Everyday conditional reasoning with working memory preload

See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/228778039 Everyday conditional reasoning with working memory preload Article CITATIONS READS 9 48 2 authors, including: Niki Verschueren University of Leuven 17 PUBLICATIONS 211 CITATIONS SEE PROFILE All content following this page was uploaded by Niki Verschueren on 22 July 2014. The user has requested enhancement of the downloaded file. All in-text references underlined in blue are linked to publications on ResearchGate, letting you access and read them immediately.  Everyday Conditional Reasoning with Working Memory Preload Niki Verschueren ([email protected]) Walter Schaeken ([email protected]) Gery.d’Ydewalle (Gery.d’[email protected]) University of Leuven, Lab of Experimental Psychology, Tiensestraat 102 3000 Leuven – Belgium Abstract Examples of everyday ‘if cause, then effect’ sentences are: If you phone someone, then his telephone rings. There are two accounts explaining how background If you eat salty food, then you will get thirsty. information can affect the conditional reasoning If someone has a high income, this person will be rich. performance: the probabilistic account and the mental If a dog has fleas, then it will scratch constantly. model account. According to the mental model theory reasoners retrieve and integrate counterexample Abundant research established that when people reason information to attain a conclusion. According to the on everyday conditionals, they spontaneously bring probabilistic account reasoners base their judgments on relevant background knowledge into account (for a likelihood information. It is assumed that reasoning by review see Politzer & Bourmaud, 2002). This use of a mental model process requires more working contextualization process is characteristic for common- memory resources than solving the inference by use sense reasoning and is responsible for our ability to of likelihood information. We report a thinking-aloud adaptively cope with everyday situations. The current experiment designed to compare the role of working study focuses on how background knowledge is used memory for the two reasoning mechanisms. It is found for deriving conditional inferences. that when working memory is preloaded participants use less counterexample information, instead they are more There are two reasoning mechanisms describing how inclined to accept the inference or to use likelihood background information is used during reasoning. First, information. The present results add to the growing according to the probabilistic account reasoners derive evidence showing that working memory is a crucial the probability that the conclusion follows given the determinant of reasoning strategy and performance. categorical premise and use this probability to draw a gradual conclusion (Lui, Lo, & Wu, 1996; Oaksford, Introduction Chater, & Larkin, 2002). For MP, reasoners will confine their knowledge base to the situations where There is evidence for a general link between working the cause occurs. Based on this range of situations they memory capacity and performance in a range of then determine the likelihood that the effect follows. If reasoning tasks (see e.g., Barrouillet, 1996; Gilhooly, they can induce that a particular effect always or Logie, & Wynn, 1999; Kyllonen & Christal, 1990). mostly follows the cause, they conclude that the effect Previous studies showed that skilled reasoners will (probably) follow. The endorsement of MP is thus generally give more normative answers and follow a directly proportional to L(effect|cause). AC is solved in high demand reasoning strategy (see e.g., Copeland & analogy with MP. Reasoners activate all relevant Radvansky, in press; Gilhooly, Logie, & Wynn, 1999). situations where the effect occurs. Within this subset It is assumed that these normative answers are obtained they infer the likelihood that the cause preceded the by an analytic reasoning mechanism that hinges on occurring effect. This likelihood L(cause|effect) working memory capacity (Klauer, Stegmaier, & directly reflects the AC acceptance rate. Meiser, 1997; Meiser, Klauer, & Naumer, 2001). The According to the second reasoning mechanism the present research continues this line of research and conclusion is attained by taking possible concerns causal conditional reasoning with everyday counterexamples into account. There is a strong and sentences. reliable effect of the number of available Without labeling conclusions as (in)valid, we will counterexamples on inference acceptance (see e.g., investigate how people solve the following two Cummins, Alksnis, Lubart, & Rist, 1991). The mental conditional inferences with everyday causal sentences: models theory describes how participants reason with Modus Ponens (MP) counterexample information (Johnson-Laird & Byrne, If cause, then effect 