MURI-OPUS 1 DOD Multidisciplinary Research Program: MURI Operator Performance Under Stress (OPUS). White Paper Stress, Human Information Processing, and Performance Mediators: Application of a Descriptive Framework to Current Modeling Tools P. Ward, J.L. Szalma, and P.A. Hancock. Department of Psychology and Institute for Simulation and Training University of Central Florida Work performed under the U.S. Army Research Office – MURI: Optimizing Cognitive Readiness Under Combat Conditions Version 3 MURI-OPUS 2 Stress, Human Information Processing, and Performance Mediators: Application of a Descriptive Framework to Current Modeling Tools Abstract The effects of stress upon performance have typically been researched using either an input (e.g., environmental stressors), throughput (e.g., adaptive processes) or output (e.g., physiological response) approach rather than considering each as part of a single dynamic process. The extant research has also tended to focus upon the effects of a single environmental stressor on one specific cognitive process and has often failed to consider the multidimensionality of stress. To address these issues, Hancock and Warm (1989) outlined a dynamic model of stress and performance detailing an individual’s psychological and physiological adaptability to stressful environments. The model predicts that psychological adaptability (e.g., attentional narrowing) is exhausted in a similar manner to the way in which stress degrades physiological response capacity (e.g., homeostatic regulation). Within this model, task demand is proposed to be one of the primary sources of stress. According to the model, the effect of stress upon task performance is intimately related to the nature of the task, type of stressor and moderators of the stress-performance relationship. However, without a comprehensive evaluation of the literature, prediction of how task performance will be debilitated or facilitated by stress, or whether stress effects can be alleviated remains to be fully answered. A three dimensional matrix is proposed to detail the effects of stress on information processing as well as moderators of the stress-performance relationship. The aim of this paper is to present a descriptive framework that will inform current modeling practice in a military context. The U.S. ARL-HRED presently employ a task network modeling architecture, IMPRINT, to predict soldier-system performance under stress. Within IMPRINT, performance shaping functions and stress degradation algorithms are applied to modeled data to reflect individual differences in task performance under stressful conditions. Performance on military occupational specialty tasks is currently modeled under standardized conditions using the ACT-R cognitive architecture. Whilst IMPRINT has been successfully implemented, it is somewhat limited both by the availability of suitable stress degradation algorithms and the broad brush approach used to apply the same function/algorithm to all tasks grouped together by a taxonomic classification system. The matrix aims to provide resolution, at least in part, to a number of issues in this regard. Firstly, by clearly outlining the multitude of environmental, physical and taskrelated stressors within the matrix, and considering the transactional relationship between operator and task, a more complete understanding of stress effects on human performance can be delineated. Similarly, a comprehensive description and subsequent quantification of the effects of stressors on a range of information processing variables sub-serving, perception, cognition, and action will allow both theoretical and practical implementation issues related to current modeling approaches to be addressed. In addition, the inclusion of moderator variables within the matrix will provide an added dimension that will facilitate future modeling endeavors. Version 3 MURI-OPUS 3 Introduction A plethora of research exists which documents the effects of stress on human performance (see Hancock & Desmond, 2001). However, researchers have typically examined stress via one of three inter-related approaches. Whilst some have considered stress as a property of the environment (e.g., Hockey, 1983), others have examined the appraisal or coping mechanisms of stress (e.g., Lazarus, 1966), or the associated physiological response (Selye, 1956). Researchers have often embarked upon these approaches independently of each other. In addition, the majority of research has concentrated upon the effects of a single stressor on ‘isolated’ information processes (e.g., selective attention, working memory). To date, very few assessments have been made on a broad range of interactive stressors with the specific intention of detailing both the variable and predictable effects upon perception, cognition, and action. In one exception, Hockey and Hamilton (1983) examined the cognitive patterning of stress states (e.g., anxiety noise, narcotic, etc.) on strategic (i.e., selective attention, decisionmaking) and structural (i.e., short-term memory, alertness) processes. This research highlighted that while some commonality exists between predominantly environmental stressors, a multidimensional approach was necessary to explain the lack of response uniformity. While some of the conclusions made by Hockey and Hamilton were ‘reliable and substantiated’ the authors acknowledged that on occasion, interpretation was based upon ‘rather isolated results’ (p.350). The results were also not discriminated by the magnitude of effect further decreasing the clarity of interpretation. Notwithstanding, Hockey and Hamilton’s research provides a first step to making a concerted effort in determining a range of stressor effects on various information processes within a single investigation. Lazarus and Folkman’s (1984) focus on mechanisms of appraisal and coping has emphasized the