{0‘3 5‘ 452'5 I" i‘lvli. 'l" C e “wits—3m ,§ :L 3' g ’m z 2'. ‘i '1: n ww >- 4.51.1? at. .d- 3‘- _ my, - . «w «a, . . ,. 1 '5 imamrrwwx v 1.} '....v ”n L' 9. i "h in :1” 2.4. w W“ -. c” 3 Si 770 5 Li This is to certify that the _ L'BRAR Y dissertation entitled Michigan State Universi: ’ WHEN PERFORMANCE FAILS: EXPERTISE, ATTENTION, AND PERFORMANCE UNDER PRESSURE presented by Sian Leah Beilock has been accepted towards fulfillment of the requirements for the Ph.D. degree in Kinesiolgqy and Psychology ”(K a. or, mum Major Professor’s Signature 0 April 14, 2003 Date MS U is an Affirmative Action/Equal Opportunity Institution 7* PLACE IN RETURN Box to remove this checkout from your record. TO AVOID FINES return on or before date due. MAY BE RECALLED with earlier due date if requested. DATE DUE DATE DUE DATE DUE 6/01 c:/CIRCJDateDue.p65-p.15 WHEN PERFORMANCE FAILS: EXPERTISE, ATTENTION, AND PERFORMANCE UNDER PRESSURE By Sian Leah Beilock A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of DOCTOR OF PHILOSOPHY Departments of Kinesiology and Psychology 2003 p€ III p00: pert indl pro SUI 199 in I de\ Phi uti do ABSTRACT WHEN PERFORMANCE FAILS: EXPERTISE, ATTENTION, AND PERFORMANCE UNDER PRESSURE By Sian Leah Beilock This work explored the cognitive mechanisms underlying pressure-induced performance decrements. Performance pressure is defined as an anxious desire to perform at a high level (Hardy, Mullen, & Jones, 1996). Choking, or performing more poorly than expected given one's level of skill, tends to occur in situations fraught with performance pressure (Baumeister, 1984; Beilock & Carr, 2001; Lewis & Linder, 1997). Self-focus or explicit monitoring theories of choking suggest that pressure- induced performance decrements result from the explicit monitoring and control of proceduralized knowledge that is best run off as an uninterrupted and unanalyzed structure (Baumeister, 1984; Beilock & Carr, 2001; Lewis & Linder, 1997; Masters, 1992). Conversely, distraction theories propose that pressure creates a dual-task situation in which skill execution and performance worries vie for the attentional capacity once devoted solely to primary task performance (Lewis & Linder, 1997; Wine, 1971). To date, explicit monitoring theories have accounted quite well for the choking phenomenon (see Appendix A and B). However, the extant choking literature has solely utilized sensorimotor skills as a test bed. Well-learned, proceduralized sensorimotor skills do not possess the right task control structures to choke according to distraction theories (Allport, Antonis, & Reynolds, 1972). Furthermore, unpracticed sensorimotor skills, allhou Wierc stomg scnsc aHlE \iatl COIII‘ prol deer C00 pH dii although based, in part, on explicitly accessible declarative knowledge (Beilock, Wierenga, and Carr, 2002), may not demand the type of processing and information storage that make a task susceptible to choking via distraction. Indeed, novice sensorimotor skills do not appear to be negatively impacted by performance pressure at all (Beilock & Carr, 2001). It remains an open possibility then, that choking may occur via the mechanisms proposed by distraction theories in certain tasks — for example, complex cognitive tasks not based on an automated or proceduralized skill representation. Four experiments examined performance under pressure in the mathematical problem solving task of modular arithmetic (MA). Exp. 1 demonstrated that performance decrements in difficult, unpracticed MA problems occurred under high pressure conditions. Exp. 2 demonstrated that these pressure—induced failures only occurred for the most difficult and capacity demanding unpracticed equations. Exp. 3 further explored these performance failures both early and late in learning. Similar to Exp. 2, only difficult problems with large on-line working memory demands choked. Furthermore, these failures were limited to problems early in practice when capacity-demanding rule—based solution algorithms governed performance. In Exp. 4, participants performed MA problems once, twice, or 50 times each, followed by a high pressure test. Again, only difficult problems that had not been highly practiced showed performance decrements. These findings support distraction theories of choking in the domain of mathematical problem solving. This outcome contrasts with sensorimotor skills, such as golf putting, in which the data have uniformly supported explicit monitoring rather than distraction theories (Beilock & Carr, 2001; Lewis & Linder, 1997). This contrast suggests a taxonomy of skills based on the nature and representation of their control structures. Copyright by Sian Leah Beilock 2003 To my brother, Mark Beilock, as a symbol of the fact that we can accomplish anything we work hard at. . .and that we should not let anxiety, worry, and the “what ifs” stand in our way. ”OI Fell IESE SUp my P5} me in l in l Till tha ant IE; ACKNOWLEDGMENTS There are many people whom, without their love and support, this work would not have been possible. First of all, I would like to thank my advisors, Tom Carr and Deb Feltz. Both Tom and Deb have been instrumental in my academic development, pushing me to not only to work hard, but to have confidence in myself and my abilities as a researcher. More than that, however, they have also provided invaluable emotional support and guidance at times when my anxieties and worries were getting the best of me. In addition to my advisors, countless others helped to keep me grounded during my tenure as a graduate student. My fellow graduate students in both Kinesiology and Psychology consistently remind me that there was a world outside my lab — and pushed me to get out and enjoy it every once in a while: Karin Allor Pfeiffer and Kevin Stefanek in Kinesiology and Monica Castelhano, Mareike Wieth, Laurie Carr, and Kate Arrington in Psychology. And outside of the academic arena, my friends Scott Kopchinski and Liz Thornton were always around to make sure that I didn’t take myself, or anything else for that matter, too seriously. My family offered unconditional support — my mother, Ellen Beilock, my father, Steve Beilock, my brother, Mark Beilock, my grandparents, George and Sylvia Elber, and Phyllis Beilock and Lou Baca, and my soon-to-be-husband Allen McConnell. Without your words of encouragement and your ability to make me feel better in almost any situation, I would have been lost. I would also like to mention all the strong women in my life who, whether they realized it or not, were amazing role models and inspirations to me. Women such as my vi adx'i succ who thel thy? we! not am ' “‘01" advisor, Deb Feltz, and Janet Starkes and Rose Zacks consistently demonstrated that a successful academic life was obtainable for a woman. And my mother and grandmothers who, through their own lives and explicit teachings, made sure that I was always aware of the fact that I had the ability to navigate any hurdles standing between me and my goals. Finally, I must mention Berkeley the dog. Anyone who knows me is familiar with my love/obsession for her. And at the same time, my friends and family also realize that I would have gone mad a long time ago without Berkeley’s companionship. Berkeley is not only there to comfort me when I am stressed or sad, but she serves a purpose that I am not sure any human could — Berkeley ensures that I actually leave the lab and my work once in a while! For that, I owe her my sanity. vii TABLE OF CONTENTS LIST OF TABLES ........................................................................... xiii LIST OF FIGURES ......................................................................... xiv 1 INTRODUCTION ..................................................................... 1 1.1 Theories of Choking under Pressure ............................................ 2 1.2 Support for Explicit Monitoring Theories of Choking ....................... 3 1.3 Is the Issue Settled? Differences Due to Task Control Structure ............ 1 l 1.4 Test Anxiety Literature ............................................................ 12 1.5 Overview of the Current Research .............................................. 15 2 CHOKING UNDER PRESSURE IN MATHEMATICAL PROBLEM SOLVING ............................................................................... 22 2.1 Experiment 1 ....................................................................... 22 2.1.1 Method ...................................................................... 23 2.1.2 Results ....................................................................... 27 2.1.3 Discussion .................................................................. 30 2.2 Experiment 2 ....................................................................... 32 2.2.1 Method ...................................................................... 33 2.2.2 Results ....................................................................... 34 2.2.3 Discussion .................................................................. 39 2.3 Experiment 3 ....................................................................... 41 2.3.1 Method ...................................................................... 42 2.3.2 Results ...................................................................... 45 2.3.3 Discussion .................................................................. 51 2.4 Experiment 4 ....................................................................... 52 2.4.1 Method ...................................................................... 53 2.4.2 Results ....................................................................... 55 2.4.3 Discussion .................................................................. 62 3 GENERAL DISCUSSION ........................................................... 67 3.1 Working Memory Intensive Tasks vs. Automated Skills ..................... 69 3.2 Cognitive vs. Sensorimotor Skills ............................................... 71 3.3 Future Directions .................................................................. 73 3.3.1 Choking via Explicit Monitoring and Distraction Mechanisms in one Task ..................................................................... 73 3.3.2 Individual Differences in Choking under Pressure .................... 74 3.3.3 Stereotype Threat as a Form of Choking under Pressure ............. 77 3.4 Conclusion .......................................................................... 80 viii REFERENCES .............................................................................. 8 1 APPENDIX A ............................................................................... 87 APPENDIX B ............................................................................... 1 13 ix LIST OF TABLES Table 1. Mean Modular Arithmetic Reaction Time (ms) and Accuracy (% correct) for the Low Pressure Group and High Pressure Group for the Single-Digit Problems and the Double-Digit Borrow Problems in the Pretest and Posttest Equation Blocks in Experiment 2 .................................... 37 Table 2. Mean Modular Arithmetic Reaction Time (RT) and Accuracy for the Low and High Pressure Tests for the No Repeat Problems (NR) and Multiple Repeat Problems (MR) with a Borrow Operation (Borrow) and without a Borrow Operation (No Borrow) for Experiment 4 ............................ 59 LIST OF FIGURES Figure 1. Mean accuracy (% correct) for the low pressure group and high pressure group in the pretest and posttest equation blocks in Experiment 1. Error bars represent standard errors ......................................................... 30 Figure 2. Mean accuracy (% correct) for the low pressure group and high pressure group for the single-digit problems (SD) and the double-digit borrow problems (DD-Borrow) in the pretest and posttest equation blocks in Experiment 2. Error bars represent standard errors ..................................... 39 Figure 3. Mean accuracy (% correct) for the low and high pressure tests prior to modular arithmetic training (Test 1) and following modular arithmetic training (Test 2) in Experiment 3. Error bars represent standard errors ............................................................................................. 47 Figure 4. Mean accuracy (% correct) for the low and high pressure tests prior to modular arithmetic training for the single-digit problems (SD) and the double-digit borrow problems (DD-Borrow) in Experiment 3. Error bars represent standard errors ................................................................ 49 Figure 5. Mean reaction time (ms) for the low and high pressure tests for the multiple repeat problems (Multiple Rep.), the once repeat problems (Once Rep.), and no repeat problems (No Rep.) in Experiment 4. Error bars represent standard errors ................................................................ 56 Figure 6. Mean reaction times (ms) for the low and high pressure tests for the multiple repeat problems without a borrow operation (MR-No Borrow) and the multiple repeat problems with a borrow operation (MR-Borrow), and for the no repeat problems without a borrow operation (NR-No Borrow) and the no repeat problems with a borrow operation (NR-borrow) in Experiment 4. Error bars represent standard errors ....................................... 61 xi pCIS Hun resu oi‘tei has 1 than “he & C bask golf in in Him Supc; donn Bros 1993 1995. relau. Situat im0 tl 1 INTRODUCTION The desire to perform at an optimal skill level in situations with a high degree of personally felt importance is thought to create performance pressure (Baumeister, 1984, Hardy, Mullen, & Jones, 1996). Paradoxically, despite the fact that performance pressure results from aspirations to function at one’s best, environments fraught with pressure are often where suboptimal skill execution is most visible. The term choking under pressure has been used to describe this phenomenon. Choking, defined as performing more poorly than expected given one’s level of skill, is thought to occur across diverse task domains where incentives for optimal performance are at a maximum (Baumeister, 1984; Beilock & Carr, 2001; Lewis & Linder, 1997; Masters, 1992). We often refer to the “bricks” in basketball free throw shooting when the game winning shot is on the line, or the “yips” in golf putting when an easy 3 foot putt to win the tournament stops short, and to “cracking” in important test-taking situations where a course grade or college admission is at stake as unmistakable instances of such incentive or pressure-induced performance decrements. Surprisingly, while research concerning the cognitive mechanisms governing superior task performance is abundant across both cognitive and sensorimotor skill domains (Allard & Starkes, 1991; Anderson, 1982; 1983; Anderson & Lebiere, 1998; Brown & Carr, 1989; Ericsson & Charness, 1994; Ericsson, Krampe, & Tesch-Romer, 1993; Fitts & Posner, 1967; Logan, 1988; Newell & Rosenbloom, 1981; Proctor & Dutta, 1995; Reimann & Chi, 1989; Rosenbaum, Carlson, Gilmore, 2001; Staszewski, 1988), relatively less attention has been devoted to suboptimal skill execution — especially in situations in which optimal task performance is not only desired, but expected. Insight into the mechanisms governing execution failure is important, as it will not only serve to In! re: pH of 1.1 for [ht ab an [ht I10 a d CM “h 311: 19 me the further our understanding of the variables responsible for skill decrements, but those responsible for success as well. A careful cognitive analysis of the choking under pressure phenomenon may open a new kind of window to the organization and operation of the information processing mechanisms that underlie skilled performance. 1.1 Theories of Choking under Pressure Why do skills fail in high pressure situations? Two main theories have been put forth as explanations for the choking phenomenon. Self-focus or explicit monitoring theories propose that performance pressure increases anxiety and self-consciousness about performing correctly, which in turn enhances the attention paid to skill processes and their step-by-step control. Attention to performance at this component level is thought to disrupt the proceduralized or automated processes of high level skills that normally run outside the scope of working memory during performance (Baumeister, 1984; Beilock & Carr, 2001; Butler & Baumeister, 1998; Kimble & Perlmuter, 1970; Langer & Irnber, 1979; Lewis & Linder, 1997; Marchant & Wang, in press; Masters, 1992). Distraction theories, on the other hand, suggest that performance pressure creates a distracting environment that competes with the attention normally allocated to skill execution (Wine, 1971). In essence, pressure serves to create a dual-task environment in which controlling execution of the task at hand and performance worries vie for the attentional capacity once devoted solely to primary task performance (Lewis & Linder, 1997). Thus distraction and explicit monitoring theories offer contrasting accounts of the mechanisms responsible for performance decrements under pressure. While distraction theories suggest that pressure creates a distracting environment that draws attention away not test und bye decr COIIC ordir chok Entfi eXPO: acnxi Picssi peffigl Iheori CODsci b€lng deSIgn from primary skill execution, explicit monitoring theories suggest the opposite - that pressure prompts too much explicit attention to performance processes and procedures. 1.2 Support for Explicit Monitoring Theories of Choking Training studies. To date, explicit monitoring theories have received the most noted support in accounting for choking under pressure. For example, in an attempt to test the two main theories of choking, Beilock and Carr (2001) examined skill execution under pressure in a golf putting task (for the full text of this paper, see Appendix A). The goal was to determine whether practice at dealing with the causal mechanisms proposed by each theory (i.e., explicit attention vs. distraction) would reduce performance decrements in high pressure environments. Individuals were trained to a high putting skill level under a variety of learning conditions and then exposed to a pressure situation. The first training condition involved ordinary single—task practice, which provided a baseline measure of the occurrence of choking. The second training condition involved practice in a distracting, dual—task environment (while monitoring an auditory word list for a target word) designed to expose performers to being distracted from the primary task by execution-irrelevant activity in working memory — the specific cause of performance decrements under pressure according to distraction theories. The third training condition exposed performers to the particular aspects of high pressure situations that explicit monitoring theories of choking propose as the cause of performance decrements. In this “self- conscious” or “skill focus” training condition, participants learned the putting task while being videotaped for subsequent public analysis by experts. This manipulation was designed to expose performers to having attention called to themselves and their pcdb Folk) b)'at tangle entin ngnif innne Ihaue [his :11 innne. Hannr pfirfOr Consci OI OVe mCCha decren “figaut trained fiWaren‘ anahSIs performance in a way intended to induce explicit monitoring of skill execution. Following training, all groups were exposed to the same high pressure situation created by a performance-contingent monetary award. Choking occurred for those individuals who were trained on the putting task in a single-task isolated environment, and also for individuals trained in a dual-task environment that simply created distraction. That is, both of these groups putted significantly less accurately in the high pressure situation than they had in an immediately preceding block of putts during which no pressure had been applied. However, choking did not occur for those trained in the self-conscious condition. Indeed, this group performed slightly better in the pressure situation than they had in the immediately preceding no-pressure trials. Beilock and Carr (2001) concluded that training under conditions that prompted attention to skill parameters served to adapt these performers to the type of attentional focus that often occurs under pressure. That is, self- consciousness training served to inoculate individuals against the negative consequences of over-attending to well-leamed proceduralized performance processes — the precise mechanisms that explicit monitoring theories suggest are responsible for performance decrements in high pressure situations. Previously, Lewis and Linder (1997) had also found that learning a golf putting skill in a self—awareness-heightened environment inoculates individuals against the negative effects of performance pressure at high levels of practice. Participants were trained on a golf putting task under either normal, single-task conditions or under a self- awareness condition (in which individuals putted while being videotaped for later analysis by golf professionals) and then exposed to a high pressure situation. Similar to Beilc those the n intror press Mullt induc perfo explit EXZuni IEMC Beilock and Carr (2001 ), Lewis and Linder demonstrated that pressure caused choking in those individuals who had not been adapted to self-awareness (i.e., participants trained in the normal, single-task situation). Furthermore, Lewis and Linder found that the introduction of a secondary task (counting backward from 100) while performing under pressure helped to alleviate these performance decrements (for confirmatory data, see Mullen & Hardy, 2000). Because the secondary task served to prevent the pressure- induced instantiation of maladaptive explicit attention to well-learned proceduralized performance processes, choking under pressure was assuaged — a finding consistent with explicit monitoring theories. If explicit monitoring theories are correct, then high level sensorimotor skills such as the golf putting task described above should be more susceptible to the negative consequences of performance pressure than less practiced performances. This is due to the fact that the former, but not the latter, are thought to operate outside of working memory, largely devoid of step-by-step attentional control (Anderson, 1982, 1993; Fitts & Posner, 1967; Proctor & Dutta, 1995). To the extent that performance pressure prompts explicit attention to execution, those skills not normally attended in real time (e.g., high level sensorimotor skills) should be more harmed by pressure-induced control than less practiced skills. This finding would be consistent with the idea that a majority of the evidence for choking has been derived from well-learned sensorimotor tasks that automate via proceduralization with extended practice (Marchant & Wang, in press). To test this prediction, Beilock and Carr (2001) conducted a second study examining performance under pressure at both low and high skill levels. Participants learned a golf putting skill to a high level and were exposed to a high pressure situation bouie perfor enicrg affectc‘ pressu exphcl (1997) inocuh Ofprac mechat dnecd} Recenfl IOCUs 0 novice COmple Predicti meChan Skill-r0, Condlik both early and late in practice. Early in practice, pressure to do well actually facilitated performance. At later stages of learning, performance decrements under pressure emerged. Thus, the proceduralized performances of experts appear to be negatively affected by performance pressure. However, novice skill execution is not harmed by pressure-induced attention to execution, as less skilled performances are already explicitly attended in real time. Attentional focus studies. Both Beilock and Carr (2001) and Lewis and Linder (1997) have demonstrated that skill training that induces attention to performance may inoculate individuals against the negative effects of performance pressure at high levels of practice. While these types of training methods lend insight into the cognitive mechanisms driving skill failure in high stakes situations, it may also be possible to more directly assess the processes responsible for pressure-induced performance decrements. Recently, Beilock, Carr, MacMahon, and Starkes (2002) manipulated the attentional focus of experienced golfers while performing a putting task and the attentional focus of novice and experienced soccer players while performing a soccer dribbling task (for a complete report of this work, see Appendix B). The goal was to directly test the predictions of explicit monitoring and distraction theories regarding the causal mechanisms of choking in sensorimotor skills. In Beilock et al.’s (2002) first study, experienced golfers took a series of putts in a skill-focused attention condition and a dual-task attention condition. The order of these conditions was counterbalanced across participants. In the skill-focused condition, participants were instructed to attend to a particular component of their golf putting swing. Specifically, individuals were instructed to monitor the swing of their club and at the e to as draxx moni simu Pdflli speci inten in the puttin Condi Signif (Alldc (:arr,; pIOCeS audliol the exact moment they finished the follow-through of their swing, bringing the club head to a stop, to say the word "stop" out loud. This skill-focused condition was designed to draw attention to a component process of performance, coinciding with explicit monitoring theories. The dual-task attention condition involved putting while simultaneously listening to a series of recorded tones being played from a tape recorder. Participants were instructed to monitor the tones carefully, and each time they heard a specified target tone, to say the word “tone” out loud. The dual-task condition was intended to distract attention away from skill execution, in line with the choking mechanisms proposed by distraction theories. Results demonstrated that the experienced golfers performed significantly better in the dual-task condition in comparison to the skill-focused condition. Additionally. putting in the skill-focused condition was less accurate than a single-task practice condition used as a baseline measure. Performance in the dual-task condition did not significantly differ from this practice condition. Because well-leamed golf putting does not require constant on-line control (Anderson, 1982, 1983, 1993; Beilock, Bertenthal, McCoy, & Carr, in press; Beilock & Carr, 2001; Fitts & Posner, 1967; Proctor & Dutta, 1995), attention is available for the processing of secondary task information if necessary (such as monitoring a series of auditory tones). As a result, performance does not suffer with the addition of dual-task demands. It should be noted that this may not hold true for dual-task environments in which the tasks draw upon similar processes and hence create structural interference (e. g., looking at a golf ball while lining up a putt and visually monitoring a screen for a target object). When prompted to attend to a specific component of the golf swing, howeVi task C0 chokin‘ attendi and du; and im Additic di fferi 1‘. both qr differer SUl—‘POn anende 1967; r the um, 0P€rate Keele 5 eXecum the” in. however, experienced performance degrades in comparison to both single-task and dual- task conditions. This pattern of results coincides with explicit monitoring theories of choking suggesting that well-learned performances may actually be compromised by attending to skill execution. In a second study, Beilock et a1. (2002) again explored the impact of skill-focused and dual-task attention conditions, but in a movement skill that uses different effectors and imposes different temporal demands than golf putting — soccer dribbling. Additionally, the effects of dual-task and skill-focused attention on performance at differing levels of soccer skill proficiency were also explored. Skill acquisition is believed to progress through distinct phases characterized by both qualitative differences in the cognitive structures supporting performance and differences in performance itself. Early in learning, skill execution is thought to be supported by a set of unintegrated control structures that are held in working memory and attended to one—by-one in a step-by-step fashion (Anderson, 1983, 1993; Fitts & Posner, 1967; Proctor & Dutta, 1995). With practice, however, procedural knowledge specific to the task at hand develops. Procedural knowledge does not require constant control and operates largely outside of working memory (Anderson, 1993; Fitts & Posner, 1967; Keele & Summers, 1976; Kimble & Perlmuter, 1970; Langer & Imber, 1979). Because novices must devote attentional capacity to task performance in ways that experts do not (Fitts, 1964; Fitts & Posner, 1967), novices and experts may be differentially affected by conditions that either draw attention away from, or toward, skill execution. Specifically, the capacity-demanding performance of novices may not afford these individuals the attentional resources necessary to devote to secondary task demands amen executi condnn This is perforn perforn executi IO COIII WCI‘C Ll: second execut Rmusm reCentl Perforr el‘ipern Skill in that m if the situation requires it. However, the novice, who must attend to the steps of skill execution in order to succeed, might not be harmed or could perhaps be helped by conditions that focus attention more squarely on the skill and prevent it from wandering. This is in contrast to the impact of skill-focused and dual-task attention on experienced performance, as seen in the golf putting study described above. Here, the proceduralized performances of experts are not harmed by dual-task conditions. Yet, high level execution is negatively impacted by skill-focused conditions designed to prompt attention to component parts of performance not normally attended to in real time. In Beilock et al.’s (2002) second study, novice and experienced soccer players were asked to dribble a soccer ball through a series of pylons while either performing a secondary auditory monitoring task (designed to distract attention away from skill execution, in line with distraction theories’ proposed choking mechanisms) or a skill- focused task in which individuals were asked to monitor the side of the foot that most recently contacted the ball (designed to draw attention to a component process of performance, coinciding with explicit monitoring theories). As in the first golf putting experiment, the order of these attention conditions were counterbalanced across participants. Similar to the results from the golf putting study described above, performing in a distracting, dual-task environment did not harm experienced soccer players’ dribbling skill in comparison to the skill-focused condition. However, when the soccer players were instructed to attend to step-by-step performance (i.e., monitoring the side of the foot that most recently contacted the ball), their dribbling skill deteriorated in comparison to both the dual-task condition and a single-task practice condition used as a baseline measure. Novices showed the opposite pattern. These less skilled individuals performed at a lower level in the dual-task condition, designed to distract attention away from performance, in comparison to the skill-focused manipulation, designed to draw attention toward the task at hand. Furthermore, novices performed at a higher level in the skill- focused condition than in the single-task practice condition. Consistent with the evidence presented above in support of explicit monitoring theories of choking, step—by-step attention to skill processes and procedures does not appear to harm novice sensorimotor skill execution that is already explicitly attended in real time. However, this same type of attentional control disrupts or slows down well- learned sensorimotor skill performance thought to normally operate largely outside of working memory (Baumeister, 1984; Beilock & Carr, 2001; Lewis & Linder, 1997). The negative effects of enhanced attention to highly skilled performance can not only be seen in complex tasks such as golf putting and soccer dribbling, but in more basic skills we use everyday. For example, Wulf and colleagues have suggested that directing performers’ attention to their movements through “internal focus” feedback on a dynamic balance task interferes with the automated control processes that usually control balance movements outside of conscious scrutiny (Wulf & Prinz, 2001). It should be noted that novice soccer dribbling performance was harmed by dual- task demands. This result suggests that distraction theories might be a viable explanation for performance decrements under pressure in novel sensorimotor skill execution. However, as seen in Beilock and Carr’s (2001) work presented above, novice motor skill 10 execution does not appear to choke under pressure. It may be that explicit monitoring theories are the most appropriate explanation for performance decrements under pressure, or, alternatively, that novice sensorimotor skills do not demand the type of processing and information storage that makes a task susceptible to choking via distraction. This is an issue to which we turn to below. 1.3 Is the Issue Settled? Differences Due to Task Control Structure The work reviewed above suggests that maladaptive explicit monitoring may be responsible for choking under pressure. Given the consistency of this evidence, it may seem unlikely that distraction theories could provide any additional insight into the choking phenomenon. However, the automated or proceduralized sensorimotor skills predominantly used in the extant choking research may not possess the right task control structures to be susceptible to pressure-induced performance decrements according to distraction theories. These proceduralized sensorimotor skills are thought to run outside of working memory and are largely robust to performance decrements as a result of distracting, dual-task situations (Allport, Antonis, & Reynolds, 1972; Beilock et a1., 2002; Beilock, Wierenga, & Carr, 2002, in press; Keele & Summers, 1976; Kimble & Perlmuter, 1970; Langer & Irnber, 1979; Leavitt, 1979; Smith & Chamberlin, 1992). Furthermore, even unpracticed sensorimotor skills, although based, in part, on explicitly accessible declarative knowledge (Beilock, Wierenga, and Carr, 2002) may not require the type of sequentially dependent interweaving of processing and information storage that make a task susceptible to choking via distraction. Thus, one reason why distraction theories of choking may not have received much support is because they have not been tested in the appropriate skill domains. 