Advances in Curling Game Information Analysis by Considering

Starting Position

Hiromu Otani

, Fumito Masui

, Hitoshi Yanagi

2,3

and Michal Ptaszynski

Graduate School of Kitami Institute of Technology, 165, Kouen-cho, Kitami, Japan

Common Course, Kitami Institute of Technology, 165, Kouen-cho, Kitami, Japan

High Performance Committee, Japan Curling Association, 1-1-1, Jin-nan, Shibuya-ku, Tokyo, Japan

Keywords:

Curling Informatics, Digital Scorebook, Tactics and Strategy, Game Information, Game Score, Shot Accuracy,

Starting Position, Correlation, Test for the Equality of Several Correlation Coefﬁcients.

Abstract:

Japanese curling teams have been recently preparing for Pyeongchang Winter Olympics in 2018. In curling,

there are three factors inﬂuencing game performance: physical factor, human factor, and strategic/tactical

factor. The strategic/tactical factor is considered as the most important at top level. To support the strate-

gic/tactical factor, we proposed the concept of Curling Informatics. As the ﬁrst step of Curling Informatics we

developed a digital scorebook iCE for digital collection of game information, storing it in a database and per-

forming further analysis to improve player performance. In this article, we further contribute to this project by

analyzing game information of world national top level teams. We have previously conﬁrmed that correlation

between shot accuracy and game score could differ with the team level or position. We also found out that

selected tactics and psychological pressure on opponent team has impact on game result. However, previous

analyses disregarded the order of teams in play, which could result in confusion of strategic tendencies or play

characteristics. In this paper, we carried out analysis of correlations of shot scores considering whether the

teams started as the ﬁrst or the second. We did this to specify the process of how the team strategy/tactics

inﬂuences game results.

1 INTRODUCTION

Japanese curling teams have greatly improved their

performance in recent years. For example, Japan na-

tional curling team won second place at the Women’s

World Championship 2016 in Saskatchewan, Canada,

thus winning a medal for the ﬁrst time. In addition,

both men and women players have been achieving ex-

cellent results, which allowed them to qualify for the

Pyeongchang Winter Olympics in 2018. In the back-

ground of this success, there are various strengthen-

ing programs developed for Japanese curling (Yanagi

and Miyakoshi, 2011; Takahashi, 2011; Masui et al.,

2016).

However, it is still necessary to continue working

on strengthening such support in order to make Japan

capable of obtaining a medal at the Pyeongchang

Winter Olympics, which will take place next year.

For example, Japan played against Switzerland three

times including round robin (type of a tournament in

which every team competes against every other team

in turns). This suggests that Japan should have al-

ready captured the tactics of Swiss team which should

help in winning a gold medal or at least improve

the team’s ranking during the Pyeongchang Winter

Olympics in 2018.

An Information Science approach can be men-

tioned as an example of a method for improving the

performance of a team. In the past few years a number

of cases to support sports with ICT (Information and

Communication Technology) have been reported (Fu-

jimura and Sugihara, 2004; Kagawa, 2006), including

curling (Masui et al., 2016). As an example of that,

our group proposed a new ﬁeld called Curling Infor-

matics (Masui et al., 2016).

As Figure 1 shows, the research ﬁeld of Curling

Informatics deals with strategic and tactical factors of

top curling teams and improves the strategic and tac-

tical skills of curling players. It also focuses on re-

alizing several support environments, such as record-

ing and referring to game information, tactical naviga-

tion, and assistance for reﬂection and tactical training.

Speciﬁcally, general plan developed within this ﬁeld

aims at implementing the methods to (1) collect, (2)

Otani H., Masui F., Yanagi H. and Ptaszynski M.

Advances in Curling Game Information Analysis by Considering Starting Position.

DOI: 10.5220/0006498800890095

In Proceedings of the 5th International Congress on Sport Sciences Research and Technology Support (icSPORTS 2017), pages 89-95

ISBN: 978-989-758-269-1

Figure 1: The concept of Curling Informatics (Masui et al., 2016).

analyze, (3) visualize and (4) share game information.

