Peter Juslin, Uppsala University

2007-2012

The purpose of the project was to investigate judgment phenomena from the new research perspective offered by the metaphor of the naïve intuitive statistician (Fiedler & Juslin, 2006). From this perspective, it is assumed that people have the cognitive capacity to correctly describe their experience, but they are naïve with respect to the sampling constraints that may make the samples they experience biased and with regard to more sophisticated relations between sample properties and population properties. The aims of the project were to elucidate and reinterpret previous findings in terms of the metaphor of the naïve intuitive statistician and to test new models and hypotheses derived from the metaphor.

We attempted to address three concrete questions highlighted by the metaphor: First, how do people store knowledge of statistical distributions: are they to be likened to "parametric intuitive statisticians" who interpret their experience in terms of prior assumptions, for example, about the distribution shape? Second, what metacognitive capacities do people have to correct for the biases that arise in the samples they experience, for example, because of selective feedback? Third: to develop and test the naïve sampling model for overconfidence in the production of naïve confidence intervals, which defined an important "prototype" for implementing the metaphor of man as a naïve intuitive statistician.

Three Most Important Results 

The three most important results relate to the three concrete questions raised in the application. In a number of studies we have investigated how people store knowledge of distributions and how they use this knowledge to make point estimates for new objects from the distribution (Lindskog, 2012a, 2012b; Lindskog, Winman, & Juslin, 2012a, 2012b; Lindskog et al., 2012).

These studies suggest that people do not spontaneously induce abstract knowledge of distributions, except perhaps in regard to the central tendency, but rather reconstruct distribution properties post hoc by sampling examples from memory at the time when the property is later assessed (Lindskog et al., 2012a, 2012b). Because this sampling is constrained by working memory capacity and therefore involves small sample sizes, certain biases are introduced in peoples' conceptions about the distribution properties. For example, small samples tend to convey an illusion of a distribution shape that is normal regardless of the true distribution shape in the underlying population, something that affects both the point estimates and the ability to recognize distribution shapes. These results essentially involve a direct application of the naïve sampling model (Juslin, Winman, & Hansson, 2007) also to judgments of distribution shape and point estimates.

In regard to the metacognitive ability to correct for selective feedback probably the most interesting result is that people are not merely passively storing the external feedback they receive, but they are active and "reconstruct" and store the outcome also in those cases where no external feedback is available, based on their beliefs about the likely outcome, a sort of constructivist coding (Elwin, 2009; 2012; Elwin et al., 2007; Elwin, Henriksson & Juslin, 2010). In regard to developing and testing the naïve sampling model for intuitive confidence intervals the most remarkable aspect is how successful the model has been, both in regard to accounting for previous findings (Juslin et al., 2007) and in successfully predicting novel phenomena, including pertaining to the relationship between overconfidence and working memory capacity (Hansson, Juslin, & Winman, 2008) and the effects of aging on overconfidence (Hansson et al., 2008). Also the new research questions discussed next are important results.

New Research Questions

One of the more fruitful new research questions that have surfaced within the project concerns peoples' naïve expectations about how information should be integrated in judgment tasks. We have collected a substantial body of empirical evidence in support of the claim that people spontaneously tend to integrate information in a linear and additive way also in judgment tasks that require multiplication, and that this is an important explanation of a number of classic judgment phenomena, like the conjunction fallacy and base-rate neglect (Juslin, Nilsson, & Winman, 2009; Juslin et al., 2010, Nilsson et al., 2010). (These phenomena are traditionally explained by reference to subjective judgmental heuristics.) In this context, we have shown that linear additive integration can be surprisingly efficient when probabilities are estimated from noisy input probabilities (e.g., from small samples: Juslin et al., 2009) and developed ways to minimize the biases that arise from linear additive probability integration (Juslin et al., 2010).

Two most Important Publications

Among the articles that are already published, perhaps Juslin et al. (2007) and Juslin et al. (2009), both published in the Psychological Review, appear most important, because they provide general summaries of the research in the project. In Juslin et al. (2007) we introduce the metaphor of the naive intuitive statistician and discuss its application in regard to a number of phenomena. The naïve sampling model for intuitive confidence intervals is presented and its predictions are tested against published and new data. In Juslin et al. (2009), we review the literature and document that both the conjunction fallacy and base-rate neglect appear better explained by peoples' propensity for linear additive integration, than by their use of intensional judgmental heuristics (e.g., "representativeness"), which is the standard text book explanation of these phenomena in psychology. We moreover demonstrate that one reason why people rely on linear additive integration rather than the complex multiplicative rules of probability theory may be that when the input probabilities are approximate and noisy linear additive integration is more robust against noise and often provide at least as good estimates as the rules of probability theory. An agent that has to act on the basis of approximate knowledge of probabilities simply has very little possibility to detect the optimality of, or incentives to change to, the rules of probability theory.

Dissemination of Results from the Project

The research within the project has been published in some of the most prominent scientific journals in Psychology, including Psychological Review, Psychological Science, Cognition, Journal of Experimental Psychology: General, and Psychology and Aging. The results have moreover been reported on a regular basis at a number of the more important conferences in the field, including the conference connected to the Association for Psychological Science, Society for Judgment and Decision Making, European Association for Decision Making, and Cognitive Science Society. Finally, the research has been presented at a number of seminars and lectures within Sweden, including at the Department of Psychology at Stockholm University, at SOFI at Stockholm University, Department of Psychology at Umeå University, Department of Economy, Uppsala University, the University at Dalarna, and at the annual meeting of the Swedish Statistical Association in Stocklholm.