Attempting To Create A World Speedway Rating

[khucke]

As an only occasional follower of the Speedway scene, it always confounded me that no overall rating of riders existed (to my knowledge at least). Yes, I'm aware of Green Sheet Averages, but have been unable to find out how hey were determined. Besides they exist for riders active in Britain only. My goal is to assess the results of individual riders in the various competitions, being it league or individual meetings they take part in. The method I use is the ELO rating. Primarily known as a rating for chess players, it is also used for rating boxers (boxrec.com) or national footbal teams (eloratings.net) and certainly others. It is geared to measure the results achieved versus different sets of opponents. It is a proven system that fits the sport of Speedway very well.

I started these rating out of my own curiosity, I used all the results of 2015 that I could ger hold of, namely : Elite, Premier and National leagues in the UK, the 1st and 2nd Swedish leagues, the 1st, 2nd and 3rd Polish leagues, the Danish, Russian, Czech and German leagues, testimonials and other individual competitions including the Grand Prix, of course. So far I processed the results until May 31st 2015, I certainly plan to advance further , but cannot make any promises about the timeline, though it would surely motivate me if there is some interest in the community for this project.

I have currently rated 511 riders, among those are 292 riders with at least five results, which I consider ton be the threshold for appearing in my rankings. Here are the top 25 riders, I'm looking forwards to your opinions and comments and I'm ready to elaborate on my methodology if wished. I'm also ready to field your questions on your favorite riders position.

 

1 Hampel J 1728 2 Sayfutdinov Em 1717 3 Woffinden Ta 1697 4 Pedersen N 1695 5 Pawlicki Pi 1620 6 Zmarzlik Ba 1573 7 Vaculik Ma 1567 8 Laguta Gr 1541 9 Ulamek Se 1510 10 Hancock G 1495 11 Kildemand 1487 12 Holta Ru 1474 13 Janowski M 1453 14 Pawlicki Pr 1432 15 Doyle J 1375 16 Buczkowski Kr 1364 17 Lindbaek A 1357 18 Iverson 1356 19 Protasiewicz P 1338 20 Balinski D 1325 21 Laguta Ar 1319 22 Andersen H 1310 23 Bogdanov Ma 1310 24 Przedpelski Pa 1309 25 Cook C 1305

 

[waiheke1]

The most comprehensive rating system is I believe "dolgin's" which also uses a chess based methodology I believe. Sorry don't have the link handy.

I have my own, somewhat less comprehensive system, but my list of top 25 riders can be found on here (In the international section). Your rankings certainly look closer to dolgin's, whereas my own attach additional weighting to gps.

Edited December 24, 2015 by waihekeaces1

[arnieg]

Always very interested in these efforts. The big problem is that inputting all the data can be time consuming, and there are always various anomalies.

 

Dolgin's rankings are here:

 

 

https://docs.google.com/spreadsheets/d/18wvOs1SUYAz9O50h3xcKFBNzceM7HtRnAhwaLMKHFD0/edit?hl=ru&pli=1#gid=0

[The White Knight]

Always very interested in these efforts. The big problem is that inputting all the data can be time consuming, and there are always various anomalies.

 

Dolgin's rankings are here:

 

 

https://docs.google.com/spreadsheets/d/18wvOs1SUYAz9O50h3xcKFBNzceM7HtRnAhwaLMKHFD0/edit?hl=ru&pli=1#gid=0

Dolgin used to Post them on this Forum at one time.

 

I don't know why he stopped?

 

I always found them interesting anyway.

[khucke]

It indeed is time-comsuming! I organzied my spreadsheet to simplify data entering , but still..... Can someone give my advice how to copy a spreadsheet format in a forum post? It looked easy when I wrote my first post, but the riders list was unformatted after being posted.

Thanks for the link to the Dolgin rankings. Very up to date. Looks like a similar methodology, might make my efforts redundant.

 

Here are the top 50 ratings for June 7, 2015, several leading riders had a rough week, except Woffinden, who took the lead.

