To measure the chess rating of the respondents an ELO score is known as an instrument to

measure their chess performance rating. It investigates and predicts future performance in
matches.

Studies shows that in order to determine the ELO rating of the chess players and the
distribution of their performances measured by computer analysis of moves

This paper studies the population of chess players and the distribution of
their performances measured by Elo ratings and by computer analysis of
moves. Evidence that ratings have remained stable since the inception of
the Elo system in the 1970’s is given in three forms: (1) by showing that the
population of strong players fits a straightforward logistic-curve model
without inflation, (2) by plotting players’ average error against the FIDE
category of tournaments over time, and (3) by skill parameters from a
model that employs computer analysis keeping a nearly constant relation
to Elo rating across that time. The distribution of the model’s Intrinsic
Performance Ratings can therefore be used to compare populations that
have limited interaction, such as between players in a national chess
federation and FIDE, and ascertain relative drift in their respective rating
systems.

Regan, K.W., Macieja, B., Haworth, G.M. (2012). Understanding Distributions of Chess
Performances. In: van den Herik, H.J., Plaat, A. (eds) Advances in Computer Games.
ACG 2011. Lecture Notes in Computer Science, vol 7168. Springer, Berlin,
Heidelberg. https://doi.org/10.1007/978-3-642-31866-5_20

An Elo score is a known measure of past performance in chess and other games.
This paper investigates the impact of Elo ratings on chess game results and
whether this measure can be used to predict future performance in matches. To
achieve the aim, various machine learning classification techniques have been
evaluated using chess data sourced from an online chess server. We examine how
much influence the Elo score has on the prediction power of these techniques
based on classifiers they derive. The prime objective of this experiment is to
accurately predict the winner of a chess game from attributes that are available
before the game starts. We are particularly interested in how large an impact the
Elo score has on the prediction when compared with other features. Empirical
results reported that classifiers derived by artificial neural network (Multilayer
Perceptron), Decision Tree (J48/C4.5), Rule Induction (JRip/RIPPER) and
Probabilistic (Naïve Bayes) showed how useful the Elo is at predicting chess
results, at least on the dataset considered, improving classifiers’ performance with
respect to accuracy, precision, recall and area under curve, among others.

Journal of Information & Knowledge Management VOL. 19, NO. 02

No Access

Chess Results Analysis Using Elo

Measure with Machine Learning
Fadi Thabtah, Arun J. Padmavathy, and Andrew Pritchard
Doi: https://doi.org/10.1142/S0219649220500069
Development of Chess Skill from Domain
Entry to Near Asymptote
Robert W. Howard
The American Journal of Psychology (2018) 131 (3): 323–345.
https://doi.org/10.5406/amerjpsyc.131.3.0323

The aim of the present study was to see how well the power law of practice, studied
mainly with simple skills acquired over short periods with task time as the
performance measure, generalizes to chess skill. Chess playing is a complex
cognitive skill acquired over years, and expertise can be measured by a performance
rating based on game results and relative strengths of opponents. Participants were
75 highly skilled players who entered the domain very young and improved skill
greatly. With the number of games as the practice measure, the traditional 2-
parameter power law fit the mean curve and most individual curves quite well.
However, a new formulation of the power law found by artificial intelligence program
Eureqa often worked even better. Power models with an asymptote parameter did
not work well with chess skill unless that parameter was set to a constant. Another
study aim was to see how well various models predict future performance from
varying practice levels. For three models, the less practice the greater was the
underprediction of future performance, but the new power law formulation predicted
well from early in practice. Another study aim was to test model fits with time as the
practice measure. With time, power models fit well, but an exponential model and a
quadratic model fitted most individual curves better. The power law as traditionally
formulated does generalize well to chess skill development but is not always the best
model, and no single model always fit best for all participants.
Learning curves have been proposed as an adequate description of learning processes, no
matter whether the processes manifest within minutes or across years. Different
mechanisms underlying skill acquisition can lead to differences in the shape of learning
curves. In the current study, we analyze the tournament performance data of 1383 chess
players who begin competing at young age and play tournaments for at least 10 years.
We analyze the performance development with the goal to test the adequacy of learning
curves, and the skill acquisition theories they are based on, for describing and predicting
expertise acquisition. On the one hand, we show that the skill acquisition theories
implying a negative exponential learning curve do a better job in both describing early
performance gains and predicting later trajectories of chess performance than those
theories implying a power function learning curve. On the other hand, the learning
curves of a large proportion of players show systematic qualitative deviations from the
predictions of either type of skill acquisition theory. While skill acquisition theories
predict larger performance gains in early years and smaller gains in later years, a
substantial number of players begin to show substantial improvements with a delay of
several years (and no improvement in the first years), deviations not fully accounted for
by quantity of practice. The current work adds to the debate on how learning processes
on a small time scale combine to large-scale changes.

Introduction
Anderson (2002) drew attention to the problem of time scales in psychology with the
programmatic article Spanning Seven Orders of Magnitude. On the one hand,
acquisition of expertise is known to takes years (e.g., Ericsson et al., 1993). On the other
hand, expertise research has a strong basis in cognitive psychology paradigms wherein a
large repertoire of laboratory tasks are used to understand and chart changes in potential
subcomponents of expertise acquisition over minutes or hours. This includes component
skills such as verifying and storing chess patterns (Gobet and Simon, 1996a,b,c, 1998;
Campitelli et al., 2005, 2007; Bilalić et al., 2009a), learning to discard irrelevant
perceptual features from processing (e.g., Gaschler and Frensch, 2007; Reingold and
Sheridan, 2011) or overcoming dysfunctional bindings of knowledge structures (e.g.,
Bilalić et al., 2008a,b). Anderson suggested that while meaningful educational outcomes
take at least tens of hours to achieve, those outcomes can be traced back to operations of
attention and learning episodes at the millisecond level. He went beyond offering the
perspective that expertise acquisition should in principle be reducible to small scale
learning episodes. Rather, Anderson suggested that the problem of linking domains of
(a) laboratory cognitive psychology/neurocognitive research and (b)
educational/developmental science should be tractable, because small scale learning
episodes would sum up to large scale developmental/educational changes of the same
functional form. Increases in overall performance as well as increases in efficiency of
components (e.g., keystrokes, eye movements and fact retrieval) over time are well
described by the power function (see also Lee and Anderson, 2001). Power functions of
improvements in simple components add up to a power-function improvement at the
large scale. Scalability across time-scales would offer straightforward linking of change
taking place within minutes to change taking place over years.

The power function (as well as the negative exponential function, see Table 1 and Figure
1) describes negatively accelerated change of performance with practice. Early in
practice, the absolute improvement in performance per unit of time invested is large.
Later on, the improvement per unit of time diminishes. Apart from improvements in
hour-long laboratory learning tasks, the power function has been used to describe motor
skills in individuals differing in amount of practice on the scale of years (e.g., up to 7
years of cigar-rolling in Crossman, 1959, see Newell et al., 2001 for an overview).
Description of practice gains with the power function are widespread in the literature
(Newell and Rosenbloom, 1981; Kramer et al., 1991; Lee and Anderson, 2001; Anderson,
2002) and consistent with prominent models of skilled performance such as ACT-R
(Anderson, 1982) or the instance model of automatization (Logan, 1988, 1992).

Table 1

TABLE 1. FORMULAS OF NEGATIVE EXPONENTIAL AND POWER FUNCTION AND

THE EXAMPLE PARAMETERS USED FOR FIGURE 1.

