to the illusion in both scenarios, whereas participants with low adherence to pseudoscientific beliefs seemed to be more resistant to the illusion in the alternative medicine scenario.
... Así, paulatinamente fueron incorporándose estudios que apostaban por una visión comprensiva del aprendizaje que incluyera aspectos cognitivos en la generación de las representaciones asociativas. Prueba de ello, por ejemplo, fueron las constataciones sobre la variación en los patrones de respuesta de los sujetos motivadas por la presentación de requerimientos instruccionales en formatos diferentes (Vadillo, Miller, & Matute, 2005), lo cual dio lugar a considerar de las personas para diferenciar los juicios causales de los predictivos. Así, mientras que los juicios causales son evaluados sobre la consideración simple de la fuerza asociativa de la causal candidata, los juicios predictivos están basados en la fuerza asociativa combinada de la causa candidata más el contexto. ...
El objeto de esta tesis es estudiar el papel de la contingencia de reforzamiento en los juicios de utilidad bajo condiciones de ambivalencia compleja. En la investigación experimental que se presenta se examina el proceso por el que un sujeto humano, concebido en principio como un agente racional, decide entre alternativas de juicio que reportan ciertos beneficios o expresan utilidades. El proceso de decisión que tiene lugar se estudia manipulando las relaciones de contingencia entre ciertos indicios de estímulo y la recompensa asociada en condiciones donde la utilidad esperada se mantiene constante. La contingencia se manipula en términos de a) densidad de reforzamiento, que establece la tasa de reforzamiento a lo largo de un proceso secuencial de ensayos discretos, de manera que el sujeto se enfrenta a situaciones de combinación variable entre estímulo y recompensa, y en términos de b) probabilidad asociativa entre una clave y los efectos que esa clave anticipa - diseños de tipo predictivo-, o entre una clave y los efectos consecuentes de ésta -diseños de aprendizaje causal.
La hipótesis fundamental de este investigación es que las variaciones producidas en la densidad de reforzamiento provocan cambios en la representación de la contingencia que se reflejan en modificaciones del patrón de respuestas conductuales del agente, expresado en forma de tasa de respuestas en la presión de teclas en el ordenador y sesgo de juicio respecto a los niveles de contingencia programados. Asimismo, una segunda hipótesis plantea la existencia de patrones de respuesta a la contingencia diferentes entre sujetos conservadores –sujetos que manteniendo controlado el valor esperado, prefieren la alternativa de mayor magnitud- y sujetos arriesgados – sujetos que manteniendo inalterado el valor esperado prefieren la alternativa de mayor probabilidad. Esta hipótesis se somete a prueba mediante dos tipos de diseño: a) una situación de elección discreta en emparejamientos de estímulos concurrentes (Tobler et al. 2007) y b) un procedimiento de flujo de ensayos sucesivos bajo condiciones de sensibilidad a la señal y establecimiento del criterio de decisión, basado en la aplicación de la Teoría de Detección de Señales a la evaluación de la contingencia (Siegel et al., 2009).
... Although we cannot ascertain causal direction from this correlational study, the results could mean that people acquire beliefs about cause-effect relationships by observing relationships between health behaviours and their supposed effects. This finding parallels contingency learning research in the laboratory that uses CAM-like cover stories, where participants use information obtained through trial-by-trial learning about the likelihood of recovery given a novel treatment and no treatment to judge the efficacy of the treatment in treating the disease (e.g., [8] and [27]). ...
Beliefs about cause and effect, including health beliefs, are thought to be related to the frequency of the target outcome (e.g., health recovery) occurring when the putative cause is present and when it is absent (treatment administered vs. no treatment); this is known as contingency learning. However, it is unclear whether unvalidated health beliefs, where there is no evidence of cause–effect contingency, are also influenced by the subjective perception of a meaningful contingency between events. In a survey, respondents were asked to judge a range of health beliefs and estimate the probability of the target outcome occurring with and without the putative cause present. Overall, we found evidence that causal beliefs are related to perceived cause–effect contingency. Interestingly, beliefs that were not predicted by perceived contingency were meaningfully related to scores on the paranormal belief scale. These findings suggest heterogeneity in pseudoscientific health beliefs and the need to tailor intervention strategies according to underlying causes.