11 L4 Ike pressure and info anxiety I intrusive Eysenck Because devoted available on proce this decr. (Ashcrat‘ Hoboak. R lImited C math am [ask IIIVQ Difficult) 111 an 3119] math amu lnCrEaSed dn‘x'et)’. Vt 1.4 Test Anxiety Literature There is a literature that we can look to, however, for clues concerning how a pressure situation might influence skills with sequentially dependent on—line processing and information storage demands susceptible to capacity limitations. Within the test anxiety literature researchers have suggested that anxiety manifests itself in the form of intrusive thoughts or worries about the situation and its outcome (Ashcraft & Kirk, 2001; Eysenck, 1979, 1992; Eysenck & Calvo, 1992; Eysenck & Keane, 1990; Wine, 1971). Because these thoughts are attended to, a portion of working memory capacity normally devoted to primary skill execution is consumed by such thoughts, and therefore not available for the processing of task-relevant cues. For tasks with interdependent demands on processing and storage that rely heavily on working memory for on—line execution, this decrease in capacity is thought to result in suboptimal performance outcomes (Ashcraft & Kirk, 2001; Darke, 1988; Leon, 1989; Sorg & Whitney, 1992; Tohill & Holyoak, 2000). Recent research in the math problem solving literature has found support for this limited capacity theory. Ashcraft and Kirk (2001) examined the ability of low and high math anxious individuals to simultaneously perform a mental addition task and a memory task involving the maintenance of random letter strings in memory for later recall. Difficulty levels of both the primary math and secondary memory tasks were manipulated in an attempt to examine the effects of task difficulty on performance as a function of math anxiety. Results demonstrated that performance was lowest (mainly in the form of increased math task error rates) in instances in which individuals, regardless of math anxiety, were required to perform both a difficult math and difficult memory task 12 simultaneously. Furthermore, in comparison to less anxious individuals, those participants high in math anxiety showed an exaggerated increase in performance errors under the difficult math and memory task condition. The authors concluded that performance deficits under demanding dual-task conditions were most pronounced in individuals high in math anxiety, as anxiety, similar to the instantiation of a secondary demanding task, drains the attentional capacity that might otherwise be available for primary skill performance. Anxiety has also been shown to lead to performance decrements in capacity- demanding analogical reasoning tasks. Tohill and Holyoak (2000) explored the relationship between anxiety and the ability to make attributional and relational mappings between objects. Attributional mapping involves mappings based on the physical characteristics of individual objects (e.g., a woman in picture A might map to a woman in picture B). In contrast, relational mappings are based on relationships linking multiple objects together (e. g., a dog being held by a woman in picture A might map to a baby being held by a woman in picture B, as both the dog and the baby are being held by a woman). Because attributional mapping requires that only a single object characteristic be held in memory, while relational mapping requires that an individual maintain a number of object relations in memory, this latter form of mapping is thought to impose a heavier load on working memory (Halford, 1993; Tohill & Holyoak, 2000). If anxiety serves to limit working memory capacity through the instantiation of attention- demanding situational worries (Ashcraft & Kirk, 2001; Eysenck, 1979, 1992; Eysenck & Calvo, 1992; Eysenck & Keane, 1990; Wine, 1971), then high anxiety individuals may have difficulty performing memory-intensive relational mappings more so than 13 attributional mappings, as both the relational mapping task and anxiety should compete for attentional capacity. Tohill and Holyoak (2000) presented individuals with pairs of analogical pictures and asked them to perform mappings between a specific object in one picture to an object in a second picture. Highly anxious individuals were less likely to generate relational mappings in comparison to less anxious individuals. This pattern of results occurred even when individuals were given explicit instructions to use relational mapping techniques. Thus, similar to the effects of anxiety on the simultaneous performance of a difficult mental addition and memory task, in memory manipulation and maintenance intensive relational mapping procedures, anxiety also appears to restrict the working memory capacity required for successful skill execution. Research stemming from the test anxiety literature lends support to the notion that performance decrements may result from anxiety-induced worries that decrease task— relevant processing resources. If, as proposed by distraction theories, pressure serves to create a distracting environment via worries about the situation and its consequences, then pressure may impose constraints on working-memory-intensive tasks similar to those of anxiety as outlined above. Under this view, skills that have become proceduralized with extended practice and do not rely heavily on explicit attentional processes for successful skill execution (e. g., the well-leamed golf putting or soccer tasks mentioned above), as well as skills that do not possess the appropriate types of inter- dependent processing and information storage demands (e.g., the novice soccer task mentioned above), may not be harmed by performance pressure. The lack of support for 14 distr; under 1.5 I task it accord autom; levels < analysi accoun into p0. NIodul; number and b is (mod4 (Le.,51 equatio [he equ distraction theories in the choking literature then, may be a function of the types of skills under investigation. 1.5 Overview of the Current Research The aim of the current work was to examine skill performance under pressure in a task with working memory requirements that are likely to make it susceptible to choking according to distraction theories — at least at low levels of learning prior to skill automatization. It may be that such a task would also be susceptible to choking at higher levels of practice, but due to explicit monitoring rather than distraction. This type of analysis will not only shed light on the ability of these theories to successfully predict and account for performance decrements in high pressure situations, but may also lend insight into possible differences in their domains of applicability. I chose Gauss’s ( 1801) modular arithmetic task (Bogomolny, 1996) as a test bed. Modular arithmetic is defined as a particular sequence of arithmetic operations. Two numbers a and b are said to be equal or congruent modulo N if the difference between a and b is exactly divisible by N. For example, to verify the modular equation “51 E 19 (mod 4),” the middle number of the equation is first subtracted from the far left number (i.e., 51 - 19), and then this answer is divided by the far right number of the original equation (i.e., 32 + 4). Because the result of this division step is a whole number (i.e., 8), the equation is true. How does the modular arithmetic task change with practice? To answer this question we may need to draw on a different conceptualization of skill automaticity than the model of automaticity via proceduralization described in the introduction and often applied to sensorimotor skill acquisition (e.g., Fitts & Posner, 1967). According to 15 Logan’s (1988) instance-based theory of automaticity — which was originally developed to account for the data from mental arithmetic-type tasks — a rule-based algorithm should be initially employed to solve unpracticed modular equations. In this unpracticed state, problem solutions are dependent on the explicit application of a capacity-demanding step-by-step process that must be maintained and controlled on-line by working memory during execution. With practice on particular problems, the reliance on this procedure decreases and past instances of problem solutions are retrieved directly or “automatically” from long term memory, whereas new problems continue to engage the algorithm. Logan’s model proposes an alterative view of automaticity to traditional theories of proceduralization and has been quite successful in accounting for changes in speed and accuracy of performance with practice on cognitive tasks such as alphabet arithmetic, lexical decision, and semantic categorization. Nonetheless, contrary to Logan’s (1988) predictions, there is some evidence that unfamiliar problems still based on algorithmic computations do become more efficient with practice of the algorithm (Touron, Hoyer, & Cerella, 2001). Even practiced algorithms, however, may not be governed by the same type of control structures as, for example, proceduralized motor programs. An algorithmic solution procedure, regardless of each component’s efficiency, is based on a hierarchical and sequentially dependent task representation in which initial steps must be held and acted on in working memory in order to generate subsequent processes and final solutions. Well-learned sensorimotor skills that operate largely outside of working memory (Fitts & Posner, 1967; Proctor & Dutta, 1995) are most likely not governed by 16 the wil and men‘ press arithn harme harm u notneg On‘line. actually . execution differed Ill l9] the same type of working-memory-dependent representation. This is an issue to which we will return later. In Experiment 1, participants were assigned to either a low or high pressure group and performed a series of unpracticed modular arithmetic equations. From the standpoint of distraction theories, modular arithmetic is a good candidate for choking at low levels of practice when the attentional demands of problem solving are high and pressure- induced worries compromise task-relevant processing resources. This should be especially true for the problems utilized in Experiment 1. All of the equations consisted of double-digit numbers with a borrow operation (e.g., 43 = 28 (mod 5)). Large numbers and borrow operations are thought to increase task difficulty and on-line working memory demands (Ashcraft, 1992; Ashcraft & Kirk, 2001). Thus, if performance pressure impinges on the processing resources needed to successfully solve modular arithmetic equations, difficult problems of the type used in Experiment 1 are likely to be harmed. In contrast, explicit monitoring theories would suggest that pressure should not harm unpracticed modular equations -— as pressure-induced attention to execution should not negatively impact information that is already explicitly attended to and maintained on-line. Indeed, as mentioned above, Beilock and Carr (2001) have found that pressure actually enhances the performance of novices in golf putting, despite harming the execution of more practiced individuals. In Experiment 2, participants were again assigned to either a low or high pressure group prior to performing a series of unpracticed modular arithmetic equations. In contrast to Experiment 1, the modular arithmetic equations utilized in Experiment 2 differed in terms of the demands these problems placed on working memory. If, as 17 dist res; unds dent (Son‘ prob re gar press 'dllt‘l'lt' preSSL anthn levels induce d€pent distraction theories would propose, pressure-induced limitations on working memory are responsible for choking, then difficult equations with large capacity demands should fail under pressure. However, easier problems that do not impose such heavy attentional demands should be less susceptible to pressure-induced performance decrements. Conversely, if explicit monitoring theories are correct in the domain of mathematical problem solving, then the unpracticed modular equations utilized in Experiment 2, regardless of problem difficulty, should not fail under pressure. As mentioned above, pressure-induced attention to skill execution should not harm novel equations normally attended on-line. Experiment 3 extended the first two experiments’ exploration of choking under pressure to highly practiced equations. Distraction theories propose that modular arithmetic should be most susceptible to performance decrements under pressure at low levels of practice. Following extended practice on specific problems however, pressure- induced distraction should not harm execution as answer derivation is no longer dependent on the intermediate memorial maintenance of task information. To the extent that the control structure of modular arithmetic changes to automatic answer retrieval with practice, explicit monitoring theories might make a different prediction —— at least according to Masters, Polman, and Hammond (1993). At high levels of practice, pressure “may result in a return to an explicit, algorithmic-based control of behavior through disruption of automatic retrieval of skill-based information from memory” (Masters et al., 1993, p.664). Such a regression, if it occurred, would slow performance and increase the opportunity for error — which would create poorer performance. Furthermore, if rather than a shift to automatic answer retrieval as Logan l8 (1988 mgofi presst induo reheat dufen Indhd expos rcpeai inman Probli exper lSdUe (1988) would propose, modular arithmetic automates via proceduralization of the algorithm, explicit monitoring theories would still predict performance decrements under pressure at high levels of practice. In this case, such failures might be due to pressure- induced attentional control that increases the time or error associated with maintaining, rehearsing, or acting on the well-leamed algorithm. Experiment 4 further examined susceptibility to choking under pressure following different amounts of exposure to specific problems within the modular arithmetic task. Individuals performed over 800 modular arithmetic practice problems and were then exposed to a high pressure environment. Specific problems were presented either once, repeated twice, or repeated fifty times each during practice. According to Logan’s (1988) instance-based theory of automaticity, the task control structures of modular arithmetic problems should only change as a function of specific problem exposure, not necessarily experience at performing many different modular arithmetic problems. Thus, if choking is due to pressure-induced capacity limitations, as distraction theories would propose, then regardless of how many different problems individuals have been exposed to, only those equations that have been practiced enough to produce instance-based answer retrieval (a minimum of 36 to 72 exposures according to Klapp, Boches, Trabert, and Logan, 1991), should be inoculated against the detrimental capacity-limiting effects of performance pressure. In contrast, explicit monitoring theories would again make an opposite prediction. Namely, pressure-induced attention will harm those problems that have been repeatedly practiced to the level of automatic answer retrieval. As mentioned above, there is some evidence that unfamiliar problems still based on algorithmic computations do become more efficient with practice of the algorithm 19 (Tour impO? under com; me m- perfo probl decre the ”It possib fail uni monitor serves 1, much lil to some t memOry( maladapn problemS f Wantiatio perfOnnanCc the fact (hat 6 (Touron, Hoyer, & Cerella, 2001). However, even practiced algorithms should still impose attentional demands, and thus show susceptibility to performance decrements under pressure according to distraction theories. This is due to the fact that the component parts of any algorithmic solution procedure must still be held in working memory during on-line performance. Such a prediction can be tested by examining performance decrements under pressure as a function of problem difficulty. If novel problems based on practiced algorithms are susceptible to pressure-induced performance decrements via distraction, then the more difficult and capacity-demanding the problem, the more vulnerable the equation should be to capacity—limited failure. If practice increases the proficiency of algorithmic computations, then it is also possible that unfamiliar problems based on highly practiced algorithmic procedures might fall under pressure as a result of the attentional control mechanisms proposed by explicit monitoring theories. That is, such problems might fail via pressure-induced attention that serves to disrupt or slow down well-learned and highly efficient performance processes — much like choking under pressure in well-leamed sensorimotor skills. This should be true to some extent across all problems, regardless of equation difficulty level or working memory demands. While easier problems may be less susceptible to failure as a result of maladaptive attentional control than their more demanding counterparts, as these problems have fewer independent steps and thus less of an opportunity for the instantiation of pressure-induced control, there should still be at least some sign of performance decrements under pressure in these less demanding problems. This is due to the fact that all novel equations based on practiced algorithms require several steps to 20 mfinevr induce pressU' proper leannr achieve a problem solution and thus should be at least somewhat susceptible to pressure- induced attention that disrupts proceduralized algorithmic processes. I now turn to Experiment 1 in which I initially explored performance under pressure at low levels of practice in modular arithmetic -— a skill that should have the right properties to be susceptible to pressure-induced performance decrements during initial learning according to distraction, but not explicit monitoring theories. 21 ZCH 2.1 E high pi equafir the sec perfori was in pressu equan. prohle might C0nsis mfintit (1111ch is Yes; most 1 2 CHOKING UNDER PRESSURE IN MATHEMATICAL PROBLEM SOLVING 2.1 Experiment 1 In Experiment 1 individuals were randomly assigned to either a low pressure or high pressure group prior to performing three blocks of novel modular arithmetic equations. The first block of equations served as a pretest measure of performance and the second block served as a small amount of practice at the algorithm to stabilize performance. Immediately preceding the last block of equations, the low pressure group was informed that they would be performing another set of equations, while the high pressure group was given a scenario designed to create a high pressure environment. From the standpoint of distraction theories, unpracticed modular arithmetic equations that are dependent on a working-memory-intensive rule-based algorithm for problem solutions should be susceptible to performance decrements under pressure. This might be especially true for the problems utilized in Experiment 1, as all of the equations consisted of double-digit numbers with a borrow operation (e. g., 43 = 28 (mod 5)). As mentioned above, large numbers and borrow operations are thought to increase task difficulty (Ashcraft, 1992; Ashcraft & Kirk, 2001). Thus, if pressure-induced distraction is responsible for choking, difficult equations with large capacity demands should be most likely to fail. If, however, explicit monitoring theories are correct in the domain of mathematical problem solving, and performance pressure prompts explicit attention to skill execution processes, then the unpracticed modular equations utilized in Experiment 1 should not be harmed by increased pressure. Here, pressure-induced attention to execution should not impact information that is already explicitly attended on line. 22 IQ eran POSSH PTOhle Slandm 2.1.1 Method Participants. Participants were students enrolled at Michigan State University who were not math majors, had taken no more than 2 introductory college-level math courses, and had no previous exposure to the modular arithmetic (MA) task. Participants were randomly assigned to either a low pressure group (3:18) or a high pressure group (3:18) provided their accuracy in the pretest block was greater than 80%. A minimum accuracy criterion was implemented in order to assure that the low and high pressure groups in Experiment 1 consisted of individuals with similar levels of modular arithmetic ability. This allowed for the inference that any difference in MA performance as a function of pressure group could be attributed to our pressure manipulation rather than to random disparity in group ability. ' Procedure. Participants filled out a consent form and a demographic sheet detailing previous math experience. They were informed that the purpose of the study was to examine how individuals learn a new math skill. Participants were set up in front of a monitor controlled by a standard laboratory computer and introduced to MA through a series of written instructions presented on the computer screen. They were informed that they would be judging the validity of MA equations and provided with several examples. Participants were instructed to judge the validity of the equations as quickly as possible without sacrificing accuracy and when they had derived an answer to the problem presented on the screen, to press the corresponding “T” or “F” keys on a standard keyboard set up in front of them. The stimuli were digits, the word “mod” designed to denote the modular arithmetic statement, and a congruence sign (E). Each trial began with a fixation point 23 expo rephi \tord whet] insin sepan (fight across “met each p counts (hill pe eqUaU( Perform Snnph. POSSlbl( high We Piscedrr group w. exposed for 500 ms in the center of the screen. The fixation point was immediately replaced by an equation, which remained on the screen until the participant pressed the “T” or “F” key. When the participant responded, the equation was extinguished and the word “Correct” or “Incorrect” was displayed on the screen for 1,000 ms, indicating whether the problem had been solved correctly. The screen then went blank for a 1,000 ms inter—trial interval. All participants performed three blocks of 24 modular arithmetic problems each, separated by a short break of approximately 1 minute. All equations required a double- digit borrow subtraction operation (e. g., 51 E 19 (mod 4)) and were presented only once across the entire experiment. Half of the problems within each block were true and half were false. Equations within each block were presented in a different random order to each participant. Additionally, the equations presented in the last two blocks were counterbalanced across participants. This counterbalancing was done in order to assure that performance in the last block of equations was independent of the particular equations to which individuals were exposed. The first block of equations served as a pretest measure of modular arithmetic performance (pretest block) for both the low and high pressure groups. Individuals were simply informed to perform as best they could - solving the equations as quickly as possible without sacrificing accuracy. Similar instructions were given to both the low and high pressure groups prior to the second, practice block of equations. Immediately preceding the last block of equations (posttest block), individuals in the low pressure group were simply informed that they were going to be performing another set of 24 pmh CI'CBI formi score to the that r: panici had in ImprQ did ng Final] SIIUarj perf0[ Panic both I COmp Off [h problems while individuals in the high pressure group were given a scenario designed to create a high pressure situation. The high pressure group participants were informed that the computer used a formula that equally takes into account reaction time and accuracy in computing an “MA score.” Participants were told that if they could improve their MA score by 20% relative to the preceding practice trials, they would receive $5. Participants were also informed that receiving the monetary award was a “team effort.” Specifically, individuals were told that they had been randomly paired with another participant, and in order to receive their $5, not only did the participant presently in the experiment have to improve in the next set of problems, but the individual they were paired with had to improve as well. Next, participants were informed that their partner had already completed the experiment, and had improved by the required amount. If the participant presently in the experiment improved by 20%, both participants would receive $5. However, if the present participant did not improve by the required amount, neither participant would receive the money. Finally, participants were told that their performance would be video taped during the test situation so that local math teachers and professors in the area could examine individuals’ performances on this new type of math task. The experimenter set up the video camera on a tripod directly to the left of participants approximately 0.61 m away. The field of view of the video camera included both the participant and the computer screen. Participants in the high pressure group completed the last block of modular arithmetic equations. The experimenter then turned off the video camera and faced it away from the participants. 25 aware \& hetl' presSL seen 11 of indi acaden 1617115 C pressurr anxiety Trait A] SIdIC-T] COHSisti In time “I feel a State Ft anley HOIFOal POsttegl It is an empirical question whether different types of pressure (e.g., monetary awards, peer pressure, or social evaluation) have equivalent effects on performance, and whether all pressures exert their effects via similar processes. The goal of the current pressure manipulation was to impose a high level of pressure using sources commonly seen in everyday life. For example, in athletics, team success is based on the performance of individual athletes and this performance is often scrutinized by others. And in more academic arenas, college entrance exam performance has monetary consequences in terms of scholarships and future educational opportunities. Following completion of the MA task, participants in both the low and high pressure groups filled out a number of questionnaires designed to assess their feelings of anxiety and performance pressure. Individuals first filled out the State Form of the State- Trait Anxiety Inventory (Spielberger, Gorsuch, & Lushene, 1970). The State Form of the State-Trait Anxiety Inventory (STAI) is a well-known measure of state anxiety, consisting of 20 questions designed to assess participants’ feelings at a particular moment in time. Individuals are instructed to assign a value to questions such as, “I feel calm” and “I feel at ease” on a 4 point scale ranging from 1. (not at all) to 4 (very much so). The State Form of the STAI has been used in a number of studies investigating the impact of anxiety on complex task performance (e.g., in analogical reasoning ability, see Tohill & Holyoak, 2000). Following the STAI, participants answered a number of questions related to their perceptions of performance in the posttest. Specifically, individuals were asked on a 7 point scale (a) how important they felt it was for them to perform at a high level in the posttest — ranging from 1 (not at all important to me) to 7 (extremely important to me), 26 (bi ht“ rangir and (C pOOFI' award 2.1.2 absolu ahhoug 3.94. S ‘ Perforn more p. the 10“- Signing COmpar F(134): impona that II W aIIXIetV : (b) how much performance pressure they felt to perform at a high level in the posttest — ranging from 1 (very little performance pressure) to 7 (extreme performance pressure), and (c) how well they thought they performed in the posttest —- ranging from 1 (extremely poor) to 7 (extremely well). Individuals were then fully debriefed and given the monetary award, regardless of their performance. 2.1.2 Results Questionnaires: The high pressure group (M = 38.83, S_E = 2.73) showed higher absolute levels of state anxiety than the low pressure group (M = 33.89, S_E = 1.44), although this difference did not reach significance, F(1,34)=2.57, p<0. 12. The low pressure group (M = 4.33, SE = 0.40) and the high pressure group (M = 3.94, E = 0.33) did not significantly differ in their perceptions of the importance of performing at a high level in the posttest, F<1. Participants in the high pressure group (M = 4.33, S_E = 0.29) felt significantly more performance pressure in the posttest equation block in comparison to participants in the low pressure group (M = 3.39, S_E = 0.31), F(1,34)=4.85, p<0.04, MSE=1.66. Finally, participants in the high pressure group (M = 3.94, S_E = 0.34) had significantly worse perceptions of their performance in the posttest equation block in comparison to participants in the low pressure group (M = 5.22, SE = 0.26), F(1,34)=8.91, p<0.01, MSE=1.65. Thus, while the low and high pressure groups did not differ in terms of how important they thought it was to perform at a high level in the posttest — both believing that it was moderately important to succeed — the high pressure group reported more state anxiety and significantly heightened perceptions of performance pressure in comparison 27 to the perfor count whetl“ its oh; equati correc perfor requjr. ANO\ 105. nc Ft 1.34 block f DOStre;S groUp ( 545.40 to the low pressure group. Furthermore, the high pressure group thought that they performed at a lower level on the posttest in comparison to their low pressure counterparts. The reader can now turn to the actual performance data to determine whether participants’ self-reports of performance pressure and its consequences parallel its objectively measured impact. Reaction time and accuracy. Reaction times (RT) were computed for each equation and retained for only those equations answered correctly. Using RT’s for only correct equations helps to guard against possible speed-accuracy trade-offs in performance, allowing for the interpretation of RT data as representative of the time required for successful modular arithmetic execution (Lachman, Lachman, & Butterfield, 1979; Pachella, 1974; Stemberg, 1969). Accuracy data was analyzed separately. Furthermore, RT’s more than 3 SD below or above an individual’s mean RT for each block of equations were considered outliers and removed. This resulted in the dismissal of 12 RT and corresponding accuracy scores from the entire data set. A 2 (low pressure group, high pressure group) x 2 (pretest block, posttest block) ANOVA on RT indicated a main effect of block, F(1,34)=92.89, p<0.01, MSE=14.74 x 105, no main effect of group, F(1,34)=l .74, ns., and no block x group interaction F(1,34)=2.56, ns. Reaction times significantly decreased from the pretest to the posttest block for both the low pressure group (pretest: M = 8582.07 ms, SE = 660.48 ms; posttest: M = 6282.25 ms, SE = 430.13 ms), t(17)=6.23, p<0.01, and the high pressure group (pretest: M = 10141.88 ms, SE = 802.09 ms; posttest: M = 6925.79 ms, S_E = 545.40 ms), t(17)=7.36, p<0.01. 28 Fl. 1.34 MSE= signifi 93.03‘ accura 8 IJI .56‘ signifi 'dCCUl'ii ‘l A similar ANOVA on accuracy revealed no main effect of block, F<1, or group, F(1,34)=3.61, ns., but a significant block x group interaction, F(1,34)=6.28, p<0.02, MSE=0.01. As can be seen in Figure 1, while the low pressure group’s accuracy significantly increased from the pretest (M = 89.44%, §E = 1.14%) to the posttest (M 93.03%, SE = 1.35%), F( 1,17)=6. 10, p<0.03, MSE=0.01, the high pressure group’s accuracy declined from the pretest (M = 90.00%, S_E = 1.07%) to the posttest (M = 85.56%, $2 = 2.74%), F(1,l7)=2.42, p<0.14, MSE=0.01, although this decrease was not significant. Additionally, the low and high pressure groups did not differ in terms of MA accuracy in the pretest, F<1. This was not the case in the posttest however, in which the high pressure group had significantly worse accuracy than the low pressure group, F(1,34)=5.99, p<0.02, MSE=0.01. 29 Silt the U1]; lOO - 98 ~ 96 - 94 ~ 92 ~ 90 - 88 4 86 A 84 - 82 - 8O - 78 - 76 . 1 Pretest Posttest Test +Low Pressure Group _ .— High Pressure Group Accuracy (%) Figure 1. Mean accuracy (% correct) for the low pressure group and high pressure group in the pretest and posttest equation blocks in Experiment 1. Error bars represent standard errors. 2.1.3 Discussion Experiment 1 was designed to examine performance under pressure in a task that should be susceptible to choking according to distraction, but not explicit monitoring theories. Individuals assigned to either a low pressure or high pressure group performed unpracticed modular arithmetic equations requiring both large number manipulations and 30 a borrow i 1992; Asl‘ Rt pressure g accuracy 1 Instead. p following Pa the high p higher lex the low p] it Wag eqt This findi Variations D: amhmetit dUring prt distractiol in [he hlg group On Pregmre S If modular g a borrow operation — two variables thought to increase problem difficulty (Ashcraft, 1992; Ashcraft & Kirk, 2001). Reaction times decreased from the pretest to the posttest for both the low and high pressure groups. While this decrease in reaction time was accompanied by an increase in accuracy for the low pressure group, this was not the case for the high pressure group. Instead, participants in the high pressure group declined in modular arithmetic accuracy following the instantiation of a high pressure situation. Participants’ self-reports mirrored these performance differences. Individuals in the high pressure group felt significantly more performance pressure, had moderately higher levels of state anxiety, and thought that they performed significantly worse than the low pressure group. Furthermore, both the low and high pressure groups reported that it was equally important to perform at a high level on the posttest block of equations. This finding suggests that differences in motivation were not responsible for the variations in modular arithmetic performance reported above. Distraction theories predict that choking is most likely to occur in novel modular arithmetic problems that require a rule-based algorithm to be held in working memory during problem solution — as these are the equations most susceptible to pressure-induced distractions. This pattern of data was clearly born out in Experiment 1 in that individuals in the high pressure group showed performance decrements relative to the low pressure group on unpracticed, difficult modular equations following the introduction of a high pressure situation. If distraction is responsible for performance decrements under pressure in modular arithmetic by way of decreased capacity for task-relevant problem solving, then 31 it foll choke alterir equafi the ex. weres Expen proble: K)crea Operan auenfic and b0, AStha] lnCreaSi Ac PTODOur memo“. (,th p“ SuSCeptil it follows that equations with higher on-line attention demands should be more prone to choke under pressure than less demanding problems. Experiment 2 tested this notion by altering the working memory demands of modular arithmetic through the manipulation of equation difficulty. 2.2 Experiment 2 Individuals carried out the exact same experimental procedure as Experiment 1 with the exception that modular arithmetic problems with varying working memory demands were substituted for the difficult modular arithmetic problems solely utilized in Experiment 1. Working memory demands were manipulated through two different problem difficulty levels: Single-digit problems without a borrow operation were thought to create the least on-line capacity demands and double-digit problems with a borrow operation (the same type of equations used in Experiment 1) were assumed to be the most attention demanding. As mentioned above, problem size (single—digit vs. double-digit) and borrow operations were chosen as the means to establish these comparisons as Ashcraft (1992) has suggested that these are the two variables most associated with increasing math task problem difficulty. According to distraction theories, performance decrements should be most pronounced in equations that possess the heaviest on-line executive control and working memory maintenance demands. Under this view, difficult modular arithmetic equations (e.g., problems that require both large numbers a borrow operation) should be more susceptible to performance decrements under pressure than easier, less working-memory- intensive equations (e. g., single-di git problems without a borrow operation). This pattern 32 of res doma 24L] vi ho v COUI'SC were r (9:40 Expefi grOUps. pressUr arithmc Experin modUlar the ALA ‘ modujar bOrrow St digit born wnhnnen OperaIlOn ( of results would replicate Experiment 1 and further support distraction theories in the domain of mathematical problem solving. 2.3.1 Method Participants. Participants were students enrolled at Michigan State University who were not math majors, had taken no more than 2 introductory college-level math courses, and had no previous exposure to the modular arithmetic (MA) task. Participants were randomly assigned to either a low pressure group (g=40) or a high pressure group (11:40). Unlike Experiment I, a minimum accuracy criterion was not implemented in Experiment 2. A larger sample size in Experiment 2 reduced the initial variability across groups. Thus, a minimum accuracy criterion was not necessary to assure that the low pressure and high pressure groups consisted of individuals with similar levels of modular arithmetic ability. However, implementing the same minimum accuracy criterion used in Experiment 1 would not have significantly altered the pattern of results in Experiment 2 in any way. Procedure. The procedure was identical to Experiment 1 with the exception that modular arithmetic (MA) problems with varying levels of difficulty were substituted for the MA equations utilized in Experiment 1. Specifically, in each of the three blocks of 24 modular arithmetic problems in Experiment 2, eight equations required a single-digit no borrow subtraction operation (e.g., 7 E 2 (mod 5)) and eight equations required a double- digit borrow subtraction operation (e.g., 51 E 19 (mod 4)). An additional eight equations with intermediate attentional demands, requiring a double-digit no borrow subtraction operation (e. g., 19 E 12 (mod 7)), were also included in each of the equation blocks. 