Figure 1

FIGURE 1. (A) Schematic plot of the power function (blue lines) and negative
exponential function (red lines) over 20 time points. (B) Shows the absolute differences
in performance from one time point to the next (filled symbols) and the relative learning
rate (empty symbols).
However, the reason for the dominance of the power function in describing the
functional form of describing practice has been debated in the literature on skill
acquisition. Heathcote et al. (2000) see also Haider and Frensch (2002) argued that the
analysis of averaged data favors the power over the exponential function as a statistical
artifact. They suggested computation of power and exponential curves with non-
aggregated data, separately for each participant. They found an advantage of the negative
exponential function over the power function in 33 of 40 different re-analyzed data sets
with an average improvement in fit of 17%. Note that success of a mathematical model in
fitting data better than a competitor model might not mean that it provides a more
concise description. Potentially, one mathematical model is more flexible than the other,
and better able to accommodate systematic as well as chance features in the data. Thus,
further credence is lent to a model by accurate prediction rather than fitting (i.e., without
any further parameter adjustments; cf. Roberts and Pashler, 2000; Pitt et al., 2002;
Wagenmakers, 2003; Marewski and Olsson, 2009).

It is worthwhile considering the exact shape of the learning curve to predict future
performance. Furthermore, the differences between exponential and power function are
linked to assumptions in theories of skill acquisition (see below). Figure 1 represents
schematic examples of learning curves and derivatives. The left panel depicts a power
function and an exponential function that start at the same level in the first year of chess
tournament participation and approach similar levels in year 20 of tournament
participation. The power function shows especially strong performance gains in the first
years. For instance, the gain in rating points (e.g., Elo, 1978) in year one is about
double the size of the gain in year two. Year two still yields
considerably more performance gain as compared to year three, and so
on and so forth. Absolute gain per year is depicted in the right panel.
It is decreasing for both, the power and the exponential function. The
qualitative difference between the two types of learning curves
becomes most obvious when considering the relative learning rate
(RLR). This rate is decreasing for the power function, but remains
constant for the exponential function. In our example, the
exponential function has a relative learning rate of about 20%. In each
year, the players gain about 20% of the ELO points they have not
gained yet. If someone starts with 1000 and will end up with 1500
points (see Method for an explanation of the scale used in chess), this
would mean a gain of 100 points for the first year and 80 points in the
second year (20% of 1500 − (1000 + 100) = 80 points).

One qualitative aspect of learning curves is that they represent the diminishing absolute
payoff of practice-investment. Exponential practice functions can be derived from a
narrow set of assumptions. As Heathcote et al. (2000) explained one needs only to
assume that learning is proportional to the time taken to execute the component in case
of a continuous mechanism. First, a component that takes longer to execute presents
more opportunity for learning. Second, as learning proceeds, the time to execute the
component decreases. Therefore, the absolute learning rate decreases, resulting in
exponential learning. Similarly, for discrete mechanisms, such as chunking, exponential
learning can be explained by a reduction in learning opportunity. As responses are
produced by larger and larger chunks, fewer opportunities for further composition are
available. Time-demanding control is no longer necessary for small steps but only for
scheduling sets consisting of fixed series of small patterns. Naturally, the opportunities
for compilation of small single knowledge units into larger ones reduce, as more and
more patterns are already chunked.
Additional theoretical assumptions are needed to accommodate a decreasing RLR. For
instance, Newell and Rosenbloom (1981) see also Anderson (2002) assumed that chunks
are acquired hierarchically and that every time a larger chunk is practiced, this entails
practice of its smaller components. Thus, by practicing a knowledge unit consisting of
sub-units, the sub-units and the overall pattern are fine-tuned and strengthened.
Furthermore, at least in combinatorial environments, acquisition proceeds ordered by
chunk span. No larger span chunk is acquired until all chunks of smaller span have been
acquired.

The above research suggests that one or the other simple learning function might be
adequate to describe improvements over long time intervals (cf. Howard, 2014).
Functions known from work on short-term skill acquisition should be relevant to
describe long-term expertise acquisition. We take chess as an example to explore this
perspective. First, longitudinal data spanning years of practice are available. Second,
theories on expertise in chess can be taken to suggest that scalability between small scale
learning and large scale expertise acquisition should be especially likely to hold in this
domain. Expertise development in chess might predominantly be based on cumulatively
storing more and more patterns of chess positions (Chase and Simon, 1973; Gobet and
Simon, 1996b). Spatial (Waters et al., 2002; Connors and Campitelli, 2014; Leone et al.,
2014) and perceptual capabilities are deemed crucial (Charness et al., 2001; Reingold et
al., 2001; Bilalić et al., 2008a,b; Kiesel et al., 2009; Bilalić et al., 2010; Bilalić and
McLeod, 2014). This suggests that attentional and learning episodes taking place at the
time scale of milliseconds might together lead to expertise acquisition. This in turn
would make it likely that expertise acquisition can be described by the learning function
exhibited during learning episodes that take place within a single laboratory session.

In order to explore the potential of this conjecture in the current study, we provide a
descriptive analysis of the development of chess performance in German players who
start playing chess at an early age and continue with the activity for at least 10 years.
Relevant for theoretical as well as practical purposes, the time courses of expertise
acquisition could thus potentially be predicted. Based on the shape of the curve of
improvements during the first years of expertise acquisition, one might be able to predict
the time course of improvements over the years of practice to come (Ericsson et al., 1993;
Charness et al., 2005).

Methods
Database
We used archival data of the population of German players recorded by the German
chess federation (Deutscher Schachbund) from 1989 to 2007. Data were kindly provided
by the federation and analyzed in line with guidelines of the ethics review board at
Humboldt-Universität, Berlin. With over 3000 rated tournaments in a year, the German
chess federation is one of the largest and the best-organized national chess federations in
the world. Given that almost all German tournaments are rated, including events such as
club championships, the entire playing careers of all competitive and most hobby players
in Germany are tracked in detail. This is particularly important because we wanted to
capture the very first stage of chess skill acquisition by focusing on the very young chess
players who just started to play chess. The German database provides a perfect
opportunity to study the initial stages of skill acquisition because even school
tournaments are recorded.
The Measure—Chess Rating
Besides precise records of players, the German federation's database and chess databases
in general use an interval scale, the Elo rating, for measuring skill level. Every player has
an Elo rating that is obtained on the basis of their results against other players of known
rating (see Elo, 1978). Average players are assumed to have rating of 1500 Elo points,
experts over 2000 points, grandmaster, the best players, over 2500. Beginners usually
start at around 800 Elo points. The German database uses the same system but labels
the rating as Deutsche Wertzahl (DWZ), which is highly correlated (r > 0.90) with the
international Elo rating (Bilalić et al., 2009b).

Selection Criteria and Grouping of Data Analyzed

The German chess federation database contains records of over 124,000 players and the
average rating of these players is 1387 points with standard deviation of 389 points. For
all practical purposes, the database contains the entirety of the population of tournament
chess observations in Germany (for more information about the database, see Bilalić et
al., 2009b; Vaci et al., 2014). With interest in expertise development (rather than
maintenance), we used the subset of data from all players who entered the database
between age 6 and 20. This population consisted of 1383 players that played competitive
chess for at least 10 years. All players took part in tournaments in each of the 10 years.
To be sure that the initial observation was indeed first entry into competitive chess, we
excluded players who were already listed in the first year the federation started tracking
players. For the players starting young, there should have been little opportunity for
expertise acquisition prior to taking part in tournaments covered by the database. To
track this issue, we split the sample into age-groups (see Table 2 gender and age as well
as for means and standard deviations of games played, rating reached by year 10, change
in rating between year 1 and 10, and change in rating per game played).

Table 2

TABLE 2. SAMPLE CHARACTERISTICS AND SUMMARY STATISTICS ON GAMES

PLAYED AND RATINGS.

