Embodiment in Causal Learning? Effects of Fluency and Body Specificity Kelly M. Goedert (kelly.goedert@shu.edu) Department of Psychology, Seton Hall University, 400 South Orange Ave. South Orange, NJ 07079 USA Clinton C. Dudley (dudleyc1@mail.montclair.edu) Department of Computer Science, Montclair State University, 1 Normal Ave Montclair, NJ 07043 USA Abstract of objects are scaled by one’s own body (see Proffitt & Linkenauger’s, 2013, phenotypic expression theory). Of The body specificity hypothesis (Casasanto, 2009) predicts an association between positive conceptual information and the particular relevance here, however, is the body specificity side of space associated with the dominant hand. In the current hypothesis (Casasanto, 2009). According to body specificity, study we investigated whether body specificity may produce the fluency of movement associated with a person’s dominant states of fluency or disfluency that influence causal learning by hand spatially grounds notions of positive or good to the inducing intuitive versus analytical processing. Right-handed dominant side of space. Thus, right-handers implicitly participants learned about two potential causes of a common associate good with the right side of space and left-handers outcome in a trial-by-trial contingency learning task. We manipulated the spatial location of the causes (left, right) and associate it with the left (Casasanto, 2009). These the hand participants used to make responses (left, right). handedness-based fluency effects may also produce some of Consistent with expected fluency for a strong cause on the the bodily influences on size and distance perception. For right, when using their right hand, participants better- example, right-handers perceive their right arm as longer and discriminated between strong and weak contingencies for a believe they can reach farther with their right hand versus right-located cause. Eye tracking revealed that this increased their left (Linkenauger, et al., 2009). accuracy was not associated with an increase in overt visual attention. Rather, RT and eye tracking suggest that the body- specificity effects were associated with fluency differences Embodiment, Fluency, and Judgment across conditions. The concept of fluency may be one mechanism for uniting Keywords: causal learning; causal inference; embodiment; at least some embodiment effects with work in memory, eye tracking; body specificity hypothesis; fluency; judgment judgment, and decision making. Fluency refers to individuals’ subjective perception of the relative ease or Introduction difficulty of their own on-going cognitive processing (Oppenheimer, 2008). According to a recent model, fluency Systematic and reliable biases emerge from the functional serves as a cue to the status of distal events about which organization of our perception and action systems, affecting individuals do not have direct information (Unkelbach & high-level cognitive processes such as memory retrieval and Greifeneder, 2013). For the phenotypic expression and body problem solving (e.g., Thomas & Llera, 2007). Yet, work in specificity effects discussed above, action-based fluency may causal inference has largely ignored the embodied reasoner serve as a cue for perceptual and affective judgments (i.e., (cf. Wolff, Holmes, & Ritter, 2014). embodied fluency; Alter & Oppenheimer, 2009) In using the phrase embodied, our intention is not to claim What then of fluency’s effects? A consistent finding in that cognition occurs within the body. Rather, we use the reasoning and decision making is that individuals in a state of phrase to reflect those instances in which higher-level fluency are more likely to engage in faster, intuitive cognitive processes are influenced by seemingly irrelevant reasoning processes, and those in a state of disfluency in stimulus and response characteristics of a task. That is, when slower, analytic processing – i.e., system 1 versus system 2 cognition is affected by being carried out in a body. (Sloman, 1996), respectively (Alter, Oppenheimer, Epley, & Embodiment may affect cognition in a variety of ways. For Eyre, 2007; Alter, Oppenheimer, & Epley, 2013; Thompson example, making goal-directed upward movements with the et al., 2013). In the current study, we used a paradigm in hands leads to faster retrieval of positive memories, while which participants learn about two potential causes of a downward goal-directed movements leads to faster retrieval common outcome to investigate the potential for body of negative memories (Casasanto & Dijkstra, 2010). In a specificity to produce states of fluency or disfluency that Stroop task, placing the response buttons representing influence causal learning by inducing intuitive versus conflicting answers farther apart induces faster responding on analytical processing. incongruent trials – i.e., less conflict (Lakens et al. 2011). While the above examples reflect the influence