6

Developing expertise

Stage models of expertise

There is a mathematical model of learning and expertise development that appears to cover a range of measures and a wide range of domains. This is the Power Law of Practice (Newell & Rosenbloom, 1981). Newell and Rosenbloom referenced a number of studies (mirror tracing, cigar manufacturing, learning to read inverted texts, scanning for targets in a display and so on), all of which showed the same learning pattern that could be displayed as a version of the one in Figure 6.1 which gives an abstract version of the graphs based on the law. In the graph on the left, as practice continues (x-axis), measures such as the number of errors or reaction times (RTs) drop but the rate at which this happens changes over time. So, for example, RTs might be very slow to begin (a in the figure) but speed up rapidly with relatively little practice (b), but this speed up slows so that a lot of practice is needed to make a small difference in RT (c). The same effect is shown in the right-hand graph, but this time the spacing in the x-axis is logarithmic. For example a might be 1 hour, b 10 hours, c 100 hours, d 1,000 hours and e 10,000 hours. The law is in the form RT = aP^−b + c, where

• RT = Trial completion time
• P = Trial number, starting from 1 (for exponential functions the P − 1 argument is used)
• a, b and c are constants

Other models of the acquisition of expertise include stages of development leading to expertise beginning with a stage where one’s knowledge is declarative and verbalisable and where general knowledge becomes more and more specialised. Fitts and Posner (1967) provided an early list of the phases of learning and expertise development and these have often been incorporated in various guises into other learning models (e.g., Anderson, 1982, 1983; Schneider & Shiffrin, 1977; Tenison & Anderson, 2015). These are listed in Information Box 6.1.

Given the nature of the Power Law of Practice and the general agreement about stages of expertise development, it has been unclear, according to Tenison and Anderson (2015), whether this speed-up pattern is due to continuous improvement based on a power function or if there are abrupt qualitative changes during learning corresponding to the stages of learning. In fact several researchers have found evidence for qualitative changes within the stages listed in Information Box 6.1 (Kim, 2015; Siegler, Thompson, & Opfer, 2009), and Tenison and Anderson have found evidence for different cognitive processes in three stages as well as evidence for a role for the power function within each stage that accounted for speed-up of learning within the stage.

Alexander’s model of domain learning

In Patricia Alexander’s model of domain learning (Alexander, 2003; Alexander, Jetton, & Kulikowich, 1995; Alexander, Murphy, Woods, Duhon, & Parker, 1997; Alexander, Sperl, Buehl, Fives, & Chiu, 2004) there is a different take on the stages which she calls acclimation, competence, and proficiency. At each stage there is an interplay between knowledge, strategic processing and interest, thereby capturing both individual differences among learners and addressing the difficulty of translating traditional models of skill acquisition such as ACT-R into everyday educational practice in academic domains. Acclimation is the initial stage where the novice’s knowledge is limited and fragmentary; the strategies employed tend to be surface level; and performance is maintained by relying on “situational interest” such as the strategies teachers might use to sustain students’ interest. Competence comes about through qualitative and quantitative changes in knowledge that is now principled and much more cohesive in nature; familiarity with typical problem types leads to a move from surface-level strategies to deeper processing strategies based more in underlying principles; at this stage there is likely to be a greater intrinsic individual interest in the domain. At the stage of proficiency (expertise), experts have a broad and deep knowledge base and can contribute new knowledge to the domain through problem finding; there is a high level of deep processing strategy use; and being experts, they show a high level of individual interest in their field.

Dreyfus’s model of expertise development

Other researchers have listed various stages people go through as they move from novice to expert. For example, Dreyfus (1997) and Dreyfus & Dreyfus (2005) proposes a five-stage model from novice at stage 1 to advanced beginner in stage 2, competent at stage 3, proficient at stage 4, and leading to expertise at stage 5. The stages in the acquisition of a procedural skill proposed by Anderson and his co-workers (Anderson, Fincham, & Douglas, 1997; Tenison & Anderson, 2015) have already been discussed and roughly correspond to Dreyfus’s first two stages. The remaining three stages are:

Stage 3 competence

Dreyfus and Dreyfus (2005) state:

At this stage the huge variety of subtly different situations could cause performance to be “nerve-racking and exhausting”. Competence comes from the necessity and ability to prioritise and the development of rules to allow the student to decide which plan to follow. Dreyfus and Dreyfus dwell at some length on the emotional impact of the learner’s performance at this stage.

Stage 4 proficient

Dreyfus and Dreyfus (2005) state:

Some of the description of this stage is much like that of the second associative stage described in Information Box 6.1.

Stage 5 expertise

By this stage the expert has developed a vast repertoire of discriminations that distinguish him or her from the proficient performer. Furthermore, apparently, “he or she also sees immediately how to achieve the goal.”

This is a strange assertion to make given that researchers such as Anderson have provided an explanation over many decades of why skills are not verbalisable. For example, Lovett and Anderson (2005) in their description of one of four features of production rules, state: “Production rules cannot be directly verbalized.”