1991; Markovits & Barrouillet, 2002). When given a Cause occurs. Does the effect follow? problem based on a causal rule, for instance, ‘If you water a plant well, the plant stays green’, reasoners Affirmation of the Consequent (AC) will start by representing the content of the conditional If cause, then effect Effect occurs. as a possibility: It is possible that a plant is well Did the cause precede? watered and green. Active consideration of the problem content will then lead to an automatic activation of of relevant situations have to be represented. The larger relevant background information. This information is the number of mental models that participants have to used to complement the initial model. For MP and MT, represent and maintain, the heavier the load on working the categorical premise triggers the retrieval of memory during reasoning (Barrouillet & Lecas, 1999). disablers. Some examples of disablers are: ‘the plant Additionally, it has been found that counterexample caught a disease’ or ‘the plant was deprived of retrieval efficiency suffers from dual task loads, which sunlight’. When reasoners retrieve at least one disabler, indicates that working memory is also involved in the they do not conclude that the effect follows. For AC an retrieval of counterexample information (De Neys, automatic search for alternative causes starts, for 2003). In case the reasoners have a representation of example, ‘the lack of water was compensated by both the conditional sentence and at least one adding fertilizer’ or ‘the plant is a succulent’. When counterexample, they subsequently have to integrate reasoners retrieve an alternative cause, their mental this information to see that there are two different models inform them that there are two conclusions conclusions for the same problem. This information possible (watered and not watered). As a result, they do manipulation and integration is considered as a crucial not accept the default conclusion. task of working memory. It is clear that the probabilistic and the mental model For the reasoning process based on likelihood reasoning mechanisms both rely on available information, the demands on working memory are far background information, but they focus on a different less. The situations used for attaining a likelihood type of background knowledge: probabilities versus estimate are not actively represented in working exemplars. Both information types have already been memory, but rather briefly accessed. There is neither an brought together by, e.g., Weidenfeld & Oberauer active controlled search process nor a need for premise (2003); Verschueren, Schaeken and d’Ydewalle (2003; integration. The likelihood estimate is based on all 2004a) integrated the two theories that explain how the relevant situations at a time and the final conclusion information is taken into account in a dual process directly mirrors the obtained likelihood estimate. perspective. They label the probabilistic mechanism as When reasoners are asked to think aloud during heuristic and the mental model mechanism as analytic. reasoning, we can monitor which information they use Heuristic processes are generally considered as fast, for deriving conclusions. By concurrently checking the automatic mechanisms that operate at a low cognitive information that people use we get a direct indication cost and at the periphery of awareness. Analytic of the underlying reasoning process. Only in case processes are generally slower, more demanding where people do not provide extra information but reasoning mechanisms that operate in a conscious and accept the conclusion without further argumentation, strategic manner (Stanovich & West, 2000). this procedural aspect is unclear. It can be that Verschueren et al. (2004a) manifest three reasons for participants did use their background knowledge and linking the two reasoning processes to a heuristic- found that the likelihood that the conclusion follows is analytic polarity. (1) The heuristic reasoning process is sufficient to grant acceptance or that there are no mainly implicit - reasoners have no recollection of the counterexamples available. Or else it can be that they range of situations that are taken into account to did not rely on background information and just calculate a likelihood estimate whereas people satisfied the conclusion by restating the given reasoning by use of mental models are conscious of the information. counterexample(s) they retrieve. (2) The process based In a previous thinking-aloud study Verschueren, on likelihood information yields relatively fast results Schaeken and d’Ydewalle (2004b) showed that whereas using counterexamples requires a sequential participants with low working memory capacity more thus slower reasoning process. (3) The heuristic often use likelihood estimates to solve an inference, conclusion is overwritten when a more analytical