Campitelli, G., & Gobet, F. (in press).

The role of practice in chess: A longitudinal

study. Learning and Individual Differences.

The Role of Practice in Chess: A Longitudinal Study

Guillermo Campitelli and Fernand Gobet*

Centre for the Study of Expertise
Centre for Cognition and Neuroimaging
Brunel University
Uxbridge, Middlesex, UB8 3PH
United Kingdom

*Corresponding author
Email: [email protected]
Phone: +44 (1895) 265484
Fax:

+44 (1895) 237573

Running head: The role of practice in chess: A longitudinal study

Abstract

We investigated the role of practice in the acquisition of chess expertise by submitting

a questionnaire to 104 players of different skill levels. Players had to report their
chess rating, the number of hours of individual and group practice, their use of
different learning resources and activities, and whether they had been trained by a
coach. The use of archival data enabled us to track the rating of some of the players
throughout their career. We found that there was a strong correlation between chess
skill and number of hours of practice. Moreover, group practice was a better predictor
of high-level performance than individual practice. We also found that masters had a
higher chess rating than expert players after only three years of serious dedication to
chess, although there were no differences in the number of hours of practice. The
difference that may explain the variation in rating is that masters start practising at an
earlier age than experts. Finally, we found that activities such as reading books and
using computer software (game databases, but not playing programs) were important
for the development of high-level performance. Together with previous data and
theories of expert performance, our results indicate limits in the deliberate practice
framework and make suggestions on how best to carry out learning in chess and in
other fields.

Keywords:

chess, chunking, coaching, expertise, individual differences, learning,

variability, practice

The Role of Practice in Chess: A Longitudinal Study

Learning high-level cognitive skills such as those of senior scientists, virtuosic
composers, and chess masters typically requires one to acquire a vast and meaningful
knowledge base and engage in situations in which this knowledge base is used. For
example, senior scientists should acquire knowledge of the results, theories, and
methodologies related to their field of research by reading scientific journals;
virtuosic composers should acquire knowledge related to previous compositions with
their different styles by reading scores and listening to masterpieces; and chess
masters should acquire knowledge of previous master games by replaying them.
Moreover, they all have to use their knowledge in performance situations. For
example, senior scientists have to design experiments, carry them out, analyse the
results, and write reports; virtuosic composers have to create music; and chess masters
have to play games. Participating in these situations reorganises the knowledge base
and incorporates new information to it.
Research into the acquisition of high-level cognitive abilities has raised a
number of issues that are still waiting for a definite answer. Probably the first and
most debated of these issues is whether the acquisition of high-level skills is only a
matter of the amount of dedication to the domain of expertise or whether innate
factors are involved as well. This debate has two extreme views: the first states that
talent is the main factor (Galton, 1869/1979), and the other maintains that all skills are
acquired through the interaction with the environment (e.g., Watson 1924). Ericsson,
Krampe and Tesch-Romer (1993) have revived the latter view with their deliberate
practice framework, which postulates that not every type of practice leads to expert
performance, but only practice that is performed with the deliberate purpose of
improving ones own skill. Deliberate practice consists of activities deliberately

designed to improve performance, which are typically effortful and not enjoyable.
These activities cannot be extended throughout long periods and must therefore be
limited to a few hours a day. Moreover, Charness, Krampe and Mayr (1996),
Charness, Tuffiash, Krampe, Reingold and Vasyukova (2005), and Ericsson et al.
(1993) identified individual practice as critical to the development of expertise. On
the other hand, Gobet and Campitelli (2007) found that group practice is also relevant
to acquire chess expertise.
Assuming that practice is an essential aspect of the acquisition of high-level
cognitive skills, a second issue arises: do all individuals benefit to the same extent
from practice? This issue was investigated nearly 100 years ago by Thorndike (1908),
who suggested a relation between inter-individual variability in performance and the
nature of the task. He suggested that if inter-individual variability in performance
increases with increasing practice in a particular domain, then the task performed is
one that requires factors not related to practice (i.e., innate factors). On the other hand,
if inter-individual variability in performance decreases with increasing practice, then
this is a sign that practice is the main factor to perform the task. Ackermans (1987)
literature review showed that, in most studies, inter-individual variability in fact
decreased, a pattern of results that supports the role of practice.
A third issue revolves around the type of activities that are useful for acquiring
high-level cognitive skills. In particular, is it important first to engage in activities
leading to the creation of a vast internal knowledge base, and only then to participate
in performance situations? Or, alternatively, should the participation in performance
situations start early on in the learning process? Perhaps surprisingly, Ericsson et al.
(1993) excluded activities in a performance situation from their definition of
deliberate practice. In a study with chess players, Charness, Krampe and Mayr (1996)

followed Ericsson et al.s (1993) decision and suggested that individual practice is
more related to deliberate practice than group practice. (Tournament games were
considered as group practice in Charness et al.s definition.).
Research into expertise has enriched practice in education, for example
through the concepts of adaptive expertise, pattern recognition, chunking, and
selective search (Anderson, Reder & Simon, 2000; Baxter & Glaser, 1998; Gobet,
2005; Gobet & Wood, 1999; Hatano & Inagaki, 1986; Simon, 1980). Similarly,
understanding the three issues introduced above would have strong implications for
education. For example, it would help teachers and coaches to choose appropriate
teaching and training activities.
In this paper, we address these three issues by studying practice behaviour in
chessplayers. Chess has already considerably enriched our understanding of problem
solving (e.g., Campitelli & Gobet, 2004; Charness, 1981; Gobet, 1998a), imagery
(e.g., Campitelli & Gobet, 2005; Saariluoma, 1991), perception (Charness, Reingold,
Pomplun, & Stampe, 2001; De Groot & Gobet, 1996), memory (e.g., Chase & Simon,
1973; Cooke, Atlas, Lane & Berger, 1993; Gobet & Simon, 1996a, b) and other
psychological phenomena (see Gobet, de Voogt & Retschitzki, 2004, for a review).
Chess has several advantages that make it a powerful task for studying cognitive
processes and learning (see Gobet, 1998b). For the purpose of this study, three of
these advantages are essential. First, it is a complex game that requires many years of
practice to attain high levels. Second, the existence of an international rating scale
allows researchers to know the level of expertise of their participants with precision.
Third, the existence of archival data of chess players ratings makes it possible to
track their level of expertise throughout their careers.