Dreyfus and Dreyfus make a number of assertions that do not seem to be backed up by anything other than anecdotal evidence. They state for example that “a skill is never produced by interiorizing the rules that make up the theory of a domain” (p. 790). As a result it is difficult to assess their conclusions. Nevertheless some researchers have found the stages useful in describing where people stand on a continuum of expertise (Lyon, 2015; Ramsburg & Childress, 2012) and have argued that the framework could be potentially useful to assess career development (Hall-Ellis & Grealy, 2013). Field (2014) found it useful but limited and Peña (2010) did not find that it adequately explained the development of clinical skills, and doubted that expert clinicians work from intuition rather than reason.

Glaser’s change of agency for learning

Glaser (1996) has referred to three general stages in what he has termed a “change of agency” in the development of expert performance. The stages are termed external support, transitional and self-regulatory. The first stage is one where the novice receives support from parents, teachers, coaches, and so on, who help structure the environment for the novice to enable him or her to learn. The second stage is a transitional stage where the “scaffolding” provided by in the first stage is gradually withdrawn. The learner at this stage develops self-monitoring and self-regulatory skills and identifies the criteria for high levels of performance. In the final, self-regulatory stage the design of the learning environment is under the control of the learner. The learner might seek out the help of teachers or coaches or other sources of information when he or she identifies a gap or shortcoming in performance, but the learning and the regulation of performance is under the control of the learner. For example, Chi, Glaser and Rees (1983) found that physics experts were better than novices at assessing a problem’s difficulty. They also have knowledge about the relative difficulty of particular schemas.

The intermediate effect

In some domains at least there is evidence of an intermediate effect as one of the stages that learners go through on the way to becoming experts. Lesgold and colleagues (Lesgold, 1984; Lesgold et al., 1988) have found evidence for such an intermediate effect where there is a dip in performance on certain kinds of tasks such that novices outperform the intermediates despite the greater experience of the latter. Lesgold (1984) found that third- and fourth-year hospital residents performed less well at diagnosing X-ray films than either first- or second-year residents or experts. The same phenomenon was found by Patel and Groen (1991) (see also Boshuizen & Schmidt, 1990; Schmidt, Norman, & Boshuizen, 1990). Lesgold argues that this should not be surprising, as the same phenomenon can be found in the intermediate phase of language learning that children go through where they produce over-regularisations. For example, a child might start off by saying “I went …” but by absorbing the rules for the formation of English past tenses, the child goes through a stage of saying “I goed …” before learning to discriminate between regular and irregular past tenses. Patel and Groen (1991) and have argued that the intermediates have a lot of knowledge but that it is not yet well organised. This lack of coherent organisation makes it hard for them to encode current information or retrieve relevant information from long-term memory. Experts have a hierarchically organised schematic knowledge that allows them to pick out what is relevant and ignore what is irrelevant (Patel & Ramoni, 1997). Novices don’t know what is relevant and stick to the surface features of situations and base their representations on them. Intermediates, on the other hand, try to “process too much garbage”. As a result, the novices in Lesgold’s (1984) study rely on the surface features of the problem (which are usually diagnostic of the underlying features), the experts take the context into account, but the intermediates try to do so and fail.

Raufaste, Eyrolle and Mariné (1999) have argued that the intermediate effect can be explained by assuming that some kinds of knowledge are only weakly accessible. Much of an expert’s knowledge is implicit and experience adds this implicit knowledge to the structures originally acquired through the academic study of the domain. Furthermore they distinguish between two types of experts: “basic experts” and “super experts”. Basic experts are typical of the population (at least, of radiologists in France) whereas “super experts” refers to the very small number of world class experts. This distinction is similar to that between expert chess players and Grandmaster. There is, they argue, a qualitative difference between basic experts and super experts. If Lesgold had used basic experts in his studies, then the U-shaped curve produced by the performance of intermediates would have become a straight line.

There is another form of the intermediate effect that works the other way round in that intermediates, in some disciplines at least, can perform better at some tasks than experts. For example, Boshuizen and Schmidt (1992) included a table showing example studies where intermediates outperformed experts and others where experts outperformed intermediates (p. 156, table 1). They themselves found evidence for a decreased use of biomedical knowledge with increasing expertise. Their explanation was that there appears to be a three-stage model of expertise in medicine, although not the same as the one with which this chapter began. The stages involved the acquisition of biomedical knowledge, followed by practical experience, followed by the integration of theoretical and practical knowledge leading to “knowledge encapsulation”, where the biomedical knowledge is “tacit” and encapsulated in practical clinical knowledge.

Van de Wiel et al. (1993), using a similar methodology, failed to find the intermediate effect. There was instead a linear relationship with level of expertise in diagnosis and recall of case study details.

Gobet and Borg (2011) examined the performance of musculoskeletal physiotherapy students at different levels of expertise (novices, intermediates, experts) along with controls. They found a linear relationship with expertise when the participants were asked to make a diagnosis on the basis of a case study and also when they had to make high-level inferences about them (although the difference between experts and intermediates here was not significant). However, when they tested the recall of propositions from the studies, there was evidence of an inverted U-curve with intermediates outperforming experts, thus providing evidence for knowledge encapsulation.