33 These contra problc panici Additi counte that pe equatit 2.3.2 Signific 1.21),; Pressurt peI‘CepI, Fl 1,78); least urn more Pre Pressu re F1 SignifiCan These intermediate level equations served as filler problems, intended to diminish the contrast between the easy single-digit problems and the difficult double—digit borrow problems. Half of the problems within each operation were true and half were false. Equations within each block were presented in a different random order to each participant and across the entire experiment, each problem was presented only once. Additionally, as in Experiment 1, the equations presented in the last two blocks were counterbalanced across participants. This counterbalancing was done in order to assure that performance in the last block of equations was independent of the particular equations individuals were exposed to. 2.3.2 Results Questionnaires: The high pressure group (M = 42.65, E = 1.87) showed significantly higher levels of state anxiety than the low pressure group @ = 32.13, E = 1.21), F(1,78)=22.35, p<0.01, MSE=99.15. Participants in the low pressure group (M = 4.63, S_E = 0.21) and the high pressure group (M = 5.03, SE = 0.19) did not significantly differ in terms of their perceptions of the importance of performing at a high level in the posttest equation block, F(1,78)=1.95, ns. As in Experiment 1, on average, both groups reported that it was at least “moderately important” to perform at a high level on these equations. Participants in the high pressure group (M = 5.08, SE = 0.21) felt significantly more pressure to perform at a high level in the posttest than individuals in the low pressure group (M: 3.95, SE = 0.24), F(1,78)=12.44, p<0.01, MSE=2.03. Finally, participants in the high pressure group (M = 4.03, SE = 0.20) had significantly worse perceptions of their performance in the posttest equation block in 34 conipar F(1,781 VVhHe‘ import that it ‘ ugnui perfori high p postte demor perfor increa eQUat RT“5 equat anclc comparison to participants in the low pressure group (M = 4.98, SE = 0.19), F(1,78)=11.55, p<0.01, MSE=1.56. The questionnaire results described above replicate Experiment 1’s findings. While the low and high pressure groups in Experiment 2 did not differ in terms of how important they thought it was to perform at a high level in the posttest - both believing that it was moderately important to succeed — the high pressure group reported significantly higher levels of state anxiety and significantly heightened perceptions of performance pressure in comparison to their low pressure counterparts. Additionally, the high pressure group thought that they performed at a significantly lower level on the posttest than the low pressure group. Similar to Experiment 1, the questionnaire results demonstrate that our manipulation was successful in increasing participants’ feelings of performance pressure. Again, we can turn to the behavioral data to determine if these increased perceptions of pressure parallel actual modular arithmetic performance. Reaction time and accuracy. Reaction times (RT) were computed for each equation and retained for only those equations answered correctly. As in Experiment 1, RT’s more than 3 SD below or above an individual’s mean RT for each block of equations were considered outliers and removed. This resulted in the dismissal of 18 RT and corresponding accuracy scores from the entire data set. A 2(low pressure group, high pressure group) x 2(pretest, posttest) x 2(single— digit problems, double—digit borrow problems) ANOVA on RT revealed a main effect of test, F(1,78)=48.77, p<0.01, MSE=15.03 x 105, a main effect of problem difficulty, F(1,78)=615.54, p<0.01, MSE=40.68 x 105, no main effect of group, F(1,78)=3.06, ns.. and no test x pressure group x problem difficulty interaction, F<1. 35 As can be seen in Table 1, reaction times significantly decreased from the pretest to the posttest equation blocks for the single-di git problems and the double—di git borrow problems for the low pressure group, t(39)=5.05, p<0.01, and t(39)=2.60, p<0.02, respectively, and the high pressure group, t(39)=7.71, p<0.01, and t(39)=5.20, p<0.01, respectively. 36 Lo w E Sing Dou High] Sing DQU Table Low F DOUbl EXPCF Pretest Posttest (M) (SE) (M) (SE) Low Pressure Group Single-Digit RT (ms) 2444.10 94.65 1982.75 91.22 Accuracy (%) 93.13 1.55 98.44 0.80 Double-Digit Borrow RT (ms) 8814.71 449.81 7816.94 404.85 Accuracy (%) 81.88 2.83 85.31 1.95 High Pressure Group Single-Digit RT (ms) 2530.52 109.14 1846.02 84.78 Accuracy (%) 94.69 1.33 98.44 0.66 Double-Di git Borrow RT (ms) 8117.57 423.13 6432.11 357.65 Accuracy (%) 80.00 2.32 74.06 2.66 Table 1. Mean Modular Arithmetic Reaction Time (ms) and Accuracy (% correct) for the Low Pressure Group and High Pressure Group for the Single-Digit Problems and the Double-Digit Borrow Problems in the Pretest and Posttest Equation Blocks in Experiment 2. A digit prob significan MSE:0.0 A the single .\=lSE=0.C respectivt posttest) , F5 8 5 80 - U Q < 75 - 70 - 65 - 60 r l Pretest Posttest Test Figure 2. Mean accuracy (% correct) for the low pressure group and high pressure group for the single-digit problems (SD) and the double-digit borrow problems (DD-Borrow) in the pretest and posttest equation blocks in Experiment 2. Error bars represent standard errors. 2.3.3 Discussion In Experiment 2, individuals assigned to either a low or high pressure group performed novel, unpracticed modular arithmetic equations that varied as a function of problem difficulty. Individuals in the high pressure group reported increased levels of state anxiety and heightened feelings of performance pressure following the instantiation of the high pressure scenario in comparison to participants in the low pressure group. 39 Additionally, individuals in the high pressure group performed at a significantly lower accuracy level on the MA equations after receiving the pressure scenario in comparison to both their pretest performance and the performance of their low pressure counterparts. Analysis of these performance failures as a function of problem difficulty revealed that only the most difficult equations, requiring both large number manipulations and a borrow operation, were performed at a lower accuracy level under pressure. This pattern of results replicates and extends Experiment 1’s support for distraction theories of choking in the domain of mathematical problem solving. Unpracticed modular arithmetic equations, whose solutions require the maintenance of intermediate problem steps and their products in working memory, choke under pressure. Furthermore, these performance failures are limited to those equations with the heaviest on-line maintenance demands (i.e., double-digit borrow problems). According to distraction theories, pressure-induced limitations in working memory capacity cause choking. Thus, novel equations with large on-line attentional demands should be precisely the type of equations for which performance failures under pressure are most likely to occur. The first two experiments lend support to distraction theories as an explanation for the choking phenomenon in the domain of mathematical problem solving. However, it is still possible that performance decrements under pressure may occur at high levels of practice via the mechanisms proposed by explicit monitoring theories. Experiment 3 was designed to test this notion. 40 memo extend deman proble and dc deman were I. pressu consid Suscep Here, 1 “'Orkir decren that Illl highEr are b6] related Via the baSed 0 2.3 Experiment 3 In Experiment 3, individuals performed modular equations with varying working memory demands under low and high pressure conditions both prior to and following extended modular arithmetic practice. Similar to Experiment 2, working memory demands were manipulated through different problem difficulty levels: Single-digit problems without a borrow operation thought to create the least on-line capacity demands and double-digit problems with a borrow operation assumed to be the most attention demanding. Participants had no previous exposure to the specific problems on which they were tested prior to the first high pressure situation. By the time of the second high pressure test, however, participants had received 49 exposures to each problem under consideration. As seen in Experiments 1 and 2, modular arithmetic problems should be susceptible to choking under pressure early in learning according to distraction theories. Here, pressure-induced reductions in the attentional capacity needed to carry out working-memory-intensive problem solving processes should result in performance decrements. Furthermore, these failures should be most pronounced in difficult problems that incur the highest working memory load (e. g., double-digit borrow problems). At higher levels of problem-specific practice, when answers to now well-practiced problems are being retrieved from memory rather than computed via the algorithm, such capacity- related failures should no longer occur. However, it is still possible that choking might happen at high levels of practice via the mechanism proposed by explicit monitoring theories. Well-leamed problems, based on the stimulus-driven retrieval of past problem instances from memory, may fail 41 un ('31 ht’gl prof to pr retrie still f. and hi 2.3.1 I'nit'ers level m; l detailing W35 10 e) Of a mon: MA task In. Oldej- 10 6; operation I SublractiOr inlEI‘medjal procedure 1 under pressure because pressure-induced attention disrupts automatic answer retrieval (Masters, Polman, & Hammond, 1993). If so, then choking should be observed for all highly practiced problems regardless of difficulty — at least to the extent that practiced problems are solved via automatic answer retrieval. Furthermore, even if practice serves to proceduralize the algorithm, rather than shift performance to automatic answer retrieval as Logan (1988) would propose, practiced modular arithmetic problems might still fail via pressure-induced attention that serves to disrupt or slow down well-leamed and highly efficient algorithmic processes. 2.3.1 Method Participants. Participants (N=20) were students enrolled at Michigan State University who were not math majors, had taken no more than 2 introductory college- level math courses, and had no previous exposure to modular arithmetic (MA). Procedure. Participants filled out a consent form and a demographic sheet detailing previous math experience. They were informed that the purpose of the study was to examine how individuals learn a new math skill. Individuals were set up in front of a monitor controlled by a standard laboratory computer and introduced to the same MA task used in Experiments 1 and 2. Individuals first performed 12 practice problems, presented in a different random order to each participant. Four of the equations required a single—digit subtraction operation (e. g., 7 E 2 (mod 5)) and four equations required a double-digit borrow subtraction operation (e. g., 51 E 19 (mod 4)). An additional four equations with intermediate attentional demands, requiring a double-digit no borrow subtraction procedure (e.g., 15 E 10 (mod 3)), were also included. As in Experiment 2, these 42 intermediate level equations served as filler problems, intended to diminish the contrast between the easy single-digit problems and the difficult double-digit borrow problems. Half of the equations within each operation were true, half were false, and each equation was presented only once. Following the practice problems, individuals completed a 12 equation low pressure test (LPl) and a 12 equation high pressure test (HPI), separated by a short break of approximately 1 minute. The equations in LP] and HP] were presented in a different random order to each participant. Each equation appeared only once in either LPI or HP] and the equations in LPI and HPI were counterbalanced across participants. Within both LP] and HPl there were four equations with a single-digit subtraction operation and four equations with a double-digit borrow subtraction operation. Four equations with a double-di git no borrow subtraction operation, that served as intermediate difficulty filler problems, were also included. Half of the equations within each operation were true and half were false. To the participant, LPl appeared to be just another series of practice equations. Following LPI, participants were given a scenario designed to create a high pressure situation. The same high pressure scenario and video camera situation used in Experiments 1 and 2 was presented to participants in Experiment 3, with the exception that individuals in Experiment 3 were informed that they were about to enter the first of two test situations in the experiment and that they had to improve their MA performance by the required amount in both test situations in order to receive the monetary award for themselves and their partner. 43 Following HPl , individuals were informed that they would be performing a series of practice MA equations (MA training). Participants were presented with 12 new equations. Four of these equations required a single-digit subtraction operation and four equations required a double-digit borrow subtraction operation. An additional four equations with a double-digit no borrow subtraction operation were also included. Half of the equations within each operation were true and half were false. Each equation within this MA training session was repeated 48 times for a total of 576 trials, separated into 3 blocks of 192 equations each, with a short break of approximately 1 minute after each block. Within each block, each equation was repeated 16 times. Trials were presented in a different random order to each participant. Participants then took part in the second 12 equation low pressure test (LP2) and the second 12 equation high pressure test (HP2). The equations within LP2 and HP2 were the same 12 equations that were presented 48 times each in the training session. Each participant received the equations within LP2 and HP2 in a different random order. There was no need to counterbalance these equations across the second low and high pressure tests, as the same 12 equations were used in both LP2 and HP2. As in LPI , participants were not made aware of the LP2 test situation. They were not told that this block would be shorter, nor were they given any other cues. To the participant, LP2 appeared to be just another series of practice problems. The I experimenter then informed participants that they were about to take part in the second test situation, repeated the high pressure scenario, and turned the video camera on. Participants completed HP2. Individuals were then fully debriefed and given the monetary award regardless of their performance. In total, participants were exposed to 50 44 presentations of each of the 12 training problems (48 exposures during the training session, 1 exposure in LP2, and 1 exposure in HP2). The self-report measures of performance pressure and anxiety administered in Experiments 1 and 2 were not utilized in Experiment 3. Experiment 3 was a completely within participants design. While one might imagine that the questionnaires could be administered multiple times in order to create a within participant comparison (i.e., once prior to and once following the instantiation of each high pressure scenario), pilot testing revealed that asking for such off-line measures of performance pressure prior to the introduction of a high pressure situation significantly increased participants’ skepticism regarding the validity of the pressure manipulation. Therefore, in order to present the strongest pressure manipulation possible, the questionnaires were excluded. As will be seen, however, behavioral evidence of choking made it clear that the manipulation was again effective. 2.3.2 Results Reaction times (RT) were computed for each equation and retained for only those equations answered correctly. As in the previous experiments, RT’s more than 3 SD below or above an individual’s mean RT for each experimental condition were considered outliers and removed. A total of 3 RT and corresponding accuracy measures were dismissed from the entire data set. A 2 (low pressure test, high pressure test) x 2 (before training, following training) ANOVA on RT indicated a main effect of time, F(1,l9)=61.17, p<0.01, MSE=28.89 x 105, no main effect of pressure, F(1,l9)=1.83, ns., MSE=24.41 x 104, and no time x pressure interaction, F<1. Reaction times were slower prior to MA training in LP] (M = 45 41 14.70 ms, SE = 415.27 ms) and HP] (M = 4001.97, SE = 432.09), than following training in LP2 (M = 1178.44 ms, SE = 100.87 ms) and HP2 (M = 992.7] ms, S_E = 63.04 ms). A similar ANOVA on accuracy revealed a main effect of time, F(1,l9)=1 1.22, p<0.01, MSE=0.01, no main effect of pressure, F(1,l9)=2.68, ns., MSE=0.01, and a significant time x pressure interaction, F(1,l9)=6.24, p<0.03, MSE=0.01. As can be seen in Figure 3, while MA accuracy significantly declined from LP] (M = 92.92 %, SE = 2.28 %) to HP] (M = 86.67 %, SE = 2.66 %), t(19)=2.26, p<0.04, accuracy improved somewhat from LP2 (M = 96.25 %, E = 0.95 %) to HP2 (M = 97.92 %, SE = 1.03 %), t(19)=l .45, ns., although this improvement was not significant. Thus MA performance decrements under pressure occurred prior to extended problem training, but not following it. This pattem of data supports the predictions of distraction theories as an explanation for the choking phenomenon in that only novel MA problems, requiring the instantiation of a capacity-demanding rule-based solution algorithm, choked under pressure. After extended practice of the problems being tested, behavioral evidence of choking was no longer observed. 46 --0- Test] +Test2 Accuracy (%) Low pressure High pressure Pressure Figure 3. Mean accuracy (% correct) for the low and high pressure tests prior to modular arithmetic training (Test 1) and following modular arithmetic training (Test 2) in Experiment 3. Error bars represent standard errors. If distraction theories are correct, then the pressure-induced performance decrements just observed for novel MA problems should be most pronounced in equations that possess the heaviest on-line executive control and working memory maintenance demands (i.e., double-digit problems that require a borrow operation). In 47 order to explore this possibility, we compared RT and accuracy of the least demanding single-digit problems and most difficult double-digit borrow problems across the low pressure and high pressure tests administered prior to MA, where performance decrements under pressure were shown to have occurred. Applied to RT, this 2 (single-digit, double-digit borrow) x 2 (LPI, HPl) ANOVA produced a main effect of problem difficulty, F( l,19)=27.64, p<0.01, MSE=36.87 x 107, no main effect of pressure, and no pressure x difficulty interaction, F’s<1 respectively. Reaction times increased as a function of problem difficulty during both LP] (single— digit: M = 2263.13 ms, g = 183.33 ms; double-digit borrow: M = 6393.77 ms, S_E = 884.09 ms) and HP] (single~digit: M = 2262.02 ms, SE = 165.12 ms; double-digit borrow: M = 6718.47 ms, S_E = 873.23 ms). A similar ANOVA on accuracy produced significant main effects of problem difficulty, F(1,l9)=l4.46, p<0.01, MSE=0.03, and pressure, F(1,l9)=6.17, p<0.03, MSE=0.02, and a significant problem difficulty x pressure interaction, F(1,l9)=12.67, p<0.01, MSE=0.02. This interaction is shown in Figure 4. Paired sample t-tests performed as post hocs revealed that the single-digit problems did not significantly differ in accuracy from LP] (M = 95.00 %, g = 2.29 %) to HP] (M = 96.25 %, S_E = 2.74 %), t(19)=0.44, ns. In contrast, the accuracy of the double-digit borrow problems got significantly worse from LP] (M = 91.25 %, S_E = 3.28 %)to HP] (M = 72.50 %, SE = 4.76 %), t(19)=3.30, p<0.01. Furthermore, there were no significant accuracy differences across problem difficulty levels in LP], t(19)=1.00, ns. During HP] however, an effect of problem difficulty occurred in which the double-digit borrow problems were significantly less accurate than the single—digit problems, t(19)=4.50, p<0.01. Thus, as can be seen in 48 Figure 4, the most difficult MA problems requiring a borrow operation showed performance decrements under pressure while the least difficult, single-digit problems did not. This finding parallels work in the math anxiety literature demonstrating that mental arithmetic problems possessing a carry operation are most susceptible to performance difficulties as a result of decreased working memory capacity (Ashcraft & Kirk, 2001). 100 - +SD - ° * - - DD—Borrow ._l_. Low pressure High pressure Pressure Figure 4. Mean accuracy (% correct) for the low and high pressure tests prior to modular arithmetic training for the single-digit problems (SD) and the double-digit borrow problems (DD-Borrow) in Experiment 3. Error bars represent standard errors. 49 Finally, in order to demonstrate that the MA equations were fully automated following extended training, and hence should have shown choking if explicit monitoring theories were applicable, I performed an analysis of reaction time and accuracy as a function of problem difficulty (i.e., single-digit, double—digit borrow). This kind of analysis has been used in other types of mental arithmetic tasks — for example, Logan’s (1988) alphabet arithmetic — to diagnose the extent to which the control structures of performance have shifted from a working-memory-intensive counting algorithm that produces a significant effect of problem difficulty, to automatic memory retrieval which is independent of equation difficulty level (Klapp et al., 1991). A significant interaction of problem difficulty by HP] versus HP2 for RT was found, F( 1,19):26.55, p<0.01, MSE=36.15 x 105. Prior to MA training, RT was significantly faster for the simplest single-digit problems (M = 2262.02 ms, SE = 165.12 ms) in comparison to the most difficult double-digit borrow problems (M = 6718.47 ms, SE = 873.23 ms), t(19)=5.26, p<0.01. In contrast, following MA training, there was no significant difference in RT between single-digit problems (M = 914.23 ms, SE = 55.6] ms) and double-digit borrow problems (M = 989.79 ms, SE = 67.59 ms), t(19)=1.84, ns. A similar analysis of problem difficulty by HP] versus HP2 on accuracy also revealed a significant interaction of problem difficulty by HP] versus HP2, F(1,l9)=21.11, p<0.01, MSE=0.01. Again, prior to MA training, accuracy was significantly higher for the simplest single-digit problems (M = 96.15% , S_E = 2.74%) in comparison to the most difficult double-digit borrow problems (M = 72.50%, SE = 4.76%), t(19)=4.50, p<0.01. In contrast, following MA training, there was no significant difference in accuracy between the single-digit problems (M = 97.50%, SE = 1.72%) and 50 the double-digit borrow problems (M = 98.75%, SE = 1.25%), t(19)=0.57, ns. Thus, following repeated exposure to MA problems, there appears to be a relative independence between MA performance (whether measured by reaction time or accuracy) and problem difficulty. This is a sign that performance following MA training was automated — based on the direct retrieval of answers from long term memory rather than working-memory- intensive algorithmic computation (Klapp et al., 1991). 2.3.3 Discussion The results of Experiment 3 replicate the finding of the first two experiments that novel modular arithmetic equations, whose solutions require the maintenance of intermediate problem steps and their products in working memory, decline under pressure. Furthermore, as in Experiment 2, analysis of these performance decrements prior to training revealed that only the most difficult equations requiring both large number manipulations and a borrow operation failed under pressure. In contrast, well- learned modular arithmetic problems, thought to be supported by the one-step direct retrieval of past problem instances from memory, showed no signs of choking. According to distraction theories, modular arithmetic should be most susceptible to pressure-induced performance decrements at low levels of practice when working memory demands are greatest and pressure-induced worries impinge on task-relevant processing resources. This prediction was clearly borne out. Not only was choking solely observed prior to modular arithmetic training, but similar to Experiment 2, performance decrements under pressure were limited to difficult problems that incurred the highest working memory load (i.e., double-digit borrow problems). 51 The first three experiments provide support for distraction theories and argue against explicit monitoring theories as an explanation for choking under pressure as observed in the working—memory-intensive task of modular arithmetic. The purpose of Experiment 4 was to further explore performance under pressure in this task by considering the role of general practice at the algorithm, through a comparison of infrequently practiced problems with heavily practiced equations under pressure, at similarly high overall levels of general algorithmic practice. 2.4 Experiment 4 Participants performed over 800 modular arithmetic problems over 3 days of practice prior to being exposed to a low and high pressure situation. Specific equations within this practice period were presented either once, repeated twice, or repeated 50 times each. As previously discussed, the task control structures of modular arithmetic problems should change most dramatically as a function of specific problem exposure, not necessarily experience at performing many different modular arithmetic problems (Logan, 1988). Thus, if choking is due to pressure-induced capacity limitations, as distraction theories would propose, then regardless of how many different problems individuals have been exposed to, only those equations that have been repeated enough to produce instance-based answer retrieval should be inoculated against the detrimental capacity-limiting effects of performance pressure. Furthermore, even if algorithmic computations do become more efficient with practice (Touron, Hoyer, & Cerella, 2001), implementation of the algorithm should still impose attentional demands as novel problem information must be maintained and manipulated on-line in working memory during performance. This prediction can be 52 tested by examining performance decrements under pressure as a function of problem difficulty level. If novel problems based on practiced algorithms are susceptible to pressure—induced performance decrements via distraction, then the more difficult and capacity-demanding the problem, the more vulnerable it should be to capacity-limited failure. This earmarks difficult problems repeated only a few times during practice, and still based on problem-solving algorithms, as candidates for choking under pressure according to distraction theories. It should be noted that if practice increases the proficiency of algorithmic computations, it is also possible that performance failures under pressure (for novel problems based on practiced algorithms) could be explained by explicit monitoring theories of choking. Specifically, pressure-induced attention may serve to disrupt highly efficient, proceduralized algorithmic computations. If so, this type of skill failure should be evident, at least to some extent, across all problem difficulty levels, as practiced algorithms, regardless of working memory demands, should be harmed by the instantiation of explicit attentional control mechanisms that slow down or disrupt highly efficient computations. This difference between theories concerning whether choking should depend on the capacity demands of the problems being performed gives Experiment 4 some further leverage in distinguishing distraction from explicit monitoring as a source of performance decrements under pressure in modular arithmetic. 2.4.1 Method Participants. Participants (N=22) were students enrolled at Michigan State University who were not math majors, had taken no more than 2 introductory college- level math courses, and had no previous exposure to the modular arithmetic (MA) task. 53 Procedure. Participants filled out a consent form and a demographic sheet detailing previous math experience and were informed that the purpose of the study was to examine how individuals learned a new math skill over several days of practice. Individuals performed the same MA task used in first three experiments over three days of practice. On Days 1 and 2, participants performed three blocks of 120 equations each, separated by short breaks of approximately 1 minute. On Day 3, individuals performed 90 equations, for a total of 810 practice equations over the 3 days of practice. Ten practice equations were repeated 50 times each (multiple repeats) over the 3 days of practice (22 presentations of the multiple repeat equations on Days 1 and 2; 6 presentations on Day 3), 100 equations were repeated once (once repeats) over the three practice sessions (80 of these occurred on Days 1 and 2; 10 occurred on Days 1 and 3; 10 occurred on Days 2 and 3), and 1 10 equations were presented only once (no repeats). Practice equations were presented in a different random order to each participant. Following practice on Day 3, participants took part in a 30 equation low pressure test and a 30 equation high pressure test. The low pressure test consisted of the 10 equations that were repeated 50 times each during practice (multiple repeats), 10 equations that were repeated once during practice (once repeats), and 10 equations not previously presented during practice (no repeats). The high pressure test consisted of the 10 multiple repeats, 10 new once repeats, and 10 new no repeats. Equations within these tests were presented in a random order to each participant. To the participant, the low pressure test appeared to be just another series of practice equations. Participants then completed the high pressure test, consisting of the same high pressure scenario and video camera situation utilized in the first three 54 experiments. Participants were then fully debriefed and given the monetary award regardless of their performance. 2.4.2 Results Reaction times (RT) were computed for each equation and retained for only those equations answered correctly. There were no RT and corresponding accuracy measure outliers in either the low or high pressure test for any participants utilizing the 3 SD outlier criterion established in the first three experiments. However, 5 equations in the low pressure test (2 once repeats; 3 no repeats) and 3 equations in the high pressure test (1 once repeat; 2 no repeats) were discarded from the subsequent analyses because the accuracy of these equations across participants was not significantly different from chance. I began by comparing once repeat problems to no repeat problems in order to determine if a small amount of problem-specific exposure changes performance. A 2 (low pressure, high pressure) x 2 (once repeats, no repeats) ANOVA on accuracy produced no main effects of pressure, or problem repetition, and no pressure x repetition interaction, F’s .10). As an adjunct to the analysis of assessment. those steps that involved mental imagery (i.e., imagining some sweet of how a putt ought to look or feel before executing the action) were counted Mental imagery is a topic of considerable interest in sports psychology and has been defined in that literature as “the imagined rehearsal of skill processes. procedures. and possible outcomes prior to task performance" (Woolfolk. Murphy. Gotres- feld. &. Aitken. 1985). In the undergraduate group. 0.0% of generic steps and 0.7% of episodic steps referred to mental imagery. 1n the athlete group. 0.7% of the generic steps and 0.0% of the episodic steps referred to imagery. In the golf team group. 7.0% of the generic steps and 2.0% of the episodic steps referred to imagery. Thus almost all of the reports of imagery were from golfers. and most of these were part of the genetic descriptions. One might worry that the experts‘ exclusion of assessment steps from their episodic recollections was merely an artifact of our very simple and highly repetitive situation. in which assessment was not much needed by the time episodic memory was measured. which 92 706 BEILOCK AND CARR Table I Representative Generic and Episodic Putting Description: Generic putting description Episodic putting description Undergraduates l. Feet apart I. Feet apart 2. Lean forward 2. Knees not locked 3. Aim ball 3. Learung forward 4. Swing 4. Positioning hands 5. Lining putter up with the ball 6. Look at the hole 7. Aim ball 8. Swing 9. Follow through Athletes I. Estimating distance I. Estimate distance to target 2. Bending knees 2. l placed my feet a comfortable distance apart 3. Looking back at target 3. Bent my knees 4. Relaxed backswing 4. Line up the putter with the target 5. Follow through 5. Slowly pulled the putter back 6. Follow through lightly 7. Using straight arms Golf team members I. Walk behindtheballand lookattheputt l. Lookupatputt 2.Readthegreenfrombehindtheball ZPlaoeputterbehindbalIwiththeheadsquareatthetarget 3. Make sure nothing is in its path 3. Look et target 4. book at distance of putt 4. Look at putter and ball 5. Pick a target to aim at 5. Take putter back 6. Place puner behind ball lined up with the target 6. Swing through ball 7. Move putter class to you of the ball and line up at target '7. Look up at target 8. Take a practice swing 9. Move putter back to behind the ball l0. Line up squarely with target ll. Move feet and body square with putter head I2. Look at target [3. Look down at the ball I4. Swing the putter head straight back 15. And straight through I6. book up at ball was after the 70th putt. To guard against this alternative explana- tion. we performed a reanalysis of the golf team members‘ proto- cols. dropping from each generic protocol all assessment steps that (it) did not appear in the corresponding episodic protocol and (b) were likely to be unnecessary once 69 putts had been taken in our laboratory situation. Excluded were steps such as head the green" and “read the lie of the ball." because neither the green nor the lie of the ball changed during the experiment. Steps such as “taking Table 2 Questionnaire Responses: Number of Steps (Experiment 1) Generic l Generic 2 Episodic Group M SE M SE M 88 Undergraduate 5. I9 0.39 5.63 0.38 7.69 0.58 Athlete 5.94 0.54 6.25 0.55 6.75 0.77 Golf team 8.63 0.94 8.44 0.9’7 5.56 0.60 aim.” that would always be necessary in order to execute a putt. were maintained This reanalysis of assessment produwd the same—shaped interaction between expertise and type of protocol as the original. F(2. 45) = 3.34. p < .05. M55 = 0.68. firming to meclnnics. this analysis also produced an interaction between expertise and protocol type. F(2. 45) = 7.96. p < .001. MSE = 1.68. but ofa very different nature. as can be seen in the middle panel of Figure 3. Undergraduates gave significantly more mechanics steps in their episodic descriptions than in their generic descriptions. t(15) = 3.34. p < .005. and athletes produced a nonsignificant difference in the same direction. t(15) = 0.36. In contrast. the golfers gave more mechanics steps in their generic descriptions than in their episodic descriptions. though the differ- ence was only marginally significant. t(l5) = 1.75. p < .10. The greater number of mechanics steps in the episodic protocols of undergraduates compared with golfers was significant. t(30) = 2.13. p < .05. in sum, mechanics was a category of steps that for experts tended to appear more often in generic descriptions than in episodic descriptions. but for novices appeared more often 93 CHOKING UNDER PRESSURE ‘0 —-O-Undetuld l +Atflata g. nautical-am Number of steps Garretic 1 Generic2 Episode Oucstiomaire Figure 2. Mean number of steps for the first and second generic ques- tionnaires and the episodic questionnaire for each group. Undergrad I undergraduate. in episodic descriptions than in generic descriptions. Athletes were intermediate. The analysis of ball destinations produced two main effects but no interaction. Overall. more ball destinations were included in the episodic protocols than in the generic protocols. F(1. 45) = 9.36. p < .004. MSE = 0.22. and the golf team included more destina- tion information than either the undergraduates or the athletes. F (2. 45) = 3.98. p < .026. M55 = 0.21. Thus. as shown in the right panel of Figure 3. ball destinations were more likely to appear in the episodic recollections of experts than anywhere else. though even there they were relatively infrequent. accounting for only 13% of the steps that were included. A second qualitative analysis looked for steps present in both protocols that referred to the same action or biomechanism but provided more detail in one type of protocol than in the other. For instance. a step in the episodic description of one participant was stated as “l positioned my feet so that they were shoulder length apart." This was scored as an elaboration of a step in the same participant‘s generic description that was stated as “feet position- ing." Overall. elaborations were more likely to occur in episodic descriptions relative to generic descriptions than vice versa. There- fore. greater detail in the episodic description was scored as a “positive" elaboration whereas greater detail in the generic de- scription was scored as a “negative" elaboration. 1n the undergrad- uate group. 14% of the steps in the episodic descriptions were elaborations of steps in the generic descriptions. 1n the athlete group. - 1% of the episodic steps were elaborations of generic Table 3 707 steps. In the golf team group. 5% of the episodic steps were elaborations of generic steps. A one-way ANOVA on these data produced a significant effect of expertise. F(2. 