 

Rank Rider Rating Prev Week Rank

1 Woffinden 1688 3

2 Vaculik Ma 1643 7

3 Pedersen N 1638 4

4 Pawlicki Pi 1625 5

5 Hampel J 1601 1

6 Zmarzlik BA 1594 6

7 Sayfutdinov Em 1579 2

8 Laguta Gr 1524 8

9 Kildemand 1512 11

10 Ulamek Se 1501 9

11 Holta Ru 1474 12

12 Hancock G 1471 10

13 Janowski M 1456 13

14 Protasiewicz P 1407 19

15 Pawlicki Pr 1397 14

16 Iverson 1396 18

17 Laguta Ar 1376 21

18 Lindbaek A 1357 17

19 Buczkowski Kr 1356 16

20 Balinski D 1354 20

21 Andersen H 1339 22

22 Doyle J 1326 15

23 Przedpelski Pa 1319 24

24 Holder C 1310 26

25 Cook C 1306 25

26 Zagar Ma 1277 41

27 Drabik Ma 1272 39

28 Masters Sa 1271 28

29 Szczepaniak Ma 1271 29

30 Woodward 1266 30

31 Morris N 1264 27

32 Woelbert Ke 1257 31

33 Bogdanov Ma 1255 23

34 Kurtz Br 1246 32

35 Zengota Gr 1241 42

36 Jedrzejak To 1238 63

37 Schlein 1235 34

38 Szczepaniak Mi 1230 35

39 Karlsson P 1221 33

40 Loktajew Al 1205 36

41 Harris C 1201 37

42 Jonsson An 1198 40

43 Kennett 1182 38

44 Gollob To 1182 60

45 Wells R 1176 45

46 King Da 1171 43

47 Nilsson Ki 1163 50

48 Ljung 1158 44

49 North Da 1157 52

50 Woryna Ka 1151 56

Edited December 25, 2015 by khucke

[khucke]

Not much movement among the top riders, Woffinden had one 15 in 5 effort, so he expanded his lead considerably. Frederick Lindgren is the biggest riser going from 59 to 45. At the moment there are 322 riders ranked., Who's last?. Thank you for asking! It's Ellis Perks of Scunthorpe at 202 points.

 

Rank Name Rating Prev Rat

1 Woffinden Ta 1740 1

2 Vaculik Ma 1643 2

3 Pawlicki Pi 1625 4

4 Hampel J 1601 5

5 Pedersen N 1587 3

6 Kildemand 1580 9

7 Sayfutdinov Em 1579 7

8 Zmarzlik BA 1527 6

9 Laguta Gr 1524 8

10 Janowski M 1512 13

11 Ulamek Se 1501 10

12 Holta Ru 1474 11

13 Hancock G 1462 12

14 Pawlicki Pr 1431 15

15 Iverson 1409 16

16 Protasiewicz P 1407 14

17 Laguta Ar 1376 17

18 Buczkowski Kr 1369 19

19 Doyle J 1353 22

20 Lindbaek A 1350 18

21 Andersen H 1343 21

22 Holder C 1321 24

23 Przedpelski Pa 1319 23

24 Cook C 1314 25

25 Balinski D 1311 20

26 Zagar Ma 1277 26

27 Szczepaniak Ma 1271 29

28 Drabik Ma 1266 27

29 Woodward 1266 30

30 Jonsson An 1258 42

31 Kurtz Br 1246 34

32 Zengota Gr 1241 35

33 Morris N 1240 31

34 Harris C 1239 41

35 Jedrzejak To 1238 36

36 Schlein 1235 37

37 Szczepaniak Mi 1230 38

38 Woelbert K 1226 32

39 Masters Sa 1221 28

40 Karlsson P 1221 39

41 Loktajew Al 1205 40

42 King Da 1187 46

43 Bogdanov Ma 1182 33

44 Gollob To 1182 44

45 Lindgren F 1179 59

46 Kennett 1173 43

47 Wells R 1173 45

48 Lambert Ro 1171 55

49 Nilsson Ki 1163 47

50 Ljung 1158 48

Edited December 30, 2015 by khucke

[Gatwick Rocket]

Khucke, this is a great initiative. I too think it is time speedway adopted a ratings system. The traditional average generally worked when the league format resulted in each rider racing roughly equal times against each other, but clearly doesn't work under the current format.

 

A few months ago I started using race details from the start of the British League in 1965 to develop an ELO-based rating. So far I have input c.5000 race details to end of August 1965. As has already been pointed out it is a painstaking task.