Note that since we are working with the entirety of tournament chess performances in
Germany since 1989, we provide description of the entire population of interest—chess
players that played competitive chess in Germany for at least 10 years (means, standard
deviations, correlations that allow for an estimation of effect sizes). Generalization of
findings, beyond the internal predictions, will have to be based on replications with other
or future databases (see e.g., Asendorpf et al., 2013).

Predicting and Fitting with the Power and Exponential Function

Fits were derived with constrained optimization, requiring the A and B parameters to
take sensible values (0 < B < A < 3000) using the MATLAB Curve Fitting Toolbox. For
each participant we compared estimated and observed ratings and determined whether
the power function or the exponential function led to a smaller squared deviation. For
predictions, we only used the data of the first 5 years to extract the parameters of the
power function and the exponential function. Then we used these parameters to
extrapolate the predicted ratings for the next years (at least 5—each person in the sample
had database entries for a minimum of 10 years). The predicted values were then
compared to the ratings actually achieved. For instance, for a given participant who
played for 10 years, we took the performance in the first five, acting as if the trajectory
data of the next years were not yet available. The power function and the exponential
function were fit to the data of the first 5 years in order to obtain the parameter values
exemplified in Table 1. Next we used these values in order to extrapolate for the coming
years of tournament participation. These predicted values were than compared to the
actual ratings obtained. For each participant we could thus compare the root mean
square error (RMSE) between power function-based prediction and prediction based on
the exponential function.

Results
Descriptive Statistics
Table 2 indicates that our sample was predominantly male. Participants starting to play
tournaments younger accumulated more games as compared to those starting at an older
age. For instance, the 6 to 9 year olds played twice as many games than the 18 to 20 year
olds. The rating reached by Year 10 was similar across age groups. Yet, this implied a
much stronger improvement compared to Year 1 for the players starting young rather
than old. For instance, the youngest group showed trifold the increase of the oldest
group. The increase in rating relative to the number of games played was similar across
age groups (with the players starting oldest, who showed a reduced gain per games
played).

Exponential Better than Power Function in Predictions and Fits

Figure 2 presents a random subset of individual time courses. Despite fluctuations from
one year to the next, participants generally showed increases in skill, as measured by Elo,
over years of chess played. Some participants showed large gains especially in their first
years. In order to systematize such observations, we tested the capability of the power
function and the exponential function to fit and predict the observed trajectories.
Prediction is interesting for practical purposes as we can infer the skill level someone will
have after ten years of activity based on the pattern of performance in their first years.
On the other hand, prediction circumvents methodological problems inherent in curve
fitting. For instance, one mathematical function might fit better than another, because it
is flexible enough to mimic the competitor.

Figure 2
FIGURE 2. A RANDOM SAMPLE OF INDIVIDUAL TIME COURSES.

Across individuals from all age groups, the exponential function provided better
prediction and fit to the data than the power function (Table 3). The average RMSE and
its standard deviation were smaller for the exponential function than for the power
function (with exception of the prediction among those starting chess at ages 18–20).
For 88% of the players, the exponential function was better in fitting the first 5 years the
skill acquisition process, and for 62% it was better in predicting the skill level in later
years. As shown in Figure 3, the distribution of RMSE values was heavily left-skewed.
For a substantial proportion of participants neither the exponential nor the power
function provided an account of the dynamics of individuals' skill development.

Table 3
TABLE 3. AVERAGE AND STANDARD DEVIATION OF RMSE PER AGE GROUP.

Figure 3

FIGURE 3. DISTRIBUTION OF ROOT MEAN SQUARE ERROR (RMSE) OF RATING

POINTS OF THE POWER FUNCTION (BLUE) AND EXPONENTIAL FUNCTION (RED)
IN FITTING THE TIME COURSE OF THE FIRST 5 YEARS (FILLED LINES) VS.
PREDICTING THE TIME COURSE FOR THE NEXT YEARS BASED ON THE
PARAMETERS DERIVED FROM FITTING THE FIRST YEARS (DOTTED LINES).

Increasing Gains in Participants Starting Young

We grouped players by the age they started to play tournaments in order to explore
reasons for the substantial problems in fit and prediction encountered with both the
power and exponential functions. Potentially, players starting tournament participation
at older ages might have profited from substantial opportunities to practice chess before
they entered our window of analysis. Thus, the expected learning gains might manifest
more readily in those players that started at younger ages. Figure 4A indeed shows that
players entering tournament chess at older ages demonstrate higher skill levels by the
end of their first year, while players entering at younger ages, start at lower levels. Most
notably, however, the shape of the improvements deviates systematically from the
patterns of change that would be expected based on either the exponential or the power
function. Both learning functions predict that participants should show a higher absolute
gain in rating points from the first to the second year compared to the gain from the
second to the third year, which in turn should yield a higher gain as compared to the
change from the third to the fourth year, and so on and so forth. Among the subset of
players starting young, however, the contrary seemed to be the case (see also Figure 4B
for difference values). Over the first years of tournament participation, the absolute
amount of gain per year increased rather than decreased. The deviation from the
expected learning curve might be related to year-to-year variations in practice. For
example, the players starting young may, at first, participate in very few tournaments,
and then, in the next few year, increase in the number of tournament games they take
part in. This is indeed the case (Figure 4C). Therefore, it is conceivable that the amount
of practice which increases over the years accounts for the dynamics of the skill increase
—only once the players starting young take part in more and more games, their skill
might start to increase in the manner predicted by the learning curves. As we do not
possess any further data on changes in the amount of practice per year (i.e., off
tournament practice), we cannot conclusively judge this account. However, at least we
can state that the increase in the number of tournament games played cannot fully
account for the dynamics. As shown in Figure 4D, the change in rating per year per
number of tournament games played also shows an increase over the first years for
players starting young.

Figure 4
FIGURE 4. (A) Charts the average time course in tournament performance for players
starting at 6–20 years of age. (B) Shows the time course of the gain in performance from
one year to the next for the different starting-age groups. In (C) the time course of
number of tournament games played per year is charted for the different starting-age
groups. (D) Shows the time course of the improvement in tournament performance from
year to year relative to the number of tournament games played in the respective year.

Fluctuations in Games Played Per Year Relate to Misfit with Power

Function
It is conceivable that the misfit and inaccurate prediction of the power and the
exponential functions are related to variability in the number of games played per year.
While we only examined the subset of players who played in tournaments in each of the
10 years tracked, the number of games per year might have fluctuated. We computed the
within-person (intraindividual) standard deviation of games played per year, assuming
that fits should be optimal if the number of games a player takes part in does not change
over the years. This index is equivalent to computing the deviation from a zero-slope line
in numbers of games played. Table 4 shows the Spearman rank order correlation of
intraindividual variability in number of games played per year with the RMSE obtained
from fitting and predicting ratings based on the power function and the exponential
function. The correlations suggest that larger intraindividual variability in number of
games played per year was weakly but consistently related to worse fit in case of the
power function (while the pattern was less consistent for the exponential function). A
similar pattern was observed when correlating the overall number of games played to
accuracy in prediction and fitting. Participants who played more games showed worse
fits compared to participants who played less games. This was likely the case, because
the number of games played over the ten years (a count variable) was closely linked to
the intraindividual variability in games per year (Spearman correlations ranging between
0.84 and 0.88 across the four age groups).

Table 4

TABLE 4. SPEARMAN RANK ORDER CORRELATIONS OF (A) INDICES OF

REGULARITY IN NUMBERS OF GAMES PLAYED PER YEAR WITH (B) THE FITTING
AND PREDICTION ERROR.