Although basic experts have typically had less practice than super experts, an appeal to weakly associated memory and hence implicit or intuitive knowledge is probably not enough to explain a qualitative gap between basic experts and the super experts. There still seems to be some magic involved in producing the gap.

What distinguishes experts and novices

Although the various stages and transformations between one stage to another are often revealing, it can be useful to compare stages of expertise development at a fairly crude dichotomous level – typically by distinguishing novices from experts.

In studying simple puzzle problems psychologists assumed that what was learned there about the psychology of problem solving could be generalised to other real-life problem situations; indeed, this was one of the criticisms of traditional views of expertise development by Alexander and her colleagues (Alexander et al., 1995; Alexander et al., 1997; Alexander et al., 2004). However, most problems people face are ill-defined and usually make demands on at least some relevant prior experience and knowledge. The more knowledge a person has about a domain (however defined), the more that person is equipped to deal with complex problems in that domain. Thus one can expect a broad range of individual differences in problem solving ability in a particular domain.

As a result of research over several decades, a number of qualitative and quantitative differences in the performance of experts and novices have been revealed. Chi, Glaser and Farr (1988, pp. xvii–xx) have listed seven “key characteristics of experts’ performance”:

1. 1  Experts excel mainly in their own domain.
2. 2  Experts perceive large meaningful patterns in their domain.
3. 3  Experts are fast: they are faster than novices at performing the skills of their domain, and they quickly solve problems with little error.
4. 4  Experts have superior short-term and long-term memory.
5. 5  Experts see and represent a problem in their domain at a deeper (more principled) level than novices; novices tend to represent a problem at a superficial level.
6. 6  Experts spend a great deal of time analysing a problem qualitatively.
7. 7  Experts have strong self-monitoring skills.

These characteristics have emerged from the variety of ways in which expertise has been examined. Expertise has been examined in terms of differences in predispositions, such as intelligence, personality, thinking styles, motivation and so on. The assumption here is that there is more to the differences between experts and novices than can be accounted for by the knowledge they possess.

Another general characteristic shown up by these seven characteristics is the way experts represent a task. It is not just quantity of knowledge but the way the knowledge is structured that affects how an individual represents a problem. Experts may have developed reasoning or problem solving strategies and heuristics to help them deal with novel or difficult problems in their domain of knowledge. These include variations in the way resources or time are allocated to planning or representing a problem in the first place. Differences in knowledge can also affect cognitive processes such as perception and the role played by working memory. Expertise can change how one perceives a situation visually (at least in some domains). These dimensions are summarised in Figure 6.2.

Are experts smarter? Are there differences in abilities?

Is knowledge the only factor or the main factor that leads to expertise, or are there other factors (such as “ability”) to be taken into account? For over a century there have been many diverging explanations for exceptional performance in a particular domain. They have generally tended to take a stance somewhere along two dimensions: innate versus acquired ability, and domain-specific versus domain-general ability (Ericsson & Smith, 1991). Some explanations have included the role played by personality, motivation and thinking styles.

One specific ability that has long been assumed to play a part in exceptional performance is “intelligence” (the quotation marks are meant to represent the slipperiness of this concept). In other words one could ask the question, do you need to be intelligent to be an expert? (Which is essentially asking, is there a domain-general ability that leads to expertise in a chosen field?) Some studies have suggested that expert chess players also performed well in other fields such as chess journalism and languages (De Groot, 1965; Elo, 1978). On the other hand, some studies have found remarkably little correlation between intelligence and other measures such as subsequent occupation, social status, money earned and so on, despite a high correlation with success in school tests (Ceci, 1996; Sternberg, 1997a). Wolfgang Schneider, Körkel and Weinert (1989) found that children who were highly knowledgeable about football but who were low on measured IQ scores could nevertheless outperform other children with high IQ scores in reading comprehension, inferencing and memory tasks if those tasks were in the domain of football. Ceci and Likert (1986) compared two groups of racetrack goers, one of whom was expert at predicting what the odds would be on a horse at post time. The expert group, unlike the other non-expert group, used a complex set of seven interacting variables when computing the odds. Each bit of information (one variable) they were given about a real horse or a hypothetical one would change the way they considered the other bits of information (the other six variables). Despite the cognitive complexity of the task the correlation between the experts’ success at the racetrack and their IQ scores was −.07 – no correlation at all, in fact.

Swanson, O’Connor and Carter (1991) divided a group of schoolchildren into two subgroups based on their verbal protocols while engaged in a number of problem solving tasks. One subgroup was designated as having “gifted intelligence” because of the sophistication of the heuristics and strategies they employed. The subgroups were then compared with each other on measures of IQ, scholastic achievement, creativity and attribution (what people ascribe success or failure to). No substantial differences were found between the two groups. Swanson et al. concluded that IQ, among other measures, is not directly related to intelligence defined in terms of expert/novice representations. If a person shows “gifted intelligence” on a task this does not mean that that person will have a high IQ.

