1
Skill Acquisition and Training in Context
All learning from experience, all thinking, all inference, is transfer: there are only differences in degree.
C. K. Lyans (1914, p. 384)
When Asiana Airlines flight 214 crash-landed in fair weather at the San Francisco International Airport on July 6, 2013, fatally injuring three passengers and severely injuring many more, one question raised was how the flight crew could have allowed the flight speed to become dangerously slow in their final landing approach without reacting and correcting it. In fact, the U.S. National Transportation Safety Boardâs (2014) report on the crash identified pilot mismanagement of the descent as well as the complexity of the aircraftâs autoflight system as contributing factors in the incident. Among the Boardâs recommendations were improved training for the pilots on the autoflight system and increased time flying the aircraft manually during training.
In contrast to the apparent failure of the pilots, the skill with which the flight attendants carried out their duties to evacuate the aircraft to minimize casualties was praised in the news media. The cabin manager for the flight, Lee Yoon-hye, told journalists, âI wasnât really thinking, but my body started carrying out the steps needed for an evacuation. I was only thinking about rescuing the next passengerâ (quoted by Briggs, 2013). Candace Kolander, head of air safety, health, and security for the Association of Flight Attendants, commented,
She reacted automatically. When that happens your brain says: âOK, itâs time to do what Iâm trained to do.â And your body just does it because through training you build those motor skills, the motions youâre supposed to do, the voice commands youâre supposed to do ⌠You become a robot. Thatâs why I can open that (aircraft) door in my sleep because in training you get that adrenaline going and you do it the same way every year. So when it does happen you know what adrenaline feels like and you know you can do it.
(Quoted by Briggs, 2013)
These quotes capture many of the characteristics of highly practiced skills: They are automatized as a consequence of extensive practice in contexts similar to those under which the skills will ultimately have to be performed. Such incidents emphasize the importance of designing training programs to ensure that necessary skills are acquired.
Although the flight attendants operated under an emergency situation, skilled behavior is fundamental to virtually all human activities. Some skills, such as driving, reading, and typing, are so common that they can be taken for granted. Other skills, such as serving tennis balls, playing bridge, or baking bread, are more specialized, yet most of us can, with some practice, master them. Skills may have large perceptual (e.g., reading a medical image), cognitive (e.g., reading or remembering large amounts of information), or motor (e.g., typing or skiing) components. A basic question regarding skill acquisition is whether skills in perceptual, cognitive, and motor domains share common mechanisms. Some researchers emphasize commonalities across the different domains, whereas others point to differences (see, e.g., A. Johnson, 2013). Almost all skills, however, require coordinated processes of perception, cognition, and action. In acquiring skill we learn to select relevant information and link it to actions in a smooth, integrated fashion. Skill can thus be defined as goal-directed, well-organized behavior that is acquired through practice and performed with economy of effort.
One skill that illustrates how the processes of perception and cognition, as well as the ability to react and to interact with others, jointly determine performance is air traffic control (ATC). ATC can be considered a dominantly cognitive task (Corver & Aneziris, 2015; Vu, Kiken, Chiappe, Strybel, & Battiste, 2013). The air traffic controller must keep track of large amounts of information (weather conditions, the registration numbers and types of aircraft under the controlled space, as well as the speeds at which they are traveling, the positions of the aircraft in the air and on the runway, etc.) and make complex decisions. The controller must also communicate and coordinate actions with pilots and other controllers in the ATC system, as well as learn and follow rules but still be flexible when necessary. Even though ATC is predominantly a cognitive task, understanding how skill in the task develops and predicting who will excel at the task require a broad approach. Research in this field has shown that training for ATC should be perceptual (e.g., learning to make a quick scan across aircraft in the airspace), cognitive (e.g., developing decision-making skills), and social (e.g., focusing on communication and teamwork skills) in nature (Durso & Manning, 2008).
Although understanding and training complex skills such as ATC can be considered an essential goal of skills research, much of the research on skill acquisition has been carried out in laboratory settings. Laboratory settings offer convenience, but more importantly allow the researcher to control the environment so that operative variables can be identified. It can be argued that the basic processes involved in the execution of controlled laboratory tasks are required in more complex tasks (e.g., Raymond, Healy, & Bourne, 2012). However, it remains essential to question whether principles derived from studying skill in simple laboratory tasks are applicable to the more complex real-world tasks. Complex tasks, by definition, consist of a number of processes, and it is necessary to understand how these processes compete and are coordinated. Performing complex tasks will likely require processes of attention and cognitive control that are not required when component tasks are performed in isolation. Moreover, real-world task performance is affected by a wide range of variables and subject to fewer constraints than laboratory task performance (Beier & Oswald, 2012). The competing desires to impose experimental control and to consider a wide range of factors thus trade off in using laboratory versus real-world tasks.