Figure 4C suggests that variability in number of games played per year is not purely
random. Instead it can be based on an ordered pattern (inverted U-shape). Separately
for each age group, we took the average profile in number of games played per year
(displayed in Figure 4C) as a prototypical pattern. Then, we determined for each
participant the profile correlation between his/her pattern of numbers of games played
with the average pattern of the respective age group. Our analyses suggested that there
was substantial variability, with some participants following the pattern represented in
the group mean and others deviating from it. Median within-person correlations per age
group were r = 0.58, 0.5, 0.47, and 0.19. The percentage of individuals showing a
negative correlation with the prototypical pattern was 9.8, 15.2, 19.6, and 31.8%.
However, as suggested by Table 4, the extent to which the dynamics of an individual's
number of games played per year was represented by the average pattern of the age
group was not systematically related to the accuracy in power function or exponential
function fits and predictions.
Off-the-Curve Patterns in 2/3rds of the Sample
We sought to provide descriptive data on the number of participants who deviated from
the predictions of the learning curves by showing smaller rather than larger rating gains
during their early as compared to their later years of tournament participation. For this
we sorted individuals into tertiles based on the total gains achieved during the first 3
years (lowest, medium, and highest rating gains). As shown in Figure 5A, the third of
players with the lowest gains even showed small decreases in rating during the first
years, while only the individuals with the largest gain yielded performance changes in
line with the predictions by the learning curves (i.e., larger gain per year in early rather
than late years, compare Figure 5B). Players that did not improve in their first three
tournament years caught up to some extent in later years, but did not reach the same
level by year 10 as those players with a steep increase early on. Thus, irrespective of
complex dynamics of the shape of the performance curve, the first years do seem to offer
a proxy for predicting the level a player will eventually reach.

Figure 5
FIGURE 5. (A) Charts the average time course in tournament performance for players
starting at 6–20 years of age grouped by improvement over the first three years. (B)
Shows the change in ELO per year.
Cohort Differences
There have been many changes in resources available for chess players since 1989. We
analyzed the time course in development of chess ratings separately for different cohorts
in order to explore whether deviations from the pattern predicted by the learning curves
varied in relation to the historical period that a chess career was started. Deviations from
the learning curve were not accounted for by cohort. Rather, for all 5-year cohorts from
1970 to 1990 and age-groups displayed in Figure 6, the increase in rating during the first
years of performance was linear or positively accelerated. The pattern of negative
acceleration (larger gains in earlier as compared to later years, compatible with the
learning curves) was not observed.

Figure 6

FIGURE 6. AVERAGE TIME COURSE IN TOURNAMENT PERFORMANCE FOR

PLAYERS STARTING AT 6–20 YEARS OF AGE GROUPED BY BIRTH COHORT (1970–
1990). Due to small n (11) we dropped the 14–17 year old starters born between 1970 and
1975.
Age-groups and cohorts differed more with respect to the rating level they started out
with (i.e., reached by end of their first year) than with respect to their level of
performance in Year 10. As already observable in Figure 5A, people starting to play
tournaments at younger age, started out at a lower level. In addition, Figure 6 shows that
later cohorts started at lower levels. This might be taken to suggest that players starting
young in late cohorts are the best candidates to track trajectories in chess performance
based on tournament ratings, while ratings of players starting older and earlier cohorts
might be shaped more strongly by off-tournament practice.

Gain in Rating from Games Played

The above analyses suggest that the success of the power function and the exponential
function in predicting development of chess performance might be rather limited due to
quantitative and qualitative misfit. Furthermore, the number of tournament games
played seemed to be linked to deviations from the learning curve. Therefore, we sought
to describe the extent to which early vs. late years in playing tournament chess are
related to gain in rating as well as performance level reached by Year 10. For this we used
games played per year and gain in rating per year. We applied Spearman rank order
correlations separately for each age group and year of tournament participation. Figure
7A shows that number of games played per year is related to between-person differences
in gain in rating. Participants playing more tournament games in a year tend to show a
larger increase in rating compared to those playing less games. This holds consistently
across age-groups and especially so for early years of tournament participation.
However, diminishing returns seem to be observable with respect to the extent to which
more tournament games can lead to an increase in rating. Figure 7B shows that the
relationship between (a) games played per year and (b) gain per games played per year
can become negative. Thus, overall it does not seem to be the case that playing more
tournament games can lead to an increase in efficiency in taking gains in rating from a
tournament game. For instance, those players starting tournament participation at age
10–13 who played more tournament games, seemed to show a reduced gain in rating per
tournament game played in their middle years.

Figure 7
FIGURE 7. SPEARMAN RANK ORDER CORRELATIONS PER AGE GROUP BETWEEN
PERFORMANCE VARIABLES TAKEN FROM SINGLE YEARS AND RATING GAIN
PER YEAR (A,B) OR OVERALL RATING GAIN (C,D).

The gain in rating that players show from Year 1 to Year 10 can be predicted by gain in
rating per year in early years of tournament participation. As depicted in Figure 7C, gain
in later years is less predictive of the overall increase in rating. While the power and the
exponential function would have predicted that we can observe large gains in rating in
early years, we thus, somewhat analogously, observer a larger predictive power of
between-person differences in early as compared to late years of chess tournament
participation. Apart from the gain per year, also the gain per year relative to the number
of games played per year could be used to predict the overall increase in rating between
Year 1 and 10 (Figure 7D). Participants who, during the first years of tournament
participation, efficiently increased their rating per games played, ended up at a higher
performance level than those, who did not show a large gain per games played during
early years.

Selective Attrition
Finally, we checked for selective attrition. While in our main analyses we only used 10
years of subsequent tournament participation, some participants provided records for
additional years (up to 19 years overall). Rank order correlations indicated that the
number of overall years of tournament participation per age group was neither
systematically related to gain between Year 1 and Year 10, nor the gain in rating within
the first 3 years (rs between −0.10 and 0.16).

Discussion
In the current work we have explored the potential of the power- and the negative
exponential learning functions to account for the development of chess performance
measured in ratings based on tournament outcomes. In line with re-evaluations of the
power law of practice (Heathcote et al., 2000; Haider and Frensch, 2002), we
documented that the exponential function was better than the power function in fitting
and predicting the time course of chess ratings over years of practice. However, a crucial
aspect shared by both of these mathematical functions and the underlying theories of
skill acquisition was not reflected in the data. While according to the power- as well as
the negative exponential function players should achieve large absolute gains early in
practice and small gains later in practice, this was not the case for many of the
participants. Rather, many players started to show substantial improvements only after
their first years of tournament participation. They were playing off the learning curves
suggested by skill acquisition theories. If expertise acquisition is not well described by
learning functions used to describe skill acquisition, the linking of underlying cognitive
processes of attention and learning that proceed on time-scales measuring milliseconds
to hours with learning processes that proceed on time-scales measured in years seems
much less straightforward than one could have hoped for (i.e., Anderson, 2002).