Despite some studies showing a stronger role for deliberate practice (DP) over intelligence, other studies have found different results although a simple correlation between intelligence and expertise is hard to find. One will obviously find a high correlation between the mathematical reasoning tests of IQ and success in mathematics. Bilalic´, McLeod and Gobet (2007) gave four subtests of the WISC-III intelligence test to 57 children with an average of 4 years’ chess playing experience. They then presented them with a chess test, a chess recall test similar to that given by De Groot (1978; see later in the chapter), and the Knight’s Row Task, where the participants had to move the knight from one corner to the other on the same row. They found a moderate positive correlation between chess skill and intelligence but the results were not that straightforward, as an elite subsample of chess players did not show a correlation between chess skill and intelligence – in fact there was a small negative association. For the best (young) players intelligence does not have an impact on chess skill. For the rest, intelligence played a small role but practice was the best predictor of skill.

So far there have been examples from specific domains. More recently, a very large longitudinal study in the United States examined students who had performed exceptionally well on the College Board Scholastic Aptitude Test (SAT) before age 13 over more than two decades (Lubinski, 2009; Lubinski & Benbow, 2006; Lubinski, Benbow, Webb, & Bleske-Rechek, 2006; Lubinski, Webb, Morelock, & Benbow, 2001; Wai, Lubinski, Benbow, & Steiger, 2010). “Exceptionally well” here means performing between the top 0.5% or 0.01% in mathematical reasoning and verbal reasoning. Lubinski and colleagues found that the test results at age 12 were a strong predictor of career success 20 to 25 years later, measured in terms of gaining a doctorate (over 50% of the top 0.01% – mainly in highly ranked institutions – compared to 1%, which is the base rate in the United States), publishing novels, earning tenure at a top university, generating patents and so on. Such large-scale longitudinal evidence suggests an effect of exceptional intelligence in gaining expertise and career success.

Is expertise due to talent or deliberate practice?

As we have seen the role of intelligence in expertise is somewhat equivocal. Alternatives might be a great deal of experience in a domain or some form of innate talent. In trying to account for where expertise comes from, Ericsson, Krampe and Tesch-Römer (1993) have stated that “the search for stable heritable characteristics that could predict or at least account for the superior performance of eminent individuals has been surprisingly unsuccessful.” Simon and Chase (1973) argued that expertise (in chess at least) was the result of 10 years’ practice, and this figure has been presumed to apply to a range of domains. Ericsson et al. (1993) refer to 10,000 hours’ practice over at least a decade to produce an expert.

Ericsson et al. (1993) argued that expert performance was a function of the amount of DP in a particular skill such as violin playing. Ericsson and Charness (1994, 1997) have argued that comparing experts and novices can only take us so far. A more useful and valid task is to examine those aspects of a person’s superior performance that are reproducible. That is, expertise can be examined and levels of expertise differentiated by looking at a set of representative tasks that experts do well under standardised conditions. De Groot’s study of middle games and the next move from random positions provides an example. Ericsson and Charness argue that superior performance is not best understood in terms of “incremental refinements of pre-existing capacities and processes” but that the mechanism that produces expertise is deliberate, individualised, intensive practice. This kind of individualised training, or “effortful adaptation” to the demands of a domain, allows experts to restructure their performance and acquire new methods and skills.

De Bruin et al. (2008) list a number of studies in a range of different areas (from sports to music to academic disciplines) where the 10-year rule seems to apply. However, van de Wiel and Van den Bossche (2013) found that the development of expertise among physicians was due to a combination of patient care and DP, and that the organisations for which physicians work should ensure that there is adequate opportunity for them to engage in intentional learning. “The interaction between learning-by-doing and learning-by-intention can be reinforced, helping physicians to adapt to the requirements of their dynamic working environment.” (van de Wiel & Van den Bossche, 2013, p. 154).

Gobet (2016) points out that chess is nearly unique in having a reliable and quantitative measure of expertise (the Elo rating). It is thus possible to compute how much variance is accounted for by DP. Three studies with adult players found correlations of .42, .48 and .54 between DP and skill (Charness, Tuffiash, Krampe, Reingold, & Vasyukova, 2005; Gobet & Campitelli, 2007). Thus, between 17.6% and 29.2% of variance in skill is accounted for by the amount of DP.

Grabner (2014) presents an overview of studies in chess skill examining the role of cognitive abilities and DP. His review shows up the inconsistency of findings but overall, although practice is very important, “chess expertise does not stand in isolation from intelligence” (p. 32).

It is likely to be the case that deliberate practice is more appropriate in some domains than in others. You can deliberately practice serves in tennis if that is an area that needs improvement, if appropriate, and it is clear where practice can play a part in areas such as music. In other areas, practice may simply involve just doing your job (although you could increase your declarative knowledge) in fields such as law or computer programming.

Does expertise cross domains?

Does becoming an expert in a domain help you in understanding or developing expertise in another? In a review of the literature, Frensch and Buchner (1999) have pointed out that there is little evidence for expertise in one domain “spreading” to another. Ericsson and Charness (1997) have also stated (although with specific reference to memory) that “experts acquire skill in memory to meet specific demands of encoding and accessibility in specific activities in a given domain. For this reason their skill is unlikely to transfer from one domain to another” (p. 16, emphasis added).