person-environment-stress interaction. This viewpoint suggests that the transaction between an individual and their appraisal of the environmental task as taxing, or exceeding available resources, moderates the effect of a stressor. Consistent with Hancock and Warm (1989), Matthews (2001) indicated that this transaction occurs at multiple levels (e.g., biological, psychological). The negative effects of stress occur when individuals appraise an event as threatening and when they consider that available coping strategies would be ineffective in dealing with the stressor. Consequently, those variables that moderate the relationship between sources of stress and the nature of task-related information processing are likely to be influenced by the transaction between operator and task. Although a wealth of literature exists on coping (e.g., Matthews & Campbell, 1998; Smith & Lazarus, 1993) and effort allocation (e.g., Eysenck & Calvo, 1992), surprisingly, only limited empirical research has specifically attempted to identify those variables that moderate stress effects on task-related information processing in a realworld setting. Hancock and Warm (1989) suggested that the three approaches described as the ‘trinity of stress’ are, in reality, three aspects of a single dynamic process which could be used to describe responses at multiple levels of analysis (Hancock & Warm, 1989). More specifically, these authors suggested that the study of the physical characteristics of Version 3 MURI-OPUS 4 environmental stressors was a deterministic, input approach to examining the influence of stress upon performance, whereas coping mechanisms reflected a nomothetic, compensatory or adaptive approach, and bodily response, an ideographic, output approach. The Hancock and Warm (1989) model outlined three modes of operation that are reflective of adaptational success, ranging from dynamic stability to dynamic instability. Hancock and Warm predicted that at minor levels of input stress, behavior remains relatively stable and is not adversely affected. Thus, no output stress would be observed. These low levels of stress are adequately dealt with by an individual’s adaptive capability. On a continuum ranging from hypostress to hyperstress, the central point, the zone of comfort, reflects such adaptive capability to minor stress levels. However, as the intensity of, or exposure to, stress rises (i.e., as individuals transition towards the region of maximal adaptability), and adaptive capability is eventually surpassed, output is affected. The result is increasing discomfort, a reduction in performance efficiency and then a rapid decrease in adaptive capability. Importantly, the model indicates that input stressors affect both psychological (i.e., resources allocation) and physiological adaptability (i.e., homeostatic regulation). Whilst, an input stressor may appear to affect psychological processes alone, physiological functioning is also typically indirectly affected. Moreover, the region of maximal adaptability is larger for physiological functioning. Although attentional resources are depleted when maximal psychological adaptability is exceeded, a swift decline in physiological adaptability closely follows. The model is defined by two different constituents of task-related information, information rate; the temporal flow of environmental information, and information structure; the perceived meaning of the information’s content. The latter is specifically influenced by the task, stressor and future expectations. Although many forms more of input stress can be observed, the primary source is task demand (Hockey, 1986; Hancock & Warm, 1989). Moreover, Hockey and Hamilton’s research illustrates that variable effects of stressors are likely to be observed as the nature of the information processing requirements of the task change. In a normal information-rich environment and under low levels of stress, multiple solutions to a task may be available, and different individuals are likely to interpret information in a number of ways to achieve a successful outcome. The result is an array of functionally operational goal-directed behaviors. However, as stress increases, the strategies available to an operator that lessen task demand are reduced, and the number of solution paths with respect to the goal is diminished. Consequently, the stereotypical nature of behavior is increased as attention narrows to finding the most appropriate and available solution to the task. During task performance, individuals seek to maximize their capacity to retain optimum information throughput using such compensatory strategies as ‘attentional narrowing’ to modify the perceived rate and structure of information. A number of theoretical predictions can be made both from the model and recent works elaborating on this theme (Hancock, Szalma & Weaver, 2003; Szalma & Hancock, 2003; Ward, Szalma & Hancock, 2003). Of particular interest here, however, is the way in which performance degrades over time under stressful conditions and how performance Version 3 MURI-OPUS 5 is restored to within a normative range. The model predicts that the drain on attentional resources (e.g., narrowing) occurs in a similar manner to the way in which stress degrades physiological response capacity. However, recent predictions derived from the model indicate that stress may affect tasks differently, depending upon the nature of the task and/or how well the task is learnt. One hypothesis is that performance may degrade quicker for those tasks that are more knowledge-driven than those that are more skilloriented. The degree to which a task is automated or proceduralized (see Anderson, 1983; Fitts & Posner, 1967); a product of learning and skill development, may well determine the relative effect of stress on performance. In contrast, recent evidence suggests that stress acts to bring normally automated tasks