Because of the need to balance experimental control with applicability to complex tasks, it is often necessary to use a range of tasks to study skill acquisition in any particular domain. A good example of how laboratory and more naturalistic tasks have been combined to understand skill is reading radiological images for medical diagnosis. Reading medical images is a skill that develops over many years (van der Gijp et al., 2014). Due to its importance in diagnosing serious medical conditions, this skill has received considerable investigation, much of which is discussed in Chapter 6. Diagnostic processes have been studied naturalistically by comparing the performance of medical students, residents, and practicing radiologists. To gain a fuller understanding of particular factors influencing diagnostic skill, such as prior knowledge of patient anatomy, more detailed analyses have been performed using constrained diagnostic tasks with people of varying levels of skill and training. Because diagnosis from medical images depends on basic skills such as visual inspection and decision on an interpretation (Krupinski, 2010), findings from laboratory studies of these basic abilities have also been applied to the specific problem of medical diagnosis to develop a better picture of how skill in this domain develops.
Historical Overview of Skills Research
Research relating to skill began in the early years of experimental psychology with the publication by Ebbinghaus (1885/1964) of a monograph on learning and memory that, although not directly concerned with skill acquisition, set the stage for the research on skill that would follow. In Ebbinghausâs time, the generally accepted view was that higher mental processes such as thinking and memory could not be studied experimentally. Given this prevailing atmosphere in early scientific psychology, the breadth and depth of Ebbinghausâs investigations are remarkable. Using himself as the subject, Ebbinghaus demonstrated beyond question that it in fact was possible to obtain objective, quantifiable measures of the mental processes underlying memory.
Ebbinghaus introduced a set of materials, consonant-vowel-consonant nonsense symbols, that provided him with many different items from which lists of varying length could be composed. He used serial recall, learning the lists to a criterion of one or two completely correct recalls in correct order. One fairly unsurprising finding resulting from Ebbinghausâs explorations was that the number of repetitions of a list required for learning increased with the length of the list.
More important, rather than relying on recall of the lists at a later time to measure retention, Ebbinghaus developed the method of relearning the list to a criterion of correct recall and measuring a savings score relative to the initial learning. This savings paradigm provides a measure of the benefit of prior learning when something is learned again at a later time, even when the original list could not be consciously recalled. Hilgard (1964) emphasizes the significance of this contribution in his introduction to a republication of Ebbinghausâs monograph, stating, âEbbinghaus took the ease of relearning something once knownâa fact so plausible that it must have been often observedâand made it a part of science by developing the quantitative saving score, in which the saving in relearning is scored as a per cent of the time (or trials) in original learningâ (p. viii). Formally,
where P is some measure of performance.
For Ebbinghaus, performance was usually the number of trials to learn series of nonsense syllables to a criterion of correct recall as a function of variables such as retention interval and number of successive days on which learning was repeated. He demonstrated that much forgetting occurred over the first hour after initial learning and that forgetting continued to occur in decreasing amounts as the retention interval increased over hours and days (Wixted & Carpenter, 2007). In one study, Ebbinghaus memorized stanzas of Byronâs Don Juan and found that savings increased over 3 successive days of relearning. Most remarkably, when he relearned the same stanzas 17 years later, long after they could no longer be recalled, those stanzas showed savings of approximately 20% compared to learning of stanzas that had not been learned previously (Verhave & van Hoorn, 1987).
One of Ebbinghausâs most theoretically important findings was that items could act as retrieval cues for other items in learned lists even when the cued and retrieved syllables were separated by other elements in the original list. He demonstrated this by showing savings for lists composed of the same syllables as the originally learned lists but with adjacent syllables in the new lists having been separated by 1, 2, 3, or 7 intervening syllables in the original lists. Assumptions about how associations are formed between elements that are relatively remote from each other distinguish many models of learning and memory (e.g., Cleeremans, 1993). The problem of how associations between items that occur together are formed, as well as remote associations linking items that are separated by others, is treated in more detail in Chapter 4.