 

I'd be interested to compare notes if you are happy to share your spreadsheet.

[arnieg]

15 Iverson 1409 16

16 Protasiewicz P 1407 14

17 Laguta Ar 1376 17

18 Buczkowski Kr 1369 19

19 Doyle J 1353 22

20 Lindbaek A 1350 18

21 Andersen H 1343 21

22 Holder C 1321 24

23 Przedpelski Pa 1319 23

24 Cook C 1314 25

25 Balinski D 1311 20

26 Zagar Ma 1277 26

 

Nice to see a second string doing so well!

[khucke]

As I continue to to update the 2015 rider's rating, I find it surprising how big the differences among the even the best riders are. These ratings are not mere abstrct numbers. For example take Woffinden with a rating of 1760, he's 132 points ahead of the 2nd-placed guy, Nicki Pedersen at 1638. This difference of 122 points can be translated into a little less than one full points (133 is the actual number for one point) in a five-heat event. So Woffinden can be expected to score 1 point more in an event than Nicki, on average.

(if you're interested in the math-y details pm me!)

 

 

 

 

Here are the ratings as of June 21st, 2015:

 

 

 

 

 

 

 

Rank Name Rating Nation Pr Week Rank

1 Woffinden Ta 1760 UK 1

2 Pedersen N 1638 DEN 5

3 Pawlicki Pi 1611 POL 3

4 Hampel J 1601 POL 4

5 Zmarzlik BA 1576 PÖL 8

6 Sayfutdinov Em 1569 RUS 7

7 Laguta Gr 1561 LAT 9

8 Kildemand 1557 DEN 6

9 Vaculik Ma 1547 SLK 2

10 Janowski M 1529 POL 10

11 Ulamek Se 1465 POL 11

12 Hancock G 1464 USA 13

13 Protasiewicz P 1437 POL 16

14 Holta Ru 1410 POL 12

15 Iverson 1406 DEN 15

16 Laguta Ar 1404 RUS 17

17 Buczkowski K r 1403 POL 18

18 Lindbaek A 1375 SWE 20

19 Przedpelski Pa 1361 POL 23

20 Pawlicki Pr 1357 POL 14

21 Doyle J 1349 AUS 19

22 Holder C 1341 AUS 22

23 Balinski D 1283 POL 25

24 Jonsson An 1282 SWE 30

25 Szczepaniak Mi 1276 POL 37

26 Drabik Ma 1270 POL 28

27 Woodward 1266 AUS 29

28 Cook C 1263 AUS 24

29 Masters Sa 1259 AUS 39

30 Zagar Ma 1253 SLO 26

31 Zengota Gr 1249 POL 32

32 Woelbert K 1240 GER 38

33 North Da 1239 AUS 51

34 Szczepaniak Ma 1236 POL 27

35 King Da 1234 UK 42

36 Karlsson P 1221 SWE 40

37 Jedrzejak To 1221 POL 35

38 Gollob To 1216 POL 44

39 Loktajew Al 1205 UKR 41

40 Harris C 1203 UK 34

41 Kurtz Br 1203 AUS 31

42 Ljung 1188 SWE 50

43 Kennett Ed 1184 UK 46

44 Kroner T 1184 GER +

45 Jensen Mi Je 1181 DEN 53

46 Lawson 1179 UK 56

47 Andersen H 1177 DEN 21

48 Morris N 1173 AUS 33

49 Lindgren F 1167 SWE 45

50 Bogdanov Ma 1164 LAT 43

Edited January 1, 2016 by khucke

[khucke]

@Gatwick Rocket

 

It's interesting to find someone else working in the same directio, only from a historical perspective. I'll give you a brief overview of my methodology:

 

1) What events are rated?

I approached this rather pragmatically, I selected the races, whose results I could get hold of. In my initial post I named the selected leagues. Additionally I select events, when at least 50 percent of the riders are already in my database (currently some 550 riders).

 

2) How are the ratings calculated?

 

First I determine the quality of an event by calculating the average rider rating of all participators. Then I set the point per heat result of a rider in relation to the event's average.