45) = 3.53. p < .038. MSE = 1.23. Fisher’s LSD test showed that undergraduates elaborated their episodic recollections relative to their generic descriptions significantly more often than the athletes and margin- ally more often than the golf team (p < .06). Athletes and golfers did not significantly differ from one another. Although the athlete group consisted of novice golfers. their elaborations were more similar to the golf teams' than to the undergraduates'. Similar to the athletes' pattern of mechanics steps. the athletes' pattern of elaborations suggests that sport training and participation lead athletes to approach novel skill situations in certain ways that resemble the approach of more experienced perfomters. This occurs despite the fact that the ath- letes' measured achievements in golf putting perfomtance are no better than those ofother novices. Discussion The results of Experiment 1 demonstrate an effect of level of expertise on the content of generic knowledge and episodic mem- ories of golf putting. Experts gave longer. more detailed generic descriptions of the steps involved in a typical putt compared with the accounts given by novices and shorter. less extensive episodic recollections of a particular putt. These quantitative differences were accompanied by qualitative differences between experts and novices in the nature of the steps included in each type of description. Expert golfers' generic descriptions dealt considerably more with assessing and planning a putt than did novices'. This finding is consistent with research on expert performers across a wide range of task domains (Chi. Feltovitch. & Glaser. 1981; Lesgold et al.. 1988'. Priest & Lindsay. 1992; Proctor & Dutta. 1995; Voss & Post. 1988). In areas as diverse as physics problem solving and radiological X-ray diagnosis. experts spend more time evaluating a situation and deciding how to approach or formulate a problem before they actually begin to work on it than do novices. Expert golfers" episodic recollections included fewer assess- ment steps than did their generic descriptions. Expert golfers also made fewer references to putting mechanics in their episodic recollections than did novices. This pattern follows the prediction of expertise-induced amnesia derived frorn current theories of skill acquisition and automaticity. According to this idea. cxperts' ex- tensive generic knowledge of putting is declaratively accessible during off-line reflection. but it is not used during real-time per- Assessment. Mechanic. and Destination Descriptions by Questionnaire Type—Experiment I Generic Episodic Assessment Mechanics Destination Assessment Mechanics Destination steps steps description Total steps steps steps description Total steps Group M SE M SE M SE M SE M SE M SE M SE M SE Undergraduate 1.44 0.32 4.19 0.56 0.00 0.00 5.63 0.38 1.62 0.24 5.88 0.74 0.19 0.14 7.69 0.58 Athlete 1.25 0.27 5.00 0.58 0.00 0.00 6.25 0.55 1.25 0.27 5.12 0.63 0.38 0.15 6.75 0.77 Golf team 3.69 0.48 4.50 0.84 0.25 0.11 8.44 0.97 1.37 0.30 3.63 0.75 0.56 0.16 5.56 0.60 94 708 BEILOCK AND CARR 7 Assessment Mechanics Destination —+—wm 6 1 +AM. ' - i- - -Gottteam 5 l 8 . 2 4 . ' - . f.’ A. I S 3 ‘ B \ 1 . o 4 0A . . . _1 J l 1 1 I L Generic Episodic Generic Episodic Generic Episode Questionnaire Figure 3. Mean number of steps in each category for the second generic questionnaire and the episodic questionnaire for each group. Undergrad = undergraduate. forrnance. which is controlled by automated procedural knowl- edge. Because proceduralization reduces the need to attend to the processes by which skill execution unfolds. episodic recollection of step-by-step real-time performance is impoverished. How are the details of these declarative reports related to the accuracy of performance? A significant negative correlation was found between the length of the undergraduates‘ generic descrip- tions and their pretest putting accuracy (r = -.52. p < .03). Because the measure of golf putting accuracy in the present study was an error score (i.e.. mean distance from the target). it appears that the more detailed the generic descriptions supplied by the undergraduate novices early in practice. the better they performed This correlation is consistent with an additional idea derived from current theories of skill and automaticity stipulating that novices' real-time performance is controlled by declaratively accessible knowledge concerning skill execution. Furthermore. this correla- tion was the only significant individual—differences relationship found between the contents of the declarative protocols and the accuracy of putting within any of the groups. This pattern suggests that a more extensive generic representation aids putting perfor- mance in the very earliest stages of skill learning but loses its impact as practice proceeds. once again consistent with expecta- tions generated from theories of skill and automaticity. The dis- appearance of the correlation between undergraduates' generic knowledge and their performance accuracy appears to have oc» curred rapidly in the present situation. disappearing by 60-70 putts in the posttest scores. Thus procedural control structures may be established and begin to come to the fore quite quickly. at least in certain task domains (see Brown & Carr. 1989;K1app et al.. 1991'. Raichle et al.. 1994). Although experts gave less elaborate episodic recollections of putting mechanics than did their novice counterparts. they gave more extensive recollections of ball destinations. This result sug- gests that performance outcomes are more salient to expert golfers. paralleling findings in other. more cognitive domains. It has been shown that expert physicists allocate more attentional resources to assessing and monitoring specific goal outcomes during problem solving titan do less experienced physicists (Voss & Post. 1988). Of course. in our simple and repetitive situation. outcomes were generally similar to one another in both form and importance. The very low rate of inclusion of outcome information. even by ex- perts. should increase as competitive motivations and conse- quences of success or failure become greater. We now turn to the second experiment in the present study. which was designed to replicate and extend the findings of Experiment 1. Experiment 2 In Experiment 2 expert golfers' generic knowledge of golf putting and episodic recollection of specific putts were again compared with the generic knowledge and episodic recollection of novice golfers. Knowledge and recollection were assessed during either a standard golf putting task using a normal putter (i.e.. the same task as in Experiment 1) or an altered putting task using a “funny putter" that consisted of a regular putter head attached to an S~shaped curved and arbitrarily weighted putter shaft. The design of the funny putter required experienced golfers to alter their well-practiwd putting form in order to compensate for the dis- torted club. forcing them to allocate attention to the new skill execution processes If experts‘ golf putting skill is proceduralized. then the disruption caused by the novel putter should not only lead to a lower level of performance in comparison to regular putter use but should also produce more elaborated episodic memory proto- cols—possibly similar to those of the novice golfers—as a result of the md to attend to the specific processes of skill execution under the constraints of the new putter. However. novice performers should not be affected by the funny putter in the same way as more experienced golfers. Because novices have not yet adapted to 95 C HOKING UNDER PRESSURE 709 putting under normal putter constraints. performance should not depend as heavily on the type of putter used Furthermore. accord- ing to the theories of skill acquisition we have reviewed. novices' on-line representations of golf putting are explicitly monitored in real time. Therefore. attending to novel putter constraints should not produce different episodic memory protocols in comparison with regular putter use. because in both cases novices attend to their performances in a way that should support explicit episodic memory.I The design of Experiment 2 was similar to that of Experiment 1. with three exceptions. First. in order to ensure that individuals were not adapting to the highly repetitive task of putting from one specific spot on the green. all participants alternately putted from nine different spots. located at varying angles and distances from the target. Second, the experienced golfers in Experiment 2 were university students with 2 or more years of high school varsity golf experience rather than intercollegiate golf team members. Last. in Experiment 2 participants filled out two episodic protocols. As in Experiment 1. the first episodic questionnaire was unexpected. Prior to the last putt taken before the second episodic question- naire. however. individuals were instructed to monitor their per- formance carefully for later recall. Method Participants Participants (N = 72) were undergraduate students eruolled at Michigan State University and consisted of experienwd golfers with 2 or more years of high school varsity golf experience (n = 36) and introductory psychol- ogy students with no golf experience (n == 36). Participants were randomly assigned within skill level to either a regular putter or funny putter condition in a 2 (novice golfer. experienced golfer) x 2 (regular putter. funny putter) experimental design with 18 participants in each group. Procedure After giving informed consent and filling out a demographic sheet concerning previous golf experiences. precipants wae told that the pur- pose of the study was to examine the accuracy of golf putting over sevaal trials of practice. Participants were instructed that the object of the task was to putt a golf ball as accurately as possible from nine locations on a carpeted indoorputtinggreen(3 X 3.7 m) that wereeither 1.2. l.4.or 1.5m away from a target. marked by a square of red tape. on which the ball was supposed to step. All participants followed the same random alternation of putting from the nine different location A standard golf putter and golf ball were supplied for those participants who took part in the regular putter condition. and the funny putter and a standard golf ball were supplied for those participants in the funny putter condition. All groups participated in identical pretest. practice. and posttest condi- tions. though the panicipants were not made aware of the separate condi- tions. To the participant. the golf putting task appeared to involve four blocks of putts with a short break after each block during which a ques- tionnaire was filled out. Pretest condition. Participants were set up at the first putting spot. They were asked whether they preferred to putt right-handed or left-handed and were given the appropriate putter. Participants were then informed that they would be putting from nine different locations on the green. each with a corresponding number. The experimenter reviewed the numbers associ- ated with each putting location and asked participants to repeat back the numbers corresponding to each putting spot. Participants were informed that the expaimenter would call out a number corresponding to a particular spotonthegreenfmmwhichdteywaetoexecutetheirnextpun. Participants then took a series of 20 putts. After completing the putts. participants filled out a questionnaire eliciting a description of the steps involved in a typical golf pun (Appendix A. first paragraph). Practice condition. Participants were again set up at the first putting spot. Participants took a series of 30 putts. After convicting the putts. participants filled out an identical questionnaire to the one that they had previously filled out in the pretest condition eliciting a description of the steps involved in a typical golf putt (Appendix A. first paragraph). Posttest I condition. Participants were set up at the first putting spot Participarls then took a series of 20 putts. Immediately following the first posttest condition. participants filled out a questionnaire designed to access their episodic recollection of the last putt they hadjust taken (Appendix A. third mph)- Posuerr 2 condition. Participants were again set up at the first putting spot. Participants then took a series of IO putts. lrnrnediately prior to the 10th putt in the trial block. the experimenter instructed participants that they should pay close attention to the processes involved in their next putt because after it was complete. they would be asked to fill out another questionnaire. identical to the one they had just filled out. regarding their memories of this next putt. Immediately following the second posttest condition. participants filled out a questionnaire designed to access their episodiereeollectionofthelastputtthey hadjusttaken(Appendix A.third mar-phi. Results Putting Performance Accuracy of putting was measured by the distance (in centime- ters) away fromthecenterofthetargetatwhichtheball stepped aftereach putt. As in Experiment I. the mean distance from the targetofthelast l0 putts inthepretest condition was usedasa measure ofpretest golf putting skill. The mean distance from the target of the middle 10 putts in the practice condition was used as a measure of practice putting skill. The mean distance from the target ofthe last 10 putts in the first posttest conditiort was used as a measure of Posttest l golf putting skill. The mean distance from the target of the IO putts in the second posttest condition was used as a measure of Posuest 2 golf putting skill. Means and standard errors for putting performance appear in Figure 4. As can be seen from Figure 4. the experienced golfers showed superior putting performance in comparison with the novice golf- ers. regardless oftypc of putter used. This was true both before and after the practice phase. This pattern was confirmed by a 2 (expe- rienced golfer. novice golfer) X 2 (funny putter. regular putter) x 2 (pretest. Posttest 1) ANOVA. which revealed signifi- cant rrtain effects of experience. F(1. 68) = 42.73. p < .001. M35 = 51.55. and test. F(1. 68) = 4.04.p < .048. MSE ? 25.25; no significant main effect of putter. F(1. 68) = 1.47. p < .229, M55 = 51.55; and no interaction of Test x Experience x Putter (F < l). in order to assess puuing performance from the pretest condition to the second posttest condition. a three-way ANOVA similar to the one reported above was computed using the mean distance from the target of the last 10 putts in the pretest condition as a measure of pretest skill and the mean distance from the target of the 10 putts in the second posttest condition as a measure of ' We thank Claudia Carello for suggesting the funny putter as a diag- nostic tool. 96 ,1!- [fitmflhq 710 BEILOCK AND CARR “i -o—m 32‘ "unit +en Moan distance from target (cm) 8 r V ‘r ‘ Pretest Practice Posttest i Posttest 2 Putting condition Figure 4. Mean (t SE) distance from the target at which the ball stopped after each port in the pretest. practice condition. Posttest I. and Posttest 2. NR = novice golfer-regular putter: NF = novice golfer-funny putter: ER = experienced golfer-regular putter. EF = experienced golfer-funny putter. Posttest 2 golf putting skill. The results of this analysis did not differ from those reported above. Thus. as can be seen from Figure 4. the experienced golfers. regardless of type of putter used. outperformed the novice golfers at all stages of practice. In addition. experienced golfers using the funny putter were less accurate than the regular putter- experienced golfers—especially during the practice condition and posttests. Independent sample t tests within the experienced golfers revealed no significant differences between putter type during the pretest. t(34) = 0.74. p > .47. but significant differences during the practice condition. t(34) = 2.08. p < .05. and the first posttest. t(34) = 2.87. p < .007. and marginally significant differences during the second posttest. :04) = 2.0. p < .054. In contrast. the novice golfers did not significantly differ by putter type at any point in the experiment. although novices using the funny putter geneme performed at a slightly lower level than their regular putter counterparts. Thus. although the funny putter produwd differences in performance within higher levels of experience. it did not significantly affect the less experienwd golfers. It should be noted that although experienced golfers using the funny putter performed at a lower level than regular putter experts during the pretest. this difference was not statistically significant. It may be that in the pretest condition. expert golfers—regardless of putter type—were adjusting to the novel experimental demands of having tolandtheballonthetargetratherthaninahole.‘l‘hus. regular putter experts were not performing up to their potential in the pretest. The difference between the regular and funny putter ex- perts widened quickly. however. appearing as early as the practice condition. Because experiemd golfers often encounter novel put- ting green environments and must adapt to these situations in order to maintain a low handicap. it is not surprising the regular putter experts were able to rapidly adjust to our indoor green. In fact. several of the experienced golfers mentioned adjusting to the “fast green" or having to “land the ball on the tape" in their episodic protocols. suggesting that these individuals were able to identify and adapt to our somewhat irregular putting environment. In contrast. as can be seen from Figure 4. those experts using the funny putter were unable to adapt to the demands of the new putter within the time frame of the experiment. performing at a similar level of accuracy across experimental conditions. Generic and Episodic Memory ProtocoLr As in Experiment 1. questionnaire responses were analyzed quantitatively. in terms of the number of golf putting steps in. cluded in each type of protocol. and qualitatively. in terms of the relative frequencies of different categories of steps. Quantitative analysis. Analysis of number of golf putting steps given by participants was performed in the exact same manner a in Experiment 1. Two experimenters independently coded the data. lnterexperimenter reliability was extremely high (r = .95). Table 4 and Figure 5 present the results A 2 (experienced golfer. novice golfer) x 2 (funny putter. regular putter) x 2 (first generic protocol. second generic protocol) ANOVA on the two generic protocols revealed a marginally significant main effect of test. F(1. 68) = 3.03. p < .086. M55 = 1.67. and no interaction of Expertise x Putter x Test (F < l).1hus. as in Experiment I. the second generic protocol was used in a 2 (experienced golfer. novice golfer) x 2 (funny putter. regular putter) x 2 (second generic protocol. first episodic protocol) ANOVA to compare the lengths of the generic and episodic protocols produced at each level of expertise. This analysis revealed an interaction of Expe- rience x Putter X Questionnaire. F(1. 68) = 9.63. p < .003. M55 = 2.77. A 2 (experienced golfer. novice golfer) x 2 (funny putter. regular putter) general factorial ANOVA on the second generic Table 4 Questionnaire Responses: Number of Steps (Experiment 2) Generic 1 Generic 2 Episodic r Episodic 2 Group H SE M SE M SE M SE NR 6.39 0.5l 6.69 0.59 9.28 0.94 9.78 0.96 NF 6.l l 0.65 6.67 0.75 9.l l 0.72 9.83 0.8l ER 8. l7 0.8l 8.79 0.76 7.” 0.57 8.60 0.72 BF l0.22 0.8l l0.30 0.85 ".89 0.75 ”.78 0.89 Nate. putter. EF = experienced golfer—funny putter. 97 NR - novice golfer-regular putter: NF 2 novice golfer-funny putter. BR = experienced golfer-regular CHOKING UNDER PRESSURE 711 +NR "'1 ‘°."NF +ER ”CHEF tn ‘2‘ ... """" 0 8 . is 104 . ..... . 0 § .. 3 Z 64 4 Y ‘ Generic t Generic2 Episodiel Episodc2 Questionnaire Figure 5. Mean number of steps for the first and second generic ques- tionnaires and the first and second episodic questionnaires foreach group. NR = novice golfer—regular putter; NF =- novice golfer-funny putter: ER = experienced golfer-regular putter. EF = expaienced golfer-funny putter. protocol produced a main effect of expertise. F(1. 68) = 14.72. p < .001. MSE = 10.01. with the experienced performers giving longer generic protocols than the novices; no main effect of putter. F(1. 68) = 1.01. p > .318. M55 = 10.01; and no Experience x Putter interaction. F(1. 68) = 1.01. p > .318. M55 = 10.01. In contrast. a 2 (experienced golfer. novice golfer) X 2 (funny putter. regular putter) general factorial ANOVA on the first episodic questionnaire produced an Experience x Putter interaction. F(1. 68) = 10.70. p < .m2. 1455 = 10.28. independent sample t tests revealed that whereas the two novice groups did not differ in terms of the number of steps given in their episodic protocols. ((34) = 0.14. p > .89. the funny putter-experienced golfers gave significantly more steps in their episodic protocol than the regular putter-experienced golfers. t(34) = 5.09. p < .001. In addition. both novice groups and the funny putter—experienced group gave significantly more steps than the regular putter—experienced golf~ crs in the episodic questionnaire (ps < .05). Direct comparisons of the number of generic versus episodic steps within each group showed that. similar to Experiment 1. both the regular and funny putter novices gave significantly more steps in their episodic than their generic protocols. t(l7) = 4.10. p < .001. and t(17) = 6.27. p < .001. respectively. In addition. the experienced golfers using the funny putter gave significantly more steps in their episodic than in their genetic protocols. t(l7) = 2.64. p < .017. In contrast. the experienced golfers using the regqu putter gave significantly more steps in their genetic than in their episodic protocols. t(17) = 3.04. p < .007. Furthermore. as can be seen from Figure 5. the experienced golfers using the funny putter gave longer generic and episodic putting descriptions than any other group. Increased attention to the novel constraints of the funny putter most likely prompted these golfers to allocate more attention to skill execution processes. enhancing generic descrip- tions and leaving explicit episodic memory traces of performance. If the funny putter-experienced golfers gave more elaborate episodic descriptions as a result of increased attention to the specific processes involved in novel skill execution. then instrum- ing these individuals to pay close attention to a particular instance of a putt. as did the instructions given prior to filling out the second episodic questionnaire. should not significantly change episodic descriptions in comparison to the first unexpected episodic ques- tionnaire. That is. if the constraints of the funny putter serve to increase attention to skill execution. instructing experienwd golf- ers to explicitly monitor performance should not alter attentional allocation and thus should not affect episodic memory protocols. In contrast. if those experts using the regular putter are asked to monitor performance for a later recall test. their episodic descrip- tions should increase in comparison to their first episodic proto- col—especially if the first recollection was truly based on an unrnonitored proceduralized instance of performance. A 2 (experienced golfer. novice golfer) x 2 (funny putter. regular putter) x 2 (first episodic protocol. second episodic pro- tocol) ANOVA was performed. producing a significant Protocol x Experience x Putter interaction. F(1. 68) = 4.08. p < .047. M85 = 1.86. As can be seen in Figure 5, the novices. regardless of putter type. gave marginally longer putting descriptions in the second episodic questionnaire than in the first. :07) = 1.7. p < .108. and :07) = 1.83. p < .085. respectively. The experienced golfers using the funny putter did not differ in putting description length from the first to second episodic questionnaire. t(17) = 0.11. p > .92. In contrast. the experienced golfers using the regular putter gave longer protocols in the second episodic questionnaire in comparison to the first. 107) = 2.82. p < .012. Furthermore. a 2 (experienced golfer. novice golfer) x 2 (funny putter. regular punter) general factorial ANOVA on the second episodic questionnaire produced a marginally significant Experi- ence x Putter interaction. F(1. 68) = 3.35. p < .071. MSE = 12.99. Thus. instructing participants to monitor skill exe- cution did not affect the funny putter experts' episodic rccollec~ tions and only margimlly influenced the novice golfers‘ episodic descriptions. However. although instructing regular putter experts to monitor performance did increase their episodic recollections. they still did not reach the level of either the novice group or the funny putter-experienwd golfers. Qualitative analysis: Types of steps The qualitative analysis was perforated in the same manner as in Experiment 1. Steps were divided into three categories (assessment. mechanics. and ball destinations). and a 2 (experienced golfers. novice golfers) x 2 (funny putter, regular putter) x 2 (second generic protocol. first episodic protocol) ANOVA was conducted on the number of steps given in each of these three categories (see Table 5). The analysis of assessment steps produced a significant inter- action of expertise and type of protocol. F(1. 68) = 14.53. p < .001. M55 = 1.2. which is displayed in the left panel of Figure 6. along with a nonsignificant interaction of Expertise x Protocol x Putter (F < 1). A one-way ANOVA on the generic protocol with putter collapsed within skill level produced a main effect of experience. (’0. 70) = 23.47. p < .001. MSE = 2.98. Assessment steps appeared more ofien in the generic descriptions of experi- enced golfers. regardless of putter type. than anywhere else. in terms of the episodic protocol. a one-way ANOVA with putter collapsed within skill level produced a marginally significant main effect of experience. F( 1. 70) = 3.58. p < .063. M55 = 1.71. with experienced golfers continuing to give more assessment steps in their episodic recollections than the novices. Paired sample t tests further revealed that the regular putter—experienced golfers gave 98 712 BEILOCK AND CARR Table 5 Assessment. Mechanic. and Destination Descriptions by Questionnaire Type—Experiment 2 Assessment Mechanics Destination Total Group H SE A! SE M SE M SE Generic NR 1.70 0.32 5.00 0.54 0.00 0.00 6.69 0.59 NF 1.56 0.29 5.11 0.74 0.00 0.00 6.67 0.75 ER 3.54 0.41 5.26 0.63 0.00 0.00 8.79 0.76 EF 3.65 0.57 6.66 0.82 0.00 0.00 10.30 0.85 Episodic 1 NR 1.72 0.37 7.33 0.71 0.22 0.10 9.28 0.94 NF 1.94 0.26 7.00 0.80 0.17 0.09 9.11 0.72 ER 2.06 0.19 4.94 0.58 0.11 0.08 7.11 0.57 EF 2.78 0.37 8.72 0.66 0.39 0.16 11.89 0.75 Episodic 2 NR 1.89 0.25 7.61 0.88 0.28 0.14 9.78 0.96 NF 1.67 0.26 7.89 0.90 0.28 0.11 9.83 0.81 ER 2.35 0.23 5.96 0.61 0.28 0.11 8.60 0.72 EF 2.56 0.35 9.06 0.70 0.17 0.09 11.78 0.89 Note. NR = novice golfer—regular putter. NF x novice golfer-filmy putter. ER = experienced golfer—regular putter; EF = experienced golfer—funny putter. significantly more assessment steps in their generic descriptions than in their episodic recollections. t(17) = 4.03.p < .001. and the funny putter- experienced golfers gave somewhat more assessment steps in their generic descriptions. though the difference was not significant. ((17) = 1.72. p < .104. In contrast. the regular putter novices did not differ in terms of the number of assessment steps given in the generic and episodic protocols. t( 17) = 0.00. while the funny putter-novice group gave more assessment steps in their episodic recollections than in their generic protocols. t( 17) = 2.12. p < .049. Thus. similar to Experiment 1. assessment steps de- creased in number from the generic to episodic protocol for the Assessmmt “cherries Destlndicn 9 —o——m 51 u...” 7< +83 36‘ never 2 m5. '6 5‘. is. 3 22. 14 0* M -1 x e 1 .11 l s GamricEptaochenencEpiaodeGanencEpiaode Questionnaire Figure 6. Mean number of steps in each category for the second generic questionnaire and the first episodic questionnaire for each group. NR '- novice golfer-regular putter. NF - novice golfer—funny putter. ER .- experienced golfer—regular putter. EF I! experienced golfer~furmy putter. two experienced groups. regardless of type of putter used. whereas the opposite pattern occurred in the two novice groups. As an adjunct to the analysis of assessment. those steps that involved mental imagery (i.e.. imagining some aspect of how a putt ought to look or feel before executing the action) were counted. 1n the regular putter—novice group. 2.7% of generic steps and 1.5% of episodic steps referred to imagery. In the funny putter—novice group. 0.6% of generic steps and 1.9% of episodic steps referred to imagery. in die regular putter-expert group. 2.2% of generic steps and 5.0% of episodic steps involved imagery. Finally. in the funny putter-expert group. 1.2% of generic steps and 1.4% of episodic steps referred to imagery. Thus. as in Experiment 1. golf putting steps involving imagery were predom- inantly reported by experienced golfers using the regular putter. Turrting to mechanics. this analysis produced an interaction of Experience x Protocol x Putter. F(1. 68) = 5.26. p < .025. MSE= 3.43.ascanbeseeninthemiddlepanelofFigure6. A2 (experienced golfer. novice golfer) X 2 (funny putter. regular putter) general factorial ANOVA on the generic protocol produced a nonsignificant main effect of experience. F(1. 68) = 1.77. p < .188. M85 = 8.55 (though experienced golfers did give more mechanics steps in their generic protocols than novices in terms of absolute number); no main effect of putter. F(1. 68) = 1.18. p > .280. MSE = 8.55; and no interaction of Experience X Putter (F < 1). In the episodic protocol. a 2 (experienced golfer. novice golfer) x 2 (funny putter. regular putter) general factorial ANOVA produced an Experience x Putter interaction. F(1. 68) = 8.82. p < .004. M55 3 8.63. As can be seen from the middle panel of Figure 6. the experienced golfers using the funny putter gave more mechanics steps than any other group. while the regular putter- experienced golfers gave fewer mechanics steps than the other three groups. The two novice groups did not differ. 99 CHOKING UNDER PRESSURE 713 In addition. both regular and funny putter novices gave signif- icantly more mechanics steps in their episodic recollections than in their generic protocols. t(l7) = 2.20. p < .001. and ((17) = 1.84. p < .001. respectively. Similarly. the experienced golfers using the funny putter gave significantly more mechanics steps in their episodic recollections as compared with their generic protocols. r(l7) = 4.27. p < .001. In contrast. the regular putter experienced golfers gave fewer mechanics steps in their episodic recollections than in their generic descriptions. although this difference was not significant. t(l7) = 0.556. p < .585. The analysis of ball destinations produced a main effect of protocol type. F(1. 68) = 15.43. p < .001. MSE = 0.12; no main effect of experience or putter (Fs < 1); and no interaction of Protocol X Experience x Putter. F(1. 68) = 2.17. p > .145. MSE = 0.12. Thus. as shown in the right panel of Figure 6. regardless of putter type or expertise. ball destinations were more likely to appear in episodic recollections than generic protocols. As in Experiment 1. a second qualitative analysis looked for elaborations—steps present in both protocols that referred to the same action or biomechanism but provided more detail in one type of protocol than in the other. Because elaborations were more likely to occur in episodic descriptions relative to generic descrip- tions than vice versa. greater detail in the episodic description was scored as a positive elaboration whereas greater detail in the generic description was scored as a negative elaboration. In the regular putter—novice group. 11.1% of the steps in the episodic description were elaborations of steps in the generic descriptions. In the funny putter—novice group. 7.5% of the episodic steps were elaborations of generic steps. In the regular putter—experienced group. -0.3% of episodic descriptions were elaborations of ge- neric steps. Finally. in the funny putter-experienced group. 3.5% of episodic recollections were elaborations of generic steps. A 2 (experienced golfer. novice golfer) x 2 (regular putter. funny putter) general factorial ANOVA on these data produced a signif- icant main effect of experience. F(1. 68) = 8.33. p < .005. MSE = 1.28; a nonsignificant effect of putter (F < 1); and no Experience X Putter interaction. F(1. 68) = 1.97. p > .165. MSE = 1.28. Regardless of putter used. the novice golfers gave more elaborations in their episodic protocols than the more expe- rienced golfers. Furthermore. all groups gave more elaborations in their episodic descriptions than in their generic protocols. with the exception of the regular putter-experienced golfers. who gave fewer elaborations in their episodic recollections. In order to assess qualitative differences between the first and second episodic protocols. a 2 (experienced golfer. novice golfer) X 2 (funny putter. regular putter) x 2 (first episodic protocol. second episodic protocol) ANOVA was also performed on each of the three categories of steps (assessment. mechanics. destination; see Table 5). The analysis of assessment steps pro- duced a main effect of experience. F(1. 68) = 6.00. p < .017. MSE a 2.34; no main effect of protocol or putter; and no Expe- rience X Putter x Protocol interaction (Fs < 1). Thus. as can be seen from the left panel of Figure 7. the experienced golfers gave more assessment steps in both episodic questionnaires than did either group of novices. who did not differ. As an addition to assessment steps. the percentage of second episodic protocol steps involving references to mental imagery was also assessed. In the regular putter-novice group. 1.3% of second episodic steps referred to imagery. In the funny putter— 1 0 Assessment Mechanics Destination ‘ —e—Nn 9 0"" --I--NF 3 ‘ —a—En 7i H newt-:1: a 5+ ‘3 / e 5‘ E “ E 3 ‘ e.... 2« a:::: 1+ 0. H -1 Jr 3 l i 5 4. Episod‘l Episod2£pieodl£pleod2£pieod1£pbod2 Questionnaire Figure 7. Mean number of steps in each category for the first and second episodic questionnaires foreach group. Episod = Episodic; NR = novice gone—regular putter. NF = novice golfer—funny putter. ER = experienced golfer-regular putter. EF =- experienced golfer-funny putter. novice group. 2.2% of second episodic steps referred to imagery. In the regular putter-expert group. 4.8% of second episodic steps involved irmgery. Finally. in the funny putter-expert group. 1.3% of second episodic steps referred to imagery. Thus. as in the first episodic questionnaire. golf putting steps involving mental imag- ery were more likely to be found in the regular putter—experienced golfers' protocols than anywhere else. 11re analysis of mechanics produced a main effect of protocol. F(1. 68) = 8.13. p < .006. MSE = 1.73. and an Experience x Putter interaction. F(1. 68) = 6.07. p < .016. MSE = 17.87. A 2 (experienced golfer. novice golfer) x 2 (funny putter. regular putter) general factorial ANOVA was then performed on the combined average of mechanics steps involved in the first episodic questionnaire and second episodic questionnaire. revealing an Ex- perience x Putter interaction. F( 1. 68) = 6.07. p < .016. MSE = 8.93. As can be seen in the middle pure! of Figure 7. mechanics steps increased from the first to second episodic pro- tocol across all groups. Furthermore. the funny putter-experienced golfers gave more mechanics steps than any other group in either episodic questionnaire. whereas the regular putter—experienced golfers gave fewer mechanics steps than any other group. The two novice groups did not differ. Thus. although the regular putter- experienced golfers ltnew that they would be asked to give an episodicrecollectionofthelastputttheytookinthesecond episodic questionnaire. these golfers still gave fewer mechanics steps than both groups of inexperienced golfers and the experi- enced golfers using the funny putter. Analysis of ball destinations was not interpretable because of inhomogeneity of variance across groups for destination steps reported in the episodic questionnaires. However. as can be seen from the right panel of Figure 7. all groups gave similar absolute numbers of ball destination steps in both the first (M = 0.22. SD = 0.48) and the second (M = 0.25. SD = 0.47) episodic questionnaires. 100 714 BEILOCK AND CARR Discussion Experiment 2 yielded results similar to those of Experiment 1. Experts using the regular putter gave more elaborate generic representations of putting than novices using either type of putter. in parallel with diminished episodic accounts of particular in- stancesofperformance.Thoseexpertswhowereaskedtousethe funny putter also gave more detailed generic representations than did novices. However. in contrast to experts using the regular putter in both experiments. experts using the funny putter did not show diminished episodic memories for specific performances as would be expected if on-line performance was executed automat- ically and without leaving an explicit memory trace. In fact. funny putter experts gave more elaborate episodic descriptions of partic- ular instances of skill execution than did the experts using the regular putter and both novice groups. The results of Experiment 2 suggest that real-time putting performance for experienced golfers is supported by proceduralized knowledge that may be disrupted through the addition of novel task constraints. When this disrup- tion occurs and experts are forced to attend to step-by-step per- formance in the same way as novices. their expertise allows them to remember more of what they have attended to than do less skilled performers. This outcome resembles findings of superior episodic memory in chess and computer programming experts for the stimuli to which they have applied their knowledge (Chase & Simon. 