Early Studies of Skill Acquisition
The advent of the study of skill acquisition, per se, can be dated to the work of Bryan and Harter (1897, 1899) on the acquisition of skill at telegraphy. Telegraphy was a major form of communication at the turn of the century, when the broadcast of news and other current events depended on the accurate transmission and reception of telegraphic messages. In telegraphy, a skill that can take up to 2.5 years to master, a sender must transcribe a written message into Morse code and rapidly tap a telegraph key, in taps of long and short durations, separated by pauses. The receiver of the code must re-transcribe the message, typically using a typewriter. Receiving high-speed Morse code depends on the (a) perceptual ability to parse the âditsâ and âdahsâ that make up the message and to group these symbols into conceptual units, (b) motor ability to quickly type the message, and (c) strategic ability to âcopy behindââthat is, to allow the typing of the message to lag behind the decoding of the message (Wisher, Sabol, & Kern, 1995). The speed with which messages are transmitted is limited by the skill of the operators at translating the messages and executing the required physical actions.
Bryan and Harterâs initial work focused on differences in performance as a function of level of experience. They found that (a) rates of receiving varied greatly across operators, (b) external disturbances had less impact on experts than on novices, and (c) variability in sending time decreased with increasing skill level. Bryan and Harter also examined the individual learning curves of several operators. Measures of operator performance were taken weekly, starting with the operatorsâ initial experiences as telegraphers and continuing for up to as many as 40 weeks (see Figure 1.1).
As shown in Figure 1.1, Bryan and Harter found that the sending rate improved more rapidly than the receiving rate but also reached asymptote sooner. Eventually, the receiving rate approached or even exceeded the sending rate, so that a skilled receiver could easily handle the most rapid incoming message. Moreover, the sending curves showed continuous improvement with practice up until the point that the asymptote was reached, whereas the receiving curves for some operators showed distinct plateaus, or periods during which there was relatively little change in performance. These plateaus led Bryan and Harter to characterize skill acquisition as the development of a hierarchy of habits. For telegraphic skill, this hierarchy was proposed to involve letters, words, and higher-language units. According to Bryan and Harter, receivers are concerned with identifying individual letters during their first few days as telegraphers. Over this period they begin to identify words, and it is words that then become the primary units for the next few months. Only after the letter and word âhabitsâ have been acquired does the operator fully benefit from the higher-level organization of the language. Thus, the plateaus represent periods where a lower-level habit is at asymptote but the next higher habit is not yet affecting performance. Bryan and Harter emphasized that automatizing lower-level habits in the hierarchy is necessary for skill to be acquired, due to the freeing up of attention, and that âin learning to interpret the telegraphic language, it is intense effort which educatesâ (1897, p. 50), a sentiment that is mirrored in contemporary studies of acquisition of expertise in other domains.
Just as Ebbinghausâs research set the stage for the investigation of skill by establishing paradigms for measuring learning and demonstrating fundamental learning phenomena, Bryan and Harterâs work provided a direct impetus for the systematic investigation of skilled performance. Their seminal papers can be directly linked to three major areas of investigation: (a) variability in performance, (b) whether the changes underlying the acquisition of expertise are qualitative or quantitative in nature, and (c) the nature and development of automaticity (T. D. Lee & Swinnen, 1993).
Figure 1.1 Sending and receiving curves for two telegraphers studied by Bryan and Harter.
The most visible effect of practice documented by Bryan and Harter, the learning curve, has achieved the status of a psychological law. Performance of most tasks seemingly improves continually, with the greatest changes occurring early in practice. This improvement with practice tends to follow a power function (see Figure 1.2):
Performance Time = a + bNâc,
where a and b are constants representing asymptotic performance and the difference between initial and asymptotic performance, respectively, N represents the number of practice trials, and c is the learning rate. Note that this implies that the logarithm of response time plotted as a function of the logarithm of the number of trials performed will be a straight line. Snoddy (1926) was the first to conclude that learning could be described as a linear function of the logarithms of time and trials and thus conformed to a power function. This relation is referred to as the power law of practice and has become a benchmark for models of skill acquisition (A. Newell & Rosenbloom, 1981). Although the power law provides a good description of group learning curves, it may not do so for the curves of individuals (Haider & Frensch, 2002; Heathcote, Brown, & Mewhort, 2000). Specifically, Heathcote et al. showed that for many...