Example: The average rider rating in an event happens to be 1000. Rider A races 4 heats and places 3rd, 2nd, 1st and 4th, that earns him 6 points . These six point are exactly average, therefore his rating for this event is 1000. Rider B won all his races, his rating is 2000, Rider C finished last four times his rating is, you guessed it, zero.

 

Let's assume rider A has a rating of 1200 coming into this event before that he had taken part in 4 other events, we calculate:

 

4 times 1200 plus 1000 divided by 5 equals 1160. Rider A's new rating is thusly 1160.

 

Let's assume rider B has no previous event to his credit. He is assigned 1000 , the average of all participants, as a preliminary rating. Remember he did fairly well in his debut and won all his four races, and received a rating of 2000 (for this event), we calculate:

 

1 time 1000 plus 2000 divided by 2 equals 1500. Rider B's new rating is then 1500.

 

Let's assume rider C is a grizzled veteran and has already ridden in 25 events and has a rating of 800. In today's event he sorrily finished last four times for zero points and a rating of 0. We calculate:

 

15 times 800 plus 0 divided by 16 equals 750. Rider C's rating took a hit and tumbled to 750 .

 

Wait, rider C had 25 previous events, shouldn't it be '25 times 800 plus 0 divided by 26'? Glad you asked! I set the maximum of events entering this calculation to 15. I did this to ensure that the rating 'fits' the true talent level of a rider. What does that mean ? young rider starts his career and ains very few points initially, therefore is quite low after his first year . As he gains more experience against the big boys, his results improve and his rating should too. Say 30 races of poor results in his first year, would make it very hard to improve his rating to his new true talent level. Therefore a Limit of 15 previous rsults is neccessary to avoid a constant 'lagging' of ratings to true talent. Obviously the same is true for an older championship rider who is losing his strength or interest or whatever and achieves poorer results all the time. We wouldn't want to put too much weight on his past result, but rather gain a clear view of his current 'true' talent level.

 

 

3. Are there more things that are important? Only a few: From number question 2 you learned that a rider's rating is rather volatile in the beginning. The rating tracks the ups and downs of rider's performance and eventually stabilizes as more results are included. In my current rankings I set a minimum of 5 events for a rider to be included, that might be to low, maybe 10 would be more appropriate. Also there is the question of inactivity for whatever reason, I have not yet decided how to handle it.

 

So that's it for the moment, don't hesitate to ask if something is unclear (as it well may be due to my lack of writing skills)

[pugwash]

As I continue to to update the 2015 rider's rating, I find it surprising how big the differences among the even the best riders are. These ratings are not mere abstrct numbers. For example take Woffinden with a rating of 1760, he's 132 points ahead of the 2nd-placed guy, Nicki Pedersen at 1638. This difference of 122 points can be translated into a little less than one full points (133 is the actual number for one point) in a five-heat event. So Woffinden can be expected to score 1 point more in an event than Nicki, on average.

(if you're interested in the math-y details pm me!)

Do you do company accounts?

[builderbob]

when did australia have a c cook

[iris123]

when did australia have a c cook

1770

[Gatwick Rocket]

@khucke, thanks for sharing your methodology.

 

Could you please clarify how you determine the ratings score for each new meeting? So, in your example, what would the rating be for C if he had scored 2 points instead of 0? Does it depend on the individuals C is up against in each of his races or just his score at the end of the meeting? Also, what do you do about situations where one rider finishes behind a team-mate (historically earning him bonus points)?

 

 

Regarding your point about limiting previous meetings that count towards the rating, I agree that a mechanism like this is necessary. My suggestion would be to introduce a factor based on elapsed time. This would help to avoid situations where A has a rating based on performances over the past month (say) because he races several days each week in different countries and B has a rating based on performances over the past year or more (say) because he races in one country and has been out injured for a period. It would also reflect that form is (I believe) a function of time as well as the number of previous performances.

[khucke]

when did australia have a c cook

 

Sorry, my fault UK of course. I knew I'd mix up at least one UK/AUS rider.

[khucke]

@khucke, thanks for sharing your methodology.

 

Could you please clarify how you determine the ratings score for each new meeting? So, in your example, what would the rating be for C if he had scored 2 points instead of 0? Does it depend on the individuals C is up against in each of his races or just his score at the end of the meeting? Also, what do you do about situations where one rider finishes behind a team-mate (historically earning him bonus points)?