whereas participants with a larger working memory conclusion can be produced (see Verschueren, et al., capacity rather use counterexample information. These 2004a for experimental evidence for 2 and 3). At results can be considered as an indication for the present we will investigate whether both reasoning difference in working memory demands of the heuristic mechanisms differ in their working memory demands. and analytic process. This setup provides however only If indeed the mental model account describes an correlational evidence. Indeed, it is still possible that a analytical reasoning mechanism it should pose more third factor (e.g., general intelligence, motivation, etc.) demands on working memory capacity than the explains both the performance on working memory heuristic likelihood process. tests as well as on reasoning tasks. The following experiment was designed to test whether there is a Experiment difference in the actual working demands of the two processes. It is assumed that reasoning with counterexample In this experiment we examined the effect of information draws heavily on working memory secondary task interference on the applied reasoning resources, whereas the use of mere likelihood estimates mechanisms. In the dual task methodology, a secondary imposes a far lesser demand on working memory. task chosen to burden working memory capacity has to When participants reason based on counterexample be carried out concurrently to the criterion task. The information, the problem content as well as all models degree of disruption in the criterion task under dual task conditions – as compared to single task conditions Did this person catch a cold or not? – is taken to reflect the dependence of the criterion task The participants read the premises aloud and answered on working memory. The criterion task we used was a immediately. When they found that they had completed thinking-aloud conditional reasoning task. Concurrent their answer, they pressed a key to go to the following verbalization allows us to monitor the information that problem. After the presentation of the reasoning reasoners consult for deriving conclusions. By instructions, the participants either reasoned with or checking the information that people refer to without working memory preload. In the preload (likelihood or counterexample information) we get a conditions participants started by practicing two dot direct indication of the underlying reasoning process. patterns: A pattern was presented for 500ms and Because the criterion task entails spontaneous participants were immediately asked to reproduce this verbalization, the choice of secondary tasks is limited. pattern. The overall performance on the test problems Pilot work revealed that concurrent motor, auditory or was nearly perfect. After these dot pattern practice articulatory activity interfered with the participants trials, participants were given instructions for reasoning verbalization. We therefore opted for a preload under preload. First, a dot pattern was presented for paradigm. Because a spatial load is less likely to 500ms, next the reasoning problem occurred, interfere with verbalization than a verbal or numerical participants read the premises aloud and answered load, we worked with a spatial preload set-up. The immediately. The answers participants gave were evidence that spatial storage tasks tap a working recorded on tape. When they finished their answer, they memory feature crucial for reasoning is twofold: pressed a key and a blue screen appeared where they Klauer, et al.(1997) report that a concurrent spatial load were asked to reproduce the dot pattern. When they led to a significant disruption of propositional completed the dot pattern, they pressed a key to start (including conditional) reasoning. Second, in the the next trial. It was explicitly mentioned that they had visuospatial domain simple storage tasks have a similar to memorize the dot patterns correctly; they were told correlation with executive functioning and reasoning as that an incorrect reproduction rendered the trial invalid. classic processing-and-storage tasks (Miyake, This was done to make sure that participants actively Friedman, Rettinger, Shah, & Hegarty, 2001; Suess, attended the dot pattern and tried their best in Oberauer, Wittman, Wilhelm, & Schultze, 2002). We memorizing it. In the control condition, the dot patterns can thus assume that the preload task taps working were presented for 500ms before the premise memory resources that are needed for reasoning, while presentation. Participants were told that these dot at the same time minimizing a possible interference patterns are presented as a control condition, they were with the verbalization process. The dot memory task asked to look at the dot patterns but not to memorize we used is a classic simple storage task (adapted from them. They read