A large sample of Argentinian players had to fill out a questionnaire similar to

that used by Charness et al. (1996), as well as the Edinburgh Handedness Inventory
(Oldfield, 1971; the handedness data are discussed in Gobet & Campitelli, 2007,
which also addresses the question of starting age). The results allowed us to
systematically address the three issues identified in the introduction. First, we
addressed the hypothesis that individuals current performance is directly related to
the amount of deliberate practice; we refined this question by studying individual and
group practice separately. Second, we addressed the role of inter-individual variability
in the acquisition of chess expertise, again considering both individual and group
practice. Third, we examined the detail of the practice activities used by the players,
and their impact on skill acquisition.
Method
Participants
104 chessplayers (101 males and 3 females) filled in a three-section
questionnaire. They were recruited in several ways. First, the questionnaire was left
visible at the reception desk in the Crculo de Ajedrez Torre Blanca, one of the most
important chess clubs in Buenos Aires (Argentina). In addition, posters asking for
volunteers were also put on the notice board of the club. Finally, GC went to several
tournaments, both in the Crculo de Ajedrez Torre Blanca and other chess clubs in
Buenos Aires, and distributed the questionnaires to the participating players. The
participants included three grandmasters (mean age = 31 years, SD = 3.5), 10
international masters (M = 29.1, SD = 10.7), 13 FIDE1 masters (M = 27.1, SD = 8.9),
39 untitled players with international rating (M = 30.2, SD = 13.9), and 39 players
without international rating (M = 33.2, SD = 17.8). The mean age of the sample was
30.8 (range: from 10 to 78 years, median = 28 years, SD = 14.5). Since not all players

had international rating, we used the national rating in order to measure chess skill in
some analyses. The two ratings are closely related: in our sample, for the 65 players
with both international and national ratings, the correlation between the two scales
was .89. The range of the sample was 983 points (from 1490 to 2473), with a mean of
1991 and a standard deviation of 221. Since the Elo rating has a normal distribution
with a theoretical SD of 200, our sample had a range of nearly 5 SD.
Materials
The questionnaire was divided into three sections. The first section (see
appendix 1 for an English translation) consisted of questions about date of birth, age,
profession, international rating, national rating, speed chess rating (rating of the
Crculo de Ajedrez Torre Blanca),2 chess title, chess category, age at starting to play
chess, age at starting to play chess seriously, age at joining a chess club (club age),
presence of a coach (0,1) at any time during the career, number of chess books owned,
playing speed chess (0,1), number of speed games played, and type of training
(blindfold chess, reading games without seeing the board [henceforth, blindfold
reading], use of chess databases, and use of chess programs). The second section
contained a grid in which participants had to fill out the number of hours per week
they spent studying chess alone at each age (henceforth, individual practice). They
also had to fill out a second row with the number of hours per week they spent
studying or practising with other chessplayers, including tournament games
(henceforth, group practice).3 In both cases the participants were asked to estimate the
mean number of hours per week they spent studying or practising each year. We
estimated the number of hours studied per year by multiplying the figures reported by
52, and then we calculated the sum of the total hours spent with individual and group
practice in their entire chess career. Not all players answered all questions, with the

result that the number of data points varies across our measures. The third section
contained the Edinburgh Handedness Inventory (Oldfield, 1971). As the results of the
latter were already presented in a previous paper (Gobet & Campitelli, 2007), we do
not present these data here.
Results
We present the results according to the three issues developed in the
introduction. First, we consider the results related to the importance of practice in
achieving high levels of expertise in cognitive abilities. Second, we expound the
results related to inter-individual variability in performance. Third, we discuss data
related to the relative role of different types of practice.
The Role of Practice
We performed correlations between national rating and the total cumulative
number of hours of individual practice on the one hand, and of group practice on the
other hand. The correlation between current national rating and log cumulative
individual practice was .42 (p < .0001; calculated with 90 participants), and the
correlation between current national rating and log cumulative group practice was .54
(p < .0001; calculated with 89 participants).4 These statistically significant
correlations between the current skill level and two different measures of practice
show that the amount of previous practice has a very important role in the current skill
level.
Related interesting research questions include: How many hours per week do
the players spend practising, on average? Is there a difference between the stronger
players and the weaker players? And, do the players carry out more group practice or
individual practice? In order to investigate these issues, we performed a 2x3 ANOVA
with type of practice (group and individual) as a within-subjects independent variable,

chess level (masters, experts, intermediates) as a between-subjects independent

variable, and hours of practice per week as dependent variable. The masters were 17
players with one of the FIDE titles (that is, grandmasters, international masters, and
FIDE masters); the experts were 35 players with international rating but without title;
and the intermediate players were 31 players without international rating.
In order to estimate the mean hours of practice per week, we followed two
procedures and we report both results. In the first one, we used a range from 4 years
of age until the current age of the players. In the second one, the first data point was
when the players reported starting playing chess. The first procedure takes age into
account, while the second does not. The two procedures shed a slightly different light
on the role of practice.
The results were similar for both measures. For the first procedure (i.e.,
calculation from 4 years of age), there was a main effect of type of practice, F(1, 80)
= 27.54, p < .001, a main effect of chess level, F(2, 80) = 17.40, p < .001, and a
marginally significant interaction type of practice x chess level, F(2, 80) = 3.07, p =
.052. For the second procedure (i.e., calculation since starting playing age) we found a
main effect of type of practice, F(1, 80) = 19.71, p < .001, a main effect of chess
level, F(2, 80) = 6.76, p < .005, and a non-significant interaction, F(2, 80) = 1.00, ns.
As expected, stronger players practice more hours per week than weaker players. The
interesting result is that in the three chess levels group practice is more important than
individual practice.
It can be argued that the measures used so farcumulative practice until the
present and mean hours of practice until the presentare inflated: some players may
have acquired their current skill level some years ago, and therefore the number of
hours beyond this point in time should not be counted.

INSERT FIGURE 1 ABOUT HERE

INSERT FIGURE 2 ABOUT HERE

In order to address this possibility, we show in Figures 1 and 2 the number of

hours of group and individual practice per week as a function of age and chess level.
The data point at each age indicates the mean hours of practice per week in each
group. The ages before the players started practising were coded as zero hours of
practice. The ages older than the players current age were not included in the
calculation of the mean (i.e., at older ages, the means were calculated with fewer
players than at younger ages).
It can be observed in the figures that the amount of individual and group
practice is different in the three groups. It is also apparent that, overall, the amount of
group practice is greater than that of individual practice. More importantly, both
figures allow us to visualise that the differences in time spent practising among
groups started early on, and that these differences are also present in the range of 13
to 25 years of age. Interestingly, around the 30s experts play (i.e., practice in group) a
similar number of hours per week as masters. It has been argued elsewhere (Dextreit
& Engel, 1981) that non-professional players show interest and commitment to the
game of chess until the late teens, when the amount of time devoted to chess peaks.
After this, players often start work, attend university, and/or get married, which
dramatically reduces the time spent playing and studying chess. In the mid-30s, when
family and work issues are more stable, players return to the game and play more
frequently.

10

INSERT TABLE 1 ABOUT HERE

With respect to the data of experts and intermediates presented in Figure 1 and
2, this hypothesis predicts a cubic function, with increase up to about 18 years old
where there should be a peak, then decrease to a local minimum, and then increase
again. We tested this hypothesis with linear (y = a + bx), quadratic (y = a + bx + cx2)
and cubic (y = a + bx + cx2 + dx3) regression analyses predicting hours of practice as
a function of age. Table 1 shows that, as predicted, the data of experts and
intermediates were fitted better by a cubic function than with a linear or quadratic
function. In particular, with the intermediates the cubic function explains at least an
additional 41% of the variance compared to the linear and quadratic functions. With
the masters, who were expected to be professionals, the cubic function does not
improve on the quadratic with individual practice, although, somewhat surprisingly, it
does with group practice.