Many players showed an acceleration of gain in rating in the first years of tournament
participation, followed by a deceleration. Based on the power function and the
exponential function we would have expected to only find the latter. Newell et al. (2001)
suggested to mathematically and conceptually accommodate such findings by assuming
a mixture of learning processes taking place on different time scales. Acceleration
followed by deceleration could be captured by a sigmoid function that consists of two
exponential components, a positive (acceleration) and a negative one (deceleration).
Learning opportunities and efficiency in using them might increase during first years of
tournament performance for many players, while in later years returns of investing in
chess performance are diminishing. In line with this view, year-long trajectories of skill
acquisition might be better understood from a perspective that takes lifespan-
developmental and educational changes into account (Li and Freund, 2005). For
instance, players starting to take part in tournaments at a young age are likely to
promote changes in self-regulation strategies available (Lerner et al., 2001; Freund and
Baltes, 2002) and acquire the potential to shape their social and learning environment.
Their ability to learn about chess from (foreign language) media and options to travel to
and communicate with other players will increase. Deliberate practice (cf. Ericsson et al.,
1993) might require that young players develop skills to competently use of their
motivational resources, by, for instance, scheduling work on skill acquisition such that as
many of the activities as possible are intrinsically motivating (cf. Rheinberg and Engeser,
2010 as well as Christophel et al., 2014, for training of motivational competence).
Underlining this challenge, Coughlan et al., 2014 reported that participants in the expert
group of their study rated their practice as more effortful and less enjoyable compared to
other participants. The experts were successful in improving performance, by
predominantly practicing the skill they were weaker at. However, such gains in potential
to learn might for many players no longer compensate for the physical and social
changes faced during puberty (Marceau et al., 2011; Hollenstein and Lougheed, 2013), at
the end of adolescence, during secondary education, family formation or labor force
participation. Future research should thus try to simultaneously account for
development in the individual, the opportunities provided by the environment (cf. Ram
et al., 2013) and to model different trajectories in one framework (e.g., Grimm et al.,
2010; Ram and Diehl, in press).

For skill acquisition mechanisms such as chunking, negative exponential learning can be
explained by a reduction in learning opportunities (cf. Heathcote et al., 2000). The later
in practice, the fewer chunks are yet to be learned. While a deceleration of learning
should be observed late in practice, such an account does not preclude that strong
increases in learning opportunities early in practice can lead to an acceleration of chunks
acquired per time invested. It appears that, for at least some players, opportunities and
efficiency in increasing chess performance are already fully present at the time they start
to play tournaments. They start at the turning point of the sigmoid function. The “upper”
negative exponential portion of the sigmoid is sufficient to describe their performance
gains, which are large in their early years and then diminish as performance approaches
the asymptote. For other players, both positive and negative exponential portions of the
sigmoid function are needed to represent the dynamics of their chess performance over
time. These players appear to be less saturated with respect to learning opportunities
and efficiency when starting to take part in tournaments covered by the database. They
thus first show an acceleration in rating gains per year, followed by the deceleration
when approaching asymptote.

In line with these speculations, Howard (2014) reported an average trajectory of rating
increases showing deceleration only for International Chess Federation (FIDE) players
(rather than acceleration followed by deceleration). The shape of the curve reported by
Howard matches the exponential curve from Figure 1A. Starting at an average of about
2200 points, the sample mean increased beyond 2500 points with practice. Different
from the database used in the current study, the threshold to be listed in the FIDE
database is high (cf. Vaci et al., 2014 for a discussion of problems implied by restriction
of range in chess databases). Likely, players were already taking full advantage of
opportunities to improve chess performance when entering the database so that an
acceleration in rating gain with practice was no longer possible. Descriptive analyses
suggest that the dynamic in rating improvement that players at the international level
show with practice seems consistent with the negatively accelerated exponential
function. As implied by the exponential function, the relative learning rate (RLR)
estimated based on the average data published by Howard (2014) is constant. While the
power function should lead to a decrease of RLR with practice (cf. Heathcote et al.,
2000), the RLR is fluctuating around 20%. Focusing on the first half of practice in order
to avoid inflation of RLR at the end of the practice curve, we obtained an r = 0.11
correlation of RLR with time point. Thus there was no hint toward a decrease.

Our correlational analyses suggest that interindividual variability in rating gain over the
course of ten years of tournament participation can be predicted by between-person
differences in performance during the first years. Even by taking data from single years,
number of games played, rating points gained or rating points gained per games played,
allow to predict overall gain at a moderate level. While the power and the exponential
learning curve would suggest that the first years of practice should be important because
of the large performance gains, we thus can somewhat analogously conclude that the
first years are more important than later years for predicting between-person differences
in performance level reached on the long run (cf. Ackerman and Woltz, 1994).

We focused on examining changes in rating with year of practice (rather than number of
games played, cf. Howard, 2014). This allowed us to explore changes in rating gain and
rating gain per games played with age and cohort. Yet, a direct comparison of the
capability to capture performance change is lacking so far for the two potential time
scales, (1) number of games played, (2) chronological time in years, as well as (3) a
mixture of both scales. Several issues are worth considering when exploring the
complexity of models needed to account for expertise acquisition over years, as
compared to models of skill acquisition in hour-long laboratory sessions. In the lab,
quantity and quality of practice per unit of time is usually well controlled. In skill
acquisition processes outside the lab they might vary considerably over the years of
practice an individual engages in. In addition, potential cohort differences should not be
neglected (cf. Gobet et al., 2002; van Harreveld et al., 2007; Connors et al., 2011). Future
work should consider how data on both, quantity of practice and quality of practice, can
be used to explain the time course of chess skill development (cf. Baker et al., 2003;
Charness et al., 2005; Gobet and Campitelli, 2007; Howard, 2014). Apart from obtaining
data on the amount of off-tournament learning opportunities, available data sets could
be used to gauge variability in specific aspects of the learning opportunities. For
instance, taking part in tournaments with large spread in opponent strength might
provide more opportunities for improvement as compared to tournaments with more
homogenous competitors.

Applied Cognitive PsychologyVolume 19, Issue 2 p. 151-165

Research Article

The role of deliberate practice in chess

expertise
Neil Charness, Michael Tuffiash, Ralf Krampe, Eyal Reingold, Ekaterina Vasyukova
First published: 02 March 2005
https://doi.org/10.1002/acp.1106
Citations: 161

Two large, diverse samples of tournament-rated chess players were asked to

estimate the frequency and duration of their engagement in a variety of chess-
related activities. Variables representing accumulated time spent on serious study
alone, tournament play, and formal instruction were all significant bivariate correlates
of chess skill as measured by tournament performance ratings. Multivariate
regression analyses revealed that among the activities measured, serious study
alone was the strongest predictor of chess skill in both samples, and that a
combination of various chess-related activities accounted for about 40% of the
variance in chess skill ratings. However, the relevance of tournament play and formal
instruction to skill varied as a function of skill measurement time (peak vs. current)
and age group (above vs. below 40 years). Chess players at the highest skill level
(i.e. grandmasters) expended about 5000 hours on serious study alone during their
first decade of serious chess play—nearly five times the average amount reported by
intermediate-level players. These results provide further evidence to support the
argument that deliberate practice plays a critical role in the acquisition of chess
expertise, and may be useful in addressing pedagogical issues concerning the
optimal allocation of time to different chess learning activities. Copyright © 2005
John Wiley & Sons, Ltd.

Reference in chess study no. 3

Chess as a Powerful Educational Tool

for Successful People
Authors:
Jankovic, Alojzije
Novak, Ivan
Year of Publication:
2019
Citation:
[Editor:] Tipurić, Darko [Editor:] Hruška, Domagoj [Title:] 7th International OFEL Conference on
Governance, Management and Entrepreneurship: Embracing Diversity in Organisations. April 5th
- 6th, 2019, Dubrovnik, Croatia [Publisher:] Governance Research and Development Centre
(CIRU) [Place:] Zagreb [Year:] 2019 [Pages:] 425-441

The 7-Skills Chess Training Model

In the 7 Skills Chess Training Model I will show you the important chess
skills you need to train. I will also share suggestions on how to train these
skills!

1. Tactics
The first step to improve your tactics skill is to make an in-depth study of
tactical motifs and checkmate patterns. A consistent approach to studying
tactics is key. It will be more beneficial to solve 3-5 puzzles every day than
to binge occasionally.

2. Strategy
Your ability to find a suitable plan in the position relies heavily on your
evaluation skill. Working to improve your evaluation skill will also help you
find candidate moves that serve the need of the position at hand.