under conscious control, thus competing for resources which are either otherwise deployed or already exhausted (see Beilock et al, 2002; Masters, 1992). Lastly, it is unknown whether there is a hysteresis effect with respect to depletion and recovery of attentional resources dependent upon the nature of, and experience with, the task. The implication of these hypotheses is that if stress affects tasks differently depending upon the nature or constituents of the task, then tools used to model performance effects may need to take these factors into consideration during, rather than after the modeling process. Detailing the multidimensional nature of stress and performance is an onerous task. However, without a comprehensive evaluation of the literature, prediction of how task performance will be debilitated or even facilitated by stress will be even more arduous. What is more, this is considerably amplified when one considers the vast range of potential interactions between different environmental, task-based and social stressors and the information processing requirements of tasks sub-serving perception, cognition and action. A recent review by Pew and Mavor (1998) suggested that only limited information is available on the effects of all known environmental stressors. These authors summarized known effects for heat, toxic substances, noise, vibration, as well as factors related to fatigue (e.g., sleep degradation), and cognitive workload. While the details of the effects are beyond the scope of this paper, Pew and Mavor concluded that much of the current research has been conducted with the aim of eliciting threshold limits as opposed to determining the degree of degradation per unit of time. When one includes those factors that may moderate or mitigate the stress-performance relationship, the complexity of performance prediction becomes all too evident. From a practical perspective, one often simply wants to know how stress affects performance and by how much performance will typically be degraded. Moreover, the question often arises, is there anything one can do to alleviate these effects? Of related interest are whether particular individuals are more suited to the task than others? Can we aid in the selection of best man for the job? Can we train the right person to acquire the appropriate strategies? In order to answer each of these questions the stressor-taskmoderator relationship outlined above needs to be fully delineated. To fill the current void, we propose a descriptive framework which details the effects of a broad range of stressors on information processing and examines the potential variables that may moderate this interaction. Our aim is to provide both a descriptive database and a series of meta-analytic reviews that inform current mechanisms for predicting task performance under stress. Ultimately, the purpose of this venture is to increase the current prediction Version 3 MURI-OPUS 6 capacity of existing models of human performance that incorporate some form of stress degradation algorithm, moderation process or workload assessment. To meet these goals, a three-dimensional matrix has been constructed which encapsulates the variables of interest on each dimension: sources of stress, information processes, and moderators of the stress-performance relationship. Before giving significant detail on each dimension, an overview of the current model of soldier-system and human cognitive performance prediction is provided. IMPRINT Improved Performance Research Integration Tool (IMPRINT) is a task network modeling architecture developed by the U.S. Army Research Laboratory Human Research and Engineering Directorate (ARL-HRED). It is used to represent soldier-system performance for the purposes of test and evaluation. Two of the primary aims of IMPRINT are to determine the soldier performance requirements for the acquisition of a new system and in particular, to predict soldier-system performance under a myriad of conditions (Allender, Salvi, & Promisel, 1997; Lebeire, Biefeld, Archer, Archer, Allender, et al., 2002; O’Brien, Simon, & Swaminathan, 1992). Earlier DOS-based versions of IMPRINT (e.g., HARDMAN III) were constructed from discrete modules for a range of operator-, task-, and system-related estimations. IMPRINT is now implemented using the Microsaint task network modeling environment which subdivides system missions into functions and tasks, and produces performance estimates and comparisons to a standard. Given that a multitude of performance changes are likely to occur under stressful compared to standardized conditions and that some individuals may be more suited to a particular task than others, of specific interest in these and subsequent operations are the performance shaping functions (PSFs), stress degradation algorithms (SDAs) and workload modeling tools used in performance prediction. The PSFs predict relative changes in performance for a specific set of personnel characteristics and training regimen (i.e., recency and frequency of practice) whereas the SDAs (or look up tables for stressors) modify typical/baseline measures to account for negative affects upon performance. Current methods of workload prediction will be addressed in a later section. The relationships between predictor and performance for both PSFs and SDAs are described for categories of tasks (i.e., taxons) rather than specific tasks. That is, different functions/algorithms were developed for different taxons (see O’Brien et al., 1992). Similar effects for tasks that fall into the same taxon can be accounted for using this method. The task taxonomy was largely borrowed from Berliner, Angell, and Shearer (1964, also see Fleishman & Quantance, 1984). Nine taxons were identified spanning perceptual, cognitive, psychomotor and communication modalities: visual recognition/discrimination numerical analysis information