INSERT FIGURE 3 ABOUT HERE

INSERT FIGURE 4 ABOUT HERE

It is a common practice in the deliberate practice literature to plot the

cumulative number of hours of practice as a function of years of practice since the
participants began to seriously apply themselves to a domain (e.g., Charness et al.,
2005; Ericsson et al., 1993). Figure 3 shows the estimated number of cumulative
group practice as a function of years of studying seriously, and Figure 4 shows the
estimated number of cumulative individual practice as a function of years of studying
seriously. We consider the year at which the participants reported studying chess

11

seriously as the first year of serious practice and the figures depict the estimated
number of hours of practice in the first ten years of serious practice. These figures are
in agreement with the previous ones in that the three groups differ in the number of
hours of cumulative group and individual practice. They also show that the
differences in group practice (Figure 3) arise earlier that those in individual practice
(Figure 4).5 Not all the players in our sample achieved ten years of serious practice
(either because they became serious at chess late or because they were too young at
the point of measurement); therefore, the average hours of practice for each year was
calculated with a different number of players. This explains why in Figure 4 the
estimation of average cumulative number of hours of individual practice in
intermediate players at 9 and 10 years of serious practice is lower than that of 6 to 8
years of serious practice.
Inter-Individual Variability
Addressing the second issue, that of inter-individual variability, requires
correlating a measure of amount of practice with a measure of performance
variability. To do so, we ranked the players of our sample according to the amount of
cumulative individual practice, and then divided the sample into 9 groups of 10
players. Thus, groups were homogeneous with respect to players amount of practice,
with average practice increasing from group 1 to group 9. For each group, we
calculated the mean and standard deviation for national rating (see Figure 5). Then,
we obtained the correlation of the mean national rating of the groups with their mean
number of hours of individual practice. Also, we calculated the correlation between
the standard deviation of national rating and the mean number of hours of individual
practice. We divided the sample into 9 groups in order to have enough data points for

12

the correlation analysis and enough subjects within each group for the average ratings
and standard deviations of rating to be calculated reliably.
In Figure 5, the blip occurring in group 6 with the standard deviation of
national rating is due to an outlier, which was removed in all the following analyses.
(The pattern of correlations and significances is the same when this outlier is included
in the analyses.) There was a significant positive correlation between mean national
rating and mean number of hours of individual practice, r(7) = .79, p < .02, but we did
not find any significant correlation between standard deviation of national rating and
mean hours of individual practice, r(7) = .38, ns.

INSERT FIGURE 5 ABOUT HERE

We followed the same procedure with group practice (see Figure 6). We found
a significant positive correlation between mean national rating and mean number of
hours of group practice, r(7) = .94, p < .001, and we also found a significant negative
correlation between standard deviation of national rating and mean hours of group
practice, r(7) = -.82, p < .01.

INSERT FIGURE 6 ABOUT HERE

Given that age is positively correlated with amount of practice in this sample
(see Gobet & Campitelli, 2007, p. 164), it is possible that the significant correlations
presented in the previous analyses are accounted for by this variable. In order to
investigate this issue we calculated partial correlations, controlling for the mean age
of each group. There was no significant partial correlation between mean national

13

rating and mean number of hours of individual practice, r(6) = .61, ns., nor between
standard deviation of national rating and mean hours of individual practice, r(6) = .30,
ns. However, there was a significant positive partial correlation between mean
national rating and mean number of hours of group practice, r(6) = .89, p < .005, and
a significant negative partial correlation between standard deviation of national rating
and mean hours of group practice, r(6) = -.87, p < .005. Given the relatively low
power of these analyses, some of the medium-sized but non-significant correlations
should be interpreted with caution.
Until now, we have used current chess level (measured by current national
rating) as dependent variable. An interesting additional question is how ratings change
as a function of yearly practice. Given that we did not have complete archives for the
national rating, we used archives of international rating. Using international rating had
the disadvantage that not all players were included, because the entry requirements
are higher than those of national rating. Therefore, only a sub-sample could be
included (17 masters and 35 experts); in addition, information is lacking about what
happened at young ages or during the first years of serious practice. Nonetheless, our
data are useful to show different trends in performance change in masters and experts
from the time they achieved the level of experts.

INSERT FIGURE 7 ABOUT HERE

INSERT FIGURE 8 ABOUT HERE

Figures 7 and 8 show the international rating for the current masters and
experts, as a function of years of serious practice and age respectively. Figure 7 was
constructed as follows. First, we determined the starting point for each player; for that

14

purpose, we used the year in which the players reported they started playing seriously.
Therefore age in this figure does not matter. For example, a player who reported the
age of 13 would have 1 year of serious playing at the age of 14, 2 years of serious
playing at the age of 15, and so forth. Another player that started playing seriously at
the age of 6 would have 1 year of serious playing at the age of 7, and 8 years of
serious playing at the age of 14. Then, we calculated the average international rating
of both groups at each number of years of serious playing. The number of players at
each of the x-axis values changes. For example, there were many players younger
than 20 years, so they could not possibly have data for 20 years of serious playing.
Note also that we started the analysis at 3 years of serious playing because there was a
negligible number of players with international rating after 1 or 2 years of serious
playing.
We performed a regression analysis to predict international rating in masters
and experts as a function of the number of years of practising chess seriously (the
range of years of serious practice was 3 to 20). The correlation between international
rating and the number of years of serious practice was significant both for masters
(r(16) = .95, p < .0001) and experts (r(16) = .40, p = .05). The regression analyses
showed the following results:
international rating of masters = 2257 + 7.0 * years of serious practice;
international rating of experts = 2174 + 1.03 * years of serious practice.
These results show that, after three years of practice, masters have on average
83 international rating points more than experts (i.e., 2257 vs. 2174). Moreover, from
this point masters increase at the rate of 7 international rating points per year of
serious practice, whereas experts only increase at the rate of 1 Elo point per year of

15

serious practice. The difference between the two slopes was statistically significant:
Hotellings t(15) = 5.5, p < .01.
For Figure 8, we calculated the average international rating of players at each
age for both groups. We started the analysis at the age of 14 because there were too
few players with international rating before this age. We performed the same analysis
as above with age as a predictor variable (the age range was from 14 to 33). The
correlation between international rating and age was significant with the masters,
r(18) = .85, p < .0001), and non-significant with the experts, r(18) = .02, ns. The
regression analyses provided the following equations:
international rating of masters = 2169 + 7.2 * age;
international rating of experts = 2189 + 0.05 * age.
This result is similar to the previous one for the masters but not for the experts.
The experts seem to benefit somehow from the first years of serious practice and only
if they start before the age of 14; after this age they do not seem to improve at all. The
difference between the two slopes was statistically significant: Hotellings t(17) = 4.5,
p < .01.
The analyses of the data presented in Figures 7 and 8 should be taken with
caution. The means at each value of the x-axes were calculated only with the players
that had an international rating. It was not appropriate to follow the same procedure as
with Figures 1 and 2 (i.e., to calculate the mean value including the players with 0
hours of practice) because lack of international rating is not lack of chess skill. Hence,
the absolute values of international rating at each data point do not reflect the true
chess skill of the group perfectly. On the other hand, the differences between the
slopes show a reliable pattern: experts increase their skill level very little with the pass
of time, whereas masters keep increasing theirs.

16

INSERT FIGURE 9 ABOUT HERE

INSERT FIGURE 10 ABOUT HERE

We performed a different analysis with the same data in order to show that the
pattern is reliable. We calculated the percentage of players with international rating
and the percentage of players with more than 2200 Elo points at each year of serious
practice (see Figure 9) and age (see Figure 10). The data show a clear difference
between experts and masters.