3. Calculation
Train your mind to calculate all the “forced variations” in the position, in
other words all the checks, threats and captures. The 5-step calculation
process presented in the free 10-day chess challenge is a great way to get
you started on this.

4. Openings
Don’t rely on memorization only. Aim to understand the typical
development ideas in the opening of your choice by going over master-level
games where the particular opening was played. Improve your personal
repertoire over time.

5. Visualization
Your visualization skill will naturally improve over time if you are
disciplined in your tactics and calculation training. I highly recommend
that you further boost your visualization skill by training with the
Visualwize program, available on the website.

6. Endgames
Study essential theoretical positions and important endgame ideas. For
example, the “principle of two weaknesses” is important endgame idea that
will serve you well in many endgame situations.

7. Evaluation
Study and work on your understanding of the 5 most important positional
elements. Again, the 10-day chess challenge (day 2) will get you started
toward improving your evaluation skill.
Qualitative descriptive research generates data that describe the ‘who,
what, and where of events or experiences’ from a subjective
perspective (Kim et al., 2017, p. 23). From a philosophical perspective,
this approach to research is best aligned with constructionism and
critical theories that use interpretative and naturalistic methods
(Lincoln et al., 2017). These philosophical perspectives represent the
view that reality exists within various contexts that are dynamic and
perceived differently depending on the subject, therefore, reality is
multiple and subjective (Lincoln et al., 2017). In qualitative descriptive
research, this translates into researchers being concerned with
understanding the individual human experience in its unique context.
This type of inquiry requires flexible research processes that are
inductive and dynamic but do not transform the data beyond
recognition from the phenomenon being studied (Ormston et al.,
2014; Sandelwoski 2010). Descriptive qualitative research has also
been aligned with pragmatism (Neergaard et al., 2009) where
decisions are made about how the research should be conducted
based on the aims or objectives and context of the study (Ormston et
al., 2014). The pragmatist researcher is not aligned to one particular
view of knowledge generation or one particular methodology. Instead
they look to the concepts or phenomena being studied to guide
decision making in the research process, facilitating the selection of
the most appropriate methods to answer the research question
(Bishop, 2015).

Perhaps linked to the practical application of pragmatism to research,

that is, applying the best methods to answer the research question, is
the classification of qualitative descriptive research by Sandelowski
(2010, p. 82) into a ‘distributed residual category’. This recognises and
incorporates uncertainty about the phenomena being studied and the
research methods used to study them. For researchers, it permits the
use of one or more different types of inquiry, which is essential when
acknowledging and exploring different realities and subjective
experiences in relation to phenomena (Long et al., 2018). Clarity, in
terms of the rationale for the phenomenon being studied and the
methods used by the researcher, emerges from the qualitative
descriptive approach because the data gathered continue to remain
close to the phenomenon throughout the study (Sandelowski, 2010).
For this to happen a flexible approach is required and this is evident in
the practice of ‘borrowing’ elements of other qualitative
methodologies such as grounded theory, phenomenology and
ethnography (Vaismoradi et al., 2013).

Regarded as a positive aspect by many researchers who are interested

in studying human nature and phenomenon, others believe this
flexibility leads to inconsistency across studies and in some cases
complacency by researchers. This can result in vague or unexplained
decision making around the research process and subsequent lack of
credibility. Accordingly, nurse researchers need to be reflexive, that is,
clear about their role and position in terms of the phenomena being
studied, the context, the theoretical framework and all decision-
making processes used in a qualitative descriptive study. This adds
credibility to both the study and qualitative descriptive research.

Go to:

Methods in qualitative descriptive research

As with any research study, the application of descriptive methods

will emerge in response to the aims and objectives, which will
influence the sampling, data collection and analysis phases of the
study.

Sampling

Most qualitative research aligns itself with non-probability sampling

and descriptive research is no different. Descriptive research
generally uses purposive sampling and a range of purposive sampling
techniques have been described (Palinkas et al., 2015). Many
researchers use a combination of approaches such as convenience,
opportunistic or snowball sampling as part of the sampling
framework, which is determined by the desired sample and the
phenomena being studied.

Purposive sampling refers to selecting research participants that can

speak to the research aims and who have knowledge and experience
of the phenomenon under scrutiny (Ritchie et al., 2014). When
purposive sampling is used in a study it delimits and narrows the
study population; however, researchers need to remember that other
characteristics of the sample will also affect the population, such as
the location of the researcher and their flexibility to recruit
participants from beyond their base. In addition, the heterogeneity of
the population will need to be considered and how this might
influence sampling and subsequent data collection and analysis
(Palinkas et al., 2015). Take, for example, conducting research on the
experience of caring for people with Alzheimer’s disease (AD). For the
most part AD is a condition that affects older people and experiences
of participants caring for older people will ultimately dominate the
sample. However, AD also affects younger people and how this will
impact on sampling needs to be considered before recruitment as both
groups will have very different experiences, although there will be
overlap. Teddlie and Fu (2007) suggest that although some purposive
sampling techniques generate representative cases, most result in
describing contrasting cases, which they argue are at the heart of
qualitative analysis. To achieve this, Sandelowski (2010) suggests that
maximum variation sampling is particularly useful in qualitative
descriptive research, which may acknowledge the range of
experiences that exist especially in healthcare research. Palinkas et al.
(2015) describe maximum variation sampling as identifying shared
patterns that emerge from heterogeneity. In other words, researchers
attempt to include a wide range of participants and experiences when
collecting data. This may be more difficult to achieve in areas where
little is known about the substantive area and may depend on the
researcher’s knowledge and immersion within the subject area.
Sample size will also need to be considered and although small sample
sizes are common in qualitative descriptive research, researchers
need to be careful they have enough data collected to meet the study
aims (Ritchie et al., 2014). Pre-determining the sample size prior to
data collection may stifle the analytic process, resulting in too much or
too little data. Traditionally, the gold standard for sample size in
qualitative research is data saturation, which differs depending on the
research design and the size of the population (Fusch and Ness, 2015).
Data saturation is reached ‘when there is enough information to
replicate the study, when the ability to obtain additional new
information has been attained, and when further coding is no longer
feasible’ (Fusch and Ness, 2015, p. 1408). However, some argue that
although saturation is often reported, it is rarely demonstrated in
qualitative descriptive research reports (Caelli et al., 2003; Malterud
et al., 2016). If data saturation is used to determine sample size, it is
suggested that greater emphasis be placed on demonstrating how
saturation was reached and at what level to provide more credibility
to sample sizes (Caelli et al., 2003). Sample size calculation should be
an estimate until saturation has been achieved through the concurrent
processes of data collection and analysis. Where saturation has not
been achieved, or where sample size has been predetermined for
resource reasons, this should be clearly acknowledged. However,
there is also a movement away from the reliance on data saturation as
a measure of sample size in qualitative research (Malterud et al.,
2016). O’Reilly and Parker (2012) question the appropriateness of the
rigid application of saturation as a sample size measure arguing that
outside of Grounded Theory, its use is inconsistent and at times
questionable. Malterud et al. (2016) focus instead on the concept of
‘information power’ to determine sample size. Here, they suggest
sample size is determined by the amount of information the sample
holds relevant to the actual study rather than the number of
participants (Malterud et al., 2016). Some guidance on specific sample
size depending on research design has been provided in the literature;
however, these are sometimes conflicting and in some cases lack
evidence to support their claims (Guest et al., 2006). This is further
complicated by the range of qualitative designs and data collection
approaches available.
Data collection

Data collection methods in qualitative descriptive research are diverse

and aim to discover the who, what and where of phenomena
(Sandelowski, 2000). Although semi-structured individual face-to-face
interviews are the most commonly used data collection approaches
(Kim et al., 2017), focus groups, telephone interviews and online
approaches are also used.

