this is the beginning of a new web project by Lance Wicks.

The idea here is to develop and share Judo analytical tools for the statistical analysis of Judo competitions. The initial project this site will be covering is a notational analysis of the Beinjing 2008 Olympic Judo competition. This is being hand notated and the results will be shared here.

Please visit regularly as the site shall start shanging rapidly over the next few months.

Lance.

The purpose of the notation was to do a statistical analysis looking to prove the hypothesis that the number of attacks made by a player had a correlation with winnning the match. The results are shown below:
Summary of the 2006 Commonwealth Judo Tournament.

Their 1987 paper “Structure of the contest and work capacity of the judoist” has been referenced and used as the basis for further research by the likes of Franchini and Sterkowicz. The work describes both the structure of matches and also the physiological elements and is of interest to researchers and coaches today.

The paper identified the periods of work (Hajime – Matte) and found that these were usually no longer than 30 seconds in duration (decreasing as the match progresses). The breaks between the work periods was on average 10 seconds (increasing as the match progresses). The trend of the duration of the work periods and rest periods continues until they are near equal in the final minutes of a match.

The paper also identified the most prevalent techniques:

• Seoi Nage
• Ko Soto gari
• Uchi Mata
• O Uchi gari
• Ko Uchi gari

The study also identified that the most important tactical factor was the ability to cause an opponent to be penalised for passivity.

In my pilot study in 2006 of the Commonwealth Judo Tournament, almost 20 years later, the 30s per work period was near that of Sikorski et al. Although the 30 second figure from the original paper is taken from the conclusion (abstract), where as in the actual text the average is described as being 25 seconds. Perhaps indicating that in two decades the average work period increased by 5 seconds? This may also be supported by the average length of the rest period being 14 seconds in 2006.

Another factor to consider is the rule changes between 1987 and 2006.
Also the level of competition observed, perhaps the work period is affected by the level of athletes competing.

These observations were in part the inspiration for my current analysis of the 2008 Beijing Olympic Judo Tournament. Again, there are factors that prevent an ideal comparison as the Olympics is a very different tournament/event.  The 20 year progression in the sport and the rules may have affected the results.

If the results are quite different, what does this suggest? If they are similar what does this suggest?

In summary, the Sikorski et al (1987)  paper is important reading for anyone looking at Judo analysis and metrics. The paper has been citedin many other works and seems to form the basis for much of the later work in the area. Essential reading perhaps?

Lance

As you can see from the table, athletes are awarded ranking points depending on the position they reach and the event they are competing in.

So, win the World Championships and get 500 points, win one of the new masters events and get 400. Third in my native OJUs and you’ll get 32 points.

Come 30th April 2012, the players ranked in the top 22 for men, or top 14 for women (per category) are the qualifiers for the London Olympics.

The system gets more complicated from here on in, only the top 5 results per year are counted. So if you were to win the 2009 World Cup, Grand Prix, Grand Slam, Masters,World Championships and the OJUs; you would receive 1400 points. However, these points will not actually be worth 1400 in 2012, they will in fact only be worth 350 points as the points “degrade” by 25% per year (The dividing line is the beginning of the month in which the tournament ‘s first
competition day, was held).

Now the rest of the places in the games will be chosen by the Continental unions. Each continental union gets half the number of National Federations affiliated to each that union. So the EJU which has 50 member states and would receive 25 places for athletes who do not make the top 22/14.  These will be decided basically by the top points scorer in the continent, until the fill quota is used, with a maximum of two athletes per category being qualified by this method.

Discussion
The new system starts with all athletes having zero points. So this weekend is the first opportunity for them to start qualifying for 2012.
It is however based on the result in the tournament not on the athletes you beat. If you happen to have the easiest draw in history at the World Championships it is worth just as much as if you fight all the top athletes in your category. Will this work well is to be decided, I have not seen what research was done in deciding on this system.

Personally, I am more in favour of a system like the ELO system used in Chess (and online gaming like Xbox Live). The ELO system takes into consideration the probabilities involved in one athlete beating another. If a player who is much lower on the ELO ranking beats a higher ranked player, they receive a larger number of ranking points than if they beat someone below them. It is my intention, if I can obtain all the draw sheets, to run the ELO system and monitor the rankings to see if it ranks athletes over the 4 years differently.

In both conversations my response has been that if we can identify the factors needed, then we should be able to predict future results. As per Max Cohen in the movie Pi, I decided to “re-state my assumptions” as I think it will help set the context for this website better for visitors.

Assumption 1: If we can identify the right metrics, we can predict results.

So, what I mean is if we knew what caused a player to win a Judo match, then we could track this for two players before (or during) a match and predict the victor.