Differential Role of Types of Practice

We have been analysing two types of practice: group practice and individual
practice. However, within these two categories there are different specific activities,
and we enquired about them in our questionnaire. Variables corresponding to group
practice were: coaching, blindfold chess, playing speed chess and log number of
speed chess games. Individual practice was represented by the following activities:
blindfold reading, use of databases, use of chess programs, and log number of books.
We followed Charness et al.s (1996) and Charness et al.s (2005) assumption that if
chessplayers have books they would read them, or at least some of them. Therefore,
the quantity of books owned could be a predictor of hours spent reading books.
The percentages of players who reported using these activities were as
follows: playing speed chess, 83.6%; coaching (0,1), 80.5%; use of databases, 67.3%;
use of programs, 66.3%; blindfold reading of games, 55.7%; and blindfold chess,
23%. We assessed the zero-order correlations between these practice activities and
skill. With national rating used as dependent variable, the correlations were .44 for log

17

number of books, .35 for the presence of coaching (0,1), .32 for the use of databases
(0,1), .27 for playing speed chess (0,1), and .27 for log speed games (all p < .05).
There were no significant correlations between skill and blindfold chess, blindfold
reading of games, use of chess programs.
The activities that are most practised by the players in this samplespeed
games and coaching (0,1)were correlated with skill, and thus seem to be useful as
well. Use of databases and computer programs are activities that were performed by
the same number of players, but the mean national rating of players using databases
was higher than that of players using programs. Playing blindfold games was not an
activity performed by many players and it did not have a high correlation with skill
level either. With speed chess rating used as dependent variable, the correlations were
.38 for log number of books, and .35 for coaching (both p < .05). There were no
significant correlations with the other variables.
We entered the eight variables corresponding to either group practice or
individual practice into a stepwise regression analysis, which removed all variables
except Log number of books, Presence of coach (0,1), and Log number of speed
games. A multiple regression with these three independent variables showed that they
all made a significant contribution (see Table 2). As Charness et al.s (2005) showed
that age was a moderator of the link between practice activities and skill level, we
carried out additional analyses with these three independent variables, splitting our
sample in players below 40 years of age and above (or equal) 40 years of age. The
results, shown in Tables 3 and 4, showed that with younger players, only Log number
of books was a significant predictor of skill, while, with older players, Log number of
books and Log number of speed games were predictors, the latter only marginally so.

18

Taking into account the activities measured in the questionnaire, we can

conclude that reading (as inferred by the number of books), an individual activity, is
the most important predictor of chess skill. On the other hand, coaching and log speed
games, two group activities, were also significant predictors of chess skill, although
their importance was moderated by age.

INSERT TABLES 2, 3, and 4 ABOUT HERE

Discussion
In the introduction we mentioned three issues: (a) to what extent is the amount
of practice important to achieving high levels of expertise? (b) does everybody benefit
from practice in the same way? and (c) what types of chess activities are most useful,
and in what order should they be practised? To address these three issues, we
considered both individual and group practice. We first discuss how our data stand
regarding these issues. Then, we will address the topics of generalisation and transfer.
Finally, we will consider the implications of our study for learning sciences in
general.
Before tackling these issues, we address the question of the reliability of the
questionnaire used in this study. It has been argued that individuals might not reliably
report their behaviour in the past (e.g., Davids, 2000; Starkes, Weir, & Young, 2003).
However, previous research has shown that this kind of questionnaire correlates fairly
well with independent measures (Ericsson et al., 1993). In addition, this methodology
produces results that are replicable. We used the same method as two previous studies
on chess (Charness, Krampe & Mayr, 1996; Charness et al., 2005), and, for example,
one key resultthe correlation between cumulative hours of individual practice and

19

skill level is fairly similar in the three studies, considering the fact that the players
participating in these studies come from three continents. We found a correlation of
.42, Charness et al. (1996) found a correlation of .60, and Charness et al. (2005) found
a correlation of .54 for the enlarged sample from their 1996 study, and .48 for a new
sample.
Regarding the role of practice in the acquisition of high-level cognitive skills,
our data partly support Ericsson et al.s (1993) framework of deliberate practice but
are against an extreme view of the role of practice (e.g., Howe, Davidson, & Sloboda,
1998). Moreover, our data are against an extreme view of the role of talent in the
acquisition of high-cognitive skills (e.g., Galton, 1869/1979). We found a significant
correlation between individual practice and national rating and also a significant
correlation between group practice and national rating. This result is also consistent
with previous similar studies in music (Ericsson et al., 1993), chess (Charness et al.,
1996; Charness et al., 2005) and sports (Janelle & Hillman, 2003). On the other hand,
none of these studies showed a correlation that accounts for all or even most of the
variability in skill level. There are several reasons why not all the variability is
explained by the number of hours of practice: measurement error, lack of sensitivity
of the method used, differences in quality of the activities performed (e.g., some
players may have had better coaches than others), age at which players started playing
chess or studying seriously, or non-chess related factors.
In a separate paper (Gobet & Campitelli, 2007), we explored other factors not
related to practice that may influence skill. We showed that a non-chess related factor,
handedness, differentiated chess players from the general population (i.e., there were
more non-right handed persons in a chess sample than in a non-chess sample). We
also found that the age at which players started playing seriously or enrolled in a

20

chess club was an important factor (i.e., there is a critical or sensitive period to start
practising chess seriously in order to achieve high levels of skill). This finding was
strengthened by the analyses presented in the current paper.
The novelty of the present article is that we present longitudinal data, which
affords the possibility of showing increases in rating and in amount of practice as a
function of age or as a function of years of serious practice. This new analysis shows
a difference in the number of hours of chess practice at early ages between masters,
experts and intermediate players (see Figures 1 and 2). Masters also showed an
advantage over experts in chess rating after only 3 years of serious practice (see
Figure 7), although there was no difference in amount of practice between masters
and experts after 3 years of serious practice (see Figures 3 and 4). This is a clear
indication that the first 3 years of serious practice at early ages are much more
profitable than the first 3 years of serious practice at later ages. Moreover, the present
data indicate that there was a considerable amount of dedication to chess before the
period players reported becoming serious about chess (between 10 and 12 years old in
masters and around 14 years old in experts; see Gobet & Campitelli, 2007, p. 163) and
that differences between masters and the rest of the players were already noticeable
by then (see Figures 1 and 2). This indicates that recreational exposure to chess at
early ages is also important, in contradiction to the deliberate practice framework.
Therefore, an extreme view that considers practice as a necessary and sufficient
condition to acquire high levels of expertise is not supported by our data. These new
findings highlight the advantages of using longitudinal data.
In order to address the second issue, we analysed our data from a different
perspective. Thorndike (1908) and Ackerman (1987) argued that with increased
practice the inter-individual variability in performance should diminish if the main

21

factor in mastering the task is practice. Moreover, the inter-individual variability in

performance should increase if the main factor in mastering the task is an innate
ability. Using national rating as a measure of performance and separating groups
according to their cumulative number of hours of practice, we found a significant
negative correlation between the number of hours of group practice and the standard
deviation of national rating. However, we did not find any correlation with individual
practice.
Following Thorndikes (1908) and Ackermans (1987) rationale, the decrease
in performance variability with increasing group practice (see Figure 6) shows that
practice is an important factor in the acquisition of high-level performance in chess.
On the other hand, individual practice did not show a significant negative correlation.
This suggests that no matter how much individual practice players engage in, the
initial differences in performance would be maintained, but the more they engage in
group practice, the less the differences in rating between players should be. The
partial correlation analyses showed that although age accounted for a part of the
correlations, the pattern remained the same: high negative correlation between
standard deviation of chess rating and group practice (marginally significant), and low
negative correlation between standard deviation of chess rating and individual
practice (non-significant).
These results suggest the following hypotheses. First, group practice is an
essential aspect of the acquisition of a high performance in a high-level cognitive skill
such as chess. However, it can only be considered as a necessary, but not sufficient,
condition because the variability in performance diminishes with group practice, but
does not disappear. Second, individual practice is also a necessary, but not sufficient,
condition for acquiring expert performance in chess. Individual practice is the type of