On the other hand, there can be skills developed in one domain which can be transferred to another where the skills required overlap to some extent. Schunn and Anderson (1999) tested the distinction between domain-expertise and task-expertise. Experts in the domain of the cognitive psychology of memory (with a mean of 68 publications), social science experts (mean of 58 publications – task experts) and psychology undergraduates were given the task of testing two theories of memory concerning the effects of massed versus distributed practice. As they designed the experiment they were asked to think aloud. All the experts mentioned the theories as they designed the experiment whereas a minority of students referred to them – nor did the students refer often to the theories when trying to interpret the results. Domain experts designed relatively complicated experiments manipulating several variables whereas task experts designed simple ones keeping the variables under control. The complexity of the experiments designed by the students was somewhere in between. Schunn and Anderson claim that, at least in this domain, there are shared general transferable skills that can be learned more or less independently of context.

Nevertheless, we need some way of explaining why one person can be outstanding in a field whereas someone else with the same length of experience is not. Factors that have been used to explain exceptional performance are personality and thinking styles. People vary. Some are better at doing some things than others. Gardner (1983) has argued that there are multiple intelligences which can explain exceptional performance by individuals in different domains. In this view, exceptional performance or expertise comes about when an individual’s particular intelligence profile suits the demands of that particular domain. Furthermore, people differ in their thinking styles. While some prefer to look at the overall picture in a task or domain, others are happier examining the details. While some are very good at carrying out procedures, others prefer to think up these procedures in the first place (Sternberg, 1997b).

Since people vary in their experience, predispositions and thinking styles, it is possible to devise tests in which person A will perform well and person B will perform poorly and vice versa (Sternberg, 1998). A might do well in a test of gardening and poorly in an IQ test; B may perform well in the IQ test but poorly in the test of gardening. In this scenario measures of intelligence such as IQ tests are really measures of achievement. Sternberg has therefore argued that intelligence and developing expertise are essentially the same thing and that the intelligence literature should be a subset of the expertise literature.

Nevertheless, knowledge differences are not the only measures of individual differences in expertise. There are also differences in people’s ability to gain and exploit knowledge in the first place. The wide variety of interacting variables involved in skilled performance gives rise to individual differences in performance. Sternberg and Frensch have pointed out the same thing, although their argument is based on a Sternberg’s own model of intelligent performance. They state:

To sum up this section: although there are generally poor correlations between exceptional performance and measures of ability or personality, there is a range of factors – including innate ones – that can lead to expertise and exceptional performance. These factors also help account for wide individual differences in performance on the same task despite the same amount of experience.

Cognitive processes in expertise

In a well-known study of expert/novice differences in categorisation, Chi, Feltovich and Glaser (1981) gave novice and expert physicists cards with the text and a diagram of a single elementary physics problem on each and asked them to categorise them. They found that novices tended to categorise them in terms of their surface features, such as whether they involved pulleys or ramps. Experts classified them according to the deep structure, that is, according to the laws of Newtonian mechanics they involved.

Chess expertise

The usefulness of chess as a domain for understanding the development of expertise and expert-novice differences has already been mentioned. Early influential studies of the cognitive processes involved in chess expertise were carried out by De Groot (1965, 1966). A Master chess player himself, De Groot was interested in finding out how chess players of different ranks planned their moves. He showed five Grandmasters and five expert players a number of middle games in chess and asked them to choose the next move while thinking aloud at the same time. He found that the Grandmasters made better moves than the experts (as judged by independent raters) and yet the former did not seem to consider more moves nor search any deeper than the experts. De Groot argued that this was because the Grandmasters had a greater store of games they had studied and played and of different board positions. In other words the major difference between novices and experts seemed to be in the knowledge they possessed. In another study he briefly presented board positions to the subjects and asked them to reconstruct them from memory. Masters could reconstruct the board positions correctly 91% of the time on average. Less expert players managed only 41% accuracy. De Groot argued that the Masters were encoding larger configurations of pieces than the experts.

Chase and Simon (1973) hypothesised that experts and novice chess players differed in the size of the “chunks” they could encode at any one time. They asked chess players to reconstruct configurations of chess pieces on chessboards with the model still in view. They measured the number of glances players took and the number of pieces placed after each glance. The best player managed to encode 2.5 pieces per glance and used shorter glances than the weakest player who managed only 1.9 pieces per glance. The expert player was therefore encoding more information per glance than the weaker players. There is a potential argument that the Masters had a better memory than the experts, so this was tested by presenting random chess configurations (a task attributed to De Groot (1946/1965) but which did not appear in that publication – it was performed as part of his PhD and followed up by colleagues in his laboratory). Chase and Simon (1973) report the results thus:

Similar findings have been noted in other domains. Waters, Underwood and Findlay (1997) found that this same kind of perceptual chunking occurred in sight reading from musical scores. In one experiment they got eight full-time music students, eight psychology students who had passed a music exam and eight non-musicians to compare two visually presented musical sequences. The experienced musicians needed fewer and shorter glances to encode groups of notes. Adelson (1981) found that expert programmers could recall more lines of code than novices and had larger chunk sizes for encoding information.

