Modeling the effects of argument length and validity on inductive and deductive reasoning.

Abstract

In an effort to assess models of inductive reasoning and deductive reasoning, the authors, in 3 experiments, examined the effects of argument length and logical validity on evaluation of arguments. In Experiments 1a and 1b, participants were given either induction or deduction instructions for a common set of stimuli. Two distinct effects were observed: Induction judgments were more affected by argument length, and deduction judgments were more affected by validity. In Experiment 2, fluency was manipulated by displaying the materials in a low-contrast font, leading to increased sensitivity to logical validity. Several variants of 1-process and 2-process models of reasoning were assessed against the results. A 1-process model that assumed the same scale of argument strength underlies induction and deduction was not successful. A 2-process model that assumed separate, continuous informational dimensions of apparent deductive validity and associative strength gave the more successful account.

Assessing the belief bias effect with ROCs: It's a response bias effect.

“…In terms of the signal detection account illustrated in Figure 3B, the analytic system may act to determine the position of arguments along the axis of argument strength, while the heuristic system might use knowledge-based cues such as believability to determine the location of the response criterion. Furthermore, our data do not rule out alternative, more complex, signal detection interpretations of dual process theory (e.g., the multidimensional account in Heit & Rotello, 2008, in press;Rotello & Heit, 2009), but these are not required to account for the data from Experiments 1-3.…”

“…Second, in a series of simulations comparing A z and d a , Macmillan et al (2004) showed that A z is a relatively unbiased measure of accuracy and has a smaller standard error than d a . Our experiments provide the first ROC data in the syllogistic reasoning task, but ROCs from other reasoning tasks have been curvilinear and asymmetric (Heit & Rotello, 2005Rotello & Heit, 2009). Those data suggest two key points: (a) The use of H Ϫ F as a measure of reasoning performance in the belief bias task is probably inappropriate, and (b) application of dЈ would confound sensitivity with response bias.…”

“…In the areas of memory and perception, however, the form of the empirical ROC (both ratings and binary response) is most often found to be curvilinear (Creelman & Donaldson, 1968;Egan, Schulman, & Greenberg, 1959;Emmerich, 1968;Macmillan & Creelman, 2005;Swets, 1986aSwets, , 1986bTanner, Haller, & Atkinson, 1967; but see Bröder & Schütz, 2009), and the inductive reasoning experiments reported by Rotello (2005, 2008, in press) and Rotello and Heit (2009) consistently produced curved ROCs similar in form to those observed in other domains.…”

“…An advantage of the SDT approach we have adopted in the current study is that it offers a thorough account of the decision process involved in syllogistic reasoning. As pointed out by Rotello and Heit (2009), one potentially fruitful research strategy would be to complement future processing accounts of reasoning with SDT analyses of decision making, allowing an understanding of both the accumulation of evidence and the way in which decision processes act on the evidence once it is accumulated. In general, SDT analyses can be used to constrain the development of more detailed processing accounts.…”

“…We place priority on developing theoretical accounts of reasoning that span different reasoning tasks and paradigms, rather than just narrowly addressing syllogistic reason-ing. For instance, recent work by Rotello (2005, 2008, in press) and Rotello and Heit (2009) has applied signal detection models to inductive reasoning to examine the effects of related variables that differentially cue the application of real-world knowledge.…”