22

activity needed to acquire a vast database of chess knowledge. The more individual
practice, the higher the chess level. However, differences in individual practice alone
cannot explain the inter-individual variability in performance that is observed early
on, when players have little time of practice behind them (cf. Figures 7 and 8). Thus,
third, group practice plus individual practice plus factors unrelated to practice are the
necessary and sufficient conditions to achieve high-level performance in cognitive
skills such as chess. What are these factors? As explained above, in a separate paper
(Gobet & Campitelli, 2007), we showed that the age at which players start to play
seriously, controlling for number of hours of practice, is a good predictor of chess
skill and that an innate factor (i.e., handedness) differentiated chess players from nonchess players. Therefore, we suggest that large amounts of individual and group
practice, starting serious practice as early as possible, and some innate factors
(perhaps the ones responsible for determining handedness) are important factors for
acquiring expert performance.
The third issue we tackled in our study was that of the types of activities that
are useful for acquiring high-level cognitive skills. We examined this issue in two
different ways. First, when considering the role of practice, we investigated the
differential roles of individual and group practice and, second, we enquired about
more specific activities. This is an important issue because, although we agree with
Ericsson et al.s (1993) thesis that deliberate practice is a crucial factor for the
achievement of expert performance, we consider that their description of deliberate
practice in chess is simplistic. For example, they explicitly rule out tournament games
as part of deliberate practice. However, Charness et al. (1996, p. 75) presented data
which showed that active participation in chess tournaments was considered the most
relevant activity by chess players. Given that deliberate practice is the intentional

23

implementation of activities aimed at improving performance, it is reasonable to

assume that experts are aware of these activities and the extent to which they are
beneficial to them. Thus, chess players and in particular chess experts reports that
playing competitive games is an important activity cannot be dismissed readily, as has
sometimes be done in the literature. Moreover, the development of chess theory and
computer technology changed the way chess players prepare for their games (see
Gobet et al., 2002). Chess players not only try to predict the best move in chess games
and then receive feedback, as proposed by Ericsson et al. (1993), but they also
memorize typical manoeuvres and opening variations with the aid of either books or
electronic databases, they investigate opening positions to find novelties to surprise
their opponents, and they play tournament or training games (against other players
physically present, on the internet, or against computer programs). Thus, becoming a
master calls for the acquisition of vast amounts of knowledge, which requires training
activities that go beyond the type of repetitive and feedback-informed activities
typically emphasised in the deliberate practice literature.
When dealing with the first analysis, we found a higher correlation between
group practice and national rating than that between individual practice and national
rating. Moreover, Figures 3 and 4 clearly show that the cumulative number of hours
of group practice is higher than that of individual practice from very early ages. These
results suggest that relevant activities that involve social interaction are more
important than previously thought. For example, Charness et al. (1996) showed that
individual practice was a better predictor of chess skill than group practice. How can
we explain the difference between Charness et al. (1996, 2005) and the present study?
Although we used a national measure of chess skill, which is highly correlated with
the international rating (therefore, in this respect, our results are generalisable to the

24

whole chess community), there are other factors that may be specific of our sample.
For example, the former studies were carried out with participants of Russia,
Germany, Canada and United States, whereas our study was carried out with
Argentine players. There may be cultural differences in the way Argentine players
approach chess in comparison to European and American players. Another reason that
may explain this difference is the wording of the questions. Charness et al.s (2005)
question about group practice is only restricted to serious competition, whereas our
question about group practice included both serious competition and study with other
players.
Another interesting pattern is that individual practice drastically diminishes at
the age of 18 in experts and intermediate players (see Figure 2). This is the age at
which most people in Buenos Aires either start university or start working; therefore,
these data may be showing that only masters that are having good results keep
practising at home, whereas experts and intermediates only continue playing chess for
fun at their leisure time.
Regarding the specific activities, we found that reading books, playing speed
chess, using databases, and having a coach are the most important ones inasmuch as
they showed the highest correlations with national rating, although the impact of these
activities on rating was moderated by age. This result suggests that the acquisition of
high levels of chess skill requires both individual practice (i.e., reading books and
studying with databases of chess games) and group practice (playing games and
receiving chess classes or coaching). These findings can be explained in terms of a
general theory of expertise: the template theory (Gobet & Simon, 1996a). We first
briefly explain the theory and then we relate these findings with the theory.

25

An extension of Chase and Simons (1973) chunking theory, Gobet and

Simons (1996a) template theory of expertise has received strong support from studies
in perception (Chase & Simon, 1973; De Groot & Gobet, 1996; Ferrari, Didierjean, &
Marmche, 2006), memory (Chase & Simon, 1973; Gobet & Simon, 1996, 2000),
imagery (Saariluoma, 1991; Campitelli & Gobet, 2005) and problem solving (Gobet,
1998b; Campitelli & Gobet, 2004). The support for the mechanisms underpinning
template theory comes not only from chess but also from different fields such as
language acquisition (Freudenthal, Pine, & Gobet, 2006), computer programming
(Gobet, Lane, Croker, Cheng, Jones, Oliver, & Pine, 2001), and physics (Lane,
Cheng, & Gobet, 2000).
In short, template theory states that high-level cognitive skills are acquired by
learning numerous domain-specific patterns that are stored in long-term memory. A
distinction is made between chunks, which are fixed perceptual patterns, and
templates, which are patterns made both from fixed and variable information. By
relevant interaction with the environment, the patterns are linked together forming
larger patterns. These patterns are not only static configurations but are also linked
with possible actions. Another aspect of the template theory is that receiving feedback
from the environment changes the links between the patterns and the actions linked to
them. Finally, the generation of a vast knowledge base is not acquired rapidly but
requires several years of practice. The template theory (Gobet & Simon, 1996) states
that any high-level cognitive skill is acquired by the same simple mechanisms. The
difference between different domains is the content of the learned patterns, which is a
function of the type of interaction with the environment. A discussion of the
mechanisms required to acquire high-level cognitive skills is beyond the scope of this
article; suffice it to point out here that the mechanisms were implemented in computer