In the ﬁrst step, Masui et al. developed a digi-

tal scorebook iCE which runs on a tablet computer to

collect and analyze game information and they con-

ﬁrmed validity of the system (Masui et al., 2016). As

one of the applications of iCE, it is possible to check

the shot accuracy of each team or player based on col-

lected information sequentially and visually.

In the next step, we have analyzed the data col-

lected with the iCE application focusing on the re-

lation between shot accuracy and game score. Masui

analyzed iCE data and conﬁrmed that there is a strong

correlation between the difference in shot accuracies

and the difference in game scores and that correla-

tion between shot accuracy and game score differ with

level of team or player (Masui et al., 2016).

In addition, we (Otani et al., 2016) analyzed game

information of the Sochi Winter Olympics in 2014

and conﬁrmed that correlation of the world national

top level was lower than the Japanese top level and

that the selected strategies/tactics and psychological

pressure on the opponent team had an impact on game

result .

However, because this analysis did not consider

the order of teams in play (whether the team started

as ﬁrst, or second), it could contain confused infor-

mation about tendencies in strategies or play charac-

teristics.

In this paper, we advanced the analysis of the rela-

tion of shot accuracy with game score by considering

the starting positions (later called: Play First and Play

Second).

The outline of the article is as follows. Firstly,

we introduce the relevant related research in Section

2. Secondly, we describe the notions of team strategy

and shot accuracy in Section 3. Next, we provide an

overview of applied game information, and the ana-

lytical method. The results are explained in Section 5

and discussed in Section 6. In Section 7 we conclude

the paper.

2 RELATED RESEARCH

Factors inﬂuencing team’s performance in curling

include: the physical factor (ice condition), the

human factor (condition of curling player), and

the strategic/tactical factor (knowledge and tac-

tics/strategies). Bradley (Bradley, 2009) points out

the strategic/tactical factor as the most important at

top level.

As for the human factor, investigations about mo-

tion dynamics of curling stone by Shegelski et al. and

Denny et al. (Shegelski, 2000; Denny, 2002) are some

of the most known research. In recent years, Maeno

(Maeno, 2014) also reported a new motion dynam-

ics model for sliding the stone on ice. Regarding

the human factor, various research on how to train

the players and improve their condition and power

balancing, were proposed. Behm (Behm, 2009) pro-

posed an effective training method to improve the req-

uisite motion for curling. Yanagi et al. (Yanagi and

Miyakoshi, 2011) veriﬁed effective training for col-

lege level players by an experimental approach with

a college curling team. They also developed a new

sweeping brush which conveys the power of player’s

motion efﬁciently based on the analysis of player’s

sweeping motions from a viewpoint of biomechanics.

In addition, Tanaka introduced ICT (Information

and Communication Technology) for human factor

analysis (Tanaka et al., 2006). They attempted to

carry out an analysis of motion to deliver the stone by

utilizing a virtual model of curling environment and

players.

About the strategic/tactical factor, Igarashi et al.

proposed an application to inverse the problem for

curling by policy-gradient methods in Non-Markov

decision processes (Igarashi et al., 2007), Ura et

al. reported on calculation technique for analysis

of strategies/tactics based on game tree (Ura et al.,

2008). Also, Sung et al. analyzed game informa-

tion focusing on ﬁrst rock and last rock per each end

and points out that team strategies/tactics differed in

teams which had last rock per end and other teams

(Sung, 2013).

Masui et al. suggested that there is a strong cor-

relation between the difference in shot accuracies and

the difference in the game scores (Masui et al., 2016).

In addition, they observed that Japanese national class

is stronger than Japanese junior national class, and

that the performance of Japanese national class is

rarely inﬂuenced by missed shots. It means that the

game result could be predicted before the game ends

if we knew the difference in the playing teams’ shot

accuracy. Furthermore, the correlation in world class

becomes weaker than Japanese national class because

of smaller differences in shot accuracies. It could

also be possible that the correlation is negatively in-

ﬂuenced by outliers. It can be expected that we could

expose the process of how tactics or strategies affect

game result or difference in game scores by analyzing

the games of outliers.