processing/problem solving fine motor (discrete) fine motor (continuous) Version 3 MURI-OPUS 7 gross motor (heavy) gross motor (light) communication (reading and writing) communication (oral) Tasks relevant to each Military Occupation Specialty (MOS) are user-assigned to a maximum of three taxons. Personnel characteristics such as the Armed Services Vocational Aptitude Battery (ASVAB) composite scores, reading grade level, complex perceptual speed/accuracy, simple reaction speed/accuracy, etc. are used to predict task performance for all tasks that are pre-assigned to each taxon by means of the PSFs (which translate to a series of regression equations based on aggregated and standardized test scores). Due to the differential effects of stressors on each task, tasks are weighted according to the taxon to which they have been assigned so that, when the SDA is applied (post application of the PSF), the impact of stressors on each task is accordingly weighted (for a more detailed overview of SDAs and PSFs, see Allender et al., 1997; O’Brien et al., 1992). At present, five, predominantly environmental, stressor equations exist within IMPRINT. These relate to Mission Oriented Protective Posture (MOPP) gear for biological, chemical and nuclear defense (i.e., protective clothing), heat, cold, noise, and sustained operation (i.e., sleepless hours). Each of these algorithms are applied individually or in combination via the use of a power function (see Harris, 1985). The majority of degradation factors for each stressor were derived from research prior to 1988, some of which dates back to the 1950’s. Moreover, the SDAs that have been developed are only capable of modifying tasks within a taxon for a very limited number of stressors (i.e., one to three). In addition, where degradation algorithms have been written, a comprehensive evaluation of current stressors has not followed. Allender et al. (1997) indicated that this is a high priority for research. Even a cursory glance at the literature suggests that there is much scope for updating SDAs to incorporate ‘state of the art’ knowledge and consequently bridge the gap between existing stressors and their effects on current tasks/taxons. Equally important is that this information will increase the number of stressors and the associated degradation factor that could be applied to each taxon. The U.S. ARL-HRED and similarly, the U.S. Defense Special Weapons Agency are currently conducting research in this area to develop a prioritized list stressors and their effects based upon ‘user need’ and the ‘availability of generalizable data’ (Allender et al., 1997; see also Anno, Dore & Roth, 1996). Finally, the cognitive mechanisms that underlie stress effects need to be explored. For performance under standard conditions, the IMPRINT model utilizes the ACT-R theory of cognition to model performance. It is to this cognitive architecture that we now turn. ACT-R (/PM) Adaptive Control of Thought-Rational (ACT-R) (Anderson & Lebiere, 1998) is both a theory of cognition and a hybrid cognitive architecture that models and predicts human cognition. It operates at both symbolic (e.g., as a production system) and sub-symbolic Version 3 MURI-OPUS 8 (e.g., as a neural-like activation-based system) levels. At the symbolic level ACT-R assumes two types of knowledge. Declarative knowledge is that which we are consciously aware of and is verbally describable, whereas procedural knowledge specifies how declarative knowledge is instantiated to solve a problem. Its basic cognitive operation is a production-rule, often referred to as a condition-action pair or ‘if… then…do…’ statement. ACT-R is both a parallel-matching and serial firing production system centered on the current goal or focus of attention. When the left hand side of the production (i.e., the condition) is matched with the current goal, post conflict resolution (i.e., selection of a single production from multiple productions that match information currently held in working memory), the production-rule fires and the right hand side of the production (i.e., the action) modifies the current, or deposits a new, goal on to the ‘goal stack’ (i.e., in working memory). As an alternative to a cognitive action, a physical action may also be executed. Goals are represented as chunks (i.e., encoded structures with a number of slots or attributes (≤ 6), each of which hold a single piece of information) and stored in declarative memory along with other information also represented as chunks. New productions are created from declarative chunks via a process of production compilation. That is, a new generalizable production cycle is generated from the residual execution trace of multiple production firings (Lebiere et al., 2002). At the sub-symbolic level, retrieval of chunks from declarative memory is based upon the chunk with the highest level of activation. Activation within this framework is derived from the attentional weight given to the goal based upon factors such as history of use, information contained within a chunk, and time-based decay. A set of sub-symbolic learning processes can also modify activation weights. In a similar manner, conflict resolution is determined as a rational process of both the estimate of a production’s worth and the trade-off between the probability and cost (e.g., time) of goal achievement (for more details, see Anderson & Lebiere, 1998). In an attempt to extend ACT-R’s application to tasks that involve both perception and action as well as cognition, Byrne and Anderson (1998) first proposed ACT-R/PM as a mechanism to augment central cognition. The perceptual-motor (PM) component is now comprised of visual, motor, speech, and auditory modules (see also Anderson, Matessa, & Lebiere, 1997; Byrne, 2001) and was largely influenced by the EPIC architecture (see Kieras & Meyer, 1997), particularly at the motor level. ACT-R/PM was