1973; Soloway & Ehrlich. 1984). Further- more. regardless of the type of putter used. novice golfers in the present study produced similar putting performances and ge- neric and episodic putting descriptions. thus suggesting that in contrast to experienced golfers. novel skill performance in novice golfers is not based on a proceduralized. practice- specific skill representation. The results of Experiments I and 2 speak to the nature of skill representations at various levels of expertise. With respect to the sensorimotor skill of golf putting. it appears that highly practiced. well-leamed task components are encoded in a procedural form that supports effective real-time performance without requiring step-by-step attentional control. Reduced attention leads to a re- duction in declaratively accessible episodic memory for details of the performance. However. if task constraints (e.g., a funny putter) are imposed that force experienced golfers to alter execution processes in order to adjust to the novel environment. the proce- duralized skill knowledge that once drove nomial execution is disrupted. The consequence is a reduction in putting accuracy. an extended period of adaptation in which learning appears to proceed rather slowly. and a more detailed episodic memory trace for specific instances of skill execution. The notion that well-learned sensorimotor skill performance is governed by a proceduralized representation carries implications for how this type of skill will behave in pressure or attentionodemanding situations. Specifically. the two main theories that have been proposed to account for choking make different predictions concerning the types of skills that will be susceptible to performance decrements under pressure. Next we review these theories and how the cognitive mechanisms hypothesized by each dreary to account for choking under pressure may be related to the type of knowledge representation governing task performance. Experiments 3 and 4: Choking Under Pressure As outlined in the introduction. two types of theories have attempted to explain choking under pressure. Distraction theory proposes that pressure influences task performance by creating a distracting environment. Distraction-based accounts of suboptimal performance propose that performance pressure shifts attentional focus to task-irrelevant cues—such as worries about the situation and its consequences. In essence. this shifl of focus changes what was single-task performance into a dual-task situation in which cmtrolling the task at hand and worrying about the situation compete for attention. The most notable arguments for the distrac- tion hypothesis come from research involving academic test anx- iety (Eysenck. 1979; Kahneman. 1973; Wine. 1971). Individuals who become highly anxious during test situations. and conse- quently perform at a suboptimal level. are thought to divide their attention between task-relevant and task-irrelevant thoughts more so than those who do not become overly anxious in high—pressure situations (Wine. 1971). The distraction explanation for performance decrements under pressure is consistent with the idea that complex performances are attention or capacity demanding and that removing attention will disrupt performance (Humphreys & Revelle, 1984; Norman 8: Bobrow. I975; Proctor & Dutta, 1995). However. as demonstrated in Experiments 1 and 2. there are skills that become automated or proceduralized with extended practice and thus may not require constant on-line attentional control during execution (Anderson. 1987. I993; Fitts & Posner, 1967; Kimble & Perlmuter. 1970'. Proctor & Dutta. 1995; Squire & Knowlton. 1994). Such skills should be able to withstand the attentional demands of a dual-task environment in that explicit attention to stepby-step skill proce— dures is not mandatory for successful performance. The well- learned sensorimotor task of golf putting may be one such task. However. skills that rely on declaratively accessible knowledge even in their practiced state may behave quite differently under pressure. A potential example is Zbrodoff and Logan's (1986) alphabet arithmetic task. Alphabet arithmetic is a laboratory task analogous to mental arithmetic in which skilled performance is thought to be supported by the retrieval of stored instances of particular equations to which the performer has been exposed. Answers to an alphabet arithmetic problem such as “A + 2 = ?.” whereby individuals must count two units down the alphabet to obtain the answer “C." may be achieved in two ways: either by using a rule-based system or algorithm to solve the equation or by the stimulus-driven retrieval of past instances of the problem from memory. Logan (1988) assumed that solutions are derived by a race between these two processes. As exposure to examples of problems increases. instances stored in memory increase as well. The larger the base of instances stored in memory. the higher the probability that memory retrieval will provide an answer to the problem before the rule-based algorithm reaches a solution. In either case. the answer enters working memmy and hence is declaratively accessible—what differs is how it gets there (Klapp et al.. 1991). Logan‘s model is supported by changes in speed and accuracy of performance with practice on the alphabet arithmetic task and ether rapidly performed tasks involving judgment and choice. such as lexical decision and se- mantic categorization. 101 CHOKING UNDER PRESSURE 715 Answers to Logan‘s (I988) alphabet arithmetic task are thought to be declaratively accessible at all stages of skill learning. If choking is due to distraction of attention. one might imagine that choking would be a more imminent danger in tasks based on declarative knowledge that often enters working memory during the course of performance. because distraction of attention is a primary antewdent of corruption of information and forgetting in working memory (Daneman & Carpenter. I980. Muter. I980. Peterson a Peterson. 1959; Posner & Rossman. 1965). Further- more. if the distraction hypothesis is valid. then it is possible that training in a dual-task environment would enable performers to adapt to distraction and the concurrent allocation of attention to something other than the primary task. alleviating the negative impact of pressure (Hirst. Neisser. & Spelke. I978; Spelke. Hirst. & Neisser. I976). It has also been proposed that pressure situations raise anxiety and self-consciousness about performing correctly and success- fully. The resulting focus on the self prompts individuals to turn their attention inward on the specific processes of performance in an attempt to exert more explicit monitoring and control than would be applied when a high-achievement outcome is less desired and its consequences are less important (Baumeister. I984; Lewis & Linder. I997). Note that the essence of this proposal is exactly the opposite of the distraction hypothesis. The main idea behind the self-focus. or what we would like to term the explicit moni- toring hypothesis. is that although close attention Md control may benefit novice performers in the initial stages of task learning. it will become counterproductive as practice builds a more and more automated performance repertoire. This is due to the fact that explicit monitoring of step-by-step skill processes and procedures is thought to disrupt well-leamed or proceduralized skill execution processes (Kimble & Perlmuter. I970; Langer & Imber. 1979; Lewis & Linder. I997). Masters (I992) proposed that performance disruption occurs when an integrated or compiled real-time control structure that can run as an uninterrupted unit is broken back down into a sequence of smaller. separate. independent units—similar to how the performance was organized early in learning. Once broken down. each unit must be activated and run separately. which slows performance and. at each transition between units. creates an opportunity for error that was not present in the integrated control structure. In addition to the differences in the types of knowledge that may govern task performance. variations in complexity of skills may also mediate the pressure—performance relationship. That is. it may be that complex skills. involving the integration and sequencing of multiple steps or parts. are more prone to breakdowns and perfor— mance deficits than less complex one-step skills. Certainly the skill of golf putting involves such complexity. According to the explicit monitoring theory. then. the eorrplex. proceduralized sensorimotor skill of golf putting analyzed in Ex- periments l and 2 should be extremely susceptible to performance decrements under pressure. as it is unaccustomed to being explic- itly attended in real time. However. alphabet arithmetic. in which answers to particular problems are thought to be declaratively accessible at all stages of skill learning. should not be negatively affected by pressure-induced attention to performance processes. Furthermore. if the explicit monitoring hypothesis is valid. then training in an environment that heightens self-consciousness turd achievement anxiety is likely to alleviate the negative impact of pressure. by adapting performers to conditions that entice them to pay too much attention to step-by-step execution. Experiment 3 As a first step toward determining whether type of task knowl- edge and/or complexity might influence susceptibility to choking. we conducted Experiment 3. In this experiment participants learned either the sensorimotor task of golf putting or Zbrodoff and Logan‘s (1986) more declaratively based alphabet arithmetic task under single-task. dual-task. or self-consciousneseraising training conditions. Following training. participants were exposed to single-task low-pressure and high~pressure posttest situations in their task. Testing the hypotheses concerning the distraction and explicit monitoring theories of choking under pressure requires conuol over the training environment To ensure that our manipulation was the major source of each participant's golf putting or alphabet arithmetic experience. we recruited novice golfers and individuals with no exposure to alphabet arithmetic and taught them these tasks in the laboratory. However. despite the predictions we have nudewithrespectwmviceperforrnancechokingasaconceptis primarily aimed at individuals who can be expected to perform at a relatively accomplished level. Therefore. in order to examine performance at the later stages of skill acquisition. we trained participants rather heavily. Participants performed more than 280 golf putts or alphabet arithmetic trials in our laboratory prior to being exposed to a high-pressure situation. This number of task repetitions was chosen because pilot testing revealed a leveling off in performance with this amount of practice. suggesting that per- formance on the practiced putts or alphabet arithmetic problems was reaching asymptote. Method Participants Undergraduate students (N I 108) with little or no golfexperieoce who were eruolled in at introductory psychology class at Michigan State University served as participants. Participants were randomly assigned to either a single-task. self-consciousness. or duaHask distraction training groupineithudtegolfputtingoralphabetarithrnetictask. Eighteen participantstookpartineachu’aininggroup. Procedure: Golf Putting Task Aftrgivinginformedconseatandfillingoutademographicsheet concerning previous golf experiences. participants were told that the pur- pose of the study was to examine the aearracy of golf putting over several trials of practice. Participants were instructed that the object of the task was to putt a golf ball as accurately as possible from nine locations on a carpetedindoorputtinggreen(3 x 3.7 m)thatwereeither 1.2. l.4.or LSm away from a target. marked by a square of red tape. on which the ball was supposed to stop. All participants followed the same random alternation of putting from the nine different locations. A standard golf putter and golf ball were supplied. Participants took part in a 270-putt training condition followed by an l8-putt low-pressure posttest and an lB-putt high-pressure posttest described below. Single-task group. Participants were set up at the first putting spot. They were asked whether they preferred to putt right-handed or left-handed andweregiventheappropriaeputter. Puticipantsweretheninformedthat they would be putting from nine different locations on the green. each with 102 716 BEILDCK AND CARR a corresponding number. The experimenter reviewed the numbers associ- ated with each putting location arid asked participants to repeat back the numbers corresponding to each putting spot. Participants were informed that the experimenter would call out a number corresponding to a particular spot on the greett from which they were to execute their next putt. Participants then completed a total of 270 putts consisting of three training blocksof90puttseach. withashortbreakattereachsctoltarttsOn completion of the training condition. participants were given a short break during whichtheexperirnentercomputedthememdistancefromdtetarget of their last l8 putts. Participants then completed an l8-putt single—task low-pressure posttest. though they were not tirade aware of the test situation. To the participant. the low-pressure posttest appeared I) be just another series of putts. Participants were then informed of their mean putting performance for the last l8 putts in the training condition and given a scenario designed to create a high.pressure situation. Specifically. participants were told that if they could improve their accuracy by 20% in tire next set of palm. they would receive 55. However. participants were also informed that this monetary award was a "team effort.” Participants were told that they had been randomly paired with another participant. and in order to receive their 55. not only did they themselves have to improve by 20% but the partic- ipant that they had been paired with had to improve by 20% as well. Next. participants were informed that the individual they had been paired with had already completed the experiment and had improved by 20%. There- fore. if the present participant improved by 20%. both participants would receive 85. However. if the present participant did not improve by the required amount. neither participant would receive the money. Participants then took another 18 putts constituting the high-premure posttest. Follow- ing these putts. the experimenter computed the participants' ptttting ava- age and informed them of their performance. Finally. participants were fully debriefed and given the monetary award regardless of their performance. Distraction group. Participants were set up ot the first putting spot. They were asked whether they preferred to putt right-handed or left-handed and were given the appropriate putter. The experimenter informed partic- ipantsthattheywouldbepurtingfromrunelocationsondtegreatJhe experimenter then directed participants' attention to a tiny light that had been set up next to each putting spot Participants were informed that the lights were hooked up to a switchboard controlled by the experimencr. Participants were told that before every pitt. a light would illuminate beside the location from which they were to take their next port. The experimenter that explained to participnts that wll'le they were putting they would be listening to a recorded list of spoken words being played from a tape recorder. Participants were told to monitor the words carefully. mdeachunndteyhearddewordcogniriomtorepeatitbacktothe experimenter. Words were played at the rate of one every 2 s. The target word occurred randomly once every four words. Participants then com- pleted a total of 270 putts consisting of three training blocks of 90 putts each. with a short break after each set of putts during which the tape recorder was turned off. When participants completed the training condi- tion. thetaperecorderwas turned offend participantswere givenashort break during which the experimenter computed the mean distance from the target of their last l8 putts. Participants then took part in an I8-putt lowopressure posttest and an l8-putt high-premier posttest identical to that of the single-task group. Self-consciousness group. Participants were set up at the first putting location. They were asked whether they preferred to putt right-handed or left-handed and were given the appropriate putter. The experimenta thert explained that participants would be putting from nine different locations on the green. each with a corresponding number. Once the participants undastood the number-putting spot relationships. Ill: experimenter in- formed participants tltat they would be filmed by a video camera while putting.‘lhevideocamerawassetuponatripodtltatstoodonatable directly in front of participants. approximately l.8 m away. Participants were told that they would be videotaped so that a number of golf teachers and coaches at Michigan State University could review the tapes in order to gain a better understanding of how individuals learn a golf putting skill. The experimenter adjusted the camera and turned it on. Participants then corrpleted a total of 270 putts consisting of three training blocks of 90 putts each.withashortbreakafiereachsetofputtsthtringwhidtthevideo camera was armed off. When participants completed the training condition. the video camera was tttrned off and faced away. Participants were then given a short break in which the experimenter computed the mean distance frontthetargetoftheirlast l8putts.’lheparticipantsthentookpartinan lS-putt low-pressure posttest arid an l8-putt high-pressure posttest identi- cal to that of the single-task and distraction groups Procedure: Alphabet Arithmetic Task After giving informed consent and filling out a demop‘aphic sheet concerning previous golf arid alphabet arithmetic experiences. participants were told that the purpose of the study was to ertunine how individuals learned the alphabet arithmetic task. Participants were set up in front of a monitor eorarolled by a standard laboratory conputer. Participants were informed that they would be solving alphabet arithmetic equations such as “A + 2 = C” by counting two units down the alphabet to C. Next. the experimenter verbally presented three alphabet arithmetic equations to participants and instructed them to solve the equations out loud in order to ensure proper understanding. Participants were then shown a small key- board containing two buttons labeled True and False and laid to press the appropriate button when they derived the answer to an equation presented on the screen. Participants were instructed to try to judge the validity of the equations as quickly as possible without sacrificing accuracy. “re stimuli were capital letters. digits. the plus symbol (+). and the equal sign(=). All participantswerepresented with thesamerandomorder of rtine different equations. consisting of the letters (A. G. artd S) arid the digits (2. 3. and 4) that were equally randomly repeated. Each trial began with a fixation point exposed for 500 ms in the center of the screen. The fixation poirtt was immediately replaced by an equation. which remained onthescreertuntiltheparticipantpresseddteTrueorFalse buttononthe keyboard When the participant responded. the equation was extinguished. and the screen remained blank for a l.5-s intertrial interval. All participants took part in a 270-equation training condition in which each equation was randomly repeated 30 times. followed by an lB-equation low-pressure posttest and an l8-erpation high-pressure posttest. to be described below. inwhicheachequationappearednviceinarandomorder. Single-task group. Participants were set up in front of the monitor and given the alphabet aritlunetic instructions. Participants then convicted a total of 270 equations consisting of three training blocks of 90 equations each. withashort breakaftereach set. Following completion of the training condition. participants took part in an l8-equation single-task low-pressure posttest. similar to the low- pressure posttest in the golf putting task. Participants then took a short break and were givert a scenario designed to create a high-pressure situa- tion. Participants were told that the computer used a formula that equally took into account reaction time and accuracy in computing an “alphabet arithmetic performance score." The experimenter then described the same high-pressure scenario used in the golf ptttting task. Participants next coupleted the l8 equations constituting the high-pressure posttest. were fully debriefed. and were given the monetary award regardless of their perforrmnce. Distinction group. Participants were set up in front of the monitor and given the alphabet aritltrnuic instructions. Participants were also told that wltilethey werepat‘orrningthearidtmetictaskdteywouldbe listertingto a series of words being played through a headset. Participants were instructedtornortitordtewordscarefullyand.eachtimedieyhearddte wordcognitr'on. totxeasafootpedalthatwaslocahdneartheirfeetfl‘he experimenter then instructed participants to put on the headphones. move 103 CHOKING UNDER PRESSURE 717 the foot pedal to a comfortable location. and wactioe pressing it a few times. Participantsthm completed a total of 270 coir-nuns consisting of [firm My“ ' L-flncachset dunng which the lmdset was taken 0". Following completion of the training condition. participants were In- structedtoremovethe andmovethefootpedalaway from their feet.“ r r ‘ '° “we" - It: .4 '1 ' - L , I- ' i=2.— 0 —-v '-r- Self-construed”: group. Participaan were set up in front of the monitor and given the alphabet arithmetic uumrctiotn Participants were alsowld that they wouldbefilrnedbya videounuawhilesolvingthc equations. mvi videocanu-awassetupouan'ipoddirectiytotheleftot participants. amximfldy 0.9 m away. Participants wae told that they would be videotaped so that a number of math teaches at Michigan State University could review the tapes in order to detamine how quickly and accurately individuals learn the arithmetic skill. 11" experimenter adjusted the and turned it on. Participants then completed a total of 270 equations consisting of three training blocks of 90 equations each. with a short break after each set during which the vitbo camera was turned off. Following completion of the training condition. the video camen was Participants then took part in an lchuauo stand I8—equation lighpeuure posttest iduttical“ to that of the single-tad anddtsmction gasps Result: Putting Performance Accuracy of putting was measured by the distartce (in centime- ters) away from the center of the target that the ball stopped alter each putt. All three groups improved significantly with practice as demonstrated by a 3 (single-task. distraction. self- consciousness) X 2 (mean distance from target of first l8 putts in training condition. mean distance from target of last 18 putts in training condition) ANOVA revealing a main effect of practice. F(l. SD = 85.03.]: < .001. MSE = 27.57; a nonsignificant main effect of mining group. F(Z. Sl) = 0.658.]; > .522. MSE = 90.65: and no interaction. F(2. Si) = 0.214. p > .808. M58 = 27.57. As can be seen in Figure 8. although there was not a significant effect of training group. the distraction group' a performance was slightly. but not the self—consciousness groups both early and late in the training condition. These results coincide with research In the skill perfor- mance literature demonstrating that dual-task performance may lead to a decrement in the performance of a primary task that has not become fully automatized (Proctor & Dutta, I995). In the low-pressure posttest (measured by the mean distance from the target of the l8 putts in the low-pressure test). putting accuracy was nearly identical across groups (F < 1). However. this homogeneity disappeared in the high-pressure posttest (mea- sured by the mean distance from the target ofthe 18 putts in the high-pressure test) as confirmed by a one-way ANOVA revealing a significant difference between the three groups. F(2. 51) = 4.57. p < .0l5. MSE = 23.57. The above results are further supported by a 3 (single-task. distraction. self-consciousness) x 2 (low- pressure posttest. high- pressure posttest) ANOVA revealing a sig- nificant Interaction of Trarning Group X Posttest. F(2. 51)— “ 7. 37 p < ”002 MSE: 9.."‘71 ',puuing within each group showed that both the single-task group and the distraction group significantly declined in putting accuracy from the low-pressure posttest to the high-pressure posttest. t(l7) == it: 8 Mean distance from target (cm) to (II Finite Lana Lowpooi Highpost Putting condition 22 —O—S‘ngle-Taak 21 +Soll-Conacioue 20 ~ - o ~ ~Diatraction Mean distance from target (cm) 3 Lowpoet Putting concition 0' "reach poup irrirthe ',’ \wswul). Lowpost = 5"Iii—pressure power. Highpoet= high- lac-sure posttest Figure 8. Mean(: ”" “ ' ‘ , munch train' * ‘ " ' “1; r -2.2|. p < .04. and ((17) = —3.24, p < .(XJS. respectively. in contrast. the self-consciousness group improved in putting accu- racy from the low-pressure posttest to the high-pressure posttest. although this improvement was only marginally significant. t(l7)=l.81.p < .09. Thus. as can be seen in Figure 8. whereas both the single-task and distraction groups were adversely affected by the high-pressure situation. the self-consciousness group actu~ ally improved. Alphabet Arithmetic Performance Accuracy (shown in Table 6) was relatively high and did not differ significantly between groups both early (as measured by the mean number of correct judgments of the first l8 equations in the training condition). H2. 5]) = 0.178. p > .838. MSE = 5.94. and late (as measured by the mean number of correct judgments of the last 18 equations in the training condition). F(2. 5|) = 0.735. p > .485. MSE = 2.80. in the training phase. Early in training. the single-task. distraction. and self—consciousness groups' accuracy was 90%. 89%. and 87% correct. respectively. and late in training the single—task and distraction groups' accuracy was 93% correct and the self-consciousness group's was 96% correct. Similarly. there we no significant differences in accuracy between groups during either the low-pressure posttest (as measured by the mean 718 BEILOCK AND CARR Table 6 Mean Accuracy (Percentage Correct) in Training and Posttest Conditions of Alphabet Arithmetic Task 'haining 1 Training 2 (first 18 (1ast18 Low-pressure High-pressure equations) more) posttest posttest Group H SE M 85 M SE A! SE Single taslt 89.51 2.00 93.21 296 96.30 1.56 95.06 1.61 Self-consciousness 87.04 4.42 96.30 1.19 96.60 1.20 94.75 1.38 Distraction 89.20 2.64 92.90 2.05 94.14 1 .52 94.75 1 .05 number of correct judgments of the 18 equations in the low- pressure test). F(2. 51) = 0.879. p > .421. MSE = 1.2. or the high-pressureposttest(asmeasuredbythemeanmmberofcorrect judgments of the 18 equations in the high-pressure test). F(2. 51) = 0.017. p > .983. M55 = 1.09. During the low-pressure posttest. accuracy for the single-task. distraction. and self- consciousness groups was 96%. 94%. and 97% correct. respec- tively. and accuracy during the highpressure posttest was 95% correct for all three groups. Reaction timeswerecomputedforonlythoseequationsthat were answered correctly. Mean reaction times and standard errors both early and late in the training condition. as well as for the low- and high-pressure posttests. are illustrated in Figure 9. All three groups significantly decreased their reaction times across the train- ing condition as shown by a 3 (single-task. distraction. self- consciousness) x 2 (mean reaction time of first 18 equations in the training condition. mean reaction time of last 18 equations in the training condition) ANOVA revealing main effects of practice. m. 51) = 171.63.,» < .001. MSE = 6.2 x 10’. and training group. R2. 51) = 5.91. p < .005. MSE = 9.6 x 10’. and a marginally significant interaction. F(2. 51) = 2.88. p < .066. MSE = 6.2 X 10’ (see Figure 9). Tukey's honestly significant difference (HSD) tests on the main effect of training group further revealed that the distrxtion group performed significantly worse than both the single4aslr and self-consciousness groups (who did not differ) early in the training condition (p < .035 and p < .017. respectively) and significantly worse than the self-consciousness group (p < .05) and nonsignificantly worse than the single-task group (p < .227) late in the training condition. This pattern did not change in either the low-pressure or high-pressure pmttests as shown by a 3 (single-task. distraction. self-consciousness) x 2 (low-pressure posttest. high-pressure posttest) ANOVA that re- vealed main effects ofboth training group. R2. 51) = 4.41. p < .017. MSE = 2.4 x 10’. and posttest. F(1. 51) = 43.42.p < .001. MSE = 1.3 X 10‘. with nointeraction. £12.51) = 2.27.p > .114. MSE = 1.3 x 10‘. ‘hrltey's HSD tests on the main effect of training group further revealed that in the low-pressure posttest. the distraction group produwd significantly slower reaction times than the self-consciousness group (p < .04) and nonsignificantly slower reaction times than the single-task group (p < .293). In the high-pressure posttest. Tukey's HSD tests revealed that the dis- traction group produced significantly slower reaction times than both the selliconsciousnees (p < .008) and single-task (p < .04) groups. Thus. similar to golf putting accuracy. all three groups in the alphabet arithmetic taslt significantly improved their reaction times with practice. However. in contrast to the golf putting task. the dual-task alphabet arithmetic group performed substantially worse than the single-task and self-consciousness groups bath earlyandlateinthetrainingcondition.aswellasintheposttest situations. 11're main effect of low-pressure versus high-pressure posttests observed in alphabet arithmetic also contrasts with put- ting. All three training groups improved somewhat in the high- pressureposttest.showingnosignsofchoking underpreasure in the alphabet arithmetic task. 4.5“ «M1 m4 8.0001 2.”! 2M‘ 1.500 r 5.” l O Mean reaction time (ma) First 18 Lani. Lowest rm: Condition Mean reaction time (ma) '3 1.000 . a Lowpoat Highpoat Condition I-“tgrw9. Meanreactiontlmeut saforaridrmeticequationsinthe nainingandposnestcondidomforuchgroupinthealplubetanmmetic task(bp)mdposttestperfonmnceonly(bottom). Lowpost = low- prusuepoumflighpost=high-pressmeposm. 105 CHOKING UNDER PRESSURE 719 As an addendum. we should note that we believe the absence of choking in alphabet arithmetic to be a result of the fact that this type of skill does not become proceduralized with practice but moves from a declaratively accessible algorithm to retrieval of declaratively accessible facts into working memory. However. it may be that at higher levels of practice than we have examined. pressure-induced decrements in alphabet arithmetic performance could appear. Significant differences in performance between the dual-task training group and the other two groups in the alphabet arithmetic task at later stages of practice indicate that alphabet arithmetic performance was not yet fully autonutized during the high-pressure situation (see Klapp et al.. 1991). Thus. it may be that once differences in reaction times between these groups dis- appear. indicating full automatization. alphabet arithmdic will more closely resemble golf putting performance under pressure.2 To pursue this possibility. we conducted an analysis of alphabet arithmetic reaction times as a function of digit addend—the num- ber of counting steps up the alphabet required by each equation. An effect of this variable has been used to diagnose the extent to which the control structure of alphabet arithmetic has shifted from the counting algorithm (which produces a significant effect of addend) to memory retrieval (which is independent of addend). This analysis produced a significant interaction of digit addend by pretest versus high-pressure posttest. F(2. 104) = 7.06. p < .01. Reaction time averaged across training groups increased markedly as a function of digit addend in the pretest (M = 3.240 ms for two-digit addend. M = 3.690 ms for three-digit addend. M = 3.880 ms for four-digit addend). Reaction time in the high- pressure posttest flattened considerably (M = 1.185 ms for two- digit addend. M = 1.417 ms for three-digit addend. M = 1.310 ms for four-digit addend). though there was still a significant effect of digit addend averaged across training groups. Further analysis of individual training groups showed that the significant effect of digit addend during the high-pressure posttest was a result of the single-task and distraction groups (M = 1.084 ms for two-digit addend. M = 1.286 ms for three-digitaddend. M = 1.268 ms for four-digit addend and M = 1.324 ms for two-digit addend. M = 1.679 ms for three—digit addend. M = 1.533 ms for four-digit addend. respectively). Reaction time did not differ significantly as a function of addend for the self-consciousness group. F(2. 34) = 2.73. p > .1 (M = 1.098 ms for two-digit addend. M = 1.211 ms for three-digit addend. M = 1.125 ms for four-digit addend). These data indicate that the self-consciousness training group achieved the most automated alphabet arithmetic performance. diagnosed by the relative independence they showed between alphabet arithmetic reaction time and digit addend. This is consis- tent with the prediction that increased monitoring of task compo- nents enhances skill acquisition among novices undergoing train- ing (Anderson. 1987. 1993). If. similar to golf putting performance. those individuals trained under self-consciousness- raising conditions are immune to the detrimental effects of perfor- mance pressure. whereas those trained in single—task or distraction conditions are not. and the likelihood of choking increases as performance becomes more automated. then differences in high- pressure posttest reaction times should be apparent between the single-task and distraction groups on the one hand. and the self- consciousness group on the other. However. as can be seen in Figure 9. reaction time shows the same pattern across all three training groups. Thus it would appear that neither degree of au- tomatization as measured by the effect of digit addend nor the differential impact of training condition bears on whether choking is observed in alphabet arithmetic. Increasing the amount of prac- tice so that susceptibility to choking could be assessed in a com- pletely automatized alphabet arithmetic skill would serve to further clarify this issue. Putting Versus Alphabet Arithmetic in order to further verify the differences in performance across posttests in the golf putting and alphabet aritlunetic tasks. mea- surements taken in the putting and alphabet arithmetic tasks were converted into 2 scores. A 3 (single-task. distraction. self-con- sciousness) x 2(10w-pressure posttest. high-pressure posttest) x 2 (putting task. alphabet arithmetic) ANOVA was then performed. revealing a significant three-way interaction. F(2. 102) = 6.5. p < .002. M85 = .08. This confirms the pattern of data obtained above demonstrating that performance across the golf putting and alpha- bet arithmetic posttest conditions is different. Discussion Experiment 3 yielded three main results. First. following single- task practice. choking under pressure occurred in golf putting but not in alphabet arithmetic. Second. practice under dual-task con- ditions reduced performance in both tasks and altered practice benefits in alphabet arithmetic but did not alter either task's susceptibility to choking Finally. practice under conditions in- tended to raise self—consciousness and execution~oriented achieve- ment anxiety did not harm performance or change practice benefits relative to single-task practice in either skill but did inoculate putters against choking. Thus. at least at the levels of practice examined in the present study. choking arises in a task whose underlying knowledge base is thought to be procedural. but not one in which the underlying knowledge base is assumed to be more explicitly accessible. Furthermore. in terms of the effects of the two training regimens in the proceduralized task. it appears that when choking occurs. it results from explicit monitoring in re- sponse to self-consciousness and achievement anxiety. Perfor- mance pressure appears to elicit maladaptive efforts to impose step-by-step monitoring and control on complex. procedural knowledge that would have run off more automatically and effi- ciently had such monitoring not intervened. Practice at dealing with self-consciousness-raising situations counteracts this tendency. We now turn to Experiment 4 in which we sought to replicate and extend Experiment 3‘s findings concerning the choking under pressure phenomenon. Because we were interested in the mecha- nisms governing choking. and alphabet arithmetic did not appear to show decrements in performance under pressure, in Experi- ment 4 we only examined the sensorimotor task of golf putting. Experiment 4 In Experiment 4. the two possible sources of choking were examined at different stages of practice. It has been proposed that 2 We thank Stuart Klapp for suggesting this possibility. 