Running head: R538-BF CHOKING UNDER PRESSURE Choking Under Pressure and Working-Memory Capacity: When Performance Pressure Reduces Fluid Intelligence (Gf)

Recent findings (Beilock & Carr, 2005) demonstrate that only individuals with a high working-memory capacity (WMC) “choke under pressure” on math problems with high WM demands. This suggests that performance pressure hinders those people who are the most qualified to succeed because it consumes the WMC they usually rely on to achieve superior performance, and thereby questions the ability of performance in high-pressure situations to distinguish individuals with a lesser or greater WMC potential. While addressing several limitations of past research, we offer evidence that such choking (1) occurs only in individuals high in WMC due to their anxiety-ridden perceptions of high-stake situations, and (2) is not confined to tasks involving acquired skills and knowledge but encompasses fluid reasoning abilities or intelligence (Gf). These findings have strong implications for assessments of people’s intellectual capacities in academic, clinical, work, and research settings. R538-BF_Choking under pressure 3 Choking Under Pressure and Working-Memory Capacity: When Performance Pressure Reduces Fluid Intelligence (Gf) Choking under pressure is an expression used to refer to performing more poorly than expected, given one’s skill, in situations where the pressure to perform is high (i.e., where the desire to excel is maximal ; see Baumeister, 1984; Beilock, Kulp, Holt, & Carr, 2004). As noted by Beilock and Carr (2005), the consequences of suboptimal performance, especially on examinations, include poor evaluations by mentors, teachers, and peers, lost scholarships, and relinquished educational and employment opportunities. Examining the relationship between choking under pressure and individual differences in working-memory capacity (WMC), these authors concluded that the individuals most likely to fail under performance pressure are those who, in the absence of pressure, have the highest potential for success (i.e., high-WMC individuals). Performance pressure, they reasoned, hinders the most qualified people by consuming the WMC they use in low-pressure circumstances to devise complex (resourcedemanding) strategies and produce superior performance. In their study, however, Beilock and Carr (2005) focused only on arithmetic problems (involving both subtraction and division). Here, we examine the generalizability of Beilock and Carr’s (2005) conclusion using a task (Raven’s Standard Progressive Matrices or SPMs) known to measure “fluid intelligence” (Gf). Unlike “crystallized intelligence”, which quantifies aptitudes linked to acquired skills and knowledge, Gf is defined as the ability to reason on largely unfamiliar materials and problems (Cattell, 1943; Raven, Raven, & Court, 1998). The results of our experiment reveal that choking in high-powered people is not confined to tasks involving acquired skills and knowledge but encompasses fluid reasoning abilities or intelligence. They also help understand how individuals who are high versus low in WMC perceive high-performancepressure situations, and have strong implications in the assessment of people’s intellectual capacity. R538-BF_Choking under pressure 4 Relating WMC and Choking Under Pressure WMC is typically viewed as reflecting individual differences in the amount of goaldirected attention available for temporarily keeping information of interest activated while inhibiting irrelevant thoughts and thus preventing distraction (Barrett, Tugade, & Engle, 2004, p. 560). Exploring the potential impact of individual WMC differences on susceptibility to choking, Beilock and Carr (2005) brought three preliminary findings to bear: (1) performance pressure increases reported state anxiety (Beilock & Carr, 2001; Beilock et al., 2004), (2) intrusive thoughts like worries reduce the storage and processing capacity of working memory (Eysenck & Calvo, 1992), and (3) anxiety-provoking situations reduce attentional resources, resulting in performance decrements (Ashcraft & Kirk, 2001; Beilock et al., 2004; Schmader & Johns, 2003). On this basis, they suggested that if pressure-induced consumption of working memory denies high-WMC individuals (HWMs) the resources they normally rely on to achieve superior performance, then their usual working-memory advantage may be just what makes them prone to fail under pressure. In their experiment, the HWM advantage found on the most memory-demanding problems disappeared when pressure (manipulated within subjects) was applied. THE PRESENT EXPERIMENT This new experiment tested whether Beilock and Carr’s (2005) fascinating findings can be generalized to fluid intelligence (Gf). WMC measures are indeed strongly related to various psychometric measures of intelligence, especially Gf (e.g., Kane et al., 2004). If pressure-induced consumption of working memory denies HWMs the resources needed to mobilize their advantageous strategies, then their Gf level should be suboptimal when pressure is applied. We chose Raven’s SPMs, one of the best measures of Gf, which have the added advantage of being progressively difficult, starting from problems low in working-memory demands (involving Gestalt continuation and