developed to model both the interaction with environmental stimuli and the perceptual-motor demands of most real world tasks. ACT-R and its integration with IMPRINT ACT-R and IMPRINT have recently been integrated for the purposes of modeling and predicting task performance in situated events. The most developed application to date has been in a pilot navigation context for taxiing along the runway (Lebiere et al., 2002). A similar integration of ACT-R and an IMPRINT-related task network modeling environment (TNME) has also recently been implemented in a shoot-list management task (e.g., recognition and decision making) by the U.S. Air Force Research Laboratory’s Version 3 MURI-OPUS 9 Human Performance Model Integration (HMPI) program using the Combat Automations Requirements Testbed (CART) TNME (see Craig et al., 2002). In both situations, ACT-R and IMPRINT functionally combine by focusing upon different aspects of performance. Whilst IMPRINT is predominantly concerned with the macro-components of real world tasks, ACT-R contends with a task’s cognitive, perceptual, and motor components. To direct ACT-R’s goal driven processes, task-related information is sent from IMPRINT to ACT-R via a component object model (COM) link which specifies a goal in ACT-R. This in turn creates a goal set that corresponds to the task description. Each sub-goal specified within ACT-R results in data regarding how the goal was accomplished, as well as the latency and accuracy (i.e., number of errors) of each goal’s completion. Latencies for retrieval are determined as a function of a chunk’s activation whereas the time taken to apply a production is determined by a production’s action component. Rudimentary internal actions take approximately 50 ms, while external actions typically take longer and are determined by the ACT-R/PM module. Errors occur due to the stochastic nature of the decision process and its interaction with a dynamic environment. As opposed to errors being a problem of ‘faulty symbolic logic’ (Lebiere et al., 2002, p.3), the inherent variability is more likely to reflect an accurate representation of human operators’ performance under changeable conditions than that of a purely deterministic system. Errors may originate from a number of sources within ACT-R such as retrieving the wrong information or failure to retrieve necessary information from memory (i.e., memory failures), use of an ineffective production rule (i.e., choice or selection errors), or failure of the perceptual modules to detect important information (i.e., attentional errors). One of the benefits of using this type of integration is the reduced need for endless data. Moreover, ACT-R can generate distributions of time and error data for each task within IMPRINT that appropriately reflect the actual perceptual, cognitive, and motor components of tasks experienced in the real world, albeit under standardized conditions (for detailed summary, see Lebiere et al., 2002). In addition to modeling cognitive performance, associated workload is currently being modeled in ACT-R for use within IMPRINT (see Lebiere, 2001; Lebiere et al., 2001) such that the level of effort required to perform each MOS task could be used to moderate the ‘attributes of the memory model’ (Lebiere et al., 2002, p.6). As an adjunct to the IMPRINT/ACT-R integration model, recent developments have been made in modeling related factors such as emotion (e.g., Belavkin, 2001), behavior moderators (e.g., Ritter et al., 2002), and fatigue (e.g., Jongman & Taatgen, 1999) within ACT-R. There is no current initiative to replace IMPRINT’s performance shaping functions and stress degradation algorithms with an ACT-R (/PM) model. Regardless, the potential exists to model the effects of stress in this manner. Accordingly, when determining the effects of stressors and moderators on perceptual, cognitive and motor mechanisms current and potential modeling developments need to be considered. Version 3 MURI-OPUS 10 The Matrix In addition to gaining greater understanding of the stress-performance relationship, the impetus for this research is to increase current prediction capacity under stressful conditions by informing the construction of performance shaping functions and degradation algorithms for existing and future stressors within IMPRINT. The matrix is a three-dimensional database detailing the qualitative (and quantitative) effects of stress upon performance, and where applicable, the effects of specific moderators upon that relationship. Our purpose is to detail the psychological and physiological adaptability with respect to different information processing tasks, stressors, and moderators. Such a system will allow the trinity of stress (e.g., input, adaptive, and output processes) to be fully examined and the associated effects to be evaluated. The first dimension on the matrix will catalog a range of performance stressors, including environmental (e.g., social), physical (e.g., fatigue, hunger) and task-based stress. The second dimension will detail the range of human information processes underlying perception, cognition, and action that could potentially be affected by these stressors. Moderators of the stressperformance relationship, such as personality traits, skill level and training will reside on the third dimension. Version 3 MURI-OPUS 11 The characteristics of experiments examining the stress-performance relationship are currently being descriptively coded and entered into the searchable database. Details include sample characteristics, task description, dependent measures and experimental design. To date, over 120 independent experiments from 100 articles have been entered into the matrix across a range of stress and performance indices and mediators of that relationship. Each study descriptively coded within the matrix is graphically represented as a specific x, y, z co-ordinate