the premises and pressed a key when Miyake, et al., 2001; Oberauer, Suess, Wilhelm, & their answer was complete, the next trial started Wittman, 2003). We briefly presented a 3x3 matrix immediately. The time that participants needed to read with 4 dots forming a complex pattern, afterwards and solve the reasoning problem was measured. participants were asked to reproduce this dot pattern. In the preload-condition participants had to memorize Materials and Design Based on previous research we the pattern of the dots while solving a reasoning selected 12 sentences with a maximally varying problem. We will verify whether the use of necessity and sufficiency of the cause (maximal counterexample information decreases when working variation in L(effect|cause), L(cause|effect), and in the memory is preloaded, compared to performance in the number of available disablers and alternatives). We control condition. The decreement in the use of made sure that the reading time of all 12 sentences was likelihood information should be significantly smaller comparable (Mnumber of words = 9.5, SD = .314). Twenty- than the decreement in counterexample use. six participants solved 12 AC inferences; the others solved 12 MP problems. The 12 sentences occurred Method always in the same order; the causes of the first six sentences and the last six sentences were equally Participants A total of 52 first year psychology necessary and sufficient. For both reasoning forms, half students participated in the study. of the participants solved the first six problems under preload; the other six problems were solved without Procedure and Design The participants were tested preload (control condition). For the other half of the individually. The experiment was run on computer. participants the order of the preload/control conditions Participants started by reading the instructions. They was reversed. Because we used 12 different sentences, were told that they will be asked to think aloud while transfer effects between the two conditions could be solving conditional inference problems. The reasoning excluded. instructions read that they should answer the question as in an everyday setting. Each participant then solved Results two test problems, e.g., The obtained reasoning answers were literally transcribed. Next, the condition-codes were removed If someone catches a cold, then he will cough. and the answer types were rated. It was indicated Someone coughs. whether the answer reflected a simple acceptance of the default conclusion or whether there was reference to a Effect of preload on the reasoning process. For counterexample or to a likelihood estimate. There was examining the effect of preload on the types of no overall difference in the average response time for answers, we only included the preload trials where the the preload (18.19s) and the control condition (18.53s). dot pattern was correctly reproduced. All analyses were In the control condition, there was 26% inference run on proportions; the number of times each answer acceptance, in 22% of the trials participants used type occurred was divided by the total number of likelihood information and in 64% they referred to correctly reproduced trials. We ran an analysis of counterexamples. These results are similar to those variance with sentences as the unit of analysis, and a 2 observed by Verschueren et al. (2004a; 18%, 18% and (inference type, between subjects) * 2 (preload, within 66% respectively). subjects) * 3 (answer type, within subjects) design. We In the preload condition there were 6.4% combination found a main effect of answer type. There were more trials (in a ‘combination trial’ participants refer to answers referring to counterexample information counterexample and likelihood information) whereas in (60.1%) than there was plain inference acceptance the control condition there were 23.1% combination (27.7%) or likelihood information used (5.6%), F(2, trials. The observation that combining the two types of 21) = 102, 72, p < .001 (Wilks’lambda = .08). The information becomes less prevalent when working interaction between answer type and preload condition memory is preloaded, suggest that the information was marginally significant, F(2, 21) = 3.120 p = .065 integration process that is characteristic for (Wilks’lambda = .771). Figure 2 illustrates this combination answers taps on working memory interaction. There was a clear yet marginally significant resources. For comparing the relative importance of decrease in the use of counterexample information both reasoning processes, we confined the analysis to when working memory was preloaded, F(1, 22) = 3 trials where participants either referred to a likelihood 304, MSE = 0.078, p = .082. There were significantly or to counterexample information. Combination trials more inferences accepted in the preload condition, F(1, were excluded from the analysis (14.4%). 