visuospatial abilities; see Appendix A) to highR538-BF_Choking under pressure 5 demand problems requiring verbal-analytic reasoning (see Appendix B). If pressure consumes WMC, choking should be specific to verbal-analytic/high-demand problems. Our experiment also addressed four limitations of Beilock and Carr’s (2005) study. First, Beilock and Carr’s manipulation of pressure implied a combination of factors (e.g., camera, money incentives) that prevented them from identifying the main component of performance pressure. Here, we induced pressure simply by characterizing the SPMs as diagnostic or non-diagnostic of analytic reasoning (a core faculty in the participants’ scientific curriculum). Second, although the authors associated pressure with increased state anxiety, they did not test this specific point. We measured state anxiety. In addition, when pressure is applied, HWMs may display a tendency to excuse their performance. We assessed such a tendency through a series of self-handicapping items. Third, Beilock and Carr measured WMC after testing subjects on arithmetic problems. Although they used filler tasks in between, the possibility still exists that their WMC measures captured some pressure effects. WMC was measured here before the pressure induction. Finally, because WMC predicts success on a broad range of cognitive tasks (Engle & Kane, 2004), it could be that Beilock and Carr’s findings were not due to the participants’ WMC per se but to their feelings of superiority or inferiority in the focal-task domain. Effects on performance were calculated here while controlling for this potential confounding factor. Method Participants The participants were 67 undergraduates (mean age 22, SD = 4.61; 35 females, 32 males) from the University of Provence (France) who had majored in math and science in secondary school. They agreed to take part in the study, which was presented as part of a larger project focusing on students’ memory, attention, and reasoning abilities. R538-BF_Choking under pressure 6 Procedure First, participants (tested individually) were asked to use 5-point scales ranging from 1 (very inferior) to 5 (very superior), to rate their standing relative to their classmates on mathematical ability, capacity to succeed in school, general intelligence, attention, reasoning, verbal ability, and general knowledge (hereafter called comparative evaluations). Next, their WMC was measured using Kane et al.’s (2004) computer-based version of the classic Reading Span Task (Daneman & Carpenter, 1980). Each display included a meaningful or meaningless sentence the participants read aloud while verifying whether it made sense, and a to-be-remembered letter, which they also read aloud (e.g., “We were fifty lawns out at sea before we lost sight of land. ? X”). The sentences (12 to 17 words, M = 14.4 words, SD = 1.2) were taken from the French version of the RSPAN (Desmette, Hupet, Schelstraete, & Van der Linden, 1995). Following Kane and colleagues’ procedure, the meaningless sentences were created by changing only one word (“miles” into “lawns” in the above example). The set size ranged from 2 to 5 sentence/letter problems per trial, with 3 trials per set size, making a total of 12 trials. At the end of each set, individuals had to write down the sequence of letters in the correct order. An item was scored as correct when recalled in the correct serial position. RSPAN scores (range: 0-100%) were equal to the total number of correctly recalled letters divided by the total number of letters to be recalled (42). Participants were then introduced to the SPMs. For half of them, the task was described as measuring analytic reasoning, which was explicitly associated with overall success in mathematics and science (high-pressure condition). For the other half, it was described as simply measuring attentional and perceptual capacities (low-pressure condition). All participants were then given a general description of the task via two sample exercises, an easy one (the first SPMs; see Appendix A) and a difficult one (from the Advanced Progressive Matrices; see Appendix B). The experimenter did not give the correct answers or any feedback. Next the participants had to use a 7-point scale ranging from 1 (none) to 7 (all), to R538-BF_Choking under pressure 7 say how many exercises they expected to answer correctly. Then they were given both the lowand high-demand SPMs’ exercises (see Lynn, Allik, & Irwing, 2004; van der Ven & Ellis, 2000 for this categorization). Both accuracy (percentage of correct answers) and inspection times were measured. The inspection time is the time it takes a participant to answer a problem. It runs from the start of a trial or exercise to the validation of the participant’s choice. Here, only inspection times on correct exercises are considered. When they had finished the 60 exercises or when the allotted time (15 min) was up, they rated how frequently they felt anxious, comfortable, jittery, worried, at ease, nervous, relaxed, and calm while doing the SPMs, using a 5-point scale ranging from 1 (never) to 5 (always) (questionnaire adapted from Spielberger’s State Anxiety Scale, 1970; see Schmader & Johns, 2003). Finally, participants rated how handicapped they had felt during task performance, on a 5-point