26

models that simulated human data in detail (Freudenthal et al., 2006; Gobet & Simon,
2000).
Within the framework of the template theory, playing numerous games
facilitates the acquisition of a large knowledge base of domain specific patterns in
long-term memory. (Note that playing rapid chess allows for the playing of many
more games than normal chess in the same amount of time.) However, it is necessary
to complement playing games with reading books, receiving formal coaching, and
using chess databases. Without the learning of proper actions to carry out in a given
situation, the large database of domain-specific patterns held in long-term memory
(chunks and templates) would be useless: a specific situation would be recognized,
but the player would not have knowledge as to what to do given this situation.
Activities such as coaching, studying databases with games of grandmasters or
reading books with grandmasters comments make it possible to learn specific chess
methods and sequences of moves and to link this procedural knowledge to perceptual
patterns, thus giving quality to the knowledge base. Playing rapid games may also
contribute to an ability important in the development of chess skill: the aptitude for
making decisions rapidly under time pressure, with only a minimal exploration of the
search tree and thus far less information than could be obtained after several minutes
of deliberation. This is partly made possible by the presence of chunks and templates,
which speed up the rate at which strong players can generate moves as they can rely
on automatic, pattern-recognition based processing rather than using slower heuristics
(Gobet, 1997).
Since the results reported in this article are consistent with a general theory of
expertise we consider that they can be generalised to other high-level cognitive
domains. However, we are negative about the possibility of transferring skills from

27

chess to other domains, and we do not recommend studying chess in order to improve
performance in other domains. Although there are strong claims in this respect, we
have showed elsewhere that most studies that claimed transfer from chess to other
cognitive skills had severe methodological problems (see Gobet & Campitelli, 2006,
for a review of research into chess and education). The explanation for the lack of
transfer is simple: our cognitive system operates with general mechanisms; what
differs is the type of information that we acquire from the environment. Acquiring a
vast knowledge base in one domain requires investing time and effort, and therefore
leaves less time to devote to the domain to which one wants to transfer skills. We
would predict transfer between domains that have similar environments but not
between domains that deal with different environmental information. For example, in
our sample, the correlation between national rating and speed chess rating is high (i.e.,
.83). Both skills share the same environment and follow the same rules, the only
difference being that the national rating measures chess performance in games lasting
4 hours on average and the speed chess rating measures performance in games lasting
10 minutes on average.
Finally, we draw conclusions for learning and teaching. Our data support
constructivist methods where the activity of the learner is emphasised and oppose
traditional educational methods where the students role is mostly passive. This study
showed that chess masters start engaging in group practice early on and to a larger
extent than experts and intermediate players. For example, at the age of 8 the average
number of cumulative hours of individual practice is 202, 51, and 5 in masters,
experts and intermediates, respectively; by contrast, the average number of
cumulative hours of group practice at the same age is higher in all the groups (547,
279 and 82, respectively). Given the correlational nature of our analysis, we cannot

28

decide whether the masters of our study practised in group more than the experts and
intermediates because they were winning more or that practising more led to
improved performance. In any case, from a learning standpoint our data suggest that it
is advisable to start participating in group activities from the beginning. Children
should have a strong interaction with the environment by playing against peers,
receiving qualified advice from a teacher or coach as a complement of reading books
individually. Moreover, the use of new technologies (in our case chess databases but
not chess playing programs) favours the acquisition of high-level cognitive skills (see
also Gobet, Campitelli & Waters, 2002).
An important finding of this article is that practice seems to be more profitable
at early ages than at older ages. This can be explained by the importance that the
template theory gives to the order in which knowledge is acquired (see Gobet, 2005).
Younger children tend to pay attention to concrete things; therefore it is likely that
they will prefer the perceptual configuration of chess pieces rather than conceptual
knowledge. Gobet (2005) argued that conceptual schemata (or templates) are better
acquired on the basis of perceptual chunks, because these chunks provide a cue that
would improve the efficiency of memory retrieval in the future. It is probable that
older teenagers, who have been already exposed to logical thinking at school, pay
relatively more attention to conceptual aspects of chess games before acquiring the
domain-specific perceptual chunks. As suggested by Gobet (2005) this would impact
in the speed of information retrieval in future chess games.
To conclude, our longitudinal data raise important questions with respect to
the deliberate practice framework. Contrary to its assumptions, practice is more
beneficial at early than later ages; group practice is more important than individual
practice; in spite of the absence of difference in amount of serious practice between

29

masters and experts in the first three years, the former had higher ratings than the
latter; and the experts did not improve much after three years in spite of considerable
amounts of deliberate practice. Together, these results raise serious doubts as to
whether practice is sufficient for reaching high levels of expertise. Similarly, our data
do not support traditional educational methods of acquiring a vast knowledge base by
long hours of isolation and little social interaction. Rather, reading books is still
advisable only if it is complemented with rich interaction with the environment. In
this way, the knowledge base the essential aspect of high-level skill acquisition
would not only be sizeable but it would also be of high quality.

30

References
Ackerman, P. L. (1987). Individual differences in skill learning: An integration of
psychometric and information processing perspectives. Psychological Bulletin,
102, 3-27.
Anderson, J. R., Reder, L. M., & Simon, H. A. (2000, Summer). Applications and
misapplications of cognitive psychology to mathematics education. Texas
Education Review.
Baxter, G. P., & Glaser, R. (1998). Investigating the cognitive complexity of science
assessments. Educational measurements: Issues and practices, 17, 37-45.
Campitelli, G., & Gobet, F. (2004). Adaptive expert decision making: Skilled chess
players search more and deeper. Journal of the International Computer Games
Association, 27, 209-216.
Campitelli, G., & Gobet, F. The mind's eye in blindfold chess. European Journal of
Cognitive Psychology, 17, 23-45.
Charness, N. (1981). Aging and skilled problem solving. Journal of Experimental
Psychology: General, 110, 21-38.
Charness, N., Krampe, R. Th., & Mayr, U. (1996). The role of practice and coaching
in entrepreneurial skill domains: An international comparison of life-span chess
skill acquisition. In K. A. Ericsson (Ed.), The road to excellence (pp. 51-80).
Mahwah, NJ: Erlbaum.
Charness, N., Reingold, E. M., Pomplun, M., & Stampe, D. M. (2001). The perceptual
aspect of skilled performance in chess: Evidence from eye movements. Memory
& Cognition, 29, 1146-1152.

31

Charness, N., Tuffiash, M., Krampe, R., Reingold, E., & Vasyukova, E. (2005). The
role of deliberate practice in chess expertise. Applied Cognitive Psychology, 19,
151-165.
Chase, W. G., & Simon, H. A. (1973). The mind's eye in chess. In W. G. Chase (Ed.),
Visual information processing. New York: Academic Press.
Cooke, N. J., Atlas, R. S., Lane, D. M., & Berger, R. C. (1993). Role of high-level
knowledge in memory for chess positions. American Journal of Psychology,
106, 321-351.
Davids, K. (2000). Skill acquisition and the theory of deliberate practice: It aint what
you do its the way that you do it! International Journal of Sport Psychology,
31, 461-466.
De Groot, A. D., & Gobet, F. (1996). Perception and memory in chess. Assen: Van
Gorcum.
Dextreit, J., & Engel, N. (1981). Jeu d'checs et sciences humaines [Chess and human
sciences]. Paris: Payot.
Elo, A. E. (1978). The rating of chessplayers. Past and present. New York: Arco.
Ericsson, K. A., Krampe, R. Th., & Tesch-Romer, C. (1993). The role of deliberate
practice in the acquisition of expert performance. Psychological Review, 100,
363-406.
Ferrari, V., Didierjean, A., & Marmche, E. (2006). Dynamic perception in chess. The
Quarterly Journal of Experimental Psychology, 59, 397-410.
Freudenthal, D., Pine, J. M., & Gobet, F. (2006). Modelling the development of
childrens use of optional infinitives in English and Dutch using MOSAIC.
Cognitive Science, 30, 277-310.