within the three-dimensional space. Figure 1 illustrates the potential ability of the matrix database and subsequent meta-analyses to update IMPRINT and its SDAs. Findings from the first meta-analysis of noise intensity and memory recall can be found in Ward, Szalma and Hancock (2003). Collectively, the matrix provides an integrative database that will facilitate future modeling efforts, drive theoretical development and allow existing models of stress and performance (e.g., Hancock & Warm, 1989) and their associated predictions to be tested. Issues for Concern within IMPRINT. As previously noted, only five stress degradation algorithms have been derived within IMPRINT, and in the worst case scenario, only one algorithm is available for a task that falls within a particular taxon. Clearly, the current implementation is not exhaustive and other environmental, physical and task-based stressors need to be incorporated within this dimension to make it comprehensive. More specifically, how these stressors affect a wide range of information processes sub-serving perception, cognition and action is of particular concern. One of the goals of our research was to build upon the seminal work of Hockey and Hamilton (1983) to determine both the dynamic and stochastic nature of the cognitive patterning of some stressors while potentially identifying the deterministic, or at least invariant, performance effects for others. Hockey and Hamilton’s approach fits well within the IMPRINT model in that it mirrors the taxon system (e.g., particular stress effects for certain types of tasks) devised within IMPRINT to which SDAs are applied. However, given the current delineation of taxons within IMPRINT, it is likely to prove impossible to derive a ‘common’ stress degradation algorithm for each taxon within the existing categorization scheme. That is, other than at the perceptual, cognitive, and psychomotor level only modest parity exists between taxons and the vast array of information processes upon which research has been conducted. For instance, in the current taxonomy, cognitive tasks are broken down into either those that require numerical analysis or those that require some form of information processing and problem solving. Accordingly, tasks that pose differential information processing demands such as vigilance and memory recall tasks will, for all intents and purposes, be modeled similarly by the current implementation of IMPRINT. Using the current taxonomy as a template for dimension division is likely to result in a disproportionate and asymmetrical structure that conceptualizes the majority of processes as essentially equivalent and consequently, has little explanatory power. Although some of the original research which lead to the current instantiation suggested that the existing task taxonomy was developed with Wickens’ (1980) processing structure in mind (see O’Brien et al., 1992), there is little which resembles this model and that distinguishes it from Berliner et al.’s (1964) original classification scheme. The parsimony achieved by devolving human Version 3 MURI-OPUS 12 information processing into the existing taxonomic classification system is likely to prove costly when attempting to accurately model the effects of stress on performance. Moreover, there is concern regarding the reliability of user assignment of tasks to taxons. Christensen and Mills (1967) highlighted a median inter-rater reliability of .78 (Rho range = .29 to 1.00). This has enormous implications for applying SDAs to tasks, where differences between users in task assignment could lead to different SDAs being applied and contradictory predictions on performance for the same task. On a theoretical level, very little ‘human’ research has been conducted on the effects of stressors on procedural and declarative memory or on any of the sub-processes defined within the ACT framework. Even less research exists that has successfully modeled human stress and emotion in ACT-R (for exceptions, see Belavkin, 2001). Considering the questions raised by Szalma and Hancock (2002) regarding the effects of stress on Rasmussen’s (1983) trichotomy of skill, rules, and knowledge, the application of stress degradation algorithms post modeling of performance via both procedural and declarative mechanisms may be a contentious issue. This issue is complicated further when one considers the potential shift from automatic to controlled processing of previously automated skills under stressful conditions (see Beilock et al, 2002; Masters, 1992). Stress may actually change or restrict the nature of processing and so to accurately model the effects of stress upon human information processing, algorithms may have to be included as part of the cognitive modeling process, not attached a posteriori. Further empirical and modeling research is needed to resolve these issues. At present only limited research exists on the effect of stress moderators. As mentioned earlier, the person-environment interaction is likely to play an active role in moderating the effects of stress between operator and task. Consequently, coping and appraisal are likely to make up a significant proportion of this dimension. However, related factors, for instance, perceived control (see Bowers et al., 1996), resource allocation policy in terms of compensatory effort (e.g., Eysenck & Calvo, 1992; Kahneman, 1973; Matthews & Campbell, 1998; Sanders, 1983), training (e.g., Hall-Johnston & Cannon-Bowers, 1996), personnel selection (e.g., Hogan & Lesser, 1996), and system design itself (e.g., Wickens, 1996) may also moderate the effects of this interaction. Personnel characteristics and training data in the form of performance shaping function, and workload estimates are currently modeled within IMPRINT. However, the interaction between PSFs and stress is not directly considered. The individual application of PSF and