Non‐Bayesian Inference: Causal Structure Trumps Correlation

Bes, Bénédicte; Sloman, Steven A.; Lucas, Christopher G.; Raufaste, Éric

doi:10.1111/j.1551-6709.2012.01262.x

Cited by 29 publications

(48 citation statements)

References 34 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Further, Bayesian models of cognition are generally not used to detect non-normative behavior; in fact, except in the decision making literature and, in some cases, in the reasoning literature (e.g., Bes et al, 2012; Fernbach & Sloman, 2009), it is extremely rare that a Bayesian model of a cognitive phenomenon is published because the model does not account for the phenomenon. Moreover, even if a Bayesian model did not account for a cognitive phenomenon, it would be unclear whether this was due to unfounded assumptions (e.g., that learners prefer the most specific rules), or rather to genuine suboptimality.…”

Section: Discussionmentioning

confidence: 99%

“…Despite this growing literature, various authors have criticized Bayesian approaches on theoretical grounds (Altmann, 2010; Bowers & Davis, 2012; Fitelson, 1999; Jones & Love, 2011; Marcus, 2010; Sakamoto, Jones, & Love, 2008), and where Bayesian approaches have been explicitly compared to psychological models (e.g., in the case of causal inference), the non-Bayesian approaches typically explained the data better (e.g., Bes, Sloman, Lucas, & Raufaste, 2012; Fernbach & Sloman, 2009). Here, I add to this literature by taking a model in a domain that appears particularly suitable for Bayesian learning — rule induction, spell out its underlying assumptions as well as their predictions, and confront them with empirical data.…”

mentioning

confidence: 99%

See 1 more Smart Citation

Bayesian learning and the psychology of rule induction

Endress¹

2013

Cognition

View full text Add to dashboard Cite

In recent years, Bayesian learning models have been applied to an increasing variety of domains. While such models have been criticized on theoretical grounds, the underlying assumptions and predictions are rarely made concrete and tested experimentally. Here, I use Frank and Tenenbaum's (2011) Bayesian model of rule-learning as a case study to spell out the underlying assumptions, and to confront them with the empirical results Frank and Tenenbaum (2011) propose to simulate, as well as with novel experiments. While rule-learning is arguably well suited to rational Bayesian approaches, I show that their models are neither psychologically plausible nor ideal observer models. Further, I show that their central assumption is unfounded: humans do not always preferentially learn more specific rules, but, at least in some situations, those rules that happen to be more salient. Even when granting the unsupported assumptions, I show that all of the experiments modeled by Frank and Tenenbaum (2011) either contradict their models, or have a large number of more plausible interpretations. I provide an alternative account of the experimental data based on simple psychological mechanisms, and show that this account both describes the data better, and is easier to falsify. I conclude that, despite the recent surge in Bayesian models of cognitive phenomena, psychological phenomena are best understood by developing and testing psychological theories rather than models that can be fit to virtually any data.

show abstract

Section: Discussionmentioning

confidence: 99%

mentioning

confidence: 99%

Bayesian learning and the psychology of rule induction

Endress¹

2013

Cognition

View full text Add to dashboard Cite

show abstract

“…Such an approach has been successfully applied to the explanation of key phenomena in domains such as learning about causal systems (Kemp, Goodman, & Tenenbaum, 2010;Sloman & Lagnado, 2005;Waldmann et al, 2006), reasoning (Kemp & Tenenbaum, 2009) and conceptual development (Gopnik et al, 2004;Tenenbaum, Kemp, Griffiths, & Goodman, 2011). Krynski and Tenenbaum (2007) proposed that the causal Bayes net approach could also lead to a better understanding of how people make judgments under uncertainty (also see Bes, Sloman, Lucas, & Raufaste, 2012). Krynski and Tenenbaum suggest that failures to arrive at normative probability estimates in such judgments reflect a difficulty in mapping the statistics given in the problem to the relevant components of intuitive causal models.…”

Section: Introductionmentioning

confidence: 97%

The role of causal models in multiple judgments under uncertainty

et al. 2014

View full text Add to dashboard Cite

“…We propose that the persuasive effect of confirming covariation data will be larger when the target is compared to a referent with less causal features (hereafter, different referent) than when it is compared to a referent with the same causal features (hereafter, similar referent). Our expectation builds on current theorizing suggesting that preexisting structural and specific causal knowledge affects data interpretation (see, e.g., Bes, Sloman, Lucas, & Raufaste, ; Lagnado & Sloman, ; Sloman, ). Specifically, the structural schema “control of variables” we use (Jirout & Zimmerman, ; Zhou et al, ) leads us to expect that two categories that are similar in features will perform the same and two categories that differ in features will differ in performance (Boudreaux, Shaffer, Heron, & McDermott, ).…”

Section: Introductionmentioning

confidence: 72%

The effect of previous causal knowledge on the persuasive strength of confirming covariation data

Saporta-Sorozon

Bar‐Eli

2017

J of Consumer Behaviour

View full text Add to dashboard Cite

Consumers are often exposed to causal claims (e.g., a new pill that claims to treat acne) that are occasionally accompanied by data indicating that the product (target) performed better than another product (referent). In this study, we examined the effect of such data on persuasion as a function of target-referent similarity in causal features. Consistent with current theorizing suggesting that structural and specific preexisting causal knowledge affects data interpretation, we propose that data that are consistent with expectations will be more persuasive than data that are inconsistent with expectations. Specifically, we contend that the structural schema "control of variables" we use leads us to expect that two categories (products) that are similar in features will perform the same and two categories that differ in features will differ in performance. In addition, our specific knowledge on causal powers leads us to expect the target to perform better than the referent only if it has more causal features. Thus, when confronted with data indicating a target performed better than a referent with fewer causal features, the reasoner will find it easier to explain the data, and hence, the difference in performance will be perceived as larger, and the message will be more persuasive (e.g., belief in the causal claim and willingness to purchase the product) than when the target has the same causal features as the referent. The results of three studies revealed the expected pattern for different products, promising different effects in different communication contexts. | INTRODUCTIONIn marketing communication, consumers often are exposed to causal claims stating that consuming a certain product will produce a desired effect. For example, consumers may be exposed to an ad for a new pill (the cause) that claims to treat acne (the effect).Occasionally, such claims are accompanied by data regarding cause-effect covariation, which typically compare the target product's performance to that of another product (hereafter, the referent) that aims to provide the same benefit. For example, the acne pill ad may describe a study performed on 60 teenagers who suffer from acne and provide data in which 25 out of the 30 participants who took the pill (83%) and 10 out of the 30 participants who took the referent (33%) were cured. These data indicate that there is a covariation between the cause (the pill use) and the effect (acne cure), in which the target outperformed the referent in producing the effect. This is because on the delta-p statistics (the normative measure of the degree of covariation between dichotomous variables [Cheng, 1997;Perales & Shanks, 2007]), there is a difference of 50% between the two conditional probabilities: the probability of the effect occurrence in the cause's presence and in the cause's absence (83−33%). Importantly, for the same data, the referent may vary; it could be a referent that has the same or fewer causal features than the target-features people believe have the causal power to produ...

show abstract

Non‐Bayesian Inference: Causal Structure Trumps Correlation

Cited by 29 publications

References 34 publications

Bayesian learning and the psychology of rule induction

Bayesian learning and the psychology of rule induction

The role of causal models in multiple judgments under uncertainty

The effect of previous causal knowledge on the persuasive strength of confirming covariation data

Contact Info

Product

Resources

About