32

Galton, F. (1979). Hereditary genius. An inquiry into its laws and consequences.
London: Julian Friedman. (Originally published in 1869).
Gobet, F. (1997). A pattern-recognition theory of search in expert problem solving.
Thinking & Reasoning, 3, 291-313.
Gobet, F. (1998a). Chess players' thinking revisited. Swiss Journal of Psychology, 57,
18-32.
Gobet, F. (1998b). Expert memory: A comparison of four theories. Cognition, 66,
115-152.
Gobet, F. (2005). Chunking models of expertise: Implications for education. Applied
Cognitive Psychology, 19, 183-204.
Gobet, F., & Campitelli, G. (2007). The role of domain-specific practice, handedness
and starting age in chess. Developmental Psychology, 43, 159-172.
Gobet, F., & Campitelli, G. (2006). Educational benefits of chess instruction: A
critical review. In T. Redman (Ed), Chess and education: Selected essays from
the Koltanowski conference (pp. 124-143). Dallas, TX: Chess Program at the
University of Texas at Dallas.
Gobet, F., Campitelli. G., & Waters, A. J. (2002). Rise of human intelligence:
Comments on Howard (1999). Intelligence, 30, 303-311.
Gobet, F., de Voogt, A. J., & Retschitzki, J. (2004). Moves in mind: The psychology
of board games. Hove, UK: Psychology Press.
Gobet, F., Lane, P. C. R., Croker, S., Cheng, P. C-H., Jones, G., Oliver, I. & Pine, J.
M. (2001). Chunking mechanisms in human learning. Trends in Cognitive
Sciences, 5, 236-243.
Gobet, F., & Simon, H. A. (1996 a). Templates in chess memory: A mechanism for
recalling several boards. Cognitive Psychology, 31, 1-40.

Gobet, F., & Simon, H. A. (1996 b). The roles of recognition processes and lookahead search in time-constrained expert problem solving: Evidence from grandmaster-level chess. Psychological Science, 7, 52-55.
Gobet, F., & Simon, H. A. (2000). Five seconds or sixty? Presentation time in expert
memory. Cognitive Science, 24, 651-682.
Gobet, F., & Wood, D. J. (1999). Expertise models of learning and computer-based
tutoring. Computers and Education, 33, 189-207.
Hatano, G., & Inagaki, K. (1986). Two courses of expertise. In H. Stevenson, H.
Azuma & K. Hakuta (Eds.), Child development and education in Japan (pp.
262-272). New York: Freeman.
Howe, M. J. A., Davidson, J. W., & Sloboda, J. A. (1998). Innate talents: Reality or
myth? Behavioral and Brain Sciences, 21, 399-442.
Janelle, C. M., & Hillman, C. H. (2003). Expert performance in sport: Current
perspectives and critical issues. In J. L. Starkes & K. A. Ericsson (Eds.), Expert
performance in sports: Advances in research on sport expertise (pp. 19-45).
Champaign, IL: Human Kinetics.
Lane, P. C. R., Cheng, P. C. H., & Gobet, F. (2000). CHREST+: Investigating how
humans learn to solve problems using diagrams. AISB Quarterly, 103, 24-30.
Oldfield, R. C. (1971). The assessment and analysis of handedness: The Edinburgh
Inventory. Neuropsychologia, 9, 97-113.
Saariluoma, P. (1991). Aspects of skilled imagery in blindfold chess. Acta
Psychologica, 77, 65-89.
Simon, H. A. (1980). Problem solving and education. In D. Tuma & F. Reif (Eds.),
Problem solving and education (pp. 81-96). Hillsdale, NJ: Erlbaum.

Starkes, J. L., Weir, P. L., & Young, B. W. (2003). Retaining expertise. In J. L.

Starkes & K. A. Ericsson (Eds.), Expert performance in sports: Advances in
research on sport expertise (pp. 252-270). Champaign , IL: Human Kinetics.
Thorndike, E. L. (1908). The effect of practice in the case of purely intellectual
function. American Journal of Psychology, 19, 374-384.
Watson, J. B. (1924). Behaviorism. New York: People's Institute Publishing
Company.

APPENDIX 1: CHESS QUESTIONNAIRE

(English translation from Spanish)
Answer all the questions, please. Leave a blank space only if you do not possess the
ratings requested. If you do not know your rating/ratings, you can ask the secretary
for it/them. Alternatively, you can write down your name to allow us to look up your
ratings. Moreover, fill out the form of hours of study and practice in chess following
the instructions. Thank you for your participation.
1) How old are you?____________________________________________________
2) What is your profession? ______________________________________________
3) What is your national Elo rating?________________________________________
4) What is your speed chess rating?________________________________________
5) What is your category?________________________________________________
6) What is your international Elo rating? ___________________________________
7) Do you have any title (GM, IM, FM)? Which one?__________________________
8) At what age did you learn how to play chess?______________________________
9) At what age did you start playing chess seriously?__________________________
10) How many hours per week (on average) have you studied alone during the current
year? __
11) How many hours per week (on average) have you studied or practised chess with
other chess players (including tournament games) during the current year?________
12) Have you ever joined a chess club?_____________________________________
If yes, at what age for the first time?_______________________________________
13) Have you ever received formal chess instruction from a chess coach?__________
Individual coaching: from (age)______to (age)_______________________________
Group coaching: from (age)________to (age)________________________________
14) How many books do you have? (excluding chess journals)___________________
15) Do you play blindfold chess?_________________________________________
16)

Do

you

reproduce

chess

games

from

journals

without

using

the

chessboard?__________________
17) Do you use any computer database to study chess?_________________________
18) Do you play games against chess software?_______________________________
19) Do you play speed chess games?_______________________________________
How many per week?________________________________________________

Table 1
Amount of Variance (r2) Explained by Linear, Quadratic, and Cubic Regressions, as
a Function of Skill Level and Type of Practice.

Linear

Skill

r2

Quadratic

r2

Cubic

r2

Group practice

Masters

.25

10.6**

.59

22.0***

.80

41.1***

Experts

.41

22.0***

.45

12.7***

.78

35.7***

Intermediates

.14

5.3*

.19

3.7*

.66

19.3***

Individual practice

Masters

.14

5.4*

.83

75.7***

.83

50.5***

Experts

.05

1.8

.51

15.9***

.63

17.0***

Intermediates

.09

3.2

.15

2.6

.56

12.8***

Note: The degrees of freedom are 32, 31, and 30 for the linear, quadratic, and cubic
regression, respectively.
*p < .05. **p < .01. *** p < .001.

Table 2. Multiple regression predicting national rating using practice variables

(entire sample).

Variable
Constant

SE

Beta

95% CI

1597.0

65.6

Log number of books

158.8

34.7

.403

4.58 < .001

89.9, 227.8

Presence of coach (0,1)

136.4

52.4

.233

2.60 < .02

32.4, 240.4

62.0

29.2

.190

2.13 < .05

4.1, 120.0

Log number of speed games

24.3 < .001 1466.7, 1727.4

Note. R = .557, R2 = .310, Adjusted R2 = .288, F(3, 91) = 13.64, p < .001.

Table 3. Multiple regression predicting national rating using practice variables

(players younger than 40 years).