Focus groups involve people with similar characteristics coming

together in a relaxed and permissive environment to share their
thoughts, experiences and insights (Krueger and Casey, 2009).
Participants share their own views and experiences, but also listen to
and reflect on the experiences of other group members. It is this
synergistic process of interacting with other group members that
refines individuals’ viewpoints to a deeper and more considered level
and produces data and insights that would not be accessible without
the interaction found in a group (Finch et al., 2014). Telephone
interviews and online approaches are gaining more traction as they
offer greater flexibility and reduced costs for researchers and ease of
access for participants. In addition, they may help to achieve
maximum variation sampling or examine experiences from a national
or international perspective. Face-to-face interviews are often
perceived as more appropriate than telephone interviews; however,
this assumption has been challenged as evidence to support the use of
telephone interviews emerges (Ward et al., 2015). Online data
collection also offers the opportunity to collect synchronous and
asynchronous data using instant messaging and other online media
(Hooley et al., 2011). Online interviews or focus groups conducted via
Skype or other media may overcome some of the limitations of
telephone interviews, although observation of non-verbal
communication may be more difficult to achieve (Janghorban et al.,
2014). Open-ended free-text responses in surveys have also been
identified as useful data sources in qualitative descriptive studies
(Kim et al., 2017) and in particular the use of online open-ended
questions, which can have a large geographical reach (Seixas et al.,
2018). Observation is also cited as an approach to data collection in
qualitative descriptive research (Sandelowski, 2000; Lambert and
Lambert, 2012); however, in a systematic review examining the
characteristics of qualitative research studies, observation was cited
as an additional source of data and was not used as a primary source
of data collection (Kim et al., 2017).

Data analysis and interpretation

According to Lambert and Lambert (2012), data analysis in qualitative

descriptive research is data driven and does not use an approach that
has emerged from a pre-existing philosophical or epistemological
perspective. Within qualitative descriptive research, it is important
analysis is kept at a level at which those to whom the research
pertains are easily able to understand and so can use the findings in
healthcare practice (Chafe, 2017). The approach to analysis is dictated
by the aims of the research and as qualitative descriptive research is
generally explorative, inductive approaches will commonly need to be
applied although deductive approaches can also be used (Kim et al.,
2017).

Content and thematic analyses are the most commonly used data
analysis techniques in qualitative descriptive research. Vaismoradi et
al. (2013) argue that content and thematic analysis, although poorly
understood and unevenly applied, offer legitimate ways of a lower
level of interpretation that is often required in qualitative descriptive
research. Sandelowski (2000) indicated that qualitative content
analysis is the approach of choice in descriptive research; however,
confusion exists between content and thematic analysis, which
sometimes means researchers use a combination of the two.
Vaismoradi et al. (2013) argue there are differences between the two
and that content analysis allows the researchers to analyse the data
qualitatively as well as being able to quantify the data whereas
thematic analysis provides a purely qualitative account of the data
that is richer and more detailed. Decisions to use one over the other
will depend on the aims of the study, which will dictate the depth of
analysis required. Although there is a range of analysis guidelines
available, they share some characteristics and an overview of these,
derived from some key texts (Sandleowski, 2010; Braun and Clark,
2006; Newell and Burnard, 2006), is presented in Table 1. Central to
these guidelines is an attempt by the researcher to immerse
themselves in the data and the ability to demonstrate a consistent and
systematic approach to the analysis.

Coding in qualitative descriptive research can be inductive and

emerge from the data, or a priori where they are based on a pre-
determined template as in template analysis. Inductive codes can be
‘in vivo’ where the researcher uses the words or concepts as stated by
the participants (Howitt, 2019), or can be named by the researcher
and grouped together to form emerging themes or categories through
an iterative systematic process until the final themes emerge.
Template analysis involves designing a coding template, which is
designed inductively from a subset of the data and then applied to all
the data and refined as appropriate (King, 2012). It offers a
standardised approach that may be useful when several researchers
are involved in the analysis process.

Within qualitative research studies generally, the analysis of data and

subsequent presentation of research findings can range from studies
with a relatively minimal amount of interpretation to those with high
levels of interpretation (Sandelowski and Barroso, 2003). The degree
of interpretation required in qualitative descriptive research is
contentious. Sandelowski (2010) argues that although descriptive
research produces findings that are ‘data-near’, they are nevertheless
interpretative. Sandelowski (2010) reports that a common
misconception in qualitative descriptive designs is that researchers do
not need to include any level of analysis and interpretation and can
rely solely on indiscriminately selecting direct quotations from
participants to answer the research question(s). Although it is
important to ensure those familiar with the topic under investigation
can recognise their experiences in the description of it (Kim et al.,
2017), this is not to say that there should be no transformation of
data. Researchers using a qualitative descriptive design need to,
through data analysis, move from un-interpreted participant
quotations to interpreted research findings, which can still remain
‘data-near’ (Sandeklwoski, 2010). Willis et al. (2016) suggest that
researchers using the qualitative descriptive method might report a
comprehensive thematic summary as findings, which moves beyond
individual participant reports by developing an interpretation of a
common theme. The extent of description and/or interpretation in a
qualitative descriptive study is ultimately determined by the focus of
the study (Neergard et al., 2009).

Rigor

As with any research design, ensuring the rigor or trustworthiness of

findings from a qualitative descriptive study is crucial. For a more
detailed consideration of the quality criteria in qualitative studies,
readers are referred to the seminal work of Lincoln and Guba (1985)
in which the four key criteria of credibility, dependability,
confirmability and transferability are discussed. At the very least,
researchers need to be clear about the methodological decisions taken
during the study so readers can judge the trustworthiness of the study
and ultimately the findings (Hallberg, 2013). Being aware of personal
assumptions and the role they play in the research process is also an
important quality criterion (Colorafi and Evans, 2016) and these
assumptions can be made explicit through the use of researcher
reflexivity in the study (Bradshaw et al., 2017).
 Go to:

Challenges in using a qualitative descriptive design

One of the challenges of utilising a qualitative descriptive design is

responding to the charge that many qualitative designs have
historically encountered, which is that qualitative designs lack the
scientific rigor associated with quantitative approaches (Vaismoradi
et al., 2013). The descriptive design faces further critique in this
regard as, unlike other qualitative approaches such as phenomenology
or grounded theory, it is not theory driven or oriented (Neergaard et
al., 2009). However, it is suggested that this perceived limitation of
qualitative descriptive research only holds true if it is used for the
wrong purposes and not primarily for describing the phenomenon
(Neergaard et al., 2009). Kahlke (2014) argues that rather than being
atheoretical, qualitative descriptive approaches require researchers to
consider to what extent theory will inform the study and are
sufficiently flexible to leave space for researchers to utilise theoretical
frameworks that are relevant and inform individual research studies.
Kim et al. (2017) reported that most descriptive studies reviewed did
not identify a theoretical or philosophical framework, but those that
did used it to inform the development of either the interview guide or
the data analysis framework, thereby identifying the potential use of
theory in descriptive designs.