Evidence: If player 1 has beaten player 2 ten times in ten matches, then we could fairly comfortably predict that player 1 will beat player 2 in their next match. If player 3 has won all their pre-liminary fights by Ippon we can start predicting that Player 4 will lose to player 3 by Ippon.

The “ah yes, but…“:

The problem is that Judo is very complex, (argueably) more so than other sports. In Rugby Union for example, the team that retains possesion and territory will generally win. It is not always the case, but it is a performance metric that works in a sport where there are (again argueably) more variables to consider than in Judo. There are 30 players not 2 (more if you count subsitutions), there is weather conditions etc. Yet the simple metrics of possesion and territory can give a pretty good prediction of results.

So rather than say it is not possible to predict Judo using metrics, I argue that it is possible and that we just have yet to research well enough to find the metrics that matter.

Whether we ever are able to measure enugh worthwhile information to predict results of individual matches… we shall see. Also, whether we will be able to use these metrics for anything other than academic use is questionable also. I am not sure that we can coach players to fight certain ways as the statistics say it will result in a victory. So perhaps Judo metrics will never be an “applied science”.

Personally, I suspect that performance metrics will prove useful.
Assuming we can discover the right things to measure and we are able to interpret the results appropriately. “Knowledge is power” they say, we already know that countries like Germany and France are compiling information/knowledge about the players their players will meet, in terms of throws they use etc. It is not that big a leap from this to them also collecting statistics on throw frequencies and scoring ratios etc.

I suspect that we could increase our success in Judo if we were able to analyze more knowledge (metrics) about players from other countries.

Talent ID and Judo Metrics

I believe that we can predict the result of matches through Judo metrics, I also believe that we can/could predict who would be a good Judoka; if we can assess enough variables. We may not be able to predict accurately, but we could identify likely candidates. We have some evidence to support this via the soviet era sport and more recently in China.

Summary

It may well be the case that JudoMetrics is as realistic science as Psychohistory and it may well be “bunkum“, but IMHO it is worth researching further, at least to a point where we have a better view of it and can decide if it is the future or snake oil. It may be that on this site I am pursueing an idea as mad as Max Cohen’s in Pi…  time will tell.

Summary of the 2006 Commonwealth Judo Tournament.

Introduction:
At the 2006 Commonwealth Tournament, a study of the attacks,
scores and durations of bouts was made. Eighty-nine bouts were successfully notated. The results
have been collated and analysed and this document is a summary of the findings.
Full details on the study and methodology are being developed so to make them available to
everyone. The hope is that this will encourage others to conduct similar studies.
Statistics:
Descriptive Statistics
Total Fights 89
Total Scores 139(no penalties)
Total Attacks 1305
Total Penalties 100
Total Segments 732
Total Match Time 259Minutes 101.45Recovery time in fights
Total Actual Time 360Minutes 1.14Recovery per fight
Total Possible Contest Time 432Minutes 0.14Between segments
Fights won by Blue 47 53%
Fights won by White 42 47%
Fights won by person who attacks most 61 69%
Fights won by person who attacks least 28 31%
69 33%
21 10%
61 29%
58 28%
Total: 209
Scores per fight (AVG) 1.56
Attacks per fight (AVG) 14.66
Penalties per fight (AVG) 1.12
Segments per fight(AVG) 8.22
Of Actual time on mat 72%
Of time allocated 83%
Scores per segment 0.19
Attacks per segment 1.78
Penalties per segment 0.14
Ippons Scored
Wazaris Scored
Yokus Scored
Kokas scored
(incl Penalties)
Findings for players and coaches:
By averaging out the results of the data collected, we are able to describe an average Judo bout at
the Commonwealth Tournament level.  An average fight at commonwealth level consists of the
following:
Each match is approximately four minutes long, and consists of eight “segments” of action.
Each segment of approximately 30 seconds in duration, with 14 seconds between each segment.
Within each segment we can expect 1-2 attacks only before Matte is called.
We can expect a score every 7-8 segments, this score will be Ippon 1/3rd of the time. The other
two thirds of the time it will be divided almost equally between Koka & Yoku. Wazari will be
scored infrequently (only 10% of the overall scores).
Given this description of an average match, we are able to develop coaching strategies and
training sessions to best simulate commonwealth level Judo. Sessions could be developed
following approximately the following format.
Endurance drill:
Athlete attacks at near maximal level for 30 seconds.
They then recover for 14 seconds
The above two steps are repeated 8 times.
This drill helps develop the athletes ability to give maximum effort for entire match for difficult
matches.
Tactical Drill:
Athlete fights for grip for 10 seconds
Coach calls “NOW”, and athlete must make one large attack
Athlete then continues to grip fight for 10 seconds
Coach calls “NOW”, and athlete must make one large attack
Athlete recovers for 14 seconds
Repeat above steps eight times.
This drill trains the athlete to make maximum use of the available time. Minimising risk of
passivity attacks whilst keeping energy expenditure minimal.
Active defence Drill:
Athlete is attacked constantly for 30 seconds, they must only defend.
At 10 seconds the coach shouts “NOW”, the athlete must make some form of positive attack.
At 20 seconds the coach shouts “NOW”, the athlete must make some form of positive attack.
At 30 seconds the coach shouts “NOW”, the athlete must make some form of positive attack, this
attack must be “terminal”, concluding on the floor or outside safety area.
Recover for 14 seconds
Repeat above steps eight times.
Summary
This document is a very basic analysis of a small amount of information from the 2006
Commonwealth Tournament. It is hoped that this documents shows how this form of study and
analysis can provide interesting insights which can be applied to training programme
development.
Further analysis of the data is under way and a more detailed document will follow.
The use of simple mean averages provides generalised information which provides only an
indication of general trends in the data analysed. This needs to be considered when developing
training programmes.
For example, the four minute figure mentioned in this document is a mean average of all the
fights recorded. The range of durations went from a few seconds to over twelve minutes spent on
the mat.
Similarly, the mean averaged number of segments, covers all stages in a competition. Initial
examination of the data showed a visible change in contest structure in the later stages of a
category. This included more segments, hence more attacks, but with each segment being sorter.
Your athlete may be better served by drills that followed this pattern over the average format of a
Commonwealth level Judo match
Full details of this research are available via the www.JudoCoach.com
website and/or by contact the author, Lance Wicks, directly at the email
address: lw@judocoach.com. Fellow researchers are invited to contact Lance
Wicks to source the data and digital copies of the notation forms, etc.
It is hoped that this research will act as a catalyst, encouraging further
research within the sport by researchers both with and without experience.
Kia Kaha, Kia Toa, Kia Manawanui
Be brave, Be Strong, Be Perservering
(Old New Zealand Maori Saying)
(c)2006, Lance Wicks. www.judocoach.com lw@judocoach.com