The role of perception and conception in skilled performance

According to De Groot and Chase and Simon, perception is the key to chess skill. However, this may be putting the cart before the horse to some extent. The ability to recognise perceptual patterns and to categorise problems or situations appropriately is the result of expertise. You can’t teach people to categorise problems unless they already have the requisite knowledge of principles – the conceptual knowledge. You can’t chunk stuff perceptually without the experience and concepts to do it with.

Conceptual chunking was evidenced in a study by Cooke, Atlas, Lane and Berger (1991). Meaningful board configurations were presented to chess players. A verbal description of the configurations either preceded or followed the visual presentation of the chessboard. Where a verbal description preceded the visual presentation the performance of the experts was enhanced. This suggests that higher-level (conceptual) information prepared them for the pattern recognition (perceptual) task.

Egan and Schwartz (1979) repeated the “traditional” expert-novice memory task for meaningful and meaningless displays. The domain this time was electronic circuit drawings. In one condition experts tried to recall drawings of randomly placed electronic circuit symbols in terms of functionally related units and were faster than the novices on the task. Egan and Schwartz argued that there was more of a top-down process taking place than a perceptual chunking hypothesis could account for. It was not so much perceptual chunking that was taking place but conceptual chunking. That is, higher-level concepts were being used to govern perceptual chunking of the display.

More recent research on perceptual processes in expertise development has made increasing use of eye tracking and functional magnetic resonance imaging (fMRI) scanning to assess what novices and experts attend to. For example, Jarodzka, Scheiter, Gerjets and van Gog (2010) examined the cognitive processes of experts and novices in identifying the dynamic movements of fish with a view to discovering how conceptual and perceptual processes interacted. Experts used their conceptual knowledge to attend to relevant features of the movements of the fish as well as to features that would allow them to identify the fish species. Perhaps unexpectedly, experts also showed a greater diversity of gaze patterns possibly reflecting individual case-based knowledge rather than the shared generic knowledge one might expect of experts in a particular domain.

Wong and Gauthier (2012) examined the role of expertise in reading musical notation. Experts in this area were those who had been reading music for an average of over 13 years. Novices had 0.41 years of music reading experience. Wong & Gauthier were interested in the effects of perceptual crowding, which is when visual processing of a target is disrupted by similar and closely packed distractors – which is often what one finds in music notation. They found that experts performed significantly better than novices when the crowding elements were musical notes or the lines of the musical staff. When novices and experts were compared using letters rather than notes (e.g., b, d, p, q, h), there was no significant difference between music experts and novices, so the effects of crowding were reduced only for musical stimuli for those with extensive music reading experience. Also in relation to music reading, Wong, Peng, Fratus, Woodman and Gauthier (2014) found that, contrary to previous findings, the early stages of visual processing in the primary visual cortex rather than later stages of the visual hierarchy can be selective for musical notation. That is, experts in music reading are perceptual experts and “perceptual expertise can penetrate and influence neural activity as early as 40–60ms post stimulus onset, and the C1 [an early component of the visual system] is thus the earliest perceptual expertise marker ever reported” (Wong et al., 2014, p. 16).

Johnson and Mervis (1997) performed a categorisation study on experts on songbirds. They also found that conceptual knowledge interacted with perception.

Experts’ categories can sometimes be less distinct than those of novices. Murphy and Wright (1984) asked experts and novices to list attributes of three childhood disorders. Experts listed more features for each disorder than novices and agreed with each other more but there were fuzzier boundaries between the categories of disorder than those produced by novices. An explanation for the difference is that novices learn about prototypical cases during training but that experts have had experience of exceptions to the prototype and hence developed a broader view of the category.

Similar results concerning differences in perceptual processes between novices and experts have been found in a very wide variety of other domains from computer programming (McKeithen, Reitman, Rueter, & Hirtle, 1981) to figure skating (Deakin & Allard, 1991) to shoplifting (Carmel-Gilfilen, 2013) to tree identification (Shipman & Boster, 2008) to wine tasting (Ballester, Patris, Symoneaux, & Valentin, 2008).

The role of memory in expert performance

Chase and Simon’s original studies assumed a short-term working memory capacity of around seven chunks, as did most the studies of expert memory in the ’70s and ’80s. More recently, however, a number of studies and papers have caused a re-assessment of those early findings. For example, Gobet and Simon (1996) found that expert chess players could recall the positions of more pieces than the original theory predicted. In one experiment, a Master chess player was able to increase the number of chessboards he could reproduce to nine with 70% accuracy with around 160 pieces correctly positioned. Gobet and Simon suggest that experts in a particular domain can use long-term memory templates to encode and store information.

