Test Cricket Elo rankings

[Alex Holowczak]

I'd mentioned these on here before, but now I've finally finished and checked all the results, I thought I'd put the current ratings here. The basics are:

(1) The normal Elo system is used, with differences converted into rating change.
(2) To kick start the system, England and Australia were both given ratings of 2500.
(3) To get on the list, the country must have played at least 10 matches, at least one of which must have been a win. (As opposed to scoring at least 1 point.) If it took a country > 10 Tests to win a match, then just the previous 10 results were used to calculate the initial rating.
(4) South Africa and Zimbabwe were treated as unrated when they returned from their bans (for apartheid, and for self-imposed reasons respectively).
(5) k = 25. Why? It seemed to give the best mix between volatility and "accuracy". I looked at when periods of dominance ended (e.g. WI in the 1990s, Australia more recently), and concluded that k = 25 was the best of the 5-point intervals between 0 and 50.
(6) Ratings are live, rather than over a 1-month period.
(7) No account is made for series; e.g. if you're 2-1 down in a series, you would try to win more than you did at the start of the series. This is a bit like chess; in a team environment, you might try to win more than you set out to do, and end up losing instead of drawing, for example.

These are the Test ratings my software spewed out, for the 2067 Tests that have been played since 1877:

Code: Select all

   Country         Rating
 1 South Africa    2388 
 2 Australia       2385 
 3 England         2345 
 4 India           2296 
 5 Pakistan        2287 
 6 Sri Lanka       2267 
 7 West Indies     2169 
 8 New Zealand     2151 
 9 Bangladesh      1902 

Zimbabwe are unrated because they haven't played 10 Tests since they returned to Test cricket in 2011. They've got their 1 win already, against Bangladesh.

Here are the current ICC Test Rankings for comparison. The most obvious difference seems to be that Elo fits the teams more closely together than the ICC system.

If Australia beat Sri Lanka in the Boxing Day Test, then Australia will overtake South Africa to become number 1.

[Brian Valentine]

Have you any comments on why there appears to be rating deflation in the syatem?

[John Brewitt]

Have any of the countries achieved grandmaster status?

[IM Jack Rudd]

Have you any comments on why there appears to be rating deflation in the syatem?

Probably because teams just gaining Test status tend to be rather weaker than the existing teams, but then improve quite rapidly.

[Alex Holowczak]

Have you any comments on why there appears to be rating deflation in the syatem?

Probably because teams just gaining Test status tend to be rather weaker than the existing teams, but then improve quite rapidly.

That's exactly it - every new team to the system was weaker than the teams who already played, with the exception of South Africa when they returned in 1991, and possibly India and Pakistan (who were at least as good a New Zealand).

England and Australia had a mean rating of 2500 before South Africa came along in 1889. They lost 10 of their first 11 matches, drawing once against Australia in 1902. They recorded their first ever win against England in 1906, 17 years later. Scoring 1.5/10 meant they were always going to come onto the list at something like 2250. As they improved, they took points off England and Australia.

[Kevin Thurlow]

Interesting. I'm slightly surprised Bangladesh are that high, but they are getting more competitive.

Slightly off-topic, but I've always thought snooker would benefit from an Elo system. As an extreme example, when Tony Knowles beat Steve Davis (world champion) 10-1 (?) in round one of the world championship about 30 years ago, he got exactly the same number of ranking points as if he had beaten a qualifier 10-9. This has always seemed illogical.

And you can beat exactly the same opponent in two different events and get different amounts of ranking points. Maybe that is logical...

[David Sedgwick]

Slightly off-topic, but I've always thought snooker would benefit from an Elo system.

I believe that Stewart Reuben and others developed one in the 1980s.

[Brian Valentine]

That's exactly it - every new team to the system was weaker than the teams who already played, with the exception of South Africa when they returned in 1991, and possibly India and Pakistan (who were at least as good a New Zealand).

I was expecting the reason as the one Jack suggested. You are in a position to show some statistic that demonstrates this effect over each step as a team enters. The point about India being as strong as New Zealand is interesting because MCC/England used to send near second teams to these places to start with. But it was easier to get good teams to go to New Zealand than India in those early days.

I doubt that ELO would ever be accepted in cricket. There is so much more information in a scorecard that is not available in a chess result and quite clearly a draw in cricket is not a demonstration of some sort of equality between the teams (though a tie works).

[Alex Holowczak]

Interesting. I'm slightly surprised Bangladesh are that high, but they are getting more competitive.

Bangladesh are 450+ points behind South Africa, so two things are at work:
(1) If they played as often as (say) England, they'd be even further behind because they'd inevitably keep losing.
(2) They're so far behind everyone else that a loss means they don't actually lose that many points.

I was expecting the reason as the one Jack suggested. You are in a position to show some statistic that demonstrates this effect over each step as a team enters. The point about India being as strong as New Zealand is interesting because MCC/England used to send near second teams to these places to start with. But it was easier to get good teams to go to New Zealand than India in those early days.