Method:

• Ages and results were obtained from the JudoInside website for both events.
• Data was analyzed for medalists only
• Statistics between winners and silver and bronze medalists is compared.

Results:

• 24.95    Average Age of medalist
• 34    Max oldest medalist
• 17    Min youngest medalist

Winners/Gold Medalist:

• FEMALE:  Average age: 25.14(+-5.67)    Max: 34    Min: 19
• MALE: Average age: 26.2(+-4.66)    Max: 32    Min: 21

Non-Winners/Silver and Bronze:

• FEMALE: Average age: 25.15(+-3.62) Max: 31 Min: 17
• MALE: Average age: 24.4(+-3.62) Max: 33    Min: 20

Discusssion:

The sample size for here is very small, a mere 52 data points (of a possible 56).The analysis is in no way comprehensive.

Target age of players to compete in these events:
The ideal target age to be ready to compete in these events is between  19.47 and 30.81 for women and 21.54 and 30.86 for men. How this relates to the 2012 games is a matter of debate, but these target ages suggest either that athletes winning in London will be approximately 28, or that the older athletes competing today will be retired by London 2012, thereby keeping average ages down.

The age span is quite large 17-34, with average winners age being 25-26 years of age. This perhaps indicates younger athltetes need to spend several years at this level before winning. This could be tested by a further examination of ages for those who did not make the medals potentially.

Talent ID:
For countries looking for players to medal at London 2012 this brief analysis could suggest that they need to be focusing on players who are presently around 22 years of age. For the next Olympic games in 2016 a further talent squad could/should be identified who are approximately 18 years old. This could be extended to 2020, 14 years old now.

Notation Methodology:

Take a sheet of paper, draw a line down the middle of the page and notate as follows:

On the paper based hand notation form each attack will be recorded as a vertical line “|” if the
opponent does not touch the floor with another part of the body other than the feet. A “+” will be
recorded if the attack does cause the opponent to touch the floor with another part of their body but
no score is given. The form has two columns, one for each player, the “|” is placed in the
appropriate column for each player. If a score was given, no “|” is marked,  a letter indicating the
score is written instead. These are K for Koka,  Y for Yuko, W for Wazari, I for Ippon and P for a
penalty. When Matte was called, the recorder moves down one line on the notation form. If no
attack has been made a horizontal line is recorded “-”, is neither player has attacked, then both
columns have a “-” recorded. Each period between the referee’s “Hajime” call and “Matte” call is
termed a segment and recorded on a separate line on the form.

If you could forward your notation to me (lw@judocoach.com) that would be terrific, it will be interesting to see how many people participate and also how much the data varys between people. I shall keep your data anonymous of course.

Many Thanks,

Lance.