Otani et al. analyzed correlation of 93 games in

Sochi Winter Olympics 2014 to verify the above as-

sumption (Otani et al., 2016). As a result, they con-

ﬁrmed that there is a strong correlation between the

difference in shot accuracies and the difference in

game scores. Although correlation of the world na-

tional top level was lower than the Japanese top level.

Additionally, they analyzed game of outliers by fo-

cusing on transition of teams’ shot accuracies for ev-

ery end of each game. As a result, they found out that

the winning team gave priority to the risk minimiz-

ing tactics and led to missed shots performed by the

losing team. Therefore, the selected tactics and pres-

sure on the opponent had an impact on game result in

which the team with superior shot accuracy lost due

to wrong strategical decisions

In curling, the team that plays as a second in turn

has a strong advantage. In other words, the plan taken

by each team differ depending on starting position.

Therefore, to extract strategic tendencies and speciﬁc

play styles, distinguish Play ﬁrst from Play second po-

sitions in the analysis of game information.

3 TEAM STRATEGIES/TACTICS

AND SHOT ACCURACY

Curling is a winter sport in which two teams compete

to obtain points by throwing 16 stones at the center of

a circular area called a house in an ice-based square

area called a sheet.

One team consists of four players. Each player

throws two stones (called shot) in rotation, and the

score is calculated by each rotation, till all 16 stones

are thrown. At this time, the team that has a stone at

the nearest position to the center of the house gets the

number of points equal to the number of stones on the

inside of the opponent’s stones in the house.

One rotation is called an end. One game consists

of eight or ten ends, but the game is extended if the

score is tied in the ﬁnal end (Howard, 2009; Coleman,

2014).

Curling is often called “chess on ice” because it

is a sport in which tactics plays an important role. It

requires a player to form complex strategies in search

for effective moves with consideration of the ice con-

dition, stone position in the house, etc.

Shots in curling are roughly divided into two

types: draw shot and takeout shot. In draw shots the

stone stops in the house. Takeout shots force opponent

stones out of the house. The thrown shot is given from

0 to 4 points by a team coach or a substitute player. In

the special case, such as takeout shot that hits more

opposite stones than advised by the skip, 5 points can

be given. These points are called “shot-score”, and

shot accuracy (SA) is calculated using shot-score ac-

cording to equation (1). SA has been one of the most

important measurements to estimate player’s skill and

condition.

SA = 25 ×

total of shot-scores

number of shots

(%) (1)

The team which has a higher SA is in general capa-

ble to throw more accurate shots. Therefore, it gives

advantage to the team in the game. The team which

has higher SA is more likely to succeed when per-

forming the shot in critical situations. And the team

which has lower SA would raise the probability for

scoring by opposite team.

In general, the team that plays as second has an

advantage in curling because they can perform the

last shot at an end. Accordingly, the strategies/tactics

taken by each team differ depending on whether their

playing position is Play ﬁrst or Play second. In some

cases, the strategies/tactics taken by the opposite team

are advantageous. In other cases, it is necessary to

take the opposite stone out of the house to set advan-

tageous stone position, and put the stone in the house

to get a higher score.

The iCE application compiles SA databases from

the sets of shot scores, to analyze the scores from a

variety of viewpoints. To grasp the characteristics

and tendencies of each team, we analyzed in detail

information in iCE’s database containing shot scores.

The analysis result could lead to ﬁnd the difference of

strategy between Play ﬁrst team and Play second team

in an objective way.

We formulated a hypothesis that “the relation be-

tween SA and ﬁnal game score (FGS) are different in

Play ﬁrst and Play second”, and try to verify the hy-

pothesis in the following parts of the paper based on

the game information database.