22) = 8.255, MSE = 0.131, p<.01 while there was no Task interference. Only 69% of the dot patterns were significant increase in the use of likelihood reproduced correctly. There was an effect of answer information. No other interaction effects reached type on the correct reproduction of the dot patterns, significance. The observation that there is more F(2, 21) = 6.696, p < .01 (Wilks’lambda = .611). This inference acceptance under preload corroborates interaction is displayed in Figure 1. When the dot previous effects of secondary task load on the patterns were correctly reproduced, there were fewer conditional reasoning performance (De Neys, 2003). counterexamples mentioned than when the dot patterns The explanation provided by De Neys (2003) is that were incorrectly reproduced, F(1, 22) = 11.96, MSE = under preload, the resources available to participants .458, p < .05. On the correctly reproduced trials, there are insufficient to retrieve counterexample information. were more answers where participants referred to The currently observed decrease in counterexample use likelihood information, F(1, 22) = 5.21, MSE = .037, p is in line with this explanation. The increase in < .05. There was no significant effect on the inference inference acceptance can also be - at least partially - acceptance rates. These results reflect a task related to an enhanced matching heuristic. We can interference. When participants rely on a reasoning assume that some reasoners do not engage in an active process that puts only a minor demand on working reasoning process based on counterexample retrieval, memory there are enough resources left to maintain and but simply restate the information from the conditional reproduce the dot pattern. In contrast, when participants and blindly accept MP and AC. In this case the rely on retrieval, manipulation and integration of preloading should cause more participants to accept all counterexample information, working memory capacity conclusions, even on sentences where counterexamples is severely burdened. There are then not enough resources left to actively maintain the dot patterns, can be automatically retrieved and likelihood resulting in an incorrect reproduction. These results estimations are high. In the preload condition, there support the idea that using counterexample information were indeed more participants (13.5%) who accepted at draws heavily on working memory resources. least 75% of the inferences than in the control condition (7.7%). Even for sentences with many 0,8 0,76 available counterexamples – for these sentences counterexamples can be retrieved automatically and Proportion of Responses 0,7 0,56 0,6 0,5 likelihood estimations are very low - we found an 0,4 0,33 Correct increase in the inference acceptance rates (7.1% control 0,27 Incorrect 0,3 vs. 19.8% preload). This shows that it is unlikely that 0,2 0,06 participants consulted their background knowledge for 0,1 0,01 0,0 deriving the conclusion and lends support for the Inference Likelihood Counterexample hypothesis that the working memory preload led to an Acceptance enhancement of the computationally low demanding matching heuristic. Figure 1: Difference in the proportion of the three types of In sum, as expected the resource dependent use of answers for preload trials where the dot pattern was correctly versus incorrectly reproduced. counterexample information decreased under preload, while the use of likelihood information was unaffected  converges with the observed interference of 0,8 counterexample use and correct dot pattern recall. 0,7 0,64 Proportion of Responses 0,56 In general, these results sustain the idea that using 0,6 counterexample information draws heavily on working 0,5 0,4 0,33 No Load memory resources whereas using likelihood 0,3 0,22 Load information or matching is less resource demanding. 0,2 0,1 0,05 0,06 Discussion 0,0 Inference Acceptance Likelihood Counterexample Correlational studies revealed that differences in working memory capacity relate to differences in the Figure 2: Proportion of answers of the three types for the conditional answer patterns. A possible explanation is preload versus control condition (only preloaded trials with that differences in reasoning performance do not correctly reproduced dot patterns). simply relate to differences in a single reasoning predisposition, but are mediated by differences in the by preload conditions. The decrease in use of the working memory demands of the active reasoning counterexample based reasoning process is at least mechanisms. Highlighting the distinction between more partly compensated by shifting to inference acceptance. heuristic strategies (such as matching and likelihood use) and more cognitively demanding analytical Number of counterexamples used. Does the decrease strategies (relying on counterexamples) may provide a in the use