of Causal Learning and Cue Competition irrelevant task demands on cognition, other embodiment effects reflect the nature of the specific body a person Lateralized valence-space associations may be of particular inhabits. For example, estimates of both distance and the size relevance to the situation in which two causes, appearing 471 separately in the left and right sides of space, “compete” for right), which always appeared opposite each other on the association with the outcome. When simultaneously learning computer screen. We also varied the hand participants used about two causes, participants judge a moderately effective to make trial-by-trial predictions (left, right). target cause to be less effective when it is learned about in the Figure 1 depicts the key rationale for our predictions. The presence of a highly effective alternative (e.g., Goedert & body specificity hypothesis predicts that right-handers Spellman, 2005). This general phenomenon is termed cue associate the right side of space with good. In the context of competition, reflecting that cues may compete either for learning about causes of a positive outcome (plant blooming), association with the outcome or for attention. we assumed that “good” would be a strong cause. In In cue competition, the reduction in the perceived particular, we predicted that body-specificity would establish effectiveness of the moderately effective cause is sometimes an expectation for a strong cause on the right side of space the product of controlling for alternatives – i.e., holding other (first set of shaded boxes in Figure 1). This expectation may causes constant while evaluating the effectiveness of the be strongest when participants use their right hand, as target (Spellman, 1996). However, participants also reduce opposed to their left, for responding. By manipulating the their judgments of a moderately effective cause beyond what relative locations of the target and alternative causes, we is expected from controlling for alternatives – i.e., they manipulated whether the strong alternative cause appeared on discount a moderately effective target cause when there is a the right versus whether the non-causal alternative or strong alternative (e.g., Goedert & Spellman, 2005). moderately causal target appeared on the right. As seen in Figure 1, when the strong cause appears on the right, this Current Experiment matches the expectation produced by body-specificity and We investigated how body-specificity-induced states of produces a state of fluency, which results in faster responding fluency or disfluency affect learning about causes of a and causal judgments matching that state (i.e., strong causal positive outcome. In this initial investigation, we focused on judgments for the strong alternative cause). As a result, it may right-handed individuals because of their greater prevalence. also lead to greater discounting of the target cause. However, The participants’ task was to determine the effectiveness of when either the weak alternative cause or moderately each of two liquids in causing plants to bloom. They learned effective target appear on the right, it is a mismatch to the about the liquids simultaneously on a trial-by-trial basis. On body-specificity expectation, resulting in a state of each trial, one of the liquids appeared on the left of the disfluency. In turn, this disfluency results in slower computer screen and the other on the right, with the plant responding and more analytic thinking, which we predict will centrally located. Participants first saw some combination of produce more accurate causal judgments. the liquids applied to the plant (one, neither, or both). They Figure 1 depicts the events resulting from a strong cause then predicted whether or not the plant would bloom and appearing on the right. However, the converse set of received feedback. After a series of trials, participants made predictions may follow from a body-specificity expectation separate judgments regarding the effectiveness of each of the for a weak or non-causal event on the left side of space. When liquids in causing plant blooming. the weak (i.e., non-contingent) alternative is on the left, it Critically, we manipulated the contingencies between each matches the body-specificity expectation, which produces a of the causes and blooming such that one cause – the target – state of fluency, resulting in causal judgments for the was moderately contingent with the outcome. This alternative that match that state (i.e., causal judgments of zero moderately contingent target cause was learned about in the for the non-contingent). presence of an alternative that was either strongly related to In addition to causal judgments, we collected participants’ the outcome (strong alternative) or not related to the outcome response time on the trial-by-trial predictions, as a potential (weak alternative). The occurrence of the two causes was measure of fluency (Oppenheimer, 2008). Finally, we tracked independent. Thus, if participants perceived the target to be participants’ eye movements to assess the potential less effective in the strong than in the weak alternative competing prediction that any embodiment effects we condition that would be evidence of causal discounting. We observe are due to shifts in visual attention. varied the location of the target and alternative causes (left, 472 Methods response terminated the prediction screen and initiated feedback (2500ms). Every 12 trials, participants rated how Participants effective each liquid was from -100 (completely inhibits plant blooming) to 100 (completely produces plant blooming). One hundred twenty-four undergraduate students (88 When making these ratings, participants released the female) participated. All identified as right-handed on the Revised Edinburgh Handedness Inventory (Dragovic, 2004). Logitech controller and used both hands to type on the keyboard sitting on top of the desk. Each contingency Design condition was associated with a different set of colored liquids. Prior to starting the contingency acquisition trials, The primary design of the experiment was a 2 (contingency participants first performed training trials to learn the condition: strong alternative, weak alternative) x 2 mapping of the yes and no response to the appropriate top or (alternative location: left, right) x 2 (responding hand: left, bottom buttons of the game controller. right) mixed design with contingency manipulated within- groups and location and responding hand between. We Eye-tracking Apparatus and Analysis measured objective contingency using the phi coefficient (φ). Across contingency conditions, the strength of the target We recorded the movements of participants’ left and right eyes using a Tobii x120 eye tracker, sampling at 60 Hz. cause was constant (φ = .33), but the strength of the Because eye-movements during feedback are influenced by alternative varied: In the strong alternative condition, the contingency between the alternative and the outcome was φ the accuracy of participants’ predictions (Wills, Lavric, Croft = .67 and in the weak alternative it was φ = 0. Table 1 depicts & Hodgson, 2007), we focused our analysis on the prediction the frequencies for each type of trial across the conditions. screens. We created two interest areas (left, right) by dividing Secondary, within-groups manipulations included learning the prediction screen in half vertically. Thus, each interest area encompassed either the target or the alternative cause. block (one, two, three) and trial type (target-only, alternative- Our primary measure of overt visual attention was dwell only, both, none). Finally, because participants responded yes and no with the same hand using the top and bottom trigger time in ms, which was the sum of all fixations on an interest buttons on a game-controller, we counterbalanced the area for a given trial type. We classified an eye movement as mapping of yes and no to the top and bottom buttons between- a fixation when the eyes lingered for 50ms or longer. Eye groups. movements with a minimum velocity of 30 degrees per second for 4ms or longer were classified as saccades and Table 1: Cell frequencies for contingency conditions. screened from the data. Statistical Analyses Weak Strong Alternative Alternative We performed mixed linear modeling (MLM), modeling the Target P A P A full factorial of contingency condition (strong alternative, P 6/9 6/9 9/9 3/9 weak alternative), alternative location (left, right), and A 3/9 3/9 6/9 0/9 responding hand (left, right) as fixed effects, with Note. P = present; A = absent. Cell ratios indicate the number of participants’ intercepts as the sole random effect. We only times the outcome occurred over the number of times that report the fixed effects as those address the research questions combination of causes occurred. Frequencies represent the total of interest. number of trials administered across entire experiment. Preliminary analyses revealed that mapping of the yes/no response to the top and bottom trigger buttons mattered early Procedure in learning (block 1), but not later. It is likely that participants Participants sat 60 cm from the computer screen, with their were still learning the conceptual mapping for the yes/no head and chin stabilized in a chin rest. They held a Logitech response in the first block of trials. While we present analyses game controller in their laps and used either the left or right for causal judgments across blocks, we focus our set of response buttons on the controller. The visual stimuli interpretation on the final block (block 3), after participants subtended 16.7° of visual angle. had maximal opportunity to acquire the contingencies, which The participants’ task was to determine how effective each should also minimize potential extraneous variability from of two liquids were in causing plants to