Ericsson and Kintsch (1995) provide an explanation for how experts and skilled performers can manage a tenfold increase in performance tests of short-term memory. They cite a number of studies in the memory and expertise literature that do not seem to fit well with the notion of a limited capacity working memory limited to around only seven items. Ericsson and Polson (1988) describe a well-known case of a waiter (JC) who could memorise long and complex dinner orders. He used a mnemonic strategy that allowed him to retrieve relevant information later. Furthermore, his strategy could transfer to categories other than food orders, so the strategy was not domain-specific.

Medical diagnosis also presents a problem for a limited capacity short-term working memory. Many symptoms and facts have to be maintained in memory in a form that can be retrieved readily until a diagnosis is made. Ericsson and Kintsch therefore propose a long-term working memory associated with skilled memory performance and expertise. Experts are able to store information in long-term memory rather than maintaining it in short-term memory. In order to be able to do this and to do it quickly three criteria have to be met:

1. 1  The expert has to have an extensive knowledge of the relevant information needed;
2. 2  The activity the expert is engaged in must be highly familiar otherwise it would not be possible to predict what information will be needed at retrieval;
3. 3  The information encoded has to be associated with an appropriate set of retrieval cues that together act as a retrieval structure. When the retrieval cues are activated later the original conditions of encoding are partially reinstated which in turn leads to the retrieval of the relevant information from long-term memory.

Flexibility in thinking

Frensch and Sternberg (1989) argued that, due to the size of the expert’s knowledge base, the expert’s knowledge organisation rooted in abstract principles rather than surface features of problems, and due to the assumption that the expert’s knowledge is proceduralised, there is reason to believe that experts’ performance may be inflexible in certain circumstances; that is, they may be unable to change their “mode or direction of thinking”. With regard to this, Taatgen, Huss, Dickison and Anderson (2008, p. 548) define flexibility thus:

In sum, there is an argument to be made that knowledge might prevent the expert from seeing an entirely new way of doing or producing something novel; in other words expert knowledge might hinder creativity. In Gestalt terms, too much reproductive thinking can get in the way of productive thinking (Wertheimer, 1959). Automatisation leads to effortless performance but a concomitant lack of control since it is fast, parallel and reliant on mainly unconscious processes (Anderson, 1983; Anderson & Lebiere, 1998; Schneider & Shiffrin, 1977; Shiffrin & Schneider, 1977). Thus a situation that triggers an automatic response may actually be a “garden path” problem and an unusual response is actually required.

However, there would appear to be something wrong with a conception of expertise where learned procedures, automatisation, compiled knowledge, schematisation – call it what you will – lead to degraded performance. Experts wouldn’t be experts unless they could solve problems flexibly. Hatano and Inagaki (1986) have distinguished between “routine expertise”, which refers to the schema-based knowledge experts use to solve standard familiar problems efficiently, and “adaptive expertise” that allows experts to use their knowledge flexibly by allowing them to find ad-hoc solutions to non-standard unfamiliar problems. In fact, there is an argument that the chunking process allied to proceduralisation leads to a greater range of behaviours on the part of the expert than on the part of the non-expert. The expert therefore has a greater range of problem solving methods that allow him to be flexible when faced with uncommon problems. For example, Patel, Glaser and Arocha (2000) argue that

According to Lesgold et al. (1988) expert radiologists flexibly change their representations when new problem features manifest themselves. For example, Feltovich, Johnson, Moller and Swanson (1984) gave expert and novice clinicians clinical cases to diagnose that had a “garden path” structure. That is, the pattern of symptoms indicated a “classical” (but wrong) disease. It was the novices who misdiagnosed the disease rather than the experts with their supposed abstracted-out schema for patterns of disease. Experts were more likely to reach a correct diagnosis after consultation with the patient. Feltovich, Spiro and Coulson (1997) argue that the very fact that experts have a large, well organised and highly differentiated set of schemas means that they are more sensitive to things that don’t fit. When that happens the expert is more likely to engage in more extensive search than the novice.

Information Box 6.2 shows a protocol from an expert examining an X-ray of a patient who had had a portion of a lung removed a decade earlier. As a result the slide seemed to show a chronic collapsed lung. An effect of the removed portion was that the internal organs had moved around.

The protocol in the table shows the expert testing the “collapsed lung” schema and finding that there are indications that there are other features that don’t quite fit in with that schema. He switches to a lobectomy schema which, in the final part of the protocol, he also checks. From their work, Lesgold et al. (1988) have suggested that the behaviour of the expert radiologist conforms to the following general pattern:

According to Voss, Greene, Post and Penner (1983), experts’ knowledge is flexible because information can be interpreted in terms of the knowledge structures the experts have developed and new information can be assimilated into appropriate structures. Similarly, Chi et al. (1983) stated that experts have both more schemas and more specialised ones than novices and this allows them to find a better fit to the task in hand. Experts’ extensive knowledge and categorising ability may lead to expert intuition.

Some potential side effects of expertise …

Ottati, Price, Wilson and Sumaktoyo (2015) found evidence that relative experts in a field tend to be more dogmatic in that they tend to process information in ways that reinforce their prior expectations. This is termed the “earned dogmatism hypothesis”. Novices in a particular field are by definition unfamiliar with it and so social norms require that they “listen and learn” in an open-minded fashion. On the other hand, “The expert possess [sic] extensive knowledge, and therefore is entitled to adopt a more dogmatic or forceful orientation” (p. 132).