SDAs is unlikely to provide the same result as the inclusion of the interaction between these factors. Concluding Remarks The primary purpose of this paper and the development of the matrix is to improve the capacity to predict soldier-system performance. The matrix and subsequent meta-analyses will provide much needed information on the effects of stressors on information processing and those variables that can be used to moderate the process. Empirical verification of modeled cognitive processes will facilitate integration of stress-related effects into future modeling ventures. The integration of new stressors, validation and Version 3 MURI-OPUS 13 modification of old stressor effects, and revisions to the existing taxon scheme, once integrated, are hypothesized to all provide an immediate effect upon performance prediction. However, data can only inform where there is data to inform. The glaring truth of the matrix’s construction is likely to be that there are as many gaps in the current knowledge base as there is understanding of these complex effects. These gaps need to be filled through further empirical research. Where knowledge is lacking however, system design can still be enhanced by the matrix. Where data does not exist for the effects of particular stressors on certain processes then design emphasis can be placed on taxing those systems where the stress effects are known. Identified Research Questions Can existing models of stress and human performance account for effects of stress identified by the matrix? How well do current stress degradation algorithms used within the IMPRINT model predict real-world performance under the stressful conditions to which they apply? To what extent can current stress degradation algorithms be modified by the matrix to predict known stress effects? Can stress effects on information processing be delineated in a meaningful fashion given the abundance of theoretical models of performance and cognition? As the information processing dimension of the matrix is specified, how will this information fit with the existing taxon classification scheme? How does the effect of stress on controlled and automatic (and skill oriented) processing impact the current method of modeling cognitive, perceptual and motor performance before stress degradation algorithms are applied? Can a consensus of the known stress effects on information processing tasks be reached to fit within the current taxonomic classification system? Will the differential stress effects on information processing tasks necessitate a modification of the existing taxon scheme? If so, can a series of appropriate stress degradation algorithms be derived to fit a new scheme? Can the effects of stress be modeled within the IMPRINT/ACT-R integration model? What are the main variables moderating the effects of stress on task performance? Can the effects of moderating variables be accounted for in the IMPRINT/ACT-R integration model, as per the current workload implementation? What is the commonality of moderator effects across information processes, tasks, and taxons? What gaps exist in the literature regarding the stress-performance relationship and the effects of moderator variables? Version 3 MURI-OPUS 14 Anticipated Outcomes A series of meta-analytic reviews of stress effects upon performance, including the moderating effects of several variables (e.g., training, operator traits, etc.) will inform and enhance modeling capacity and theory development. The descriptive framework can be directly applied to the mechanisms used by IMPRINT resulting in stronger prediction capacity of this model of performance under stressful conditions. Preliminary steps toward an integrative theoretical and modeled account of stress effects on MOS tasks/taxons. References Allender, L., Salvi, L., & Promisel, D. (1997). Evaluation of human performance under diverse conditions via modeling technology. In Proceedings of Workshop on Emerging Technologies in Human Engineering Testing and Evaluation, June 1997. Brussels, Belgium: NATO Research Study Group. Anderson, J.R. (1983). The architecture of cognition. Cambridge, MA: Harvard University Press. Anderson, J.R. & Lebiere (1988). The atomic components of thought. Mahwah, NJ: Lawrence Erlbaum. Anderson, J.R., Matessa, M., & Lebiere, C. (1997). ACT-R: A theory of higher level cognition and visual attention. Human Computer Interaction, 12, 439-462. Anno, G.H., Dore, M.A., & Roth, T.J. (1996). Taxonomic model for performance degradation in combat tasks. (DNA-TR-95-115). Alexandria, VA: Nuclear Defense Agency. Atkinson, R.C. & Shiffrin, R.M. (1968). Human memory: A proposed system of its control processes. In K.W. Spence and J.T. Spence (Eds.), The psychology of learning and motivation: Advances in research and theory, vol.2, (pp.89-187). New York: Academic Press. Baddeley, A.D. (1986). Working memory. Oxford, UK: Oxford University Press. Belavkin, R. V. (2001). The Role of Emotion in Problem Solving. In Proceedings of the AISB'01 Symposium on Emotion, Cognition and Affective Computing (pp. 49-57). Heslington, York, England. Berliner, D.C., Angell, D., & Shearer, J.W. (1964). Behaviors, measures, and instruments for performance evaluation in simulated environments. Paper presented at a symposium and workshop on the quantification of human performance, August. Alburquerque, New Mexico. Bowers, C.A., Weaver, J.L., & Morgan, B.B. (1996). Moderating the performance effects of stressors. In J.E. Driskell and E. Salas (Eds.), Stress and human performance. (pp.163-192). Mahwah, NJ: Lawrence Erlbaum. Byrne, M.D. (2001). ACT-R/PM and menu selection: Applying a cognitive architecture to HCI. International Journal of Human-Computer Studies, 55, 41-84. Version 3 MURI-OPUS 15 Byrne, M.D. & Anderson, J.R. (1998). Perception and action. In J. R. Anderson & C. Lebiere (Eds.), The atomic components of thought, (pp. 