scale ranging from 1 (totally disagree) to 5 (totally agree) (e.g., “I slept well last night”; adapted from Rhodewalt, 1990). The items on each scale were averaged to obtain a state anxiety index (α = .87) and a self-reported handicap index (α = .69). These indexes assessed how participants consider the SPMs as a whole, regardless of problem demand (a within-subject factor). Results SPMs Participants were assigned to low and high working-memory groups (LWMs: M =.48, SE = .01 vs. HWMs: M = .73, SE = .01), using a median split of RSPAN scores (ranging from .33 to .95) as in Beilock & Carr (2005). The performance data (SPM accuracy and inspection time) were analyzed in a 2 (WMC: LWM vs. HWM) x 2 (Problem Demands: low vs. high) x 2 (Pressure: low vs. high) mixed ANOVA, with problem demands as a within-subject factor. Accuracy (% correct). The main effect of problem demands was clearly significant, F(1, 63) = 243.65, p < .0001, ηp 2 = .81. Accuracy was better for low-demand (M = 94%, SE = .01) than for high-demand (M = 58%, SE = .03) problems. As expected, this effect was R538-BF_Choking under pressure 8 qualified by a significant three-way interaction, F(1, 63) = 4.21, p < .05, ηp 2 = .07. Planned comparisons derived from Beilock and Carr’s findings showed that on low-demand problems, HWMs did not differ from LWMs, whether in the low-pressure (M = 95% vs. 92%) or highpressure (M = 94% vs. 93%) condition. Also as expected, on high-demand problems (see Figure 1, left graph), HWMs outperformed LWMs in the low-pressure condition, t(63) = 3.72, p < .001 (unless clearly mentioned, all t-tests in the results section are two-tailed). This difference was not observed when pressure was applied, t(63) = .69, ns. The LWMs’ performance did not suffer from pressure, t(63) = -1.32, p = .19 (they even tended to benefit from it), whereas the HWMs’ performance did, t(63) = 1.67, p = .05 (one-tailed). Moreover, the probability of replicating the three-way interaction was high (prep = .89; see Killeen, 2005). To clarify the findings, performance accuracy on high-demand problems was regressed over continuous RSPAN scores, pressure, and their cross product. In line with the cognitive literature on working memory (e.g., Unsworth & Engle, 2005), RSPAN scores predicted accuracy, β = .38, p < .001. However, as expected from the above results, the RSPAN x Pressure interaction was clearly significant, β = -.31, p < .01. Whereas the RSPAN scores predicted accuracy under low pressure, β = .60, p < .0001, it did not when pressure was applied, β = .09, p = .64. This interaction remained consistently significant when the participants’ comparative evaluations were controlled, including reasoning and mathematical ability. Gender had no effect either. Inspection time (IT) on correct exercises. The main effect of problem demands was significant, F(1, 63) = 286.69, p < .0001, ηp 2 = .82. ITs were longer for high-demand (M = 24.20, SE = .93) than for low-demand (M = 10.13, SE = .31) problems. No other effects were found. Performance Expectation Ratings, Anxiety, and Handicaps No effects were found on the expectation ratings (M = 4.69, SE = .13, overall). R538-BF_Choking under pressure 9 On state anxiety, the WMC x Pressure interaction was marginally significant, F(1, 63) = 3.63, p = .06, ηp 2 = .06. HWMs and LWMs did not differ from each other under low pressure, t(63) = .40, ns, whereas the former reported significantly more anxiety than the latter when pressure was applied, t(63) = 3.07, p < .01 (see Figure 1, middle graph). Anxiety increased in HWMs under pressure, t(63) = 1.81, p < .04 (one-tailed), but not in LWMs, t(63) = .89, ns. The WMC x Pressure interaction was significant on self-reported handicaps, F(1, 63) = 5.29, p < .03, ηp 2 = .08. HWMs and LWMs were indistinguishable under low pressure, t(63) = .44, ns, whereas (contrary to our prediction) the latter reported more handicaps than the former when the pressure was great, t(63) = -2.79, p < .01 (see Figure 1, right graph). The self-handicapping tendency increased in LWMs under pressure, t(63) = -1.87, p = .06, but not in HWMs, t(63) = 1.38, p = .17. The anxiety and handicap indexes were not correlated, neither overall, nor within each condition taken separately. Mediation Did anxiety mediate (i.e., Baron & Kenny, 1986) “choking” among HWMs? It did. MacKinnon et al.’s (2002) additional tests showed that the direct path coefficient (from pressure to performance, β = -.33, p < .05) decreased significantly (p <.05) when regressing performance on both pressure and anxiety (β = -.22, ns). --------Insert Figure 1 about here --------Discussion The present findings merit special attention. They offer new evidence that on tasks with high WM demands, the individuals most likely to fail under performance pressure are those with a high WMC. Although potential confounding factors were addressed (WMC was measured before inducing pressure, pressure was manipulated between subjects via task R538-BF_Choking under pressure 10 characterization only, and comparative evaluations were controlled), our performance effects were clearly consistent with those reported by Beilock and Carr (2005). Our findings also show that choking under pressure is due to increased state anxiety, which is in line with Beilock and Carr’s (2005) explanation that WMC is jeopardized by pressure via increased anxiety. In this explanation, however, pressure should increase anxiety in LWMs as well, which was clearly not the case here. Does this mean that we simply failed to induce pressure in our LWMs? Probably not, given that the LWMs reported feeling more handicapped under pressure. This rules out the possibility that they were totally unaffected by pressure. Taken together, our results suggest that not all individuals experience high-performance-pressure situations in the same way: reactions to pressure appear to depend on WMC. It seems, then, that choking occurs in HWMs alone, due to their anxious perception of high-stake situations at the onset. Alternatively, our findings could be taken as evidence that high spans tend to use a much more controlled processing approach (to the focal task), compared with the low spans, who primarily make use of automatic processes (e.g., automatic spreading activation, see Kane & Engle, 2000; Rosen & Engle, 1997). According to this account, the additional load (here probably worries or intrusive thoughts) interfered with controlled processing and not with automatic processing. But for this alternative account to be valid, the load the low spans have to deal with must equal that of the high spans, which seems not true here (at least when considering the anxiety findings). Once more, it can be reasonably assumed that pressure induced a higher load in the HWMs than in the LWMs. Future research is needed on this crucial point, however. One may also wonder whether the way pressure was manipulated here is isomorphic with previous work (Beilock and Carr, 2005). After all, varying the diagnosticity of the task comes close to the pressure manipulation that is typically used in the Stereotype threat paradigm (see Spencer, Steele, & Aronson, 2002, for a review). In this paradigm, however, the R538-BF_Choking under pressure 11 performance deficit that is frequently observed on difficult tests – including those measuring WMC (see Schmader & Johns, 2003) – is not due to the pressure that may be associated with taking a difficult intellectual test per se. This deficit is due to the extra-pressure that is associated with the possibility that one's performance will confirm, to others, oneself, or both, a negative stereotype about one's group. Although there is some reason here to believe that the HWMs experienced more pressure than their LWM counterparts, there is no reason to think that the former felt threatened by the possibility to confirm a negative stereotype. The question arising here is exactly which stereotype could threaten the HWMs in the diagnostic condition? A stereotype threat approach of the present study would rather lead toward expecting a suboptimal performance in women under this condition, because of the well-known negative stereotype about their gender's math and scientific abilities. No gender effects were found in the present study, however. Furthermore, assuming that our pressure manipulation comes close to that used in the stereotype threat literature would be a somewhat weak criticism. There is indeed no guarantee that pressure and stereotype threat manipulations do not produce their effects through a common mediating variable (e.g. working memory). Most people in the field would accept “performance pressure” as describing any manipulation that motivates people to try harder to perform well – including stereotype threat. Finally, earlier in this paper we assumed that low-demand problems of the SPMs were more visuospatially orientated while the high-demand ones were more verbally oriented. To the extent the diagnostic or high-pressure condition was associated with increased worries about the situation and that the related consequences were verbal in nature (e.g., see Beilock et al., 2004; Cadinu et al., 2005), it could be that the reason the high-demand problems were impacted is not because of the degree of working memory reliance per se, but the type of reliance. Although Kane and Engle (2004) offered evidence that WMC captures domaingeneral information processing capabilities (see however Baddeley, 1996; Shah & Miyake, 1996), this possibility might also be explored in future research. R538-BF_Choking under pressure 12 Above all, our findings show that pressure-induced performance decrements in HWMs are not limited to tasks involving acquired skills and knowledge, but cover fluid reasoning or intelligence (Gf). As such, they may have important implications. Gf tests are widely used to distinguish high achievers from less fortunate individuals in academic, clinical, occupational, and research settings (e.g., Kane et al., 2004; Raven et al., 1998; Schmidt & Hunter, 1998). Because pressure is generally high in these settings, individuals with different capacity-related potentials may ironically display about the same fluid ability – as revealed by our results. Combined with those of other studies (Beilock & Carr, 2005; Croizet et al., 2004; Huguet, Galvaing, Monteil, & Dumas, 1999; Schmader & Johns, 2003; Steele, 1997), the present findings offer new evidence that the test situation per se matters: It is an integral part of cognitive functioning, not just its surrounding context. R538-BF_Choking under pressure 13