Another challenge around the use of qualitative descriptive research is

that it can erroneously be seen as a ‘quick fix’ for researchers who
want to employ qualitative methods, but perhaps lack the expertise or
familiarity with qualitative research (Sandelowski, 2010). Kim et al.
(2017) report how in their review fewer than half of qualitative
descriptive papers explicitly identified a rationale for choosing this
design, suggesting that in some cases the rationale behind its use was
ill considered. Providing a justification for choosing a particular
research design is an important part of the research process and, in
the case of qualitative descriptive research, a clear justification can
offset concerns that a descriptive design was an expedient rather than
a measured choice. For studies exploring participants’ experiences,
which could be addressed using other qualitative designs, it also helps
to clearly make a distinction as to why a descriptive design was the
best choice for the research study (Kim et al., 2017). Similarly, there is
a perception that the data analysis techniques most commonly
associated with descriptive research – thematic and content analysis
are the ‘easiest’ approaches to qualitative analysis; however, as
Vaismoradi et al. (2013) suggest, this does not mean they produce
low-quality research findings.

As previously identified, a further challenge with the use of qualitative

descriptive methods is that as a research design it has limited
visibility in research texts and methodological papers (Kim et al.,
2017). This means that novice qualitative researchers have little
guidance on how to design and implement a descriptive study as there
is a lack of a ‘methodological rulebook’ to guide researchers (Kahlke,
2014). It is also suggested that this lack of strict boundaries and rules
around qualitative descriptive research also offers researchers
flexibility to design a study using a variety of data collection and
analysis approaches that best answer the research question (Kahlke,
2014; Kim et al., 2017). However, should researchers choose to
integrate methods ‘borrowed’ from other qualitative designs such as
phenomenology or grounded theory, they should do so with the
caveat that they do not claim they are using designs they are not
actually using (Neergaard et al., 2009).

Go to:

Examples of the use of qualitative descriptive research in

healthcare

Findings from qualitative descriptive studies within healthcare have

the potential to describe the experiences of patients, families and
health providers, inform the development of health interventions and
policy and promote health and quality of life (Neergaard et al., 2009;
Willis et al., 2016). The examples provided here demonstrate different
ways qualitative descriptive methods can be used in a range of
healthcare settings.

Simon et al. (2015) used a qualitative descriptive design to identify

the perspectives of seriously ill, older patients and their families on
the barriers and facilitators to advance care planning. The authors
provided a rationale for using a descriptive design, which was to gain
a deeper understanding of the phenomenon under investigation. Data
were gathered through nine open-ended questions on a researcher-
administered questionnaire. Responses to all questions were recorded
verbatim and transcribed. Using descriptive, interpretative and
explanatory coding that transformed raw data recorded from 278
patients and 225 family members to more abstract ideas and concepts
(Simon et al., 2015), a deeper understanding of the barriers and
facilitators to advance care planning was developed. Three categories
were developed that identified personal beliefs, access to doctors and
interaction with doctors as the central barriers and facilitators to
advance care planning. The use of a qualitative descriptive design
facilitated the development of a schematic based on these three
themes, which provides a framework for use by clinicians to guide
improvement in advance care planning.

Focus group interviews are a common data collection method in

qualitative descriptive studies and were the method of choice in a
study by Pelentsov et al. (2015), which sought to identify the
supportive care needs of parents whose child has a rare disease. The
rationale provided for using a qualitative descriptive design was to
obtain a ‘straight description of the phenomena’ and to provide
analysis and interpretation of the findings that remained data-near
and representative of the responses of participants. In this study, four
semi-structured focus group interviews were conducted with 23
parents. The data from these focus groups were then subjected to a
form of thematic analysis during which emerging theories and
inferences were identified and organised into a series of thematic
networks and ultimately into three global themes. These themes
identified that a number of factors including social isolation and lack
of knowledge on behalf of healthcare professionals significantly
affected how supported parents felt. Identifying key areas of the
supportive needs of parents using qualitative description provides
direction to health professionals on how best to respond to and
support parents of children with a rare disease.

The potential for findings from a qualitative descriptive study to

impact on policy was identified in a study by Syme et al. (2016), who
noted a lack of guidance and policies around sexual expression
management of residents in long-term care settings. In this study, 20
directors of nursing from long-term care settings were interviewed
with a view to identifying challenges in addressing sexual expression
in these settings and elicit their recommendations for addressing
these challenges in practice and policy. Following thematic analysis,
findings relating to what directors of nursing believed to be important
components of policy to address sexual expression were identified.
These included providing educational resources, having a person-
centred care delivery model when responding to sexual expression
and providing guidance when working with families. Findings from
this qualitative descriptive study provide recommendations that can
then feed in to a broader policy on sexual expression in long-term care
settings.

The final example of the use of a qualitative descriptive study comes

from a mixed-methods study comprising a randomised control trial
and a qualitative process evaluation. He et al. (2015) sought to
determine the effects of a play intervention for children on parental
perioperative anxiety and to explore parents’ perceptions of the
intervention. Parents who had children going for surgery were
assigned to a control group or an intervention group. The intervention
group took part in a 1-hour play therapy session with their child
whereas the control group received usual care. Quantitative findings
identified there was no difference in parents’ anxiety levels between
the intervention and control group. However, qualitative findings
identified that parents found the intervention helpful in preparing
both themselves and their child for surgery and perceived a reduction
in their anxiety about the procedure thereby capturing findings that
were not captured by the quantitative measures. In addition, in the
qualitative interviews, parents made suggestions about how the play
group could be improved, which provides important data for the
further development of the intervention.

These examples across a range of healthcare settings provide evidence

of the way findings from qualitative descriptive research can be
directly used to more fully understand the experiences and
perspectives of patients, their families and healthcare providers in
addition to guiding future healthcare practice and informing further
research

Conclusion

Qualitative research designs have made significant contributions to

the development of nursing and healthcare practices and policy. The
utilisation of qualitative descriptive research is common within
nursing research and is gaining popularity with other healthcare
professions. This paper has identified that the utilisation of this design
can be particularly relevant to nursing and healthcare professionals
undertaking a primary piece of research and provides an excellent
method to address issues that are of real clinical significance to them
and their practice setting. However, the conundrum facing researchers
who wish to use this approach is its lack of visibility and transparency
within methodological papers and texts, resulting in a deficit of
available information to researchers when designing such studies. By
adding to the existing knowledge base, this paper enhances the
information available to researchers who wish to use the qualitative
descriptive approach, thus influencing the standard in how this
approach is employed in healthcare research. We highlight the need
for researchers using this research approach to clearly outline the
context, theoretical framework and concepts underpinning it and the
decision-making process that informed the design of their qualitative
descriptive study including chosen research methods, and how these
contribute to the achievement of the study’s aims and objectives.
Failure to describe these issues may have a negative impact on study
credibility. As seen in our paper, qualitative descriptive studies have a
role in healthcare research providing insight into service users and
providers’ perceptions and experiences of a particular phenomenon,
which can inform healthcare service provision.

Go to:

Key points for policy, practice and/or research

● Despite its widespread use, there is little methodological

guidance to orientate novice nurse researchers when using the
qualitative descriptive design. This paper provides this guidance
and champions the qualitative descriptive design as appropriate
to explore research questions that require accessible and
understandable findings directly relevant to healthcare practice
and policy.
● This paper identifies how the use of a qualitative descriptive
design gives direct voice to participants including patients and
healthcare staff, allowing exploration of issues of real and
immediate importance in the practice area.
● This paper reports how within qualitative descriptive research,
the analysis of data and presentation of findings in a way that is
easily understood and recognised is important to contribute to
the utilization of research findings in nursing practice.
● As this design is often overlooked in research texts despite its
suitability to exploring many healthcare questions, this paper
adds to the limited methodological guidance and has utility for
researchers who wish to defend their rationale for the use of the
qualitative descriptive design in nursing and healthcare research
Robert W. Howard. (2018). Development of Chess Skill from Domain Entry to Near
Asymptote. The American Journal of Psychology, 131(3), 323–345.
https://doi.org/10.5406/amerjpsyc.131.3.0323