Sports analytics & 3 centuries old tradition (Part 1)

I remember reading this great sport analytics piece on fivethirtyeight.com on history of Japanese Sumo. Japanese, who are notoriously pedantic, kept records of their great sporting tradition for centuries and Benjamin Morris (n.b. author of best sports analytics piece ever, on Dennis Rodman non the less) did fantastic analysis. Whole piece was great, and visualizations were beautiful… I admit, I was little bit jealous, wished that in Croatia we had something similar, and then it occurred to me, we do, it’s pretty cool and badass, and it’s called Alka.

alka_kolaž.JPG

The Alka is a knightly tournament held annually in the Croatian town of Sinj (region of Dalmatia) every first Sunday in August since 1715. Yup, since 1715., meaning competition is more than three centuries old! It commemorates a Croatian-Venetian victory over Ottomans on August 14, 1715 in which the local Christian population of around 700 Croats in cooperation with a smaller number of Venetians managed to defend Sinj against tens of thousands Ottomans (…talk about beating the odds). In 2010, the Alka was inscribed in the UNESCO Intangible Cultural Heritage Lists.

alka_punti_4

During the competition, knights (called alkars) ride horses at full gallop along a main street, aiming lances at an iron ring (called alka) hanging on a rope. The rules of competition are simple: alkars are awarded points according to which sector of the ring they pierce and the one with most accumulated points in 3 runs wins the competition. I don’t want to bother you with more detail, but before we proceed, please take a minute and check out this short (I promise!), interactive guide about Alka, to get a feeling what it’s all about… believe me, it’s much better than me wasting ink here😊 Anyway, upon this epiphany of mine, I’ve sent couple of mails and to my surprise, very nice lady from The Alka Knights Society responded with 10 years of data. After calling her on the phone and some additional pleading I got 50 years of data in total, modern period of Alka. Bare in mind, The Alka Knights Society are old-school, obviously not keeping digital archive, so I got scanned minutes of the competitions. It looks like this (picture bellow)  and suffice to say OCR wasn’t of much help, so I had no options than to go old-school myself and prescribe 50 years of data.

scan_zapisnik

Only men born in the city of Sinj and surrounding villages can take part in the Alka, participation is considered a great privilege and it certainly requires certain set of skills so pool of potential alkars is not big which is reflected in the data – in 50 years, 72 alkars participated in the tournament and 26 of them managed to win it. Ognjen Preost and Anđelko Vučković have most wins (5 of them), while Preost is also most efficient alkar having won 5 out of 11 Alka (45%) in which he participated. He’s followed by Stipe Šimundža who won 2 out of 6 Alka (33%), another Alka all-time great – we’ll talk more about both of them later… and the “mystery” surrounding them 😊 Young Alen Poljak won Alka on his debut two years ago, so his efficiency rating is excellent as well, but we’ll see whether his numbers will hold up on larger sample. There’s another curiosity involving Poljak family, Alen’s father Ante is one of only two alkars who managed to win Alka with 3 hits in the center of alka (called srida) which is worth 3 points, thus accumulating maximum number of points. Following graph depicts all winners of the competition in last 50 years along with their points total (in 3 runs).

1_winners_and_points

If we analyze number of points that winners acquired (in 3 runs) we see that most Alka have been won with 7 points (42%), followed by 8 points (28%) and 6 points (24%). In last 50 years only once Alka has been won by 5 points (by Dušan Dinarina, 1979.), while aforementioned Ante Poljak (2006.) and Ante Zorica (in 2009.) are only ones who managed to win Alka with maximum number of points. Obviously, if two or more alkars have same number of points after 3 runs, tie-breakers are needed to decide the winner. Out of 50 Alka that we analyzed 15 of them were decided after tie-breaker (30%). In 1988. 4 alkars have participated in tie-breaker and winner was decided after first one, while 10 years later, 4 tie-breakers were needed to decide the winner between Tonći Bešlić and Ognjen Preost (who emerged as winner in the end).