England managed to Tour both New Zealand and West Indies simultaneously as they entered Test cricket.

New Zealand were quite clearly very weak for the first 30 years of their existence as a Test nation. It was probably easier to find a team to go to New Zealand due to the fact that most players on tours of India in those days ended up in hospital with illness.

I doubt that ELO would ever be accepted in cricket. There is so much more information in a scorecard that is not available in a chess result and quite clearly a draw in cricket is not a demonstration of some sort of equality between the teams (though a tie works).

There are some issues with draws. For example, these results include the two Test matches abandoned as a draw due to the state of the pitch between West Indies v England (Jamaica 1998 and Antigua 2009).

I'm not sure I agree with your comment about information availability, however. The success criterion of a cricket match is the result, be it a win, draw, loss or a tie. India shouldn't be judged by the fact their innings contained three batsmen making 100s, because they're not setting out to make 100s; they're setting out to win the game for India. It'd be like comparing chessplayers by the points value of the pieces they have left on the board at the end of a game.

A few examples of draws that shouldn't have been include stuff like England v South Africa somewhere in 1998, and England v Australia at Cardiff 2009. England was saved from defeat by heroic batting to block out a draw from those well-known batting powerhouses, Angus Fraser and Monty Panesar. Another type of draw is the draw where no result was ever possible; for example, England v West Indies at Edgbaston 2012. It rained for 3 days out of 5, and there was never likely to be a result.

The issues with the two above draws involving West Indies and England are fairly minimal - they make hardly any difference to the ratings if you take them out. One of the initial errors I made was in a West Indies v England match in 1948, which I put the wrong result in for. When corrected, there was very little difference. So the system seems robust enough, in the long-term, to deal with draws that shouldn't have been.

[Brian Valentine]

When I talked about information in scorecards, I wasn't so interested in the three centurians issue. I agree the result is more important. I was thinking about "near defeat draws", but more importantly wins by big margins suggest something different to a few runs or 1 wicket margins.

My favourite severly disrupted drawn test match is the 4th test England v South Africa 1924. The entire test career of JCW MacBryan. Two sessions of cricket: did not bat, did not ball, no catches.

[Kevin Williamson]

When I talked about information in scorecards, I wasn't so interested in the three centurians issue. I agree the result is more important. I was thinking about "near defeat draws", but more importantly wins by big margins suggest something different to a few runs or 1 wicket margins.

My favourite severly disrupted drawn test match is the 4th test England v South Africa 1924. The entire test career of JCW MacBryan. Two sessions of cricket: did not bat, did not ball, no catches.

Brian, an interesting coincidence to see your posting on Test cricket.

Very recently I saw a list of the top scores on debut for England against India, following Joe Root’s 73 last week (he came in at number 6).
Top of the list is one Bryan Valentine (136 at Bombay in 1933). Valentine had a brief, but very successful Test career it would seem - a Test batting average of 64.85, including 2 centuries and 1 fifty. He only played just 7 tests in 2 series, presumably because it was very difficult to get into a strong England team in those days.

When Alex expands his work to include ratings of individual players I expect Valentine to be master level!

[Brian Valentine]

Almost back on topic: His final test innings was curtailed because England had a boat to catch!

[Kevin Thurlow]

David is absolutely right - I remembered SR's rating of snooker this morning. In that case, tennis lends itself to an Elo system. You get the same points for (theoretically) beating Nadal 6-0, 6-0, 6-0 as you do for beating a qualifier 70-68 in the fifth set.

A chap called HM Thurlow played one Test for Australia (not sure how a branch of the Thurlow family got to Australia...) opening the bowling and taking about 39-7-86-0 (spinners took the wickets). He was run out by Bradman for 0, leaving Bradman on 299 not out. For some reason our hero never played for Australia again!

[Alex Holowczak]

Valentine had a brief, but very successful Test career it would seem - a Test batting average of 64.85, including 2 centuries and 1 fifty. He only played just 7 tests in 2 series, presumably because it was very difficult to get into a strong England team in those days.

Not as brief as that of Andy Ganteaume. With regular opener Jeff Stollmeyer injured, he was thrust in to open the batting for the West Indies against England in 1948. He made 112, but his innings was considered too slow, so wasn't picked again. Rodney Redmond made 107 and 56 in his only match for New Zealand against Pakistan, but he developed a problem with his eyesight that prevented him from playing again. Stuart Law scored 54 not out in his only innings for Australia.

[David Robertson]

Andy Lloyd (Warwickshire & England). His only Test was against the West Indies in June 1984. After making ten runs, and batting for thirty three minutes, he was hit on the head by the West Indies fast bowler, Malcolm Marshall. Despite wearing a helmet, Lloyd spent several days in hospital, and played no further cricket in 1984. He never played for England again. He is the only Test Match opening batsman never to have been dismissed. (from the Wiki)

Unlucky!