167-200). Mahwah, NJ: Lawrence Erlbaum. Craig, K., Doyal, J., Brett, B., Lebiere, C., Biefield, E. et al. (2002). Development of a hybrid model of tactical fighter pilot behavior using IMPRINT task network modeling and the Adaptive Control of Thought-Rational (ACT-R). Eysenck, M.W. (1992). Anxiety: The cognitive perspective. London: Lawrence Erlbaum Associates. Eysenck, M.W. & Calvo, M.G. (1992). Anxiety and performance: The processing efficiency theory. Cognition and Emotion, 6, 409-434. Fleishman, E.A. & Quaintance, M.K. (1984). Taxonomies of human performance: The description of human tasks. Orlando, FL: Academic Press. Fitts, P.M. & Posner, M.I. (1967). Human Performance. Westport, CT: Greenwood Press. Hall-Johnston, J. & Cannon-Bowers, J.A. (1996). Training for stress exposure. In J.E. Driskell & E. Salas (Eds.), Stress and human performance (pp.223-256). Mahwah, NJ: Lawrence Erlbaum. Hancock, P.A. & Desmond, P.A. (2001). Stress, workload, and fatigue. Mahwah, NJ: Lawrence Erlbaum. Hancock, P.A. & Warm, J.S. (1989). A dynamic model of stress and sustained attention. Human Factors, 31, 519-537. Hancock, P.A., Szalma, J.L., & Weaver, J.L. (2003). The distortion of perceptual spacetime under stress. Unpublished white paper. Harris, D. (1985). A degradation methodology for maintenance tasks. Alexandria, VA. HQDA, MILPERCEN (DAPC-OPA-E). Hockey, G.R.J. (1983). Stress and fatigue in human performance. New York: Wiley. Hockey, G.R.J. (1986). Changes in operator efficiency as a function of environmental stress, fatigue and circadian rhythms. In K.R. Boff., L. Kaufman, and J.P. Thomas (Eds.), Handbook of perception and human performance (44, pp.1-49). New York: Wiley. Hockey, R. & Hamilton, P. (1983). The cognitive patterning of stress states. In G.R.J. Hockey (ed.), Stress and fatigue in human performance (pp.331-362). New York: John Wiley & Sons Ltd. Hogan, J. & Lesser, M. (1996). Selection of personnel for hazardous performance. In J.E. Driskell & E. Salas (Eds.), Stress and human performance (pp.223-256). Mahwah, NJ: Lawrence Erlbaum. Jongman, L. & Taatgen, N. A. (1999). An ACT-R model of individual differences in changes in adaptivity due to mental fatigue (pp. 246-251). In Proceedings of the twenty-first annual conference of the cognitive science society. Mahwah, NJ: Erlbaum. Kahneman, D. (1973). Attention and effort. Englewood Cliffs, NJ: Prentice Hall. Kieras, D. E. & Meyer, D. E. (1997). An overview of the EPIC architecture for cognition and performance with application to human-computer interaction. Human-Computer Interaction, 12, 391-438. Lazarus, R. S. (1966). Psychological stress and the coping process. New York: McGrawHill. Version 3 MURI-OPUS 16 Lazarus, R., & Folkman, S. (1984). Stress, appraisal, and coping. New York: Springer Verlag. Lebiere, C. (2001). A theory-based model of cognitive workload and its applications. In Proceedings of the 2001 Interservice/Industry Training, Simulation and Education Conference (I/ITSEC). Arlington, VA: NDIA Lebiere, C., Anderson, J.R., & Bothell, D. (2001). Multi-tasking and cognitive workload in an ACT-R model of simplified air traffic control task. In Proceedings of the 10th Conference on Computer-Generated Forces and Behavior Representation (pp.91-98), May 15-21. Norfolk, VA. Lebiere, C., Biefield, E., Archer, R., Archer, S., Allender, L., et al. (2002). IMPRINT/ACT-R: Integration of a task network modeling architecture with a cognitive architecture and its application to human error modeling. Matthews, G. (2001). Levels of transaction: A cognitive science framework for operator stress. In P.A. Hancock, & P.A. Desmond (Eds.), Stress, workload, and fatigue. Mahwah, NJ: Erlbaum. Matthews, G. & Campbell, S.E. (1998). Task induced stress and individual differences in coping. In Proceedings of the 42nd Annual Meeting of the Human Factors and Ergonomics Society (pp.821-825). Santa Monica, CA: Human Factors and Ergonomics Society. O’Brien, L.H., Simon, R., & Swaminathan, H. (1992). Development of the personnelbased system evaluation aid (PER-SEVAL) performance shaping functions. ARI Research Note 92-50, United States Army Research Institute for the Behavioral and Social Sciences. Pew, R.W., & Mavor, A.S. (1998). Modeling human and organizational behavior: Applications to military simulations. Washington DC: National Academy Press. Rasmussen, J. (1983). Skills, rules, and knowledge; Signals, signs, and other distinctions in human performance models. IEEE Transactions on Systems, Man, and Cybernetics, vol.SMC-13, 257-266. Sanders, A.F. (1983). Towards a model of stress and human performance. Acta Psychologica, 53, 61-97. Selye, H.A. (1956). The stress of life. New York: McGraw-Hill. Smith, C.A.& Lazarus, R.S. (1993). Appraisal components, core relational themes, and the emotions. Cognition and Emotion, 7, 233-269. Szalma, J.L., & Hancock, P.A.(2003). On mental resources and performance under stress. Unpublished white paper. Ritter, F.E., Avraamides, M.N., Councill, I., Quigley, K.S., Klein, L.C. et al. (2002). Modeling the effects of two behavior moderators in ACT-R. Paper presented at the Workshop on ACT-R Models of Human-System Interaction, January 10th. Mesa, AZ. Ward, P., Szalma, J.L., & Hancock, P.A. (2003). The stress-performance relationship and performance mediators: Evidence from a meta-analysis of noise and memory. Paper presented at the 2003 Human Factors and Ergonomics Society Conference, October, Denver, CO. Wickens C.D. (1980). The structure of attentional rersources. In R. Nickerson (Ed.). Attention and performance VIII (pp.239-257). Hillsdale, NJ: Erlbaum. Wickens, C.D. (1996). Designing for stress. In J.E. Driskell & E. Salas (Eds.), Stress and human performance (pp.223-256). Mahwah, NJ: Lawrence Erlbaum. Version 3 MURI-OPUS 17 Williams, J.M.G., Watts, F.N., McLeod, C. & Matthews, A. (1988). Cognitive psychology and emotional disorders. Chichester: Wiley. Version 3