bloom. Participants continued learning of the yes/no button mapping. acquired the contingencies depicted in Table 1 across three blocks of 12 trials each. Participants pressed a button to Results & Discussion initiate a trial, at which point a cross-hair appeared centrally on the screen until the participant fixated the cross-hair. On Causal Judgments each trial participants saw some combination of two colored Alternative. We turn first to causal judgments of the liquids applied to a plant without a bloom (i.e., one trial from alternative, for which the predictions depicted in Figure 1 are one of the cells of Table 1). They then predicted whether or most relevant because it was either a strong cause (φ = .67) not the plant would bloom using the top or bottom trigger or non-causal (φ = 0) across the contingency conditions. button of the game controller to indicate yes or no. This Consistent with our predictions, participants’ judgments of 473 the alternative cause varied not only with its objective and weak alternatives on judgments of the target was not strength, but also as a function of its location and the apparent in blocks one and two (ps > .05). This pattern of responding hand [F(1, 563) = 6.35, p = .012, for the three- results, in combination with the observation that participants’ way interaction]. Participants accurately discriminated accurate discrimination between the alternative strength for between the strong and weak alternative conditions across all strong versus weak conditions was stable by block two, three learning blocks (all ps < .001). However, the size of this suggests that discounting of the target emerged after – and in effect increased between blocks one and two (d = 0.69 for response to – recognition of the strong alternative. strong vs. weak in block 1, and d = 1.06 in blocks 2 and 3). Figure 2 depicts causal judgments of the alternative in block three. Participants most-accurately differentiated the strong and weak alternative when it appeared on the right side of the computer screen and they used their right hand to respond, F(1, 107) = 40.3, p < .001, d = 1.34. This pattern is consistent with the expectation that with the strong alternative on the right, right-handed participants in a state of fluency would rate it as strongly causal, but in a state of disfluency would more accurately judge the weak alternative as non-causal when it appeared on the right. Response Time (RT) Overall, we observed a main effect of block, such that participants’ RT decreased across blocks [F(2, 563) = 11.6, p < .001], as is typical of trial-by-trial causal learning. Block did not, however, interact with any other factors, all ps > .13. Therefore, to be consistent with the causal judgments, we focus our interpretation on RT in block 3, which is depicted in Figure 4. Comparing Figures 2 and 4, we see faster RT associated with those instances in which the strong Conversely, the judgments of participants using their left alternative was judged more effective, suggesting a role for hand to respond are consistent with the prediction that the fluency. weak cause appearing on the left would lead to a state of fluency and subsequent judgment of the alternative as weak (2nd set of bars in Figure 2). Furthermore, participants using their left hand rated the strong alternative as less causal when it appeared on the left vs the right (p < .05). Finally, the pattern of results depicted in Figure 2 suggests that the body-specificity-based expectations hold when there is a match between the responding hand and alternative location (i.e., right-right or left-left), but not when there is a mismatch. Target. In the same condition in which subjects best- discriminated the alternative, they demonstrated the greatest amount of discounting of the target (Figure 3). Despite equal objective contingencies for the target across the contingency conditions (φ = .33), they judged the target as less effective More precisely, we predicted fluency effects on RT such with the strong rather than weak alternative, F(1,107) = that right-handed participants would be in a state of fluency 16.82, p < .001, d = 0.83, when using their right hand to and have faster RT when the strong alternative cause respond and when the alternative appeared on the right (target appeared on the right, with the expectation that these effects on the left). This interaction among contingency condition, would be maximal when participants used their right hand. responding hand, and alternative location reached As seen in Figure 4, it was the responding hand, rather than significance for causal judgments of the target in block three, alternative location, that interacted with the contingency F(1, 107) = 4.72, p = .032. However, the effect of the strong condition in determining RT, F(1, 841) = 5.32, p = .021. 