Fisher and Keil (2015) looked at the relationship between a person’s assessment of their expertise and their ability to explain topics within their domain of expertise. Where that relationship was weak there was evidence of the illusion of explanatory depth (IOED); however, this depended on the nature of the expertise – whether it was passive expertise based on their position in their culture or formal expertise based on directed study in a particular domain at the college masters level. They argue “that both insight and illusion into one’s explanatory competence can co-exist and that they occur in systematic ways related to the kind of expertise involved” (p. 1). Thus, according to them, highly educated people tend not to show IOED for passive expertise but will exhibit it in their specialist domain. People with formal expertise “exhibit meta-forgetfulness within their domain of knowledge, neglecting the rate at which deliberately learned information decays from memory” (p. 17).

We should perhaps be cautious about what should be taken from these results from Fisher and Keil (2015) and Ottati et al. (2015). They are not referring to expertise in the sense of 10 years’ experience or 10,000 hours’ practice. They are “relative experts” so they know a bit more than their peers, so the word “expertise” here is used rather loosely.

Summary

1.  1 There have been various models of expertise development, most of which propose a series of stages.
2.  2 The Power Law of Practice (or Learning) shows a particular curve such that improvement on some performance is high at the beginning but slows down with increasing practice.
3.  3 Fitts and Posner (1967) suggested a three-stage model of skill development taken up by various researchers since. These are a resource intensive cognitive stage reliant on mainly declarative learning; an associative stage where performance is less reliant; and an autonomous stage where performance is highly proficient and no longer relies on conscious control.
4.  4 Alexander’s (e.g., 2003) Model of Domain Learning also has three stages – acclimation, competence and proficiency – involving a changing interaction between interest, knowledge and strategic processing.
5.  5 Dreyfus (e.g., 1997) has developed a five-stage model from novice to advanced beginner to competent to proficient to expertise. He argues that rule-based systems cannot account for the kinds of intuition that experts use.
6.  6 Glaser presents a three-stage model concentrating on change of agency starting with the help of teachers and coaches at the stage of external support, through transitional where some support is gradually withdrawn, to self-regulatory where learning is under the learner’s control.
7.  7 Some researchers, particularly in medicine, have documented an intermediate effect where those with some experience in a particular field such as radiology perform better at some tasks than experts or perform worse in some tasks than novices. Intermediates are focussing on a lot of detail and so outperform experts on recall of propositions but it also means that they “process too much garbage” and so novices can outperform the intermediates.
8.  8 Experts tend to be experts in one domain. Expertise does not seep into another domain unless the two domains share a set of skills. Thus there can be “content” knowledge specific to a domain and “task” knowledge that can be shared with closely related domains.
9.  9 Some studies have shown little correlation between expertise and IQ. These tend to show a superiority of deliberate practice over intelligence in performance on particular domains. However, longitudinal studies of highly intelligent people show a strong effect of intelligence related to successful careers in particular domains.
10. 10 Differences in thinking styles or intelligence profiles (e.g., Gardner’s theory of multiple intelligences) can lead to different people becoming expert in different domains where the domain suits their intelligence profile.
11. 11 Ericsson and Charness (1994) have argued that it is more profitable to examine what it is in a domain that experts do well. Expertise involves effortful adaptation to the demands of a domain.
12. 12 Automatisation ought to make expert performance “rigid” since routine, well-practised procedures are no longer accessible to consciousness. While automaticity can indeed lead to errors in certain circumstances, experts would not be experts if they could not use knowledge flexibly. The paradox can be overcome if one assumes routine expertise for dealing with typical problems and adaptive expertise for dealing with novel problems. Experts have schemas and strategies that cover exceptions as well as typical cases.
13. 13 Studies of chess expertise have underpinned many subsequent theorising about expert–novice differences. De Groot showed that Grandmasters differed from experts in that they could encode larger configurations of chess patterns than less proficient players. Chase and Simon also found differences in the “perceptual chunking” that expert chess players could manage compared to others. The same perceptual outcomes of expertise have been found in other domains.
14. 14 Consequences of expertise in many domains include:
    • Fast categorisation processes: experts categorise problems differently from novices;
    • Perceptual chunking; experts can “chunk” larger configurations of elements than novices;
    • Long-term working memory; experts have developed strategies within their domain of expertise for using long-term memory for tasks that novices would rely on limited capacity short-term memory to deal with.
15. 15 Some social psychologists have pointed out that there are potential downsides to expertise. Experts see themselves as being entitled to adopt a dogmatic position when it comes to their area of expertise. Others point out that experts may suffer from the “illusion of explanatory depth” as they may not have realised how much they have forgotten.
16. 16 Many results of studies suffer from an inconsistent view of what counts as expertise. There may be people who happen to know more than those around them, there may be “basic experts” and there may be “super experts”. Results may vary depending on what kinds of expertise is being tested.