Variable
Constant

SE

Beta

95% CI

1646.8

97.3

Log number of books

165.2

42.2

.421

3.91 < .001

Presence of coach (0,1)

108.5

86.0

.135

1.26

ns

-63.0, 280.0

37.6

37.9

.107

0.99

ns

-38.1, 113.3

Log number of speed games

16.9 < .001 1452.6, 1840.9

80.9, 249.4

Note. R = .487, R2 = .237, Adjusted R2 = .203, F(3, 68) = 7.035, p < .001.

Table 4. Multiple regression predicting national rating using practice (players 40

years old or older).

Variable
Constant

SE

Beta

95% CI

1525.0

104.9

185.1

67.3

.474

2.7 < .05

Presence of coach (0,1)

94.3

75.7

.214

1.2

ns

-64.2, 252.7

Log number of speed games

93.1

44.7

.334

2.08

.051

-0.44, 186.6

Log number of books

14.5 < .001 1305.5, 1744.4

44.3, 325.9

Note. R = .725, R2 = .526, Adjusted R2 = .451, F(3, 19) = 7.036, p < .005.

Figure captions
Figure 1: Group practice as a function of age and chess level. The mean and standard
deviations for each group calculated since 4 years of age were: Masters ( M = 10.6,
SD = 5.2); Experts ( M = 6.6, SD = 3.9); Intermediates ( M = 3.4, SD = 2.9). The mean
and standard deviations for each group calculated since starting playing were: Masters
( M = 13.0, SD = 5.7); Experts ( M = 9.4, SD = 5.6); Intermediates ( M = 6.2, SD =
4.8).
Figure 2: Individual practice as a function of age and chess level. The mean and
standard deviations for each group calculated since 4 years of age were: Masters ( M
= 7.3, SD = 4.6); Experts ( M = 3.7, SD = 3.6); Intermediates ( M = 2.6, SD = 3.5).
The mean and standard deviations for each group calculated since starting playing
were: Masters ( M = 8.7, SD = 4.6); Experts ( M = 6.3, SD = 6.9); Intermediates ( M =
4.4, SD = 6.4).
Figure 3: Cumulative group practice as a function of years of serious practice, for the
first ten years of serious practice. Note that the number of participants varies from
year to year. The first year of practice was calculated over 17 masters, 35 experts, and
31 intermediates and the tenth year of practice was calculated over 15, 22 and 17
participants, respectively. Bars represent the standard errors of the mean.
Figure 4: Cumulative individual practice as a function of years of serious practice, for
the first ten years of serious practice. Note that the number of participants varies from
year to year. The first year of practice was calculated over 17 masters, 35 experts, and
31 intermediates and the tenth year of practice was calculated over 15, 22 and 17
participants, respectively. Bars represent the standard errors of the mean.
Figure 5: Variability in national rating as a function of cumulated individual practice.
The mean and standard deviation of number of hours of individual practice in each

group were: group 1: M = 10.4, SD = 33; group 2: M = 473, SD = 244; group 3: M =

1,165, SD = 348; group 4: M = 2,293, SD = 286; group 5: M = 3,567, SD = 529; group
6: M = 5,317, SD = 829; group 7: M = 6,911, SD = 306; group 8: M = 9,963, SD =
2,276; group 9: M = 18,683, SD = 3,075.
Figure 6: Variability in national rating as a function of cumulated group practice. The
mean and standard deviation of number of hours of group practice in each group
were: group 1: M = 894, SD = 444; group 2: M = 2,000, SD = 517; group 3: M =
3,494, SD = 408; group 4: M = 4,841, SD = 423; group 5: M = 6,365, SD = 371; group
6: M = 8,117, SD = 784; group 7: M = 10,314, SD = 624; group 8: M = 13,744, SD =
2,126; group 9: M = 23,030, SD = 7,358.
Figure 7: International rating in masters and experts as a function of years since
starting playing seriously. Bars indicate standard deviations.
Figure 8: International rating in masters and experts as a function of age. Bars indicate
standard deviations.
Figure 9: Percentage of players with international rating as a function of years since
starting playing seriously. Note that all the masters who are in the international list
after seven years of serious playing have more than 2200 points.
Figure 10: Percentage of players with international rating as a function of age. Note
that all the masters who are in the international list after 18 years of age have more
than 2200 points.

Figure 1.
Masters

Mean hours per week

Experts
Intermediates

20
15
10
5
0
4

10

13

16

19

22

Age

25

28

31

34

37

Figure 2.

Masters
Experts

Mean hours per week

20

Intermediates
15
10
5
0
4

10

13

16

19

22

Age

25

28

31

34

37

Figure 3.

9000

Cumulative group practice

8000

Masters

7000

Experts

6000

Intermediates

5000
4000
3000
2000
1000
0
1

Years of serious practice

10

Figure 4.
9000

Cumulative individual practice

8000

Masters

7000

Experts

6000

Intermediates

5000
4000
3000
2000
1000
0
1

Years of serious practice

10

Figure 5.
300
250
2200
200
2000

150
100

1800
50
1600

Mean
SD

0
1

Individual practice

National Rating (SD)

National Rating (Mean)

2400

Figure 6.
300
250
2200
200
150

2000

100
1800
50
1600

Mean
SD

0
1

Group practice

National Rating (SD)

National Rating (Mean)

2400

Figure 7.
Masters

2500

Experts

Mean Elo

2400
2300
2200
2100
2000
1900
3

10

11

12

13

14

15

16

Years of practice since starting seriously

17

18

19

20

Figure 8.
2500
2400

Mean Elo

2300
2200
2100
Masters
2000

Experts

1900
14

15

16

17

18

19

20

21 22

23 24

Age

25 26

27 28

29

30

31 32

33

Figure 9.

Experts_Elo
100

Masters_Elo

Percentage of players

90

Experts>2200

80

Masters>2200

70
60
50
40
30
20
10
0
3

10

11

12

13

14

Years of serious playing

15

16

17

18

19

20

Figure 10.
100

Percentage of players

90
80
70

Experts_Elo

60

Masters_Elo

50
40

Expert>2200

30

Masters>2200

20
10
0
14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33

Age

Authors note
We thank Merim Bilali, Neil Charness, Richard Smith, and anonymous
reviewers for comments on this paper.
Correspondence concerning this article should be addressed to Fernand Gobet,
Centre for the Study of Expertise, Brunel University, Uxbridge, Middlesex, UB8 3PH,
United Kingdom. Email: [email protected]

Footnotes
1

FIDE stands for Fdration Internationale des checs (International Chess

Federation).
2

The speed chess rating is computed independently from the national rating. In some

cases, the calculation for the former rating is based on more than one thousand games.
3

We included tournament games under the umbrella of group practice, as players in

Charness et al.s study (1996, Table 2.4) considered active participation in chess
tournaments as the most relevant activity.
4

Charness et al. (2005) found that the relation between skill level and practice

activities was moderated by age. This was the case in our sample as well with respect
to the relation between rating and individual or group practice, as the correlations
were stronger for older players, although, critically, they were always higher for
group practice than individual practice. For the players below 40 years of age, the
correlations were r (69) = .36, p < .005, for individual practice, and r (68) = .55, p <
.001, for group practice. The respective correlations for players of 40 years of age and
above were: r(21) = .59, p = .005, and r ( 21) = .72, p < .001.
5

Note also that, although the skill levels are broken down in slightly different ways,

the data in Figure 4 are in close agreement with those presented in Figure 1 in
Charness et al. (2005). This common pattern suggests that the data of our study are
generalisable (see also the Discussion section).