Recently a colleague of mine sent me a study he conducted some time back observing British fighter Craig Fallon in action at the 2003 European Championships. Dave Elmore analysed all the fights Craig had on the day against Paischer, Uematsu, Morokhovets, Nazaryan and Khergiani. He carefully recorded the time spent in the following categories:

• Adjusting Judo Gi
• Gripping
• Ne – Waza (grappling)
• Hajime
• Matte
• Attacking – Tachi – Waza
• Defending Tachi – Waza

Dave is the Judo Development officer at the University of Wolverhampton in the UK. He also runs the very good Advanced Apprenticeship Judo Blog. He looked at the data on a fight by fight basis and also overall, which produced the following chart:

As you can see from the chart, very little time was spent by Craig Fallon attacking in Tachi Waza. A majority of the matches was taken up with Ne Waza, Kumi Kata and adjusting Judo suits. Of the 19 minutes and 23 seconds spent in the competition, only 33 seconds of that time was actual attempts at throwing. Now there are lots of reasons for this, not least of all the speeed at which attacks happen. What stands out is the percentage of time that Craig spent fighting for a grip (29.41%), which shows the importance kumi kata perhaps.

31.30% of Craig Fallon’s time was spent in Ne Waza, it would be interesting to conduct a wider study looking at an entire category or event to see if this is a feature of Craig Fallon’s Judo or of light weight Judoka, or not.

The amount of time spent adjusting the Judo suits is also an eye opener, not an area of Judo that I have seen recorded before and as the chart shows quite a large part of the fight was spent just getting dressed (26.57%). Again this may be a feature of the fighting style of Craig and his opponents, it would be interesting to see if this amount of time is consistent across weights and genders.

If these numbers were to be proven to be standard across Judo, then it would have implications for coaching. Kumi Kata is growing in acceptance as an important skill that must be practised, this study supports this idea. This study could be used to justify increasing the amount of time drilling and practising kumi kata in club training sessions.

The amount of time spent adjusting the Judo Gi may also be important in terms of recovery time between segments of action during a match. Further study might provide suggestions to train for shorter bursts of action, or perhaps more grip related training similar to that shown on the Judo-Sport blog recently (via www.planetjudo.com).

Dave has kindly provided me with his data and I shall be looking at it in more detail and will post a follow up in a future article.

The British Judo Association (BJA) maintains a record of all Dan grades (Black belts and above) within their orgainsation. They also make this  information partially available via their website at http://britishjudo.org.uk/technical/grading_dan_register.php which means that I was able to peruse that data and make some basic analysis.

Before we begin, we should consider that the website of the BJA is not kept up to date, the Dan grade register should not be considered accurate. Also, the information is not made available in a easy to access format, you can reach only one record at a time. So to conduct any analysis of the information the first step was to scrape the BJA website and pull the data into a MySQL database where queries could be run. The data scraping process can introduce errors into the information.
Finally, the register provides no information on past grades or if the people listed are alive, deceased, members or retired. There are anomalies in the data such as the total number of entries in the database not equalling the number of grades at each level, this is caused by inconsistencies/errors in the data. This data was collected on March 26th 2009.

Simple descriptive statistics about the BJA dan grade register.

Male Dan Grades:        9941 (84.22%)
Female Dan Grades:        1862 (15.78%)

If we focus on female Dan grades, we can compare this to the general population of the BJA by referring to the BJA 2007/08 Annual Report.

A quick analysis of this data shows the following:

Male:        20226 (75.56%)
Female    :      6541 (24.44%)

Now if we compare these two basic percentages, we see right away that despite women being approximately a quarter of the Judo population, less than 16% are Dan grades. We do not have enough information to make any inferences as to why this is, but we can suggest there is an issue here that needs addressing by the BJA. Why is their an imbalance between male and female when compared to the wider Judo population. Is it a case of institutional sexism, or are there other forces at work, like for example maternity.

The general population also includes children, so we should not read too much into this difference in percentages. However, it does perhaps suggest that the BJA (and Judo more widely) should research this area and try and determine if there is an actual difference and if so, why it is occurring and when.

It would also be interesting to compare these participation levels to other sports such as Tennis, Dance, Rugby, Wrestling, TKD or Karate. We might find that Judo participation levels by females and at the Dan Grade may be good rather than poor. The percentage of girls aged 14-15 that do not participate in any active sports, on a weekly basis, is around 15% to 20%  (Balding, J. (2004). Young people in 2003. Exeter: Schools Health Education Unit.), so how would this fit with our numbers?

The Dan grade register data also has information such as area, club and date of the grading, analysing this information shall be the subject of later posts. You may also wish to take a look at the data yourself, the part that formed the basis of this post is available at http://www.swivel.com/data_sets/show/1017786 it has the data and some nice charts too.