4 GAME ANALYSIS

CONSIDERING PLAY FIRST

AND PLAY SECOND

To verify the hypothesis mentioned in Section 3, we

compared the relation between difference in SA and

difference in FGS (DFGS) for game information in

Play ﬁrst and Play second. Figure 2 shows the analy-

sis procedure.

Figure 2: Analysis procedure.

Speciﬁcally, we extracted the DFGS and shot ac-

curacy for each collected game. We deﬁne the mean

value of SA regarding all ends as Total Shot Accuracy

(T SA), the mean value of SA regarding end for Play

ﬁrst teams as Play ﬁrst Shot Accuracy (1stSA) , the

mean value of SA regarding an end for Play second

teams as Play second Shot Accuracy (2ndSA).

Furthermore, we examined the correlation be-

tween DFGS and difference in SA

As the target data we used the game informa-

tion of 93 games (45 games for men, 48 games

for women) covering around 15,000 shots in Winter

Olympic Games 2014 as game data of world national

top level. We used the data of World Curling Federa-

tion as a reference

5 RESULT OF ANALYSIS

5.1 Comparison of 1stSA and 2ndSA

Here we describe the result of comparative analysis

of 1stSA and 2ndSA data. Firstly, we carried out the

Shapiro-Wilk test which is one of the tests of nor-

mality to conﬁrm normalization for data aggregate of

DTSA, D1stSA, D2ndSA, DFGS in target data.

As a result, we conﬁrmed that all data aggregate

follows a normal distribution. It means the data can

be considered as aggregate extracted at random.

Based on the above, we carried out the T-test

which is one of parametric methods directed at data

aggregate of D1stSA and D2ndSA to investigate if

the differences between the two data aggregates are

statistically signiﬁcant.

We considered that D1stSA and D2ndSA are re-

lated data because these data were distinguished from

the same team’s SA. Therefore, we carried out the

paired T-test.

The results were conﬁrmed two that the data ag-

gregates did not have statistical signiﬁcance.

5.2 Relation between SA and GS

Here we describe the relation between SA and GS

separately for Play ﬁrst and Play second cases. Figure

3 shows a diagram representing a correlation between

DTSA and DFGS and a regression line. In Figure 3,

the X axis shows DTSA for each game and the Y axis

represents the DFGS.

Pearson’s correlation coefﬁcient between the two

differences for total data was 0.670. It means that the

two differences in Figure 3 have positive correlation.

Similarly, we investigated the correlation between

P1stSA and DFGS, P2ndSA and DFGS. Figure 4

shows a diagram representing a correlation between

Hereinafter, this is called Difference in Total Shot

Accuracy (DT SA), Difference in Play ﬁrst Shot Accu-

racy (D1stSA), Difference in Play second Shot Accuracy

(D2ndSA).

https://www.olympic.org/sochi-2014/curling

Figure 3: Correlation between DTSA vs. DFGS.

D1stSA and DFGS and regression line and Figure 5

stands for a correlation between D2ndSA and DFGS.

The correlation for Play ﬁrst teams was 0.557 and for

Play second teams it was 0.530.

Figure 4: Correlation between D1stSA vs. DFGS.

Figure 5: Correlation between D2ndSA vs. DFGS.

Next, we analyzed the correlation for male and fe-

male players using a similar technique. Table 1 shows

the correlation between between each difference in

SA and DFGS separated by sex.

As shown in Table 1, there was a tendency that the

correlation between DTSA and DFGS was stronger

Table 1: Pearson’s correlation between each difference in

SA and DFGS separated by sex.

All games men women

Total 0.670 0.634 0.707

Play ﬁrst 0.557 0.564 0.545

Play second 0.530 0.532 0.547

than when a team was in Play ﬁrst and Play sec-

ond positions. Therefore, we carried out the test for

the hypothesis that several correlations are estimates

of the same correlation (Paul, 1989) to examine the

Pearson’s correlation between each difference in SA

and if DFGS were statistically signiﬁcant.