106 720 autocx AND CARR early in skill acquisition performance is supported by unintegrated control structures that are held in working memory and attended to in a step-by-step fashion (Anderson. 1987. 1993; Fitts & Posner. 1967). With practice. however. control evolves toward the type of integrated procedures assumed by explicit monitoring theory. Ac- cording to explicit monitoring. tasks that follow this developmen- tal trajectory should benefit from performance pressure early in learning yet be susceptible to choking at later stages of practice. Attention to task components is thought to be an integral part of novel skill performance (Proctor & Dutta. 1995). The explicit monitoring theory predicts that perforrmnce pressure prompts individuals to attend to skill execution processes. Thus. at low levels of practice. performance pressure should facilitate skill execution by prompting novice performers to allocate more atten- tion to the task at hand. According to distraction theory. however. performance pressure serves to create a dual-task environment. If individuals are attend- ing to step-by-step execution processes in the early stages of skill learning. a distracting environment that draws attention away from the task at hand may harm performance. One could infer from the distraction hypothesis. then. that novice performers with little or no practice under divided attention conditions would be negatively affected by performance pressure. whereas those trained to a high skill level in a divided attention environment would not. In Experiment 4 participants learned a golf putting task to a high level of skill under dual-task or self-consciousness-raising training conditions and were subjected to identical single-task low- and high-pressure situations both early and late in the training phase. 1f distraction is the reason for suboptimal performance under pres- sure. then individuals trained in either a dual-task or self- consciousness-raising environment should show performance dec- rements in pressure situations early in skill learning because. at this point. individuals in either training condition have not adapted to performing under divided attention conditions and do not possess a proceduralized skill response. Later in learning. however. those individuals trained in a dual-task environment will presumably be accustomed to performing under divided attention conditions and thus will notbe affected by pressurewhereastheperformanceof those trained under conditions designed to increase anxiety and self-consciousness should decline. In contrast. if explicit monitor- ing is the reason for skill decrements under pressure. than at low levels of practice individuals trained under either distraction or self-consciousness-raising conditions should improve under pres— sure. 1f. as the explicit monitoring hypothesis predicts. pressure induces attention and control to skill performance. then novice perfomiers may benefit from performance pressure in the initial stages of task learning. However. once the golf putting skill ha become proceduralized later in practice. only those individuals who have adapted to performance anxiety and the demands to explicitly monitor skill performance (i.e.. those trained under self. consciousness-raising conditions) should improve under pressure. Method Participants Undergraduate students (N = 32) with little or no golf experience who were enrolled in an introductory psychology class at Michigan Stat: University served as participants. Participants were randomly assigned to either a self-consciousness (n = 16) or dual-task distraction training group (n = 16). Procedure Participants completed the same golf putting task as in Experiment 3. Individuals took part in a 27-putt training condition followed by the first l8-putt single—task lowoptusure posttest and l8-putt single-task high. pressure posttest. Participants then took pan in a 225-putt training condi- tion followed by a second 18-purt single-task low-pressure posttest and a second 18-putt single-task high-pressure posttest. Disrmar'on group. Participants completed the ptrtting task in the sarrte distraction training environment used in Experiment 3. Participants com- pletedthefirstuainingconditionoffl putts. afterwhichthetaperecorder wastumedoff.Participantswerethengivenashortbreakdmingwhichme experimentercomputedthemeandistancefromthetargetofmeirlast l8 putts. Following the first training condition. participants completed the first 18-putt single-task low-pressure posttest. though they were not made aware of the test situation. Puticipants were then informed of their mean putting mforrmnce for the last 18 putts of the first training condition and given a scenario designed to create a high-treasure situation. Specifically. panic ipants were told there would be two test situations in the present experi~ ment. Participants were informed that they were about to take part in the first test situation and that the second test situation would take place toward the end of the experiment. Participants wae given the same high-treasure scenario used in the golf putting task in Experiment 3. with the exception that they were told that they needed to improve their putting accuracy in two test situations to receive the monetary award. Participants then took the 18 putts constituting the first single-task high-pressure posttest. Fol~ lowing these putts. the experimenter informed participants that their putting average would be computed at the end of the experirrient after both test situations had been completed. Participants weredieritoldthatmeywouldbetakinganodiaseriesof practice putts under the sane dual-task conditions. The experimenter turned on the tape recorder. and panieipams completed the second training condition consisting of a ton] of 225 putts broken down into one training blockof72putts.atrainingblockof81ptrnsandanodreruainingblock of 72 putts. with a short break after each block in which the tape recorder was turned off. When participann completed the second training condition thetaperecorderwasturnedoff. Participantswerethengivenashonbreak during which the experimenter computed the mean distance from the target of the last 18 putts in the training condition. Participants then took part in the second 18-putt single’task low-pressure posttest. As in the first low-pressure posttest. participants were not made aware of the test situation. The experimenter then informed participants that they were about to take part in the second test situation and repeated the high-pressure scenario. Participants completed the second 18—putt single-wk high-pressure posttest. were fully debriefed. and were given the monetary award regardless of their performance. Self-consciousness group. Participants were set up at the first putting location and given instructions similar to those given to the distraction group regarding putting from the nine different locations on the green. Participants completed the putting task in the same self-consciousness- training environment used in Experiment 3. Participants completed the first training condition of 27 putts. after which the video camera was turned off and faced away from participants. Following the first training condition. participants were given a short break during which the experimenter computed the mean distance from the target of their last 18 putts. Partic- ipants then completed the first 18-putt single-task low‘pressure posttest and high-pressure posttest identical to that of the distraction group. Participantsweretheninforrmdthattheywetegoingtocomplete another series of practice putts. again while being filmed by the video carnaa. The experimenter turned on the video camera. and participants 107 CHOKING UNDER PRESSURE 72 l convicted the second training condition consisting of 225 total putts broken down into one training block of 72 putts. a training block of 81 putts. and another training block of 72 putts. with a short break after each block during which the video camera was turned ofl’. When participants completed the second training condition. the camera was ntmed off and faced away from participants. Participants next completed the second 18-putt single-task low-pressure posttest and high-pressure posttest iden- tical to that of the distraction group. Following the second high-pressure posttest. participants were fully debriefed and were given the monetary award regardless of their performance. Result: Accuracy of putting was measured by the distance (in centime- ters) away from the center of the target at which the ball stopped after each putt. Both groups improved significantly with practice as demonstrated by a 2 (distraction. self-consciousness) x 2 (mean distance from target of first 18 putts in first training condition. mean distance from target of last 18 putts in second training condition) ANOVA revealing a main effect of practice. F(1. 30) = 58.63. p < .001. MSE 3 31.73; a nonsignificant main effect of training group (F < I); and no interaction (F < 1; see Figure 10). In the first low-pressure posttest (measured by the mean dis- tance from the target of the 18 putts in the first low-pressure test). putting accuracy was similar across groups. F(I. 30) = 1.40. p > .245. MSE = 26.15. This homogeneity continued in the first high-pressure posttest (measured by the mean distance from the target of the 18 putts in the first high-pressure test) as confirmed by a one-way ANOVA again revealing no significant difference be- tween groups (F < I). These results are further supported by a 2 (distraction. self-consciousness) X 2 (first low-pressure posttest. fust high-pressure posttest) ANOVA revealing a significant effect of test. F(1. 30) = 17.73. p < .001. MSE = 12.38; no significant effect of group. F(1. 30) = 1.00. p > .323. MSE = 34.96; and no interaction (F < 1). Direct comparisons of putting performance within each group showed that both the distraction and the self consciousness groups significantly improved in putting accuracy from the first Iow-pressure posttest to the first high-pressure post- test. t(15) = 3.76. p < .002. and ((15) = 2.30. p < .036. respectively. As can be seen in Figure 10. following the first high-pressure posttest. both training groups’ performance accuracy decreased This was confinned by a 2 (distraction. self-consciousness) x 2 (first high-pressure posttest. first 18 putts of second training con- dition) ANOVA revealing a significant effect of condition. F( 1. 30) = 4.92. p < .034. MSE = 10.83; no main effect of training group (F < 1); and no interaction (F < 1). Thus. whereas the first high-pressure posttest led to an increase in golf putting accuracy in comparison to the first low-pressure posttest in both the distraction and self-consciousness training groups. both groups showed per. formance decrements in the initial putts following the high- pressure situation. The explicit monitoring hypothesis suggests that performance pressure prompts individuals to explicitly mon- itor skill execution. Under this hypothesis. one would expect individuals in the initial stages of skill learning to improve under pressure as a result of increased attention to the novel demands of skill execution. However. once performance pressure and in- creased monitoring of performance are alleviated. a reduction in accuracy should occur. :1 27. 24a 21- 10< 15a Moan distance from target (an) 124 9 V V f V j Avl LP! HP1 Ate AW! LP2 HP2 Puttingcondition A 24 nO-oDistroction g i +Soll-Oonadoua g 214 s ”f E 184 g 154 3 12. 9 r LP2 W2 Puttingcondtion Figure 10. Top: Mean (1 SE) distance from the target at which the ball stoppedafiereachputtforticfirst pruttsinthefirsttrainingcondition (Avl). the first low-pressure posttest (LPI ). the first high-pressure posttest (HPI), the first 18 pum of the second training condition (Av2). the lat l8 putts of the second training condition (Av3). the second low-pressure poamst (LP2). and the second high-pressure posttest (HP2). Bottom: Second posttest performance only. in the second low-pressure posttest (measured by the mean distance from the target of the 18 putts in the second low-pressure test). putting accuracy was similar across groups. F(1. 30) = 1.27. p > .269. MSE = 12.11. This homogeneity disappeared in the second high-pressure posttest (measured by the mean distance from the target ofthe 18 putts in the second highpressure test) as confirmed by a one-way ANOVA revealing a significant differ- ence between groups. F(1. 30) = 10.43. p < .003. MSE = 14.87. The above results are further supported by a 2 (distraction. self- consciousness) X 2 (second low-pressure posttest. second high- pressure posttest) ANOVA revealing a significant interaction of Training Group x Second Posttest. F(1. 30) = 24.16. p < .001. MSE = 5.55. Direct comparisons of putting performance within each group showed that the distraction group significantly declined in putting accuracy from the second low-pressure posttest to the second high-pressure posttest. t(15) = -2.79. p < .014. In con— trast. the self-consciousness group improved in putting accuracy from the second low-pressure posttest to the second high-pressure posttest. r(lS) = 4.84.p < .001.1hus. as can be seen in Figure 10. both the distraction and self-consciousness groups improved from the first low- to the first high-pressure posttest. However. later in 108 722 BEILOCK AND CARR learning. those individuals trained in a dual-task environment showed decrements in performance under pressure. whereas those who learned the golf putting task under conditions designed to foster adaptation to a self-consciousness-raising environment that would increase achievement anxiety and explicit monitoring actu- ally improved. Discussion The results of Experiment 4 once again are consistent with the predictions of the explicit monitoring theory of choking under pressure. Early in practice. regardless of training environment. performance pressure facilitated skill acquisition. However. as the golf putting skill became more proceduralized at later stages of practice. only those individuals who were accustomed to perform- ing under conditions that heightened performance anxiety and the explicit monitoring of task processes and procedures were inocuo lated against the detrimental effects of performance pressure. These findings lend support to the notion that increased attention to the execution of a well-learned. complex skill may disnipt skill execution. General Discussion The purpose of the present study was to explore the cognitive mechanisms responsible for the disruption in the execution of a well-learned skill under pressure. Experiments 1 and 2 assessed the declarative accessibility of the knowledge representations govern- ing real-time performance of golf putting at various levels of expertise. Results conformed in remarkable detail to predictions derived from current theories of automaticity and proceduraliza- tion of task performance as a function of practice. In Experi- ments 3 and 4 we looked at the phenomenon ofchoking under pressure in two very different tasks: the complex. sensorimotor skill of golf putting and a simpler. declaratively based alphabet arithmetic task. Results showed that choking under pressure oc- curred in golf putting but not in alphabet arithmetic. which dem- onstrates that sequential complexity. prowduralization. or both determine susceptibility to choking at the levels of practice we have examined. Furthermore. it was found that a particular training environment can elirrunate choking when it does occur. Whereas single-task and dual-task practice left individuals in the golf put- ting task susceptible to performance decrements under pressure. self-consciousness training eliminated choking completely. In- deed. performers who experienced self-consciousness uaining ac- tually improved under pressure—a highly desirable mull. In ad- dition to supporting the explicit monitoring hypothesis about why choking occurs. these experiments lead immediately to very prac- tical ideas about training for real-world tasks in which serious consequences depend on good or poor performance in relatively public or consequential circumstances. Properties of Task: That are Susceptible to Choking The pattern of results found in the present study speaks to the kinds of task properties that should be considered in the investi- gation of the pressure—performance relationship. Evidence of choking in the complex. proceduralized sensorimotor skill of golf putting but not in the simpler. declaratively based alphabet arith- metic task suggests at least three task properties that may be involved in choking. The first is task complexity. Masters (1992) argued that performance pressure prompts attention to skill exe- cution. which results in the breakdown of task components. Well- learned complex skills my possess on-line control structures that run off as uninterrupted units. When attended to. these units may be broken down into a sequence of smaller. independent units. each of which must be run separately. As a result. performance slows and the transition between units creates an opportunity for error that was not present in the integrated control structure. Although skill breakdown may occur in complex. multistep tasks. this may not be the case in simple. one-step retrieval tasks. One- step retrieval tasks are not thought to consist of multiple integrated units and thus may not be susceptible to dismantling in the event of performance pressure. According to current theory. the auto- mated forrn of alphabet arithmetic is such a onestep task (Klapp et al.. 1991; Logan. 1988: Logan & Klapp. 1991). However. our data indicate that alphabet arithmetic was not fully automated among our participants and hence was supported by some mixture of one-step fact retrieval and the multistep algorithm bued on counting through the alphabet. Nevertheless. no hint of choking was observed in alphabet arithmetic. If task complexity is involved in choking. then performance decrements might have been expected at least on those trials supported by the multistep algorithm. In contrast. if task complexity does not affect susceptibility to choking. and instead choking occurs for alphabet arithmetic equations based completely on fact retrieval. then one might expect to see some indication of performance decrements for the portion of alphabet arithmetic equations that have switched to a fact retrieval mechanism. Either the former or latter of these possibilities should affect overall reaction time. However. as can be seen in Figure 9. alphabet arithmetic reaction time shows the same pattern both across training groups and between the low- and high-pressure conditions. This leads to the second task property that may be involved in mediating the pressure—performance relationship. This is the de- gree to which task components become proceduralized with prac- tice. Attention to the explicit processes involved in skill execution is thought to decrease as a function of skill level (Anderson. 1987. 1993; Fitts & Posner, I967). As a result. skilled performances (cg. complex sensorimotor tasks) are thought to operate largely outside of working memory. However. there may be certain skill types that rely on working memory for storage of control-relevant information during all stages of skill acquisition. Alphabet arith- metic is one such task. There is a substantial body of evidence demonstrating that performance on practiced alphabet arithmetic problems is not based on the establishment of a proceduralized version of the algorithm that controls action directly with no involvement from working memory. instead. practice results in a shift from running through the steps of the algorithm in working memory to retrieving the answer into working memory from episodic memory (Klapp et al.. 1991; Logan. 1988'. Logan & Klapp. 1991). In either case the answer enters working memory. from where it controls the choice of an overt response. if choking is due to explicit monitoring. such skills should not be susceptible to decrements in performance because the practiced version of this task does not rely on the right kind of complex but proceduralized control structure. First. as already discussed. the control structure is too simple. and. second. control-relevant information always 109 CHOKING UNDER PRESSURE 723 enters working memory and hence is always declaratively ms- sible. The alphabet arithmetic equation as a perceptual stimuhis retrieves a single piece of information from episodic memory as the answer. and the elements of the control structure. the perceived equation and the retrieved answer. enter working memory rather than remaining outside the we of attention. as does the relatively encapsulated promdure or motor program. Finally. cognitive and motor tasks may differ in their suscepti- bility to breakdowns under pressure. The present study demon- strated that the sensorimotor task of golf putting. but not the cognitive task of alphabet arithmetic. wm negatively affected by performance pressure. From these results it is tempting to conclude that choking may be confined to sensorimotor skills. However. such a conclusion is problematic in that it does not speak to the specific task characteristics that make a skill vulnerable to break- downs under pressure. As mentioned in the introduction to this study. sensorimotor skills in both real-world and experimental settings are often associated with the choking phenomenon Yet. the apparent prevalence of choking in sensorimotor domains may be nor a function of sensorimotor skills per se but instead a result of specific task characteristics embedded in sensorimotor tasks that are susceptible to performance pressure (e.g.. complexity and/or proceduralization). Furthermore. the notion that choking is limited to sensorimotor skills contrasts with research in the educational psychology literature demonstrating decrements in academic test performance under pressure in highly anxious individuals (Ey- senck. I979; Kahneman. 1973; Wine. 1971). Clearly such aca- demic test performances do not have a large sensorimotor compo- nent. yet evidence of choking under pressure still emerges. Distraction theorists have suggested that suboptimal academic test performance results from the creation of a dual-task. distracting environment in which attention is divided between the task at hand and worries about the situation and its consequences. Thus. it remains a possibility that distraction as a mechanism for choking does hold for certain task types. It may be that pressure-induced distraction is detrimental to performance in tasks in which a large amount of information must be held in working memory and is susceptible to interference when attention is allocated to secondary sources (see. e.g.. Tohill & Holyoak. 20W). This is a notion that is open to exploration in future work. However. it may also be the case that the types of problems encountered in these cognitive- based academic test situations have characteristics in common with many sensorimotor skills (e.g.. complexity and/or prowdur- alizability) and are thus vulnerable to the same type of negative performance effects associated with the explicit monitoring of task execution processes. For example. Anderson (1993) suggested that complex cognitive skills such as algebra. geometry. and computer programming may become largely proceduralized in experts. We are pursuing this possibility in our laboratory. Choking Research in Social Psychology The present findings accord with research in the social psychol- ogy literature concerning the relationship between arousal. atten- tion. and performance. It has been demonstrated that heightened anxiety and/or arousal levels induce self-focused attention (Fenig— stein & Carver. I978; Wegner & Giuliano. I980). Wegner and Giuliano postulated that increments in arousal prompt individuals to turn their attention inward on themselves and current task performance in an attempt to seek out an explanation for their aroused state. Similar results have been found for skill execution in the presence of an audience. Butler and Baumeister ( 1998) recently demonstrated that supportive audiences were associated with un- expected perfomiance decrements in the execution of complex. procedurally based tasks. The authors proposed that supportive audiences may increase attention to the processes involved in well-learned task performance. thus disrupting performance pro cesses. And. finally. in a recent study investigating the effects of pressure on golf putting performance. Lewis and Linder (I997) found that pressure caused choking when participants had not adapted to performing in self-awareness-heightened environ- ments—results similar to the present study. Furthermore. Lewis and Linder also found that decrements in performance could be alleviated through the use of a distractor (in this case. counting backward from 100) during real-time performance. Lewis and Linder suggested that attending to the distractor during on-line performance under pressure prevented participants from focusing attention inward on skill execution processes. thus alleviating the possibility of choldng due to the “self-focus mediated misregula- tion" (p. 937) of performance. As can be seen from the literature just described. the notion that performance pressure induces self- focused attention. which in turn may lead to decrements in skill execution. is now a reasonably well-supported concept for proce- duralized skills. In conclusion. the findings of the four experiments in the present study lend support to the notion that pressure-induced attention to the well-leaned components of a complex. proceduralized skill disnipts execution. Future research in this area is needed in order to illustrate the precise nature of the control structures that lead to decrements in performance under pressure. In addition. the gener- alizability of the present results to other task types and to different levels of practice must also be assessed. Further exploration of both the task and learning environments that mediate the pressure— performance relationship will serve to enhance our understanding ofthe choking under pressure phenomenon. which stands as an intriguing exception to the general rule that well-leamed skills are robust and resistant to deterioration across a wide range of conditions. References Allard. F.. & Starkes. J. L (l99l). Motor-skill expats in sports. dance and otha donnins. In K. A. Ericsson & J. Snith (Eds). Toward a general theory of expertise (pp. 126452). Cambridge. England: Cambridge University Press. Anderson. I. R. (I982). Acquisition of a cognitive skill. Psychological Review. 89. 369—406. Anderson. J. R. (1987). Skill acquisitiort' Compilation of weak-method problem solutions Psychological Review. 94. 192-210. Anderson. J. IL (1993). Rules ofmind. Hillsdale. NJ: Erlbaum. Baumeistu'. R. F. (1984). Choking under pressure Self-consciousness and paradoxical effects of incentives on skillful performance. Journal of Personality and Social Psychology. 46. 6I0—620. Bransford. .I. D.. & Johnson. M. K. (1972). Contextual prerequisites for understanding. Sorm investigations of comprehension and recall. Jour- nal d Verbal Learning and Verbal Behavior. ll. 7l7-726. Brown. T. L. & Carr. 1'. H. (I989). Autormticity in skill acquisition: Mechanisms for reducing interference in concurrent performance. Jour- nal of Experimental Psychology: Ham Perception and Perfor- mance. 15. 686-700. 110 724 BEILOCK AND CARR Butler. J. L. & Baumeister, R. F. (I998). The trouble with friendly faces: Skilled performance with a supportive audience. Joumal (1 Personality and Soaal Psychology. 75. IZI3—IZ30. Chase. W. 6.. & Simon. H. A. (I973). Perception in chess. Cognitive Psychology. 4. 55—8I. Chi. M. T.. Feltovitch. P. J.. «I. Glasa. R. (I98I). Categorization and representation of physics problems by experts and novices. Cognitive Science. 5. IZI—ISZ. Craik. F. M., Govini. R.. Naveh-Benjamin. M.. & Anderson. N, D. (I996). The effects of divided attention on encoding and retrieval processes in human memory. Journal of Experimental Psychology: General. 125. I59—180. Daneman. M.. & Carpenter. P. A. (I980). Individual differences in working memory and reading. Journal of Verbal Learning and Verbal Behav- tor. [9, 450-466. De Groot. A. (I978). Thought and choice in chess. The Hague. the Netherlands: Mouton. (Original work published I946) Dooling. D. J.. & Christiaansen. R E (I977). Episodic and semantic aspects of memory for prose. Journal of Experimental Psychology: Human Learning and Memory. 3. 428—436. Ericsson. K. A.. Krampe, R. T.. & Tesch-Romer. C. (I993). The role of deliberate practice in the acquisition of expat performance. Psycholog- ical Review. I00. 363-406. Ericsson. K. A.. & Smith. J. (l99l). Prospects and limits of the empirical study of expertise: An introduction. In K. A. Ericsson 8: J. Smith (Eds). Toward a general theory (1 expertise (pp. l-38). Cambridge. England: Cambridge University Press. Eysenck. M. W. (I979). Anxiety learning and memory: A reconceptual- ization. Journal of Research in Personality. 13. 363—385. Fenigstein. A.. & Carver. C. S. (I978). Self-focusing effects of heart- beat feedback. Journal of Personality and Social Psychology. 36. 1241—1250. Fisk. A. 0.. & Schneider. W. (I984). Memory as a function of attention. level of processing. and automatization. Journal 4' Experimental Psy- chology: Learning. Memory. and Cognition. 10. l8l-I97. Fitts. P. M.. & Posner, M. I. (I967). Human performance. Belmont. CA: Brooks/Cole. Hardy. L. Mullen. R.. a Jones. 0. (I996). Knowledge and conscious control of motor actions under stress British Journal of Psychology. 87. 62l—636. Hirst. W.. Neisser. U.. a Spelke. E. (I978). Divided attention. Human Nature. I. 54-61. Humphreys. M. S.. & Revelle, W. (I984). Personality. motivation. and performance: A theory of the relationship between individual differ- ences and information processing. Psychological Review. 9]. l5}- I84. Jones. B. T.. Davis. M., Crenshaw. 8.. Behar. T.. & Davis. M. (I998). Classic instruction in golf New York: Broadway. Kahneman. D. (I973). Attention and efl’ort. Biglewood Cliffs. NJ: Pruitice Hall. Keele. S. W. (I986). Motor control. In K. R. Boff, L Kaufman. a J. P. Thomas (Eds). Handbook of perception and human performance (Vol. 2. pp. l-60). New York: Wiley. Keele. S. W.. & Summers. J. J. (I976). The structure of motor programs. In G. E. Stelmach (Ed.), Motor control: Issues and trends (pp. l09— I42). New York: Academic Press. Kimble. G. A.. Jr Perlmuter. L. C. (I970). The problem of volition. Psychological Review. 77. 36I—384. Klapp. S. T.. Bochcs. C. A.. Trabct‘l. M. L. & Logan. 0. D. (I991). Automatizing alphabet arithmetic: II. Are there practice effects after automaticity is achieved? Journal of Experimental Psychology: Learn- ing. Memory. and Cognition. 17. l96-209. Linger, 5.. & Imber, 6. (I979). When practice rmkes imperfect: Debili- tating effects of overlearru'ng Journal of Personality and Social Psy- chology. 37. 20l4-2024. Leagold. A.. Robinson. H.. Feltovitch. P.. Glaser. R.. Klopfer. D.. a Wang. Y. (I988). Expertise in a complex skill: Diagnosing X-tay pictures. In M. T. H. Chi. R. Glaser. & M. J. Farr (Eds). The nature ofexpertise (pp. 3| l-342). Hillsdale. NJ: Erlbaum Lewis. 8.. & Linder. D. (I997). Thinking about choking? Attentional processes and paradoxical performance. Personality and Social Psychol- ogy Bulletin. 23. 937-944. Logan. 0. D. (I988). Toward an instance theory of automatization. Psy- chological Review. 95. 492—527. Logan. G. D.. & Klapp. S. T. (I991). Automatizing alphaba arithmetic: I. Is extended practice necessary to produce automaticity? Journal of Experimental Psychology: Learning. Memory. and Cognition. 17. I79- I95. Masters. R. 8. (I992). Knowledge. knerves and know-how: The role of explicit versus implicit knowledge in the breakdown of a complex motor skill under pressure. British Journal of Psychology. 83. 343—358. McCandIiss. 3.. & Cart. T. H. (I994. Novenber). Washing clothes twice.- Do repetition and referential interpretation influence the same processes duringreading??apapreaentedattheannualmeetingofthe?xy- chooomic Society. St. Innis. M0. McCandIiss. 8.. & Carr. T. H. (I996. November). Repetition. comprehen- sion. and intention to learn as sources of memory for text. Paper presented at the amual meeting of the Psychononuc Society. Chicago. Mater. P. (I980). Very rapid forgetting. Memory ii Cognition. 8. 174-179. Naveh—Benjarnin. M., Craik. F. l.. Guez. J., & Dori. H. (1998). Effects of divided attention on encoding and retrieval processes in human memory: Further support for an asymmetry. Journal of Experimental Psydrology: Learning. Memory. and Cognition. 24. I091—IIO4. Norman. D. A.. & Bobrow. D. J. (I975). 0n data-limited and resource- Iimited processes. Cognitive Psychology. 7. 44—64. Peterson. L. R.. & Peterson. M. J. (I959). Short-tam retention of individ- ual verbal item. Journal of Experimental Psychology. 58. l93—I98. Posner, M. L. & Rossmn. E (I965). Effect of size and location of inforrnatioml transforms upon short-term retention. Jourrml of Experi- mental Psychology. 70. 496-505. Priest. A. G.. 5: Lindsay. R. 0. (I992). New light on novice-expert differences in physics problem solving. British Journal of Psychol- ogy. 83. 389-405. Proctor. R. W.. & Dutta, A. (I995). Skill acquisition and human perfor- mance. Thousand Oaks. CA: Sage. Raichle. M. E. Fiez. J. A.. Videen. T. 0.. Macleod. A. M. K.. Pardo. J. V.. Fox. P. T.. a Petersen. S. E. ( I994). Practicemelated changes in human brain functional-anatomy during nonmotor learning. Cerebral Cortex. 4. 8-26. Reimann. P.. & Chi. M. T. (I989). Human expertise in complex problem solving. In K. J. Gilhooly (Ed.), Human machine and problem-solving (pp. l6I-l89). New York: Plenum. Soloway. 8.. & Ehrlich. K. (I984). Empirical studies of programming knowledge. IEEE Transactions on Software Engineering. SE-IO. 595- 609. Spelke. P.. Hirst. W., & Neisser. U. (I976). Skills of divided attention. Cognition. 4. 215—230. Squire. L. R.. 8: Knowlton. B. J. (I994). The organization of memory. In H. Morowitz & J. L. Singer (Eds). The mind. the brain. and complex adaptive systems (Vol. 22. pp. 63—97). Reading. MA: Addison-Wesley. Tohill. J. M., & Holyoak. K. (2000). The impact of anxiety on analogical reasoning. Thinking and Reasoning, 6. 27-40. Van Dehn. K. (I989). Problem solving and cognitive skill acquisition. In M. I. Posner (Ed). Foundations of cognitive science (pp. 527—580). Cambridge. MA: MIT Pleas. 111 CHOKING UNDER PRESSURE 725 Voss. J. P.. Jr Post. T. A. (1988). On the solving of ill-structured problems. in M. T. H. Chi. R. Glaser. & M. J. Farr (Eds). The nature ofexpertise (pp. 26l—285). Hillsdale. NJ: Erlbaum. Wegner. D. M. & Giuliano. T. (I980). Arousal-induced attention to self. Journal of Personality and Social Psychology. 38. 719—726. Wine. J. (l97l). Test anxiety and direction of mention. Psychological Bulletin. 76. 92- I 04. Woolfolk. K L. Murphy. 8. M., Gottesfeld. D.. a Aitken. D. (I985) Effectsofmemalreheanalofnskmotoracdvityandmentaldepiction of task outcome on motor skill performance. Journal of Sport Psychol- ogy. 7. l9l—l97. Zbrodoff. N. J.. & Logan. 0. D. (1986). On the autonomy of mental processes; A case study of arithntetic. Journal of Experimental Psychol- ogy: General. 115. ll8—l30. Appendix A Questionnaires Generic Questionnaire—Experiments l and 2 Certain steps are involved in executing a golf putt. Please list as many steps that you can think of. in the right order. which are involved in a typical golf putt. Episodic Questionnaire—Experiment l' Pretend that y0ur friend just walked into the room. Describe the last putt you took. in enough detail so that your friend could perform the same putt you just took. Episodic Questionnaire—Experiment 2h Pretend that your friend just walked into the room Describe the last putt you took. in enough detail so that your friend could duplicate that last putt you just took in detail. doing it just like you did. 