... On a continuum going from hot emotions to cold cognition, one could view the associative process as being closer to the hot emotional pole, whereas the propositional process would be closer to the cold cognitive one. One could also argue that, rather than reflecting two processes, patterns such as the one reported here reflect a single source of knowledge the effect of which on behavior depends on how it is probed by the question asked of the participant (i.e., Vadillo, Miller, & Matute, 2005;Whittlesea & Price, 2001). This is a complex issue which, we think is not wholly empirical because it hinges on one's definition of an association. ...
Following cue-outcome (X-O) pairings, 2 procedures that reduce conditioned responses to X are extinction, in which X is presented by itself, and counterconditioning, in which X is paired with a different outcome typically of valence opposite that of training. Although studies with animals have generally found counterconditioning more efficient than extinction in reducing responding, data from humans are less clear. They suggest counterconditioning is more efficient than extinction at interfering with emotional processing, but there is little difference between the two procedures regarding their impact on the verbal assessment of the probability of the outcome given the cue. However, issues of statistical power leave conclusions ambiguous. We compared counterconditioning and extinction in highly powered experiments that exploited a novel procedure. A rapid streamed-trial procedure was used in which participants were asked to rate how likely a target outcome was to accompany a target cue after being exposed to acquisition trials followed by extinction, counterconditioning, or neither. In Experiments 1 and 2, evaluative conditioning was assessed by asking participants to rate the pleasantness of the cues after treatment. These studies found counterconditioning more efficient than extinction at reducing evaluative conditioning but less efficient at decreasing the assessment of the conditional probability of the outcome given the cue. The latter effect was replicated with neutral outcomes in Experiments 3 and 4, but the effect was inverted in Experiment 4 in conditions designed to preclude reinstatement of initial training by the question probing the conditional probability of the outcome given the cue. Effect sizes were small (Cohen's d of 0.2 for effect on evaluative conditioning, Cohen's d of 0.3 for effect on the outcome expectancy). If representative, this poses a serious constraint in terms of statistical power for further investigations of differential efficiency of extinction and counterconditioning in humans. (PsycInfo Database Record (c) 2020 APA, all rights reserved).
... As already mentioned, the hypothesis that the present thesis was aimed to test is that confirmation relations represent a viable explanatory account for apparently suboptimal probabilistic reasoning, both with verbal material (Tentori, Chater and Crupi, 2016) which is a confirmation measure itself. Most studies on contingency judgment do not adopt a completely abstract and blank experimental backstory, but they often involve some fictitious scenario (see Chapman & Robbins, 1990;Exp.2 in Hannah et al., 2009;Mandel & Lehman, 1998;and Vadillo, Miller & Matute, 2005). Because of this, it is possible to speculate that semantic background knowledge could still be involved. ...
For decades, works in psychology of thinking and decision making have been reporting suboptimal performance and systematic departures from the axioms of probability theory in people’s probability judgments. In these first works, poor performance was often attributed to people making normatively wrong intuitions because of their limited cognitive resources and lack of statistical skills. Over the last years, studies that considered various Bayesian models of inductive reasoning but also other high and lower-level cognitive processes provided a more optimistic picture by showing that, despite departing from the normative benchmark, people’s reasoning skills lead to adaptive and sound performance in everyday life. Different explanatory accounts for this suboptimal but sound reasoning have been proposed, some being more compelling than others. The present thesis is aimed at exploring one of these accounts that is based on confirmation relations and suggests that human inductive ability might rely more on estimating evidential impact than posterior probability. So far, this account has been applied to classical probabilistic reasoning errors, linguistic and psycholinguistic phenomena and probabilistic inferences with verbal stimuli. In this study, we tried to see whether the implicit estimation of confirmation relations can affect probability judgments also when the link between evidence and hypotheses is operationalized as the arbitrary association between visual features in briefly presented figures. First, we expected participants to consider confirmed hypotheses more probable than corresponding (in terms of posterior probability) disconfirmed ones; second, we expected them to choose the more likely option (i.e. the normatively correct one) more often when it was confirmed by the evidence provided than when it was disconfirmed. Four computer-based experiments were conducted using the same methodology. Experimental stimuli consisted of inductive arguments concerning 40 sets of figures composed of two features with two possible values each. By varying the probabilistic association between the two values of the features, sets were generated to have, for each possible combination of the two features, two arguments with the same posteriors and opposite impacts. In each trial, participants first looked at a set of figures. One of these figures was then randomly drawn. Participants were informed about the value of one feature of the drawn figure (e.g., that it was a “circle”) and had to guess the value of the other feature (“white” vs. “black”).