As a result of the test, there were no signiﬁcant

differences between the three correlations (DTSA and

DFGS, D1stSA and DFGS, and between D2ndSA and

DFGS).

Similar results were obtained from game informa-

tion for men and women, there were no signiﬁcant

differences in all cases.

These results mean that the relation between

DTSA and DFGS, D1stSA and DFGS, D2ndSA and

DFGS were correlated to the same degree in 93 games

from the applied data and the same was true of 45

games for men and 48 games for women.

6 DISCUSSION

In this section, we describe the discussion on analy-

sis results. Primarily, there were no signiﬁcant differ-

ences between correlations when teams were in Play

ﬁrst and Play second positions. Therefore, it dis-

missed the previously proposed hypothesis that “the

relation between SA and GS are different in Play ﬁrst

and Play second”. Below we present a detailed dis-

cussion to explain this situation.

We focused on SA in target data. Figure 6 shows

SA in analysis subject. In Figure 6, the X axis shows

SA and the Y axis represents the TSA, 1stSA, 2ndSA.

Figure 6: SA divided by sex and playing position.

As Figure 6 shows, there was a tendency that

1stSA were higher than 2ndSA through the whole

match. For men, the 1stSA was 84.26%, the 2ndSA

was 82.17% and difference between Play ﬁrst and

Play second was 2.09 point. In the SA for women,

the 1stSA was 81.84%, the 2ndSA was 78% and dif-

ference between Play ﬁrst and Play second was 3.84

point. These results mean there is a difference of per-

forming 1 or 2 shots per a game.

In addition, SA exceed 80% in all items, except

2ndSA for women. As a reason for 2ndSA for women

being lower than others, men players can better adjust

the rapidity and curl width of a stone by sweeping

than women. This indicates that men could be more

adjustable than women.

It suggests that there is a difference depending on

the starting position and it may be related to shot op-

tion and its degree of difﬁculty.

Next, we investigated the data by focusing on how

many of which types of shots were thrown. Figure 7

shows the Ratio of each types of shot in Play ﬁrst and

Play second positions.

Figure 7: Ratio of shots by types in Play ﬁrst and Play sec-

ond positions.

The ratios of each type of shot were different be-

tween Play ﬁrst and Play second. In particular, the

ratio of Guard shot which is one o the draw shots for

Play ﬁrst were performed more often than in Play sec-

ond. In the Guard shot the stone stops in the area out-

side the house to block opponent stones from entering

the house.

On the other hand, in Play second, the ratio of

the Come - around shot which goes around the guard

stone and stops in the house was higher than in Play

ﬁrst.

It can be guessed that the shot requested and its de-

gree of difﬁculty differ depending on starting position

If a shot is weak or turns aside from the desired course,

other players sweep the ice surface ahead of the moving

stone to adjust the course.

because the player has to perform a shot considering

the state of one’s own team or the match situation.

This supports our hypothesis that the strate-

gies/tactics taken by each team differed in Play ﬁrst

and Play second starting positions.

From the above, we can propose the following rea-

sons for the lack of signiﬁcant differences between

the relation of shot accuracies and game scores when

teams were in Play ﬁrst and Play second positions.

Firstly, game information for world national top level

has sufﬁciently high SA. Secondly, the SA was not

parametric which characterizes starting position (Play

ﬁrst or Play second).

In the near future, it is necessary to compare the

analysis toward game information other than World

national top level. Additionally, it is necessary to ex-

tend the analysis on other effective parameters (e.g.

throwing number by type of shot) other than SA.

7 CONCLUSION

In this paper, we performed an analysis of game in-

formation by considering Play ﬁrst and Play second

positions from a number of curling game matches.

We applied the test for the equality of several cor-

relation coefﬁcients directed at each difference in shot

accuracies and game scores revealed that, there were

no signiﬁcant differences in all cases.