of counterexample information under preload more differentiated picture of the specific role of reflect a decrease in a strategic validation tendency? If working memory in conditional reasoning. We found participants retrieve counterexample information to evidence for two conditional reasoning mechanisms merely check whether the default conclusion can be with a differing working memory demand: a falsified (see e.g., Schroyens, Schaeken, & Handley, probabilistic account relying on likelihood information 2003) they would need to retrieve only one and a mental model account relying on counterexample information. counterexample to falsify the given conclusion. The results reveal that using counterexample However, we did not find a difference in the number of information to attain a conclusion taps heavier on trials where participants referred to only one working memory resources than deriving the counterexample (preload: 73% vs. control-condition: conclusion based on likelihood information. This 82.4%). This raises doubt on the validation-hypothesis. provides additional support for considering the In contrast, we observed a decrease in the proportion of reasoning process based on likelihood information as trials where more than one counterexample was heuristic and the reasoning process based on mentioned, t(23) = 2.77, p < .05 (preload: 17% vs. counterexample information as analytic. The control-condition: 26%). This underscores the idea that differences in use of counterexamples/likelihood on in tasks without deductive instructions reasoners participants with varying working memory capacity retrieve counterexample information to provide an observed by Verschueren et al. (2004b) may thus be adequate and informative conclusion rather than to attributed to the working memory demands of the two merely falsify a default conclusion. When looking at reasoning mechanisms. the total number of specific counterexamples used, We found a large effect of working memory preload there were significantly more counterexamples used in on the inference acceptance rates. When relating the control condition (1.09) than when working inference acceptance to the two reasoning strategies, it memory was preloaded (0.86), t(23) = 3.97, p < .01. can reveal that either no counterexamples can be If counterexample retrieval, representation and retrieved or that the likelihood estimation is sufficiently integration demand effort, we should observe an effect high. However, because we also observed an increase of counterexample retrieval on the secondary task in inference acceptance on sentences for which pretests performance. We tested whether there was a difference revealed many available counterexamples as well as in the number of counterexample answers for the trials likelihood estimates that are well below 1, it rather seems that the inference acceptance rates show that where the dot pattern was correctly versus incorrectly under preload some reasoners do not consult their reproduced. We included the number of available background knowledge. When working memory counterexamples (few/many; measured by the capacity is burdened by preload, these participants are generation task) because it is a strong predictor of discouraged to engage in a demanding retrieval counterexample use. There was a marginally significant process. Instead they provide an answer that satisfies interaction between the number of counterexamples the inference question, simply by restating information used and the (in)correct reproduction of the dot pattern, from the premises. This strategically placed escape F(1.20) = 4.120, MSE = 2.866, p = .056. Pairwise hatch can explain the increase in inference acceptance comparisons revealed that for sentences with many rates under preload. available counterexamples there were significantly Taken this together, we found evidence for the more counterexamples produced when the dot patterns involvement of working memory in conditional were not recalled correctly, F(1, 20) = 6.946, MSE = reasoning. By analyzing the answers participants gave 4.832, p<.05 (not significant for few-sentences). This we were able to pinpoint which information participants used to attain their conclusion. We found support for distinguishing two heuristic reasoning skill. European Journal of Cognitive Psychology, 11, strategies -use of likelihood information and matching- 473-498. and for an analytic strategy that takes counterexamples Johnson-Laird, P. N., & Byrne, R. M. J. (1991). into account. Working memory preload yielded an Deduction. Hillsdale, NJ: Laurence Erlbaum. increase in the use of heuristic strategies whereas the Klauer, K. C., Stegmaier, R. & Meiser, T. (1997). use of the analytical strategy decreased. Working memory involvement in propositional and The present study is one of the first to combine a spatial reasoning. Thinking and Reasoning, 3, 9-47. secondary