Throughout the four experiments, we used three different combinations of features: color and shape (exp.1: black/white; exp 2: light/dark grey), pattern and shape (exp 3) and type and orientation of line (exp 4).
In all four experiments, participants systematically chose the confirmed alternative over the equally probable, but disconfirmed one, and chose the normatively incorrect (i.e. less likely) alternative more often when it was confirmed (vs. disconfirmed) by the evidence provided. These results provided a first empirical evidence of the effect of confirmation relations on probability judgment with perceptual stimuli, but also highlighted a significant influence of the experimental material itself on choice patterns. In fact, in experiments 1 to 3 the obtained results showed that color (or pattern) was a more compelling evidence than shape in determining participants’ choices. The combination of line curvature and orientation used in experiment 4 proved to be the more balanced among those employed in the present research. Only in this last experiment, indeed, the type of evidence did not affect the choice for the confirmed alternative, nor the amount of errors. The results we found supported our experimental claims showing that confirmation relations can affect probability judgments even in absence of any semantic element, but also suggested the existence of a mutual influence between perceptual features and probability judgments. Our experimental results have theoretical as well as applied implications. On a theoretical level, they extend the results coming from works involving verbal and linguistic material to perceptual stimuli with no semantic background. Additionally, they show that high-level relations, which are completely unknown to the subject, affect the way people perceive relations within a visual set of perceptual items. This might have interesting and noteworthy implications for studies on visual cognition, and, on a broader level, contingency learning and stereotypical judgments.
A large amount of literature demonstrates that social behaviour can be triggered by environmental cues. A long-standing debate involves the question of whether such stimuli trigger behaviour directly (i.e. habits) or whether these effects mediate goals. As studies on automatic goal pursuit typically use real-world cues that are already associated with the behaviour and potentially the goal, it is impossible to make strong claims about the nature of the effects. In the present paper, we use a paradigm inspired by the Pavlovian-to-instrumental transfer (PIT) literature to examine how the environment can trigger goal-directed behaviour. Building on the essence of pro-self and pro-social motives in humans, two experiments explored the PIT effect when the outcomes were framed in terms of self- versus other-interest. Participants performed actions to earn money for themselves or a charity. Each outcome was linked to a different cue. The results showed that a cue predictive of self-interest outcomes facilitated responses instrumental in gaining the outcome, while such specific PIT effect for other-interest outcomes only emerged when participants were free to donate the money. We briefly discuss these findings reflecting on whether the PIT effect in our paradigm is indeed sensitive to the value of social goals.
Deviations from the predictions of covariational models of causal attribution have often been reported in the literature. These include a bias against using consensus information, a bias toward attributing effects to a person, and a tendency to make a variety of unpredicted conjunctive attributions. It is contended that these deviations, rather than representing irrational biases, could be due to (a) unspecified information over which causal inferences are computed and (b) the questionable normativeness of the models against which these deviations have been measured. A probabilistic extension of Kelley's analysis-of-variance analogy is proposed. An experiment was performed to assess the above biases and evaluate the proposed model against competing ones. The results indicate that the inference process is unbiased.
Associative and statistical theories of causal and predictive learning make opposite predictions for situations in which the most recent information contradicts the information provided by older trials (e.g., acquisition followed by extinction). Associative theories predict that people will rely on the most recent information to best adapt their behavior to the changing environment. Statistical theories predict that people will integrate what they have learned in the two phases. The results of this study showed one or the other effect as a function of response mode (trial by trial vs. global), type of question (contiguity, causality, or predictiveness), and postacquisition instructions. That is, participants are able to give either an integrative judgment, or a judgment that relies on recent information as a function of test demands. The authors concluded that any model must allow for flexible use of information once it has been acquired.