474 When using their right hand, participants were faster in the 0.27). This pattern is consistent with the prediction that strong (M = 1464.3, SD = 725.1) than in the weak alternative responding with the right hand would produce a fluency for (M = 1740.0, SD = 859.2) conditions, p < .001, d = 0.34. This the strong alternative, and a relative disfluency for the weak pattern is consistent with a prediction that right would induce alternative. When participants used their left hands, there was an expectation for a strong cause and thus induce fluency no difference in the dwell time to the alternative in the strong when that expectation was met. However, when responding (M = 748.1, SD = 525.1) and weak alternative (M = 790.3, with their left hand, there was no difference in participants’ SD = 604.1) conditions, d = 0.07. response times across contingency conditions (M = 1553.9, SD = 856.9 for the strong and M = 1587.6, SD = 826.2 for the weak alternative). While no other effects reached significance, the three-way interaction of location, responding hand, and contingency condition was marginal, F(1,118) = 3.09, p = .081. Comparing Figures 2 and 4, we see the same condition that witnessed the greatest differentiation between the strong and weak alternative causes, also witnessed the greatest difference in RT. Also, as seen in Figure 4, there was a general tendency to respond quickly in the strong alternative conditions, except when that strong alternative appeared on the left and participants were responding with their left hands: Participants may have responded more slowly because the strong alternative was inconsistent with the expectation for a weak cause. Target As with dwell time to the alternative, participants To be consistent with the causal judgments, we have looked at the target more when it appeared on the left versus focused on RT for block 3. However, if body-specificity right, F(1, 118) = 10.35, p = .002. Note that in Figure 6 the induced a state of fluency/disfluency, then effects on RT may x-axis label indicates the alternative location, and the target be even stronger in block 1. This is indeed what we observe. In block 1, participants using their right hand were faster in the strong (M = 2056.7, SD = 1053.5) vs. weak (M = 2833.2, SD = 1231.1) alternative conditions, d = 0.64 (a larger effect than in block 3). Furthermore, consistent with a violation of expectations, there was a small tendency for participants using their left hand to respond more slowly in the strong (M = 2312.6, SD = 1002.2) vs. weak (M = 2160.7, SD = 1333.8) alternative conditions, d = 0.13. Dwell Time on Prediction Screens While RT may serve as an indicator of fluency, RT in this particular paradigm is a product of how much time individuals spend processing both the target and alternative causes. We can tease apart these contributions with the eye tracking data and how long participants spent looking at each of the causes. Average total dwell time to the alternative appeared opposite the alternative. No other effects reached appears in Figure 5, and that to the target in Figure 6. The eye significance. tracking results do not support the potential competing prediction that embodiment effects result from increased General Discussion visual attention. Rather, dwell time analyses echoed RT. As depicted in Figure 1, we predicted that body-specific Alternative Overall, participants spent more time looking associations of the right space and right hand with “good” at the alternative when it was on the left (M = 925.7, SD = would produce an expectation for a strong cause on the right 558.6) versus right (M = 502.1, SD = 611.9), F(1,118) = 8.12, to produce a good outcome. Furthermore, we predicted that p = .005, d = 0.69. The analysis also revealed a condition by meeting this expectation would produce a state of fluency, hand interaction, p = .049, which echoed that observed in RT. resulting in faster, intuitive responding, while violating this When participants used their right hand to respond, they spent expectation would result in a state of disfluency, resulting in less time looking at the alternative when it was strong (M = slower, analytical responding. Causal judgments of the 641.5, SD = 473.8) than when it was weak (M = 832.1, SD = alternative are largely consistent with these expectations, but 794.4). This effect was larger when the stronger alternative only when the responding hand and alternative location was located on the right (d = 0.34) as opposed to the left (d = 475 match. Furthermore, the effects are stronger for the right hand Casasanto, D. (2009). Embodiment of abstract concepts: and right space than they are for left hand and left space. Good and bad in right- and left-handers. Journal of A remarkable consistency emerged among participants’ Experimental Psychology: General, 138, 351-367. judgments of the alternative, those of the target, and the RT Casasanto, D., & Dijkstra, K. (2010). Motor action and and dwell time: Those conditions best-discriminated in emotional memory. Cognition, 115, 179-185. causal judgments (right-right) also showed the greatest Dragovic, M. (2004). Towards an improved measure of the discrimination in RT and dwell time. Suggestive of a Edinburgh Handedness Inventory: A one-factor congeneric disfluency for the weak alternative on the right, participants measurement model using confirmatory factor analysis. spent more time looking at the weak than the strong Laterality, 9, 411–419. alternative on the right side of the screen when using their Goedert, K. M., & Spellman, B. A. (2005). Nonnormative right hand. Eye tracking analyses revealed that these effects discounting: There is more to cue interaction effects than were not due to an overall increase in visual attention to the controlling for alternative causes. Learning & Behavior, right. Indeed, overall, participants spent more time looking at 33, 197-210. the left than the right side of space, an effect consistent with Lakens, D., Schneider, I. K., Jostmann, N. B., & Schubert, leftward biases often observed in visuomotor tasks (e.g., T. W. (2011). Telling things apart: The distance between Nicholls, Bradshaw & Mattingley, 1999). Rather, the eye response keys influences categorization times. tracking and RT analyses are most consistent with a fluency Psychological Science, 22, 887-890. explanation. Participants using the right hand responded Linkenauger, S. A., Witt, J. K., Bakdash, J. Z., Stefanucci, J. quicker in strong relative to weak alternative conditions. K., & Proffitt, D. R. (2009). Asymmetrical body perception: A possible role for neural body representations. Limitations Psychological Science, 20, 1373-1380. This investigation is just a first step in identifying possible Nicholls MER, Bradshaw JL, Mattingley JB. (1999). Free- embodiment effects, and body specificity effects, on causal viewing perceptual asymmetries for the judgement of learning. Were body specificity mechanisms truly driving the brightness, numerosity and size. Neuropsychologia, 37, effects observed here, we would expect these effects to 307-314. reverse in left-handers. We might also expect these effects to Oppenheimer, D. M. (2008). The secret life of fluency. reverse for people learning about causes of negative Trends in Cognitive Sciences, 12, 237-241. outcomes (e.g., disease). Finally, RT, as used here, is only Sloman, S. A. (1996). The empirical case for two systems of one potential measure of fluency. Future work could use reasoning. Psychological Bulletin, 119, 3-22. confidence ratings for the causal judgments to confirm the Spellman, B. A. (1996). Acting as intuitive scientists: fluency predictions and RT results. Contingency judgments are made while controlling for alternative potential causes. Psychological Science, 7, Conclusion 337-342. Thomas, L. E., & Lleras, A. (2007). Moving eyes and moving We provide evidence suggestive of embodiment effects in thought: On the spatial compatibility between eye causal learning. Thus far, these effects are consistent with the movements and cognition. Psychonomic Bulletin & prediction of fluency and disfluency resulting from body- Review, 14, 663-668. specific space-valence associations (Casasanto, 2009). More Thompson, V. A., Turner, J. A., Pennycook, G., Ball, L. J., work is needed to verify this account of the effects and to Brack, H., Ophir, Y., et al. (2013). The role of answer further understand the underlying mechanisms. fluency and perceptual fluency as metacognitive cues for initiating analytic thinking. Cognition, 128, 237-251. Acknowledgments Unkelbach, C., & Greifeneder, R. (2013). A general model of This work was funded by an Independent College Fund of fluency effects in judgment and decision making. In C. New Jersey Research Symposium Award to CCD. Unkelbach, & R. Greifender (Eds.), The experience of thinking: How the fluency of mental processes influences References cognition and behaviour. (pp. 11-32). New York, NY, US: Alter, A. L., & Oppenheimer, D. M. (2009). Uniting the tribes Psychology Press. of fluency to form a metacognitive nation. Personality and Wills, A. J., Lavric, A., Croft, G. S., & Hodgson, T. L. Social Psychology Review, 13, 219-235. (2007). Predictive learning, prediction errors, and Alter, A. L., Oppenheimer, D. M., & Epley, N. (2013). attention: Evidence from event-related potentials and eye Disfluency prompts analytic thinking--but not always tracking. Journal of Cognitive Neuroscience, 19, 843-854. greater accuracy: Response to Thompson et al. (2013). Wolff, P., Ritter, S., & Holmes, K. (2014). Causation, force, Cognition, 128, 252-255. and the sense of touch. In P. Bello, M. Guarini, M. Alter, A. L., Oppenheimer, D. M., Epley, N., & Eyre, R. N. McShane, & B. Scassellati (Eds.), Proceedings of the 36th (2007). Overcoming intuition: Metacognitive difficulty Annual Conference of the Cognitive Science Society (pp. activates analytic reasoning. Journal of Experimental 1784-1789). Austin, TX: Cognitive Science Society. Psychology: General, 136, 569-576. 476