'Additional explamtion was given inorderto make it clear that what was being asked for was a “recipe“ or “set of instructions" that would allow theputttobeduplicatedinallitsdetailsbysomeonewhohadnotswnit. Golfteammcmbasweretolddtatthcfriendwasnotanothergolfteam member but someone with an ordinary knowledge ofthe game. This was done to prevent excessive use of jargon or “in-group" shrn'thand. in an attempt to equate the need for knowledge that would be assumed by the describers across groups. ’Wsepisodicquestionnairewaschangedslightly from Experiment I tn an attenpt to elicit the most detailed episodic descriptions possible from participants Appendix B Steps Involved in a Typical Golf Putt . Judge the line of the ball. . Judge the grain of the turf. Judge the distance and angle to the hole. . Image the ball going into the hole. . Position the ball somewhere between the center of your feet. You should be able to look straight down on top of the ball. . Align shoulders. hips. knees. and feet parallel and to the left of the target (e.g.. image railroad tracks from the ball to the cup—feet outside the tracks. the ball in the huddle). 7. Grip—thumbs should be pointed straight down. palms facing each Other. a light gn'p. 8. Posture—stand tall enough so that if you were to practice putting for 30 minutes you would not experience a still or sore back. 9. Arms—should hang naturally and be relaxed. IO. Hands—should be relative to ball position. Hands should be slightly in front of the ball. I I. Head position—eyes should be positioned directly over the ball. 12 Weight— distribute weight evenly. about 50-30. or with a little more weight on the left foot. masterw— 0 l3. Backswing—swing the club straight back. The distance back that the club goes must equal the through stroke distance. l4. Stroke-the club must accelerate through the ball. finish with the “face" of the club head painting directly at the target. l5. lengtltofdtestroko—itisbenertoetrtoashomrmorecompect stroke rather than a longer stroke. l6. Stroke direction—straight back and straight through. 1?. Stroke rhythm—not too fast and not too slow. 18. Keep head and lower body statiortary throughout stroke and swing with the mm 19. Wrists—should not break dining the stroke. 20. Arms and shoulders—should do most of the work. 21. Had/ounkntipsnegr—should remain still during the stroke. 22. Watch the ball go into the hole. Received February ll. 2000 Revision received June 28. 2000 Accepted January 29. 2001 I 112 APPENDIX B Beilock, S. L., Carr, T. H., MacMahon, C., & Starkes, J. L. (2002). When paying attention becomes counterproductive: Impact of divided versus skill-focused attention on novice and experienced performance of sensorimotor skills. Journal of Experimental Psychology: Applied, 8, 6-16. Copyright © 2002 by the American Psychological Association. Reprinted with permission. 113 Journal of Experimental Psychology Applicd Illll. Vol K. NU l. b~|6 Copyright 2002 by the American Psychological Association. Inc line-riva‘Xflil/SS (it) Dot it) ltll7/lltl7h-lWliX it I 0 When Paying Attention Becomes Counterproductive: Impact of Divided Versus Skill-Focused Attention on Novice and Experienced Performance of Sensorimotor Skills Sian L. Beilock and Thomas H. Carr Michigan State University Clare MacMahon and Janet L. Starkes McMaster University Two experiments examined the impact of attention on sensorimotor skills. In Expcnmcnt I. experienced golfers puttcd under dual-task conditions designed to distract attention from putting and under skill- focused conditions that prompted attention to step-by‘stcp putting performance. Dual-task condition putting was more accurate. In Experiment 2. nght-footcd noVicc and experienced soccer players dribblcd through a slalom course under dual-task or skill-fixuscd conditions. When usmg their dominant right foot. experts again performed better in the dual-task condition. However. when usrng their less proficrcnt left foot. experts performed better in the skill-focused condition. Novices performed better under skill-focus regardless of foot. Whereas novrccs and the less-proficient performances of experts benefit from onlinc attentional monitoring of step-by-stcp performance. high-level skill execution is harmed. What drives the performance of a well-learned skill? Knowl- edge structures (Chi. Feltovich. & Glaser. 1981). memory capac~ itics (Chase & Simon. 1973; dc Groot. 1978; Starkes & Dcakin. I984). problem-solvmg abilities (Priest & Lindsay. 1992; Tcncn- baum & Bar—Eli. 1993). and individual differences (Ackerman. 1987; Ackcnnan & Ciancrolo. 2000; R. Kanfer & Ackerman. I989) involved in high-level performance have been extensively examined. as well as compared across skill levels. in an attempt to shed light on the variables mediating exceptional task execution. Although work in this area has produced a number of important findings in both cognitive and sensorimotor skill domains (sec Ericsson & Lehmann. l996). there remain aspects of high-level perfonnancc that have not yet received adequate analysis. One such area centers around the attentional mechanisms sup- porting skill execution in real time. That is. the manner in which experienced performers allocate attention to skill processes and procedures as actual skill execution unfolds. as well as differences in the attentional requirements of low- and high-level perfor- mances. are not yet fully understood. The purpose of the present study was twofold: (a) to assess the attentional mechanisms sup porting performance of two sensorimotor skills in real time and (b) to explore the relationship between the attentional demands of online skill execution and degree of task proficiency. This knowl- edge will not only aid in developing the most appropriate tech- niques for optimal skill acquisition (Singer. Lidor. & Cauraugh, Sian L. Beilock. Departments of Psychology and Kinesiology. Michigan State University; Thomas H. Carr. Department of Psychology. Michigan SlillC Universrty; Clare MacMahon and Janet L. Starkes. Department of Kinesiology. McMastcr University. Correspondence concerning this article should be addressed to Stan L. Beilock. Department of Psychology. Michigan State University. Room 236. Psychology Research Budding. East Lansing. Michigan 48824. E- mail: bci'lockstémsuedu ' 1993; Wolf. Hob, & Prinz. I998; Wulf. McNevin. Fuchs. Rittcr. & Toolc. 2000) but may also help to explain the suboptimal perfor- mance of well-learned skills in situations such as high-pressure environments thought to stress attentional capacity or interfere with its effective deployment (Baumeister, 1984; Beilock & Carr. 2001; Lewis & Linder. 1997). Attention and the Proceduralization of Skill Skill acquisition is believed to progress through distinct phases characterized by both qualitative differences in the cognitive struc- tures supporting performance and differences in performance it- self. Researchers have proposed that early in learning. skill exe- cution is supported by a set of unintegrated control structures that are held in working memory and attended to one-by-one in a stepby—stcp fashion (Anderson. 1983. l993: Fitts 8: Posncr. 1967; Proctor & Dutta. 1995). As a result. attention is committed to controlling task performance and hence is largely unavailable for the interpretation or processing of nontask-relatcd stimuli. Willi practice. however. procedural knowledge specific to the task at hand develops. Procedural knowledge does not require constant control and operates largely outside of working memory (Ander- son. 1993; Fitts & Posner. 1967; Keele & Summers. 1976; Kimble & Perlmuter. 1970; Langcr & Imber, 1979). Thus. in contrast to earlier stages of performance. once a skill becomes relatively well learned. attention may not be needed for the step-by-step control of execution and may be available for the processing of extraneous stimuli. The impact of secondary task demands on novel skill acquisition and performance. as well as the ability of high-level performers to successfully operate in environments with substantial attentional load. have been widely demonstrated. Nisscn and Bullemer (1987) examined individuals' performance on a novel sequence-learning task under both single-task and dual-task conditions. The sequence task was a four-choice reaction time task in which stimuli were 114 ATTENTION AND PERFORMANCE 7 presented in a consistent pattern in one of four locations on a computer screen. Participants were instructed to respond to the presentation of each stimulus by pressing a key at a corresponding spatial location. In the single—task condition. individuals practiced the sequence in isolation. In the dual-task condition. participants practiced the sequence while performing an attention-demanding. secondary auditory—monitoring task. In contrast to the single-task sequence condition. individuals performing under added secondary task demands showed no eVidence of sequence learning during perfomiance in the dual-task situation or during a transfer test in which individuals received the same sequence in isolation. Subse- quent work by A. Cohen. lvry. and Keele (1990) showed that the severity of the dual-task decrement in sequence learning varied with the complexity of the sequential pattern. These findings suggest that at the initial stages of performance. attention may be a necessary ingredient of skill learning. and the more so the more complicated the skill. Although attention to task components may be important for novel skill execution and learning. this may not be the case at higher levels of practice. Allport, Antonis, and Reynolds (1972) found support for this notion through the examination of skilled pianists' ability to sight-read music while performing a secondary auditory-monitoring task. When skilled pianists were asked to Sight-read in addition to shadowing a series of auditorially pre- sented words. their sight-reading performance was not signifi- cantly altered in comparison with sight-reading in isolation. A11- port et a]. suggested that well-leamed sight-reading does not demand constant attentional control. As a result. attention is avail- able to devote to secondary task demands without significantly disrupting primary task performance. Fisk and Schneider (1984) have also argued that well-leamed performances are not based on explicit attentional control mech- anisms. Individuals were trained on a visual search task in which they learned to search through arrays of visually presented words for members of a target category. With practice. both reaction time and accuracy in finding targets increased. whereas recognition memory for the words that had been searched through declined in comparison with memory at lower levels of practice. Fisk and Schneider suggested that practice automated performance. and automating performance improved real-time skill execution while decreasing attention to specific task components. Thus. novel and well-leamed performances appear to require different levels of attentional resources for successful execution. This experience by attentional demand interaction has also been demonstrated in complex sensorimotor skills drawn from the real world. For example. Leavitt (1979) examined novice and experi- enced ice hockey players' ability to complete a hockey task while performing a secondary visual shape-identification task. Individu- als were required to skate and stick-handle a puck through a slalom course of pylons in isolation and while performing a monitoring task in which they identified geometric shapes projected onto a screen they could see from the ice. Leavitt found that the addition of the secondary visual shape-identification task to the primary skating-and-stick-handling task did not affect experienced hockey players' skating and stick-handling performance. However. when nowces were required to perform the primary skating-and-stick- handling task in addition to the secondary monitoring task. their performance on the primary task declined markedly in comparison with skating and stick handling in isolation. 1n a similar study. Smith and Chamberlin (1992) also found differences across skill levels in performers' abilities to attend to multiple tasks simulta- neously. Experienced and less skilled soccer players dribbled a soccer ball through a series of cones set up on a gymnasium floor. Individuals dribbled in isolation or while performing a secondary visual-monitoring task similar to that used in Leavitt's study mentioned above. Adding the secondary task harmed the dribbling of less skilled players in comparison with dribbling in isolation but did not significantly affect experienced soccer players’ dribbling performance. The results of Leavitt (1979) and Smith and Chamberlin (1992) support the notion that well-leamed skill performance does not require constant online attentional control. However. it should be noted that these findings are potentially confounded. The second- ary tasks used in the hockey and soccer tasks were both in the visual modality. Low-skill players often spend a considerable amount of time looking down at the puck or the ball while performing these skills. Thus. skill level differences in these stud- ies may not have been due to experts‘ more automated control processes per se but instead may have been the result of less skilled individuals‘ higher need for visual information and feedback from the objects they were attempting to manipulate (i.e.. a hockey puck or soccer ball). In essence. differences in “structural interference" (Kahneman. 1973; Wickens. 1980. l984). rather than differential demands of novice and experienced skill performance on attention. may have led to the above mentioned findings. For novices. both the primary task of stick handling or dribbling and the secondary visual-monitoring task require visual input. and the information- gathering stnictures of the visual system cannot be directed toward both the primary and secondary tasks' stimuli simultaneously. Hence. the two tasks compete for visual information. This means that novice task performance under dual-task conditions may have suffered as a result of structural interference. whereas experienced individuals' performance. which presumably does not demand constant visual contact with the objects under control. did not. Recently. Beilock. Wierenga. and Carr (in press a. b) addressed this confound in the exploration of the attentional mechanisms governing novice and well-leamed golf putting. Novice and expe- rienced golfers took a series of golf putts in a single-task putting condition (involving putting in a quiet. isolated environment) and in a dual-task condition. In the dual-task condition. individuals putted while monitoring verbally presented words for a specified target word. Because auditory capacity should not differ as a function of golf-putting experience. and auditory input should not create structural interference with the visual input of putting. this secondary task was designed to be free of the confound present in Leavitt ([979) and Smith and Chamberlin’s (1992) work. Results demonstrated that experienced golfers' putting accuracy was not affected by the addition of the secondary monitoring task. in comparison with single-task putting. Furthermore. when experi- enced golfers were given an unexpected recognition memory test for a subset of the words contained in the monitoring task. their performance did not differ from a single-task word recognition test given as a baseline measure. Because attention is known to influ- ence recognition memory (Craik. Govini. Naveh-Benjamin, & Anderson. 1996), this result indicates that experienced golfers were able to pay just as much attention to the words while putting as when attending to the words was their only task. and that domg so did not harm their putting performance. In contrast. novice 115 8 BEILOCK. CARR. MACMAHON. AND STARKES golfers showed both putting and word recognition decrements from single— to dual-task conditions. Consistent WIlh Leavitt and Smith and Chamberlin. Beilock et al. (in press a. b) concluded that expertise leads to the encoding of task components in a procedur- alized form that supports effective real-time performance. without the need for constant online attentional control. When Attention to Performance May Be Counterproductive Well-leamed skills do not appear to require constant attentional control during execution. However. the notion that these skills are based on a proceduralized or “automated" representation carries even stronger implications for attending to practiced performances. Researchers have proposed that attending to the stepoby-step com- ponent processes of a proceduralized skill may actually disrupt execution (Baumeister, I984; Beilock & Carr. 2001; Kimble & Perlmuter. I970; Langer & Imber. 1979; Lewis & Linder. 1997). Therefore. attention to performance may become counterproduc- tive as practice builds an increasingly automated performance repertoire. Masters and colleagues (Masters. 1992'. Masters. Pol- man. & Hammond. I993) proposed that attention to high-level skills results in their ”breakdown." in which the compiled real-time control structure of a skill is broken down into a sequence of smaller. separate. independent units—similar to how performance may have been organized early in learning. Once broken down. each unit must be activated and run separately. which slows performance and. at each transition between units. creates an opportunity for error that was not present in the “chunked” control structure. Researchers have proposed that this process of break- down contributes to the suboptimal performance of well-leamed skills in high-pressure situations (Baumeister, I984; Beilock & Carr. 2001; Lewis & Linder. 1997). This brings us to Experi- ment 1. in which we attempted to assess the attentional mecha- nisms governing the real-time execution of a well-learned golf- putting skill. Experiment I If high-level skill execution is supported by procedural knowl- edge that does not mandate and. furthermore. may be harmed by continuous online control. then. as demonstrated by Leavitt (1979). Smith and Chamberlin (1992). and Beilock et al. (in press a; b). experienced performers should not be negatively affected by a dual-task environment that draws attention away from the task at hand. In contrast. attending to an explicit component of a well- learned skill may actually serve to disnipt or degrade automated performance procedures. In Experiment I. experienced golfers performed a golf-putting task in a skill-focused attention condition in which individuals were prompted to attend to a specific com- ponent of their performance (i.e.. the exact moment that their club head stopped its follow-through) and a dual-task attention condi- tion in which experienced golfers executed the putting task while perfonning a secondary auditory-tone-monitoring task. Method Participants PartiCipants (N = 2|) were undergraduate students (7 women. 14 men). ages 18—22 (M = 19.86 years. SD = 0.96 years). who were enrolled at Michigan State University With 2 or more years of high school varsity golf experience or a Professional Golfers‘ Association (PGA) handicap less than 8. Task Individuals performed the golf-putting task on a carpeted indoor putting green (3 m X 3.7 m). The task required participants to putt a golf ball as accurately as possible from nine different locations marked by squares of red tape. Three of the locations were 1.2 in. three locations were 1.4 in. and three locations were 1.5 m away from a target. also marked by a square of red tape. on which the ball was supposed to stop. All paniCipants followed the same random alternation of putting from the nine different Iocanns. A standard golf putter and golf ball were supplied. Participants took part in both the skill-focused and the dual-task attention condition. Skill-focused condition. In the skill-focused condition. participants were instructed to attend to a particular component of their golf-putting swing. Specifically. indiwduals were instructed to monitor the swing of their club and at the exact moment they finished the follow-through of their swing. bringing the club head to a stop. to say the word “stop" out loud. Dual-task (uniform. The dual-task attention condition involved putt- ing while listening to a series of recorded tones being played from a tape recorder. Panicipants were instructed to monitor the tones carefully and each time they heard a specified target tone to say the word “tone“ out loud. The target tone was played three times prior to the start of the dual-task condition to ensure that participants were familiar with this tone. Tones (500 ms each) occurred at a random time period once within every 2-s time interval. The target tone occurred randomly once every four tones. The random placement of the tones Wllhln the Z-s time intervals. as well as the random embedding of the target tone Within the filler tones. was designed to prevent the golfers from anticipating secondary task tone presentation. Studies in our laboratory revealed that experienced golfers perform a golf putt (from beginning the initial putt assessment to completing the actual mechanical act of implementing the putt) in roughly 10 s (M = 10.40 3. SD = 1.69 s. for 420 putts taken by 21 experienced panicrpants). Tones occurred on average once every 2 s. and target tones occurred once every four tone presentations. Thus. indiVIduals received about five tone presen- tations per putt. including a minimum of one target tone presentation. Procedure After giving consent and filling out a demographic sheet concerning previous golf experiences. participants were instructed that the purpose of the study was to examine the accuracy of golf putting over several tnals of practice. Participants were set up at the first putting spot and asked whether they preferred to putt nght-handed or left-handed. The experimenter in- fomed partiCipants that they would be putting from nine locations on the green. The experimenter then directed participants‘ attention to a any light that had been set up next to each putting spot. Participants were informed that the lights were booked up to a switchboard controlled by the experi- menter. Participants were told that before every putt. a light would illumi- nate beside the location from which they were to take their next putt. Individuals then performed one set of 20 putts. These putts constituted the practice trials. The order of the attention conditions was counterbalanced between participants. Individuals performed one set of 20 putts in the dual-task condition and one set of 20 putts in the skill~focused attention condition. PaniCipants were given ti short break in between the two attention condi- tions. Accuracy of putting was measured by the distance in centimeters away from the center of the target that the ball stopped after each putt. The measurement was made by the experimenter while the participant was setting up for the next putt. 116 ATTENTION AND PERFORMANCE 9 Results Putting Pelformance The mean distance from the target of the 20 putts in each condition was used as the measure of performance. In the practice condition. M = 15.09 cm (SD = 3.27); in the dual-task condition. M = 13.74 cm (SD = 2.65). and in the skill-focused condition. M = 19.44 cm (SD = 5.42). A Bonferroni adjustment was performed on the critical p value of the following comparisons of experienced performers’ putting accuracy to guard against inflation of Type I error rates as a result of multiple comparisons. The resulting critical p value was .017. Cohen’s d was used as the measure of effect size (for equation. see. Dunlap. Cortina. Vaslow. & Burke. 1996). J. Cohen (1992) sug- gested that 0.20 is a small effect size. 0.50 is a medium effect size. and 0.80 is a large effect size. Experienced golfers performed significantly better during the dual-task condition in comparison with the skill-focused condition. t(20) = 5.22. p < .01. d = 1.26. Additionally. putting in the skill-focused condition was significantly less accurate than in the singlestask practice condition. ((20) = 3.94. p < .01. d = 0.94. whereas performance in the dual-task condition did not signifi~ cantly differ from the practice condition. t(20) = 1.71. d = 0.45. This pattern of results coincides with predictions derived from the skill acquisition and automaticity literature. Specifically. high- level skill execution is thought to be governed by proceduralized knowledge that does not require explicit monitoring and control. Thus. a dual-task environment should not degrade performance in comparison with skill execution under single-task conditions. as attention should be available to allocate to secondary task demands if necessary without detracting from control of the primary skill. The above results demonstrate precisely this notion. It should be noted. however. that these findings may not hold true for dual-task environments in which the tasks draw on similar processes and hence create structural interference (e.g.. looking at a golf ball while lining up a putt and visually monitoring a screen for a target object). Attention Condition Secondary Task Performance Skill-focused condition. Each trial in which individuals failed to say “stop" at the cessation of their golf swing was recorded. On average. failure to follow instructions occurred in 2.9% of the 20 putts taken in the skill-focused condition by each individual (M = 0.57 "stops." SD = 0.87 “stops"). or 0.14% of the 420 skill-focused putting trials across all participants. Dual-task condition. Each instance in which individuals failed to identify a target tone was recorded. Failure to identify target tones occurred infrequently (M = 0.62 target tones. SD = 0.86 target tones). On average. individuals received 1.25 target tone presentations per putt for a total of 25 target tones per 20 putts taken in the dual-task condition. This led to an error rate of .025 per tone. Given that each participant's errors were distributed over 20 putts. failure to identify a target tone occurred in 3.1% of the 20 putts taken in the dual-task condition by each individual. Additionally. a significant positive correlation was found be— tween the number of target tone identification errors and our measure of putting accuracy (i.e.. mean distance from the target that the ball landed after each putt) in the dual-task condition (r = .47. p < .05). That is. individuals who performed at a lower accuracy level in the putting task were also more likely to miss identifying a target tone. The fact that individuals with poorer putting accuracy were also less able to identify target tones sug- gests that putting performance under dual-task conditions was not the result of a simple trade-off between primary putting and secondary tone-monitoring performance. It should be noted that there may be individual differences in performance ability such that certain individuals are less accurate at both putting and sec- ondary task target tone detection. Therefore. the possibility of a more complex trade-off cannot be completely ruled out. However, the positive correlation between putting accuracy error scores and tone detection error rates does suggest that even though average performance was at least as good in the dual—task condition as in the single-task practice condition. there remained a degree of variation in how much attention experienced golfers paid to their putting. which could be detected in accuracy of tone detection. The less accurate the putting was. the greater the amount of attention was paid to it as indexed by decreased accuracy of tone detection. Finally. a comparison of target tone detection errors between the skill-focused and dual-task condition illustrates that secondary task performance was not significantly affected by our manipulations of attention. t(20) = 0.21. d = 0.05. ns. This observed effect size is substantially smaller than the standard for a small effect size (d = 0.20; J. Cohen. 1992). suggesting that if there is a difference in secondary task performance across conditions. it is trivial. Discussion The results of Experiment 1 demonstrate that well-leamed golf putting does not require constant online control. As a result, attention is available for the processing of secondary task infor- mation if necessary (such as monitoring a series of auditory tones). However. when prompted to attend to a specific component of the golf swing, experienced performance degrades in comparison with both single-task practice performance and dual-task conditions. Although the negligible effects of divided attention on well- learned performance have been previously demonstrated (Beilock et al.. in press a; b: Leavitt. 1979; Smith & Chamberlin. 1992). the consequences of explicitly attending to automated or procedural- ized performance processes have not received so much investiga- tion. The present findings suggest that well-learned performance may actually be compromised by attending to skill execution. This result complements recent evidence on “choking under pressure." Researchers have proposed that pressure to perform at a high level prompts attention to the step-by-step components of a well-learned skill. This attention is thought to disrupt or slow down skill execution. resulting in a less than optimal performance outcome (Baumeister, 1984; Beilock & Carr. 2001; Lewis & Linder. 1997). In the present study. directly instructing experienced golfers to attend to a specific component of their swing produced just this result—a less than optimal performance. Experiment 2 Experiment 2 was designed to replicate the results of Experi- ment 1 in a movement skill that uses different effectors and imposes different temporal demands. as well as to examine the effects of dual-task and skill-focused attention on performance at 117 10 BEILOCK. CARR. MACMAHON. AND STARKES differing levels of skill proficiency. Experiment 1 demonstrated that it can be disadvantageous to explicitly attend to a specific component of an automated or proceduralized well-learned skill pert‘onnance. However. by analogy to ballistic versus nonballistic movements (Banich. 1997). it might be that explicit attention plays a different role in continuous tasks extended in time. such as soccer dribbling. than in the discrete golf-putting task with a defined beginning and ending point just reported. Furthermore. in addition to task differences. there also may be expertise differences in the impact of attention. Researchers have suggested that close atten- tional monitoring and attentional control benefits novice perfonn- ers in the initial stages of task learning (Anderson. 1983; Fitts & Posner, I967). This notion has received both empirical and anec- dotal support (see Curran & Keele. 1993: Proctor & Dutta, 1995). Recently. however. the benefits of attention to specific task com- ponents in novel sensorimotor skill performance have been chal- lenged (Singer et al.. I993; Wulf et al.. 1998. 2000). Instead. researchers have suggested that instructing novices to attend to task properties during online motor skill performance may actually hinder skill acquisition. Wulf et al. (I998) examined the effects of both an internal focus of attention (defined as attention to specific body movements. much like our skill-focused condition) and an external focus of attention (defined as attention to the effects or outcomes of body movements) on the learning and retention of a ski simulator and stabilometer task. Results demonstrated that an external focus of attention led to more effective learning than an internal focus. Wulf and colleagues proposed that explicitly at- tending to skill execution at the initial stages of skill learning may actually hinder performance. Thus. a controversy remains over the types of attentional mechanisms thought to support less experi- enced or less practiced performance processes. Experiment 2 was designed to assess the attentional mechanisms supporting soccer dribbling performance at different levels of skill proficiency. This was accomplished in two ways: First. the influ- ence of dual-task and skill-focused attention on both novice and experienced soccer dribbling performance was examined. Second. the effects of these attentional manipulations on dominant and nondominant foot performance within soccer skill level were assessed. lf attention to well-learned skill execution disrupts performance. then one might expect explicit attention to experienced perform- ers' dominant right-foot dribbling skill to compromise perfor- mance in comparison with dual-task conditions. However. this may not be the case for experienced players' nondominant left foot. That is. although soccer players must be skilled with both feet to compete at a high level. these athletes admit foot preferences and are often more skilled with one foot than with the other (Helsen & Starkes. 1999; Peters. 1981. 1988). Comments from the experienced soccer players in the present study concerning foot preference are consistent with this evidence. For example. one experienced participant stated that in comparison with right-foot dribbling. “dribbling with my left foot is the worst." and another stated that "when I use my left foot performance suffers." As with all introspective reports about task performance. these comments must be viewed With caution. Nevertheless. these comments do indicate that experienced players may not perceive their dominant right-foot and nondominant left-foot dribbling skills as equivalent. lf II is true that experienced performers' right- and left-foot drib- bling skills do not support the same level of task proficiency. then current theories of automaticity in skilled performance predict that these skills are likely not to be supported by the same attentional mechanisms (Anderson. 1983; Fitts & Posner. 1967; Logan. 1990). Therefore. nght- and left-foot performance may be differentially affected by the skill-focused and dual-task attention manipulations in the present study. Put another way. if experienced performers' nondominant foot is not supported by a proceduralized knowledge structure. and explicit attention to less practiced performances serves to enhance skill execution. then in contrast to dominant right—foot dribbling. left-foot performance may actually benefit from explicit attention to skill execution. Similarly. skill-focused attention may lead to a higher level of performance than the dual-task condition for novices. regardless of foot—as novices should not be skilled dribblers with either foot. In Experiment 2. right-foot dominant novice and experienced soccer players performed a dribbling task in which they dribbled a soccer ball through a slalom course made up of a series of pylons. Individuals performed the task under a dual-task condition involv. ing an auditory word-monitoring task (dual-task condition) and a condition in which individuals were prompted to focus on a specific component of the dribbling task—the side of the foot that last made contact with the ball (skill-focused condition). As with golf putting in Experiment I. the combination of auditory word monitoring and soccer dribbling should not create structural interference. Participants took part in both attention conditions while drib- bling with their dominant right foot and again while dribbling with their nondominant left foot. The attention and foot manipulations afforded the comparison of dribbling performance between novice and experienced soccer players under the different attentional manipulations in the soccer-dribbling task. as well as within- individual comparisons of dominant and nondominant foot performance. Method Participants Participants (N = 20) were self-proclaimed right-handed and right- footed undergraduate students at McMaster University. ages III—26 (M = 20.20 years. SD = 1.85 years). The novtce partiCipants (8 women. 2 men) had less than 2 years of organized soccer experience (M = llO years. SD = 0.74 years). The experienced participants (8 women. 2 men) had 8 or more years of competitive soccer experience (M = I330 years. SD = 2.75 years). Task Individuals performed the soccer-dnbbling task on an indoor gymnasium-type surface. The task required participants to dribble a soccer ball as rapidly as possible through a slalom course that consisted of six cones set 1.5 m apart for a total of 10.5 m from start to finish. Pnor to each dribbling trial. participants were instructed to dribble the ball through the Cones with either their right foot or their left foot. Individuals were also given instructions concerning the skill-focused and dual-task attention manipulations. Skill—focused condition. In the skill-focused attention condition. indi- viduals dribbled through the slalom course while a single tone occurred at a random time period on a blank tape once during every 6-s interval. The tone was temporally aligned with the occurrence of the target word in the dual~task condition so that the tone in the skill-focused condition appeared 118 ATTENTION AND PERFORMANCE ] l at the same rate as the target word in the dual-task condition. Individuals were instructed to attend to the side of their foot that was in contact With the ball throughout the dribbling trial. so that upon hearing the tone. indivrduals could verbally indicate whether they had JUSI touched the ball with the outside or inside of their foot. The random placement of the tone was designed to prevent participants from anticipating its occurrence. Dual-task condition. We dual-task condition involved dribbling through the slalom course while performing a secondary auditory-word- tnonitoring task. Individuals heard a series of single-syllable concrete nouns spoken from a tape recorder. Words were presented at a random time period once Wllhln every 2-s time interval. The target word. thorn. occurred randomly. averaging once every three words (6 s). Participants were instructed to monitor the list of words and to repeat the target word out loud every time it was played. The random placement of the words within the 2—s time intervals. as well as the random embedding of the target word within the filler words. was designed to prevent participants from antici- paling secondary task word presentation. Procedure Participants completed a consent form and demographic sheet detailing previous soccer experience. Individuals were also asked to report their dominant hand and foot. The experimenter further explored individuals' toot pretcrcnce by asking participants "Which foot would you normally kick a ball with?" This specific question was asked because it is relevant to the predictions made in the present study and is included on several measures of footedncss (Day & MacNeilage. 