Ciiven the task of di the source of a patient's aUer^'ic reav-tion. college students jiuigcii the causal efficacy of common (A') and distinctive (A and Bj elements of compound stimuli: AX and BX. As the differential correlation of AX and BX with the occurrence and nonoccurrence ofthe allergic reaction rose from .00 to 1.00. ratings of ihe distinctive A and B elements diverged; most importantly, ratings ofthe common X element fell. These causal judgments of humans closely parallel the conditioned responses of animals in associa-tive learning studies, and clearly disclose that stimuli compete with one another for control over behavior.
This chapter describes the potential explanatory power of a specific response rule and its implications for models of acquisition. This response rule is called the “comparator hypothesis.” It was originally inspired by Rescorla's contingency theory. Rescorla noted that if the number and frequency of conditioned stimulus–unconditioned stimulus (CS–US) pairings are held constant, unsignaled presentations of the US during training attenuate conditioned responding. This observation complemented the long recognized fact that the delivery of nonreinforced presentations of the CS during training also attenuates conditioned responding. The symmetry of the two findings prompted Rescorla to propose that during training, subjects inferred both the probability of the US in the presence of the CS and the probability of the US in the absence of the CS and they then established a CS–US association based upon a comparison of these quantities. The comparator hypothesis is a qualitative response rule, which, in principle, can complement any model of acquisition.
In the first experiment subjects were presented with a number of sets of trials on each of which they could perform a particular action and observe the occurrence of an outcome in the context of a video game. The contingency between the action and outcome was varied across the different sets of trials. When required to judge the effectiveness of the action in controlling the outcome during a set of trials, subjects assigned positive ratings for a positive contingency and negative ratings for a negative contingency. Furthermore, the magnitude of the ratings was related systematically to the strength of the actual contingency. With a fixed probability of an outcome given the action, judgements of positive contingencies decreased as the likelihood that the outcome would occur without the action was raised. Correspondingly, the absolute value of ratings of negative contingencies was increased both by an increment in the probability of the outcome in the absence of the action and by a decrement in the probability of the outcome following the action. A systematic bias was observed, however, in that positive judgements were given under a non-contingent relationship when the outcome frequency was relatively high. However, this bias could be reduced by giving extended exposure to the non-contingent schedule (Experiment 2).
This pattern of contingency judgements can be explained if it is assumed that a process of selective attribution operates, whereby people are less likely to attribute an outcome to some potential target cause if another effective cause is present. Experiments 2 and 3 demonstrated the operation of this process by showing that initially establishing another agent as an effective cause of the outcome subsequently reduced or blocked the extent to which the subjects attributed the outcome to the action.
Finally, we argue that the pattern and bias in contingency judgements based upon interactions with a causal process can be explained in terms of contemporary conditioning models of associative learning.
This article reviews recent findings that violate a broad class of descriptive theories of decision making. A new study compared 1,224 participants tested via the Internet and 124 undergraduates tested in the laboratory. Both samples confirmed systematic violations of stochastic dominance and cumulative independence; new tests also found violations of coalescing. The Internet sample was older, more highly educated, more likely male, and also more demographically diverse than the lab sample. Internet participants were more likely than undergraduates to choose the gamble with higher expected value, but no one conformed exactly to expected value. Violations of stochastic dominance decreased as education increased, but violations of stochastic dominance and coalescing were still substantial in persons with doctoral degrees who had read a scientific work on decision making. In their implications, Internet research and lab findings agree: Descriptive decision theories cannot assume that identical consequences can be coalesced.
College students rated the causal efficacy of Elements X, A, and B of food compounds AX and BX in producing the allergic reaction of a hypothetical patient. The results of a 16-day allergy test were presented to subjects in a serial, trial-by-trial manner. The response format used was a running estimate, in which subjects were asked to rate all of the three foods after each of the 16 trials. Ratings of distinctive Elements A and B diverged and ratings of common Element X decreased as the difference in the correlation of AX and BX with the occurrence and nonoccurrence of the allergic reaction increased. These human causal judgments closely correspond with stimulus selection effects observed in the conditioned responses of animals in associative learning studies. The experiment also directly demonstrated the fact that significant changes in the causal ratings of a stimulus occur on trials in which the cue is not presented. Associative theories such as that of Rescorla and Wagner (1972) predict changes in associative strength only for those stimulus elements that are presented on a particular trial. A modification of the Rescorla-Wagner model is described that correctly predicts immediate changes in the associative strengths of all relevant cues on each trial—whether presented or not.