The result conﬁrmed that there were no signiﬁcant

differences between shot accuracies and game scores

when teams were in Play ﬁrst and Play second posi-

tions because players kept a high performance in ei-

ther case, and retained a stable shot accuracy through-

out all ends of the game.

Furthermore, we discussed why it was not possi-

ble to conﬁrm a signiﬁcant difference and indicated a

need for other inﬂuential parameters than shot accu-

racy.

In the near future, we plan to record game infor-

mation of world national top level and analyze it fo-

cusing on parameters other than shot accuracy (e.g.

throwing number by type of shot) in detail. In ad-

dition, we will aim to propose techniques for strate-

gies/tactics analysis considering strategic characteris-

tics.

ACKNOWLEDGEMENTS

This work was supported by JSPS KAKENHI (grant

number: 15H02797).

REFERENCES

Behm, D, G. (2009). Periodized training program of

the canadian olympic curling team. Journal of Na-

tional Strength and Conditioning Association JAPAN,

Vol.16, No.3, pp.40-47.

Bradley, L, J. (2009). The sports science of curling. A Prac-

tical Review, Journal of Sports Science and Medicine

vol.8, pp.495-500.

Coleman, G. (2014). Introduction to curling strategy (en-

glish edition). Amazon Services International.

Denny, M. (2002). Curling rock dynamics: Towards a real-

istic model. Canadian Journal of Physics, Vol.80, pp.

1005-1014.

Fujimura, A. and Sugihara, K. (2004). Quantitative eval-

uation of sport teamwork using generalized voronoi

diagrams. IEICE Transactions D J87-D2:818-828.

Howard, R. (2009). Curl to win: Expert advice to improve

your game. HarperCollins Publishers Ltd.

Igarashi, H., Ishihara, S., and Kimura, M. (2007). A study

of policy-gradient methods in non-markov decision

processes : Curling game application. IEICE Tech-

nical Report, NC2006-148, pp.179-184 (in Japanese).

Kagawa, M. (2006). Effect of multimedia information on

web pages in physical training class of university.

Journal of Japan Society for Educational Technology

29 37-40.

Maeno, N. (2014). Dynamics and curl ratio of a curling

stone. Sports Engineering Vol.17, pp. 33-41.

Masui, F., Hirata, K., Otani, H., Yanagi, H., and Ptaszynski,

M. (2016). Informatics to support tactics and strate-

gies in curling. Int J of Automation Technology10(2).

Otani, H., Masui, F., Hirata, K., Yanagi, H., and Ptaszyn-

ski, M. (2016). Analysis of curling team strategy and

tactics using curling informatics. 4th International

Congress on Sport Sciences Research and Technology

Support.

Paul, S, R. (1989). Test for the equality of several correla-

tion coefﬁcients*. The Canadian Journal of Statistics,

Vol.17, No.2, pp.217-227.

Shegelski, M. (2000). The motion of a curling rock: Analyt-

ical approach. Canadian Journal of Physics, Vol.78,

pp. 857-864.

Sung, G, P. (2013). Curling analysis based on the possession

of the last stone per end. 6th Asia-Paciﬁc Congress

on Sports Technology, Procedia Engineering, Vol.60,

pp.391-396.

Takahashi, S. (2011). Support the japan womens curling

national team by a trainer. Journal of Training Science

for Exercise and Sport 23(1):7-12 (in Japanese).

Tanaka, Y., Tsubota, H., Takeda, Y., and Tsuji, T.

(2006). Analysis of human hand movements using

a virtual curling system. Proceedings of Confer-

ence on Robotics and Mechatronics, 1A1-3F-E1 (in

Japanese).

Ura, M., Endo, S., Miyazaki, S., and Yasuda, T. (2008).

Curling game simulation and strategies evaluation.

Proceedings of the Virtual Reality Society of Japan,

Annual Conference, Vol.13, 2B3-5 (in Japanese).

Yanagi, H. and Miyakoshi, K. (2011). Training science on

college curling team. Journal of Training Science for

Exercise and Sport 23(1):13-19.