1996; Searleman. 1980). Only those individuals who were self-proclaimed righbhanded and right- footed were used. Participants were instructed that the purpose of the task was to dribble a soccer ball as quickly and accurately as possible through the series of cones set up in front of them. IndiViduals were also informed that prior to each dribbling attempt. the experimenter would instruct them as to which foot to use. Finally. participants were told that each dribbling trial would be timed by the experimenter. If an error in dribbling performance occurred or the proper foot was not used. the dribbling trial was repeated. This was done to ensure that participants completed the entire slalom course with the specified foot. Because we were interested in making specific predictions concerning the dnbhling performance of each foot under the various attentional manipulations. it was extremely important to ensure that par. ticipants were solely using the correct foot for each attention condition. Thus. trials containing dribbling errors were repeated. However. errors as a result of failure to use the specified foot were quite infrequent and did not significantly differ across the attention or foot conditions (novice right-foot practice: M = 0.05. SD = 0.22 errors; both skill-focus and dual-task: M : 0 errors; novice left-foot practice. skill-focus. and dual-task: M = 0 errors; experienced right-foot practice. skill-focus. and dual«task: M = 0 errors; experienced left-foot practice: M = 0.05. SD = 0.22 errors; skill- focus and dual-task: M = 0 errors). The dependent measure was the time taken to complete each error-free trial. measured wrth a stopwatch to the nearest tenth of a second. Parttcr- pants performed two dribbling trials With their right foot only and two dribbling trials with their left foot only. These four dribbling trials consti- tuted the practice trials. The order of the remaining dribbling trials was counterbalanced between participants. Individuals perfonned four sets of two dribbling trials (8 total dribbling trials). alternating feet (i.e.. right foot only. left foot only) and attentional focus manipulations (i.e.. dual-task or skill-focused attention) every two trials. All participants performed the dribbling task with all possible foot and attentional focus combinations. After every two trials in a specific attention condition had been completed. individuals were given a short break during which time they were asked to verbally count back- ward from 100 by 7s. This manipulation was designed to limit the influence of persisting thoughts about the preVious attention condition on subsequent skill performance. Results Dribbling Performance We used the mean of the two error-free dribbling trials per- formed with each foot under each condition as a measure of dribbling performance for that specific foot and condition. Table 1 presents means and standard deviations for left- and right-foot dribbling performance in the practice. skill-focused. and dual-task attention conditions for both novice and experienced participants. Bonferroni adjustments on the critical p value of dribbling time comparisons in the practice condition were performed to control for the inflation of Type I error rate as a result of multiple betweenoskill-level and within-skill-level comparisons. The result- ing critical p value was .025. The experienced soccer players were significantly faster than novices during practice when instructed to dribble with either their right foot. F(1. 18) = 52.54. p < .01. MSE = 1.11. d = 3.24. or their left foot. F(1. 18) = 13.47. p < .01. MSE = 3.55. d = 1.64. Direct comparisons within skill level demonstrated that the novices did not significantly differ in dribbling time between their right and left feet during the practice condition. r(9) = 1.16. d = 0.35. us. However. this null effect exceeds .1. Cohen's (I992) criterion for a small effect size. and thus this nonsignificant result most likely reflects the fact that we do not have adequate power to detect an effect this small. With a medium effect size of 0.50. power is equal to .18 (J. Cohen. 1988). Thus. given the low power of this comparison. it may be unwise to conclude from the lack of a significant difference that the null hypothesis is true. However. it should be noted that the similarity in dribbling times between novices' right and left feet in the practice condition parallels other findings in our laboratory concerning novel skill performance. In golf putting. for example. novices have been found to putt at a similar accuracy level while using a standard golf putter or an S-shaped and arbitrarily weighted “funny putter" (Beilock & Carr. 2001; Beilock et al.. in press a; b). Because novices are not accustomed to performing with either type of putter. the distorted funny putter does not significantly alter their putting accuracy. This is in contrast to experienced golfers. whose performance is degraded by the altered golf putter. In the present dribbling task. novices should not have been accustomed to drib- bling with either foot. Thus. despite expressed foot preferences. it may not be too surprising that novices were not Significantly more Table 1 Novice and Experienced Panicipants' Mean Dribblmg Times and Standard Deviations Across Conditions for Both Right and Left Feet Condition Practice Skill-focused Dual-task Group M SD M SD M SD Novice Right foot 10.26 1.29 8.81 1.23 9 70 191 Left foot 11.02 2.48 9.30 1.90 10.47 2.04 Experienced Right foot 6.85 0.74 8.38 1.23 6.55 0 88 Left foot 7.93 0.97 7.01 0.85 8.21 1.01 119 12 BEILOCK. CARR. MACMAHON. AND STARKFS skilled with one foot in comparison with the other. However. this was not the ease for the experienced soccer players in the present study. Experienced soecer players were significantly faster With their right foot in comparison to their eft foot during practice. r(9) = 2.90. p < .03. d = 1.25. This result is conststent with the earlier documented notion that high-level soccer players are often more skilled with one foot than With the other. Turning to the attention conditions. we performed a 2 (novice. experienced) X 2 (right foot. left foot) X 2 (skill-focused attention. dual-task attention) repeated measures analysis of variance (Table 2) This analysis revealed a main effect of experience. in which the experienced participants dribbled faster than the novrce partici- pants across all foot and attention conditions. Furthermore. there was a significant Attention X Expertise interaction and a signifi~ cant Attention X Foot interaction. However. these two»way inter actions are qualified by a significant Experience X Foot X Atten- tion Condition interaction. In terms of fight—foot dribbling. shown in the upper panel of Figure I. experienced performers were faster than the novices during the dual-task condition (d = 2.12). In contrast. experienced and flower: par‘ttcrpants dribbled at a more similar speed in the skill-focused attention condition (d = 0.35). It should be noted. however. that this effect does exceed .1. Cohen's (1992) criterion for a small effect size. and thus the similarity in novice and experienced players' right-foot dribbling speed in the skill-focused attention condition should be interpreted with caution. Thus. it is clear that experienced performers were markedly faster than nov- ices in the dual-task condition. whereas in the skill~focused con- dition their advantage was substantially reduced. Funhennore. experienced soccer players dribbled faster in the dual-task condi< tion in comparison with the skill-focused condition (d = 1.62). whereas a tendency toward the opposite pattern occurred in nov- ices. who dribbled faster in the skill focused condition than in the dual- task condition td= 0.56) In terms of left-foot dribbling performance. the lower panel of Table 2 Anulisis of Variance for Mean Dribb/ing Times in the Attention Cuirdm irinJ Source J] MS F f Between subiect Experience (E) | 1‘12 51 17.09” .558 Enl‘f 18 4 K3 Within SUMCCI Attention condition (A) 1 2.61 1.94 .098 A s E I 9.02 6.71‘ .282 Error (A) 18 1.34 Fool (F) I 2.99 2.80 .135 F s E I 1.92 1.12 .058 Error (Fl 18 1.07 A r F l 13.67 13.82“ .482 I 9 46 9.56" .370 Error (A 7- F) 18 i) 99 Note (‘ohen'sfwas used as the measure of effect sue (for equation. see J Cohen. 1988) Cohen suggested that 0 It) I\ a small effect sire. 0.25 is a medium effect sire. and 040 is a large effect size. Ii .115 . 1. 3:? E 10 Du ask ‘5. 9 E 8 g 7 U 6 § 5 2'5 4 g 3 2 § 1 E o Novice Experienced Expertise A 12 f 11 lSkil Focused g 10 DDual—Task a 9 .E 5 g 7 u 6 °‘ 5 g 4 g 3 C 2 3 1 2 0 . Experienced Expertise Figure I. Mean right and left foot dribbling times in the skill-focused and d ' " ‘ "" fr"nrwice ‘ , ' ‘_ Error bars represent standard errors. Figure 1 illustrates that experienced performers were faster than novices during both the dual-task condition ((1 = 1.40) and skill- focused condition (d = 1.56). Additionally. novice and experi- enced soccer players performed better in the skill-focused condi- tion than in the dual-task condition (d = 0.59 and d = 1.25. respectively). Thus. regardless of skill level. in left-foot dribbling. a higher level of performance occurred in the skill-focused condi» tion. designed to draw attention to skill execution. than in the dual-task condition. designed to distract attention away from skill execution. This is in contrast to dominant right-foot dribbling. in which experienced and novice soccer players were differentially affected by the skill-focused and dual-task attention manipulations. Finally. separate post hoc comparisons of novice and experi- enced dribbling performance in the attention conditions in contrast to dribbling in the practice condition were ormed. Novrces‘ right-foot dribbling during the skill-focused condition was faster 120 ATTENTION AND PERFORMANCE 13 than their right-foot practice condition performance (d = 1.19). In contrast. experienced soccer players' right-foot dribbling during the skill-focused condition was slower than their right-foot prac- tice condition performance (d = 1.44). In terms of left-foot per— formance. both novice and experienced participants dribbled faster during the skill-focused condition in comparison with their respec- tive left-foot practice condition performances (d = 0.76 and d = 1.01. respectively). Thus. although attention to dominant right-foot performance in the skill-focused condition led to an improvement in dribbling speed in comparison with the practice condition for novices. this same condition led to a decrement in experienced soccer players’ dribbling skill. With the nondominant left foot. however. both novice and experienced performers im- proved in dribbling speed from the practice to skill-focused condition. Attention Condition Secondary Task Performance Skill-focused condition On average, novice participants drib- bling with their right foot heard 2.80 tones (SD = 0.42). whereas experienced participants heard 2.60 tones (SD = 0.52). During left-foot dribbling. novices heard an average of 2.80 tones (SD = 0.42). whereas experienced players heard 2.10 (SD = 0.32). Each instance in which individuals failed to verbalize the side of the foot that had just touched the ball following tone presentation was recorded. These errors occurred infrequently across both foot and experience level (M = 0.10 foot identifications. SD = 0.31 foot identifications). Overall. there were just two instances of participants failing to verbalize the side of the foot after tone presentation (1 novice and I experienced participant in the right- foot dribbling condition). Therefore. analysis of failures to identify the foot that had just touched the ball following tone presentations across foot and experience level was not interpretable because of the infrequency of these errors. Finally. if a skill-focused dribbling condition was completed in which no tones were heard. this trial was counted as an error and repeated. However. this occurred only on one trial across all participants (a novice right-foot dribbling trial). Dual-task condition. On average. novice participants heard 5.30 words (SD = 0.82; M = 2.90 target words. SD = 0.32) while dribbling with their right foot. whereas experienced partic- ipants heard 3.80 words (SD = 0.63; M = 2.10 target words. SD = 0.32). In terms of left-foot dribbling. novices heard an average of 5.6 words (SD = 1.08; M = 2.90 target words. SD = 0.32). whereas experienced players heard 4.60 words (SD = 0.70; M = 2.50 target words. SD = 0.53). Each instance in which individuals failed to identify a target word was recorded. As in the skill—focused condition. errors were infrequent (M = 0.20 target words, SD = 0.41 target words). There were five instances of failure to identify a target word across both foot and expertise level (three target word identification failures in novice right-foot dribbling. one target identification failure in experienced right-foot dribbling. and one target identification failure in novice left-foot dribbling). Analysis of target identification differences across foot and experience level was not interpretable because of the infre- quency of these errors. Thus. similar to Experiment 1. errors in secondary task performance were infrequent across both attention condition and level of expertise. Discussion The purpose of Experiment 2 was to explore differences in the attentional mechanisms supporting online sensorimotor skill exe- cution in novice and experienced soccer players. as well as to assess differences in the attentional requirements of dominant and nondominant foot performance within level of expertise. Theories of skill acquisition have proposed that distinct cognitive processes are involved at different stages of skill execution. Early in learn- ing. individuals are thought to attend to the step-by-step processes of performance. However. once a high level of performance has been achieved, constant online attentional control may not be necessary (Anderson. 1983. 1993; Fitts & Posner. 1967; Logan. 1988). One could infer from this framework of skill acquiSition that novices might benefit from conditions that prompt attention to task properties yet not profit to the same extent in environments that divert attention away from the primary task at hand. In contrast. experienced performers may be harmed by explicit atten- tion to skill processes that normally run as uninterrupted programs or procedures. yet they may not be adversely affected by condi- tions that draw attention away from performance. However. this may hold true only for those aspects of an experienced performer's skill execution repertoire that are indeed governed by a procedur- alized or automated knowledge representation. If particular aspects of a skill are not as well learned or as highly accomplished. then experienced individuals—like less practiced performers—may benefit more from conditions that prompt attention to the task at hand rather than take it away. The results of the present study conform quite well to these predictions derived from theories of skill acquisition. For right- foot dribbling. novices performed at a lower level in the dual-task condition. designed to distract attention from task performance. in comparison with the skill-focused manipulation. designed to draw attention toward the task at hand. Furthermore. novices substan- tially improved in dribbling speed from the single-task practice condition to the skill-focused condition. Experienced soccer play- ers showed an opposite pattern of results. Experienced individuals performed at a lower level in the skill-focused condition compared with either the dual-task or practice condition. These results coin- cide with those of Experiment 1 and. as mentioned above. are consistent with current theories of choking under pressure (Beilock & Carr. 2001). Performance with the left foot differed. In contrast to right-foot dribbling. novice and experienced soccer players alike performed better in the skill-focused condition than in the dual-task or prac- tice condition. In the present study. there were significant differ- ences in experienced performers‘ right- and left—foot dribbling speed in the practice trials. This pattern of results suggests that experienced players' left-foot dribbling skill was not at the same performance level as their dominant right-foot skill. The fact that the differential impact of the attentional manipulations in the present study was evident not only between skill levels but within experienced perfomiers' dominant and nondominant feet perfor- mance as well speaks to the robust nature of the impact of attention on skill performance. 121 l4 BEILOCK. CARR. MACMAHON. AND STARKES General Discussion When Attention to Performance Becomes Counterproductive The findings of Experiments 1 and 2 demonstrate that skill- focused attention benefits less practiced and less proficient perfor- mances yet hinders performance at higher levels of skill execution. High-level skills are thought to become proceduralized or auto- mated with extended practice. ”Ihe encoding of task components in a proceduralized form supports effective real-time performance. without the need for constant online control. As a result. skill performance decrements occur in conditions that impose step-by- step monitoring and control on complex. procedural knowledge that would have operated more automatically and efficiently had such monitoring not intervened. Therefore. experienced perform- ers suffer more than novices from conditions that call their atten- tion to individual task components or elicit step-by-step monitor- ing and control. However. experienced performers are better able than novrces to spare a portion of their attention for other stimuli and task demands. and hence are better able than novices to deal With conditions that create dual-task environments (e.g.. taking a series of golf putts or dribbling a soccer ball while performing an auditory-monitoring task). As shown by the contrast between right-foot and left-foot dribbling. however. this may hold only for that portion of an experienced performers‘ skill repertoire that is supported heavily by proceduralized knowledge structures. The findings of the present study confirm results generated in LeaVitt's (I979) hockey study. Smith and Chamberlin's (1992) soccer-dribbling task. and Beilock et al.'s (in press a; b) golf- putting study. Furthermore. the present findings expand previous results by examining the consequences of explicitly attending to both novel and well-learned performances. Researchers have re- cently suggested that attention to the step-by-step components of a novel skill may be detrimental to performance (Singer et al.. 1993: Wulf et al.. 1998. 2000). However. the present findings demon- strate that attention benefits both novel skill performance and performance that is not based on a heavily proceduralized knowl- edge representation. even though carried out by an experienced performer (e.g.. experienced soccer players' nondominant foot performance). In contrast. at higher levels of learning and profi- ciency. increased attention to the step-by-step execution of a well-learned skill appears to have the opposite effect—disrupting skill execution processes. It should be noted that the present study examined the perfor- mance of a golf-putting task and a soccer-dribbling skill under different attentional manipulations at approximately constant lev- els of performance. rather than examining the learning or transfer of these skills to novel task situations. For this reason it remains possible that under conditions commonly used to assess skill learning (e.g.. transfer tests). a different pattern of performance may arise. Future research in this area would serve to shed light on this issue. Not All Forms ofAttention Are Counterproductive to Well-Learned Skill Performance The above mentioned results indicate that attention to step—by- step skill execution—what we term skill-focused attention—may benefit novel performances yet hinder well-leamed and highly proficient task execution. However. this relationship may not extend to other forms of attention to task-related information. R. Kanfer and Ackerrnan (1989) demonstrated that “self-regulatory" activities, including the allocation of attention to performance outcomes and goal attainment. self-evaluation. and self-reactions. detract from the lower level performances of novices yet enhance skill execution at later stages of learning and higher levels of proficiency. Self-regulatory activities are thought to require atten- tional capacity for successful initiation and implementation. Thus. self-regulation may disrupt novel skill execution by recruiting attentional resources needed for control of task performance (F. H. Kanfer & Stevenson. 1985). However. this may not be the case for more experienced performance that does not rely on constant attentional control. Instead. selforegulatory functions may be implementable in parallel with proceduralized control processes. serving to store information about the outcomes and evaluations of performance (rather than the unfolding of their step-by-step com- ponents) that is needed for subsequent cognitions about ones‘ abilities. effort. and strategies for task control (R. Kanfer & Ack- ennan. 1989; Kluwe, 1987). We propose. then. that self-regulatory attention and skill-focused attention differ in a crucial way: Self— regulatory attention is metacognitive and aimed at the plans that precede skill execution and the products that follow skill execution (Brown. 1987). whereas skill-focused attention is cognitive and aimed at the component steps that constitute execution itself (Beilock & Carr. 2001). If attention devoted to self-regulation is different from skill focus yet depends on some of the same attentional resources. then self-regulatory activities may provide a secondary. unintended benefit to experienced skill execution: Specifically. self-regulation applied to the plans. the outcomes. and the feelings accompanying performance may prevent individuals from paying too much at- tention to the step-by-step control of that performance as it unfolds in real time. It may be that individuals involved in self-regulatory functions do not have the resources available to explicitly attend to. monitor. or try to control particular steps or components of online performance—a practice that was shown in the present study to disrupt high-level execution. Thus. although explicit at- tention to component steps of proceduralized performances may disrupt or dismantle optimal task execution at high levels of learning and proficiency. attentional processes that serve higher level. more metacognitive roles may instead promote optimal performance. both by focusing attention on plans and outcomes and also by preventing attention from focusing on step-by-step control of execution. Furthermore. skill-focused attention may not always be detri- mental to well-leamed performances. The present study demon- strated that skill-focused attention applied to current real-time performance disrupts execution. However. if applied in other cir- cumstances. such as practice situations. in which performers are consciously attempting to dismantle their skill and modify certain pans in accord with data collected by self-regulatory activities such as those mentioned above. skill-focused attention may actu- ally be helpful. That is. when the goal is not to maximize real-time performance but instead to explicitly alter or change performance processes to achieve a different outcome. skill-focused attention may be beneficial. In this manner. skill-focused attention may become embedded in the metacognitive activities of self- 122 ATTENTION AND PERFORMANCE 15 regulation. Specifically. individuals may attend to specific com- ponents of their skill (i.e.. implement skill-focused attention) to alter control strategies and execution processes that. through self- regulatory actions. have been deemed unproductive or maladaptive to progress toward a desired goal state. Although this monitoring of perfomiance may be temporarily detrimental to skill execution. as performers will most likely have to slow down and break down previous execution procedures to attend to and alter these pro- cesses. and then readapt to and proceduralize the new execution parameters. ultimately these changes should produce performance benefits as skill execution becomes more closely aligned with desrred outcomes. Implications for Skill Training and Performance Coaches and teachers have long believed that different teaching styles are required at various stages of learning to address the changing attentional mechanisms of the performer. The findings of the present study begin to lend empirical support to this notion. For example. the results of the present study suggest that it may be benefiCial to direct performers' attention to step-by-step compo- nents of a skill in the early stages of acquisition. This might be achieved through instructions that draw learners‘ attention to task. relevant kinesthetic or perceptual cues. However. at later stages of performance. this type of attentional control may be detrimental. at least in situations where maximum performance is the desired real-time outcome. In the present study. experienced golfers and soccer players showed decrements in performance under condi- tions designed to prompt attention to stepcby-step execution. Thus. it may be beneficial for experienced individuals to allocate atten— tion to aspects of performance that are not directly involved in the online control of skill execution. McPherson (2000) demonstrated that successful. experienced tennis players spend a significant amount of time examining their own performance outcomes. as well as those of their opponents. as a tool for diagnosing and updating performance strategies and maintaining focus on the task at hand. At higher levels of practice. then. when attention may not be necessary for the online control of performance. such functions as strategizing about choices of actions not only may help to achieve desired goal-states but also may prevent individuals from over attending to well-learned performance processes and procedures. References Ackcrman. P L (1987). Individual differences in skill learning: An inte- gration ol psychometric and information processmg perspectives. Psy- chological Bulletin. I02. 3-27. Ackcrman. P. L.. & Cianciolo. A. T. (2000). Cognitive. perceptual-speed. and psychomotor determinants of individual differences during skill acquisition. Journal of Experimental Psychology: Applied. 6. 259 -—290. Allport. D. A.. Antonis, 13.. & Reynolds P. (I972). On the division of attention: A disproof of the single channel hypothesis. Quarterly Journal of Experimental Psychology. 24. 225-235. Anderson. J. R. (1983). The architecture of cognition. Cambridge. MA: Harvard University Press. Anderson. I R. (1993). Rules of mind. Hillsdale. NJ: Erlbaum. Banich. M T. (1997). Neuropsychology. New York: Houghton Mifflin. Baumeister, R. F. ( I984). Choking under pressure: Self-consciousness and paradoxical effects of incentives on skillful performance. Journal of Persona/tr) and Social Psychology, 46. filo-620. Beilock. S. L.. & Carr. T. H. (200!) On the fragility of skilled perfor- mance: What governs choking under pressure? Journal of Experimental Psychology: General. I30. 70l-725. Beilock. S. L.. Wierenga. S. A.. & Carr. T. H. (in press-a). Expertise. attention. and memory in sensorimotor skill execution: Impact of novel task constraints on dual~task performance and episodic memory. Quar- terly Journal of Experimental Psychology: Human Experimental Psy- dialog): Beilock. S. L.. Wierenga. S. A.. & Carr. T. H. (in press-b). Memory and expertise: What do experienced athletes remember? In I. L. Starkes & K. A. Encsson (Eds). Recent advances in research on sport expertise. Champaign. IL: Human Kinetics. Brown. A. (1987). Mctacognition. executive control. self-regulation. and other more mysterious mechanisms. In F. E. Weinert & R. H. Kluwe (Eds). Metacognition. motivation. and understanding (pp. 65-116). Hillsdale. NJ: Erlbaum. Chase. W. (3.. & Simon. H. A. (1973). Perception in chess. Cognitive Psychology, 4. 55—81. Chi. M. T.. Feltovich. P. 1.. 6r Glaser. R. (l98l). CategoriLation and representation of physics problems by experts and nrwices. Cognitive Science. 5. 121—l52. Cohen. A.. Ivry. R. 1.. & Keele. S. W. (1990). Attention and structure in sequence learning. Journal of Experimental Psychology: Llafnmg, Memory. and Cognition. I6. l7—30. Cohen. .I. (1988). Statistical power analysis for the behavioral sciences (2nd ed.). New York. Erlbaum. Cohen. 1. (I992). A power primer. Psychological Bulletin. [12. l55-l59. Craik. F. M., GoVini. R.. Naveh-Benjamin. M., at Anderson. N. D. (1996). The effects of diVided attention on encoding and retrieval processes in human memory. Journal of Experimental Psychology: General, 125. 159—180. Curran. T.. & Keele. S. W. ( I993). Attentional and nonattcntional forms of sequence learning. Journal of Experimental Psychology: Learning. Memory, and Cognition. 19. l89—202. Day. L. 8.. & MacNeilage. P. F. (1996). Postural asymmetries and Ian- guagc Iateraltzation in humans (Homo saptens). Journal of Comparative Psychology. I [0, 811-96. de Groot. A. (I978). Thought and choice in chess. The Hague. the Neth- erlands: Mouton. (Original work published 1946) Dunlap. W. P.. Comm. J. M., Vaslow. I. 8.. & Burke. M. .l. ([996). Meta-analysis of expenmcnts with matched groups or repeated measures designs. Psychological Methods. 1. l70-l77. Ericsson. K. A.. & Lehmann. A. C. (1996). Expert and exceptional per- formance: Evidence of maximal adaptation to task constraints. Annual Review of Psychology. 47. 273—305. Fisk. A. D.. & Schneider. W. (1984). Memory as a function of attention, level of processmg. and automatization. Journal of Experimental Psy- chology: Learning. Memory. and Cognition. I0. 181—l97. Fitts. P. M., & Posner. M. l. (1967). Human performance. Montcrcy. CA: Brooks/C ole. Helscn. W. P.. & Starkes. I. L. (1999). A multidimensional approach to skilled perception and performance in sport. Applied Cognitive Ps)‘('hol- ogy. I3. 1-27. Kahncman. D. (1973). Attention and eflort. Englewood Cliffs. NJ: Prentice Hall. Kanfer. F. H.. & Stevenson. M. K. ( I985). The effects of self-regulation on concurrent cognitive processmg. Cognitive Therapy and Research. 6. 667-684. Kanfcr. R.. & Ackerman. P. L. (1989). Motivation and cognitive abilities: An integrative/aptitude treatment interaction approach to skill acqm- sition. Journal of Applied Psychology. 74, 657—690. Keele. S. W., & Summers. i. J. (I976). The structure of motor programs, In G. E. Stelmach (Ed.), Motor control: Issues and trends (pp. 109— 142). New York: Academic Press. 123 l6 BEILOCK. CARR. MACMAHON. AND STARKES Kimble. G. A.. & Perlmuter. L. C. (1970). The problem of volition. Psychological Renew. 77. 361-384. Kluwe. R H. (1987). Executive decisions and regulation of problem- solving behavior In F E. Weinert & R. H. Kluwc (Eds). Metacognition. motivation and understanding (pp. 31—64). Hillsdalc. NJ: Erlbaum. Langer. E. at lmber. (i. (1979). When practice makes imperfect: Debili- tating effects of overlearning. Journal of Personality and Social Psy- chology. 37. 2014-2024. Leavitt. J. (1979). Cognitive demands of skating and stick handling in ice hockey. Canadian Journal of Applied Sport Sciences. 4. 46—55. LcWis. B.. & Linder. D. (1997). Thinking about choking? Attentional processes and paradoxtcal performance. Personality and Socral Psycho!» ogy Bulletin. 23, 937-944. Logan. G. D. (1988). Toward an instance theory of automatization. Psy- chological Review, 95. 492—527. Logan. G. D (1990). Repetition priming and automaticity: Common un- derlying mechanisms. Cognitive Psychology. 22. 1—35. Masters. R. S W. (1992). Knowledge. knerves. and knowshow: The role of explicit versus implicit knowledge in the breakdown ofa complex motor skill under pressure. British Journal of Psychology, 83, 343-358. Masters. R. S. W., Polman. R. C. J.. & Hammond. N. V. (1993). "Rem- vesimentz" A dimensmn of personality implicated in skill breakdown under pressure. Personality & Individual Differences. I4. 655-666. McPherson. S. L. (2000). Expert-noVice differences in planning strategies during collegiate Singles tennis competition. Journal of Sport and Ex- ercise Psychology. 22. 39—62. Nisscn, M. J., & Bullcmer. P. (1987). Attentional rcqmrcmcnts of learning: Ei'idence from performance measures. Cognitive Psychology, [9. 1-32. Peters. M. (1981). Handedness; Effect of prolonged practice on between- hand performance differences Neuropsvchologia. I9. 587-590. Peters. M (1988). Footedness: Asymmetries in foot preference and skill and neuropsychological assessment of foot movement. Psychological Bulletin. 103. 179—192. Priest. A. G. & Lindsay. R. O. (1992). New light on novice-expert differences in phySICS problem solving. British Joumal of Psychol- ogy. 83. 389-405. Proctor. R. W.. & Dutta, A. (1995). Skill acquisition and human perfor- mance. Thousand Oaks. CA: Sage. Searlcman. A. (1980). Subject variables and cerebral organization for language. Cortex. [6. 239—254. Singer. R. N.. Lidor. R.. & Cauraugh. J. H. (1993). To be aware or not aware'.’ What to think about while learning and perf'orrning a motor skill The Sport Psychologist. 7. 19—30. Smith. M. D.. & Chamberlin. C. J. (1992). Effect of adding cognitively demanding tasks on soccer skill performance. Perceptual and Motor Skills. 75. 955-961. Starkes. J. L.. & Deakin. J. (1984). Perception in sport: A cognitive approach to skilled performance. In W. F. Straub & J. M. Williams (Eds). Cognitive span psychology (pp. 115-128). Lansmg. MI: Sport Science Associates. Tenenbaum. G.. & Bar-Eli. M. (I993). Decision~making in sport: A cog- nitive perspective. In R. N. Singer. M. Murphey. & L. K. Tennant (Eds). Handbook of research on sport psychology (pp. 171—192). New York: Macmillan. Wickens. C. D. (1980). The structure of processmg resources. In R. Nickerson (Ed.), Attention and performance VII (pp. 239 -257). Hills- dale. NJ: Erlbaum. Wickens. C. D. (1984). Processmg resources in attention. In R, Parasura- man & D. R. DaVICS (Eds). Varieties ofattention (pp. 63-102). Orlando. FL: Academic Press. Wulf. G.. Hob. M.. & Prinz. W. (1998). Instructions for motor learning Differential effects of internal versus external focus of attention. Journal of Motor Behavior, 30. 169—179. Wulf. 0.. McNevin. N. H. Fuchs. T.. Ritter. F.. & Toole. T (2000) AttenUonal focus in complex skill learning. Research Quarterly for Exercise and Sport. 7]. 229—239. Received March 2. 2001 Revision received August 15. 2001 Accepted August 20. 2001 I r————————————-——-————————————————--‘——1 ORDER FORM Psychology: Applied! ISSN: 1076-898X $32.00, APA MEMBER/AFFILIATE $63 .00, INDIVIDUAL NONMBMIBR $144.00, INSTITUTION In DC add 5.7596 lln MD add 5% sales to: Tom. AMOUNT ENCLOSED 5 Subscription orders must be prepaid. (Subscriptions are on a calendar basis only.) Allow 4-6 weeks for delivery of the first issue. Call for international subscription rates. SEND THIS ORDER FORM TO: American Psychological Association Subscriptions 750 First Street. NE Mm“ Washington. DC 200024242 Psvcnaoooi. Assocmon Or call (800) 374-2721, fax (202) 336—5568. TDD/ITY (202)336-6123. Email: subscriptlonsOapaorg r._____________.=____ Start my 2002 subscription to Journal of Experimental 69 Send me a FREE Sample Issue I 0 Check Enclosed (make payable to APA) I Charge my. CD VISA ® MasterCard .American I Express I Cardholder Name I Card No. Exp. date | | Signature (Required for Charge) I BILLING ADDRESS: I Citv State Zip I Daytime Phone I SHIP TO: I Name I Address I I City State Zip | APA Member 8 mar: ' ______ ______________________________.J 124 «IWjgmygumjI