task paradigm with a verbalization criterion Liu, I., Lo, K., & Wu, J. (1996). A probabilistic task. Using a preload-paradigm is probably the best interpretation of ‘if-then’. Quarterly Journal of way to investigate the working memory demands of Experimental Psychology, 48, 828-844. tasks involving verbalization. Although we cannot be Markovits, H. & Barrouillet, P. (2002). The entirely conclusive on a possible secondary task development of conditional reasoning: A mental interference on verbalization processes (the answers model account. Developmental Review, 22, 5-36. were structurally similar to baseline results) this Miyake, I., Friedman, N. P., Rettinger, D. A., Shah, P., procedure enabled us to experimentally test the & Hegarty, M. (2001). How are visuospatial working difference in working memory demands. memory, executive functioning and spatial abilities The effect of working memory capacity on inference related? A latent variable analysis. Journal of making is at present only discussed on an intensive Experimental Psychology, 130, 621-640. level: We investigated the global effect of a working Meiser, T., Klauer, K. C., & Naumer, B. (2001). memory dependent secondary task on the use of Propositional reasoning and working memory: the likelihood and counterexample information. Whether role of prior training and pragmatic content. Acta the working memory demands of the two processes Psychologica, 106, 303-327. coincide with the assumed differences in Oaksford, M., Chater, N., & Larkin, J. (2000). representation, retrieval and manipulation cost cannot Probabilities and polarity biases in conditional be decided upon based on the present results. The data inference. Journal of Experimental Psychology, 26, may also reflect the cost of determinacy: Giving a 883-899. gradual uncertain answer may be overall less Oberauer, K., Suess, H.-M., Wilhelm, O., & Wittmann, demanding than providing a determinate conclusion. W. W. (2003) The multiple facets of working There is also no information about the relative memory: Storage, processing, supervision and functional involvement of the different working coordination. Intelligence, 31, 167-193. memory components. Specific research with different Politzer, G. & Bourmaud, G. (2002). Deductive types of well-chosen secondary tasks may reveal this reasoning from uncertain conditionals. British crucial information. Journal of Psychology, 93, 345-981. In sum, distinguishing different reasoning Schroyens, W. Schaeken, W., & Handley, S. (2003). In mechanisms that can be used to solve conditional search of counterexamples: Deductive rationality in inferences can enhance our comprehension of how human reasoning. Quarterly Journal of Experimental working memory mediates the reasoning performance. Psychology. The specific working memory demands of different Stanovich, K. E. & West, R. F. (2000). Individual reasoning strategies co-determine the robust effect of differences in reasoning: Implications for the working memory capacity on the conditional reasoning rationality debate? Behavioural and Brain Sciences, performance. 23, 645-726. Suess, H.-M., Oberauer, K., Wittmann, W., Wilhelm,, Acknowledgments O., & Schultze, R. (2002). Working memory explains reasoning ability – and a little bit more. Intelligence, This research was conducted thanks to funding of the 30, 261-288. Fund for Scientific Research (F.W.O-Vlaanderen). Verschueren, N., Schaeken, W., & d’Ydewalle, G. (2003). Two reasoning mechanisms for solving the References conditional ‘fallacies’. Proceedings of the 25rd Barouillet, P., Lecas, J. F. (1999). Mental models in Annual Conference of the Cognitive Science Society conditional reasoning and working memory. Thinking Mahwah: Erlbaum. and Reasoning, 5, 289-302. Verschueren, N., Schaeken, W. & d’Ydewalle, G. Copeland, D. E. & Radvansky, G. A. (in press). (2004a). A dual process theory on causal conditional Working memory and syllogistic reasoning. reasoning. Manuscript submitted for publication. Quarterly Journal of Experimental Psychology. Verschueren, N., Schaeken, W. & d’Ydewalle, G. Cummins, D.D., Lubart, T., Alksnis, O., & Rist, R. (2004b). Working memory capacity determines (1991). Conditional reasoning and causation. Memory which reasoning process is used for solving and Cognition, 19, 274-282. conditional inferences. Accepted for publication in De Neys, W. (2003). In search of counterexamples: A Memory and Cognition. specification of the memory search process for stored Weidenfeld, A. & Oberauer, K. (2003). Reasoning counterexamples during conditional reasoning. from causal and non-causal conditionals: Testing an Unpublished doctoral dissertation, University of integrated framework. Proceedings of the 25rd Annual Leuven, Belgium. Conference of the Cognitive Science Society Gilhooly, K.J., Logie, R.H., & Wynn,V. (1999). Mahwah: Erlbaum. Syllogistic reasoning tasks, working memory and