2_alkas_by_winner_points

Let’s calculate the probability of winning the competition based on the number of points after 3 runs. In observed period, there were 112 alkars with 5 points after 3 runs, while only one of them (very, very lucky Dušan Dinarina) won the competition. Thus, implied probability of winning Alka with 5 points is basically 0%, or 0.9% to be precise. Probability of winning the competition with 6 points is 14.5%, while 7 points gives you 39% chance of winning. Probability drastically increases with 8 points and it is 87.5% – 16 alkars managed to gather 8 points and 14 of them were victorious. If you want to be safe it’s best to hit 3 centers… and be called Ante I guess. As already mentioned, only two alkars (Ante Poljak and Ante Zorica) managed to accumulate 9 points and yeah, both of them won the competition.

3_probability_of_winning

Let’s call good ol’ reverend Bayes and calculate some conditional probability, i.e. what are the chances of winning the Alka based on gathered number of points in the first run? We have following situation; with 0 points in the first run, alkar has 1% chance of winning the competition. With 1 point chances don’t get much better and stand at only 2.2%, but probability raises considerably with 2 points – 9.8%. Chances of winning the competition after hitting srida (center worth 3 points) in the first run are almost 20%. Let’s provide some additional illustration, in 50 years of the competition 204 alkars had 0 points after the first run, but only 2 of them came out victorious in the end, in the same time 116 alkars got maximum 3 points from the first run and 23 of them won the competition.

4_probability_of_winning_1run

Analyzing all runs we can see that in 92.5% of cases alkars hit their target, i.e. alka ring, meaning in only 7.5% of cases they miss completely (P in the picture below, while 0 points mean that alka was hit but not pierced, it did not remain on lance). This shows fantastic accuracy, considering the size of the target (ring diameter is only 13.1cm) and the fact that alkar approaches alka on a galloping horse with speed greater than 50km/h and in full body “armor” weighing considerably.

5_distribution_of_points

If we count P as being worth 0 points (which it is), we get point distribution from all runs in 50-year period. We see that 1 point is most “popular”, alkars hit it in 37% of cases, while 0 and 2 points are more or less evenly distributed and account for 24% of hits. Mythical srida (small center ring worth 3 points) is by far hardest to hit and alkars hit it in only 15% of cases, meaning on every seventh run on average.

6_distribution_of_points_2

If we are talking about srida, there are 7 of them on average per competition, however there were more and less fruitful years. Alka with most 3-point hits was 294. Alka from 2009 with fifteen of them (double the average!). Apart from this one there were 6 more Alka in 50-year period with 10 or more srida and as we can see 80’ very especially productive. However, there were years when “steady right hand and hawk eye” where not that coordinated; in 1999 and 2000 there were only 2 srida with couple more years with 5 or less. If we are talking about 0 points (complete misses and when alka is hit but not pierced), alkars were extremely precise in 2002 when we had just 4 occasions of 0 points. On the other hand, 1982 with 22 and 1979 with 18 occasions of 0 points were not good years for alkars, to say at least.

7_distribution_of_points

Let’s take a step back from particular point combinations and their distribution and calculate average points per run that we will use as quality metric, of both competitions and individual alkars. In observed 50-year period average number of points per run was 1.3. In 2009 alkars on average scored 1.72 points per run and along with aforementioned 15 srida we can conclude that it was best year of Alka to date. Alkars had lowest points per run in 1979, just 0.94 points. It was definitely the worst year of Alka if we remember there were only 3 srida and 18 occasions when alkars didn’t manage to get a single point. We checked weather conditions on that day, expecting maybe rain, but nope, it was bit humid with temperatures around 34°C.

8_points_per_run.JPG

Post has gotten bit long so this is it for Part 1, and yeah, we are saving best parts for sequel so you’ll need to visit us again. We’ll talk about golden period of Alka, alkar rankings and three of them that stand out with peculiar stories involving them. To conclude the post I’ll quote 18th century Croatia poet and Franciscan friar Andrija Kačić Miošić who wrote about the Battle of Sinj:  the Town of Sinj, the mace of gold, heroic megdan from days of yore…

Addendum: This peculiar analytics exercise, something that started as a fun thing to do, got us on national TV (statistics in bottom right corner)…

alka_na_TV.jpg

…and #1 regional/#3 national newspaper 🙂

clanak_zajedno.jpg