 

 

Regarding your point about limiting previous meetings that count towards the rating, I agree that a mechanism like this is necessary. My suggestion would be to introduce a factor based on elapsed time. This would help to avoid situations where A has a rating based on performances over the past month (say) because he races several days each week in different countries and B has a rating based on performances over the past year or more (say) because he races in one country and has been out injured for a period. It would also reflect that form is (I believe) a function of time as well as the number of previous performances.

 

 

 

 

I add up the ratings of all riders in a meeting to determine the quality of opposition. (to be precise: I calculate a seperate meeting's rating for each rider, as he won't compete against himself). There are inaccuraccies of course which I look upon as minor ones: some riders take part in 3 heats while others ride in 5. I only include riders that take part in > 50 percent of scheduled heats.. I ignore the opposition a rider faces in each heat. It woud add a layer of precision, at the cost of enormous amount of work tht I'm not ready for.

I ignore bonus points as well as doubled points. These are gimmicks that doesn't help in measuring a rider's result or do they?

 

Your question concerning rider C: C scores 2 points. I divide 2 by 6 (in a four heat meeting, 7.5 in a five heat meeting) and multiply with 1000 (the meeting's average rating) and get a rating of 333 for rider C in this single event.

 

Your suggestion of adding the elment of elapsed time, do I understand it correctly.you want to reward a rider for maintaiing a rating over a larger number of meetings? Can you give me a brief example? As I said I'm unsure yet how to manage inactivity in the ratings. One f my ideas is to leave a rider out of the ratings after, say, six months, put an asterisk on him as an inactive rider and reenter him as soon as he becomes active again.

[khucke]

Here are the rating as of June 28, 2015:

 

Rank Name Rating Nation Prev Week Rating

1 Woffinden Ta 1770 UK 1

2 Pawlicki Pi 1652 POL 3

3 Pedersen N 1640 DEN 2

4 Zmarzlik BA 1608 PÖL 5

5 Hampel J 1601 POL 4

6 Laguta Gr 1563 LAT 7

7 Sayfutdinov Em 1559 RUS 6

8 Kildemand 1546 DEN 8

9 Janowski M 1528 POL 10

10 Vaculik Ma 1519 SLK 9

11 Iverson 1516 DEN 15

12 Hancock Ga 1466 USA 12

13 Ulamek Se 1463 POL 11

14 Protasiewicz P 1443 POL 13

15 Lindbaek A 1421 SWE 18

16 Zagar Ma 1390 SLO 30

17 Laguta Ar 1383 RUS 16

18 Buczkowski Kr 1375 POL 17

19 Pawlicki Pr 1371 POL 20

20 Przedpelski Pa 1361 POL 19

21 Doyle J 1356 AUS 21

22 Holta Ru 1356 POL 14

23 Drabik Ma 1350 POL 26

24 Holder C 1299 AUS 22

25 Cook C 1272 UK 28

26 Balinski D 1271 POL 23

27 North Da 1267 AUS 33

28 Karlsson P 1254 SWE 36

29 King Da 1249 UK 35

30 Jonsson An 1248 SWE 24

31 Harris C 1246 UK 40

32 Jedrzejak To 1238 POL 37

33 Jensen Mi Je 1237 DEN 45

34 Bogdanov Ma 1233 LAT 50

35 Andersen H 1225 DEN 47

36 Woelbert K 1222 GER 32

37 Szczepaniak M i 1220 POL 25

38 Zengota Gr 1207 POL 31

39 Batchelor 1204 AUS 67

40 Kurtz Br 1190 AUS 41

41 Szczepaniak Ma 1183 POL 34

42 Larsen Ke 1176 DEN 84

43 Loktajew Al 1173 UKR 39

44 Nilsson Ki 1170 SWE 53

45 Masters Sa 1169 AUS 29

46 Dryml A 1166 CZE 82

47 Kylmkorpi Jo 1160 SWE 68

48 Schlein Ro 1158 AUS 51

49 Miedzinski Ad 1151 POL 78

50 Morris N 1149 AUS 48

Edited January 10, 2016 by khucke