Edo Historical Chess Ratings

by Rod Edwards (edo - dot - chess - at - yahoo - dot - ca)

Updates

  • 12 Apr. 2013 - A few additional tournament (or team match) results, mainly dated 1887-1890 from Sergeant's A Century of British Chess and further digging in primary sources, as well as some matches I came across, mostly of R.J. Loman and Nellie Showalter, but also of Harriet Worrall and some early match results between Henderson and Shaw in Canada.
  • 3 Mar. 2013 - Additions and changes:
    • As new results are added, the ratings shift slightly and I monitor the consistency of the average of the top few players, at least from about 1860, when we have enough of the top players rated. Previously, I had seen a slight deflation over time in the average of the top 10 (or the average rating of players 6 to 10), and had corrected by putting the self-game percentage at 50.1% in favour of the later year. This time, that deflation seemed to have disappeared between 1860 and 1900 at least, which makes me think that the slight decline between 1900 and 1911 might be real and not a deflation effect. Thus, I have gone back to an exact 50% self-game score. This means that very early players are rated slightly higher again, and players towards the end of the period covered (after 1900, say) are rated slightly lower.
    • Additional results on events in Britain up to 1887 added from Sergeant's A Century of British Chess, in some cases supplemented by more complete reports from original sources, including match results for Steinitz against E. Healey.
    • Some results from the first number of the Chess Player's Quarterly Chronicle of 1868, including match results of M.E. Werner and a team match between the North and West Ridings of Yorkshire, all in 1867 or 1868.
    • A result for Zukertort against Adair, from an article by Hilbert at ChessCafe.
    • A single game result for Morphy against F.H. Lewis in 1858, from the Chess Monthly.
    • I have decided that it is likely that three of the players named 'J. Wilson' I had previously listed separately were identical and have combined there results and references (under 'Wilson, J. (1)').
    • I also tried to distinguish appearances of the players W.S. Gover and F.F. Gover, F.C. Howell and A. Howell, C.G. Puller and A.G. Puller.
  • 3 Jan. 2013 - Results for some major events for 1911 added.
  • 26 Dec. 2012 - Additions and changes:
    • Date of Marco-Kaufmann match corrected to 1892 (from Kaufmann biography by Urcan and Braunwarth).
    • New Austalian tournament results from OzBase as of 25 Oct. 2012 entered.
    • The first few chapters of Sergeant's 'A Century of British Chess' mined for results.
    • Casual match between Anderssen and Löwenthal in 1851 removed, as result is too uncertain.
  • 24 Oct. 2012 - Additions and changes:
    • I have mined Owen Hindle's biography of Edward Pindar, which has turned up quite a few additional match results for him, and, with some additional detective work, the result of a match between the Dundee and Glasgow clubs in 1865.
    • I have searched through the Liverpool Mercury from 1811 to 1853 for results and information on players, which turned up a couple of new match results from 1839, Walker-Mongredien and Alexandre-Samuel, an unratable result between de Lavalette and Tascher in 1815 (famous in connection with the French revolution and Napoleon, not really chess players), and a single datable game between Löwenthal and H. Stanley in 1853.
    • I have now mined La Régence through 1851, where I obtained a few additional match results from 1851 by Delannoy, Journoud and Kieseritzky, as well as a new, somewhat uncertain, result of Alexandre against the 'Automaton', probably directed by Weyle, in 1818. This latter brought into play a result I had already found in Le Palamède for play between Weyle and Saint-Elme-le-Duc in about 1824, for which only approximate results can be used. A closer examination of that report allowed me to extract a soft result between Mouret and Saint-Elme-le-Duc in about 1826. Finally, although I am reluctant to use too many soft results, I decided to include another between Mouret and Saint-Amant in about 1835, from a later report in the Illustrated London News. These results all give some indication at least of the strengths of Weyle, Alexandre and Mouret.
    • Bell's Life in London, the Home Circle and the Era have now been mined for results through to 1853, which did not produce much in the way of new results (one datable game in the Era between Staunton and E. Cronhelm, also appearing in the Chess Player's Chronicle), but I also came across a result between Petrov and Salter in the Era of 1863.
    • I mined the four volumes of The Chess Player (1851 - 1853), turning up an additional ratable game between Anderssen and Lange, and the Chess Player's Chronicle up to 1853, obtaining a ratable game between Staunton and E. Cronhelm, some games or brief matches of Löwenthal, against Harden, H. Stanley and F.D. Zachary, and a match between R.B. Brien and Forshall.
    • Mining of the first volume (1853) of the British Chess Review turned up a number of additional match results of the early 1850s, particularly by Harrwitz, but also two new results for J.A. Laroche.
    • Information from Tim Harding provided some additional results for Paulsen in 1861, an additional casual match result between Blackburne and Burden in 1862, a second match result between Blackburne and Bird in 1879, an additional casual match result between Blackburne and Owen in 1881 and the result of a team match between the City of London Club and the St. George's Club in London in 1881.
    • A reference from the Westminster Papers, via Tim Harding, seems to make it clear that the only match between Blackburne and Steinitz in 1870 was a 2-game match with score +1-0=1 for Steinitz. The result +5-0=1 given in some sources seems suspect and I have replaced it. However, Harding also sent me a result from the Westminster Papers of 1869 giving another formal 1-game match result of +1-0=0 for Steinitz over Blackburne in that year. This slightly changes our assessment of the ratings of these two players around those years.
    • I added the result of another Potter-Heywood match, finishing in 1880, which I came across in the Illustrated London News.
    • Updates to Alan McGowan's historical pages at the Chess Scotland website led to my inclusion of the minor tournament at Edinburgh 1895.
  • 28 Aug. 2012 - Additions and changes:
    • The Hodges biography by Hilbert and Lahde has now been fully mined for information up to 1910, with additional material on New York chess events obtained from original chess columns and chess magazines (particularly in the early 1890s).
    • Some additional results on New York events were sent to me by Eduardo Bauzá Mercére.
    • Eduardo Bauzá Mercére also sent me a column from the Turf, Field and Farm of 1877 in which G.H. Mackenzie mentions the result of a match he played with Bannerjee (or Bonnerjee) when he as in India (about 1857). This result is very significant as it helps to compare Bannerjee, and therefore Cochrane, more directly with later players, even though Mackenzie was still young. It dramatically increases the assessment of Bannerjee's strength made by the Edo ratings, and indirectly increases the assessment of Cochrane's rating in the early 1850s. There is then a slight increase in the rating of other early players connected to Cochrane in the early 1840s.
    • Some results of additional Yorkshire events of the 1870s have been added, from the compilation of results of events in which Thomas Stokoe participated sent to me by his eponymous descendant.
    • Bell's Life in London for 1851 and the first half of 1852 has been mined, producing a few new results: offhand sets of games by Anderssen against Medley and Mongredien, and an additional game from the London Club tournament of 1851 (a draw between Löwenthal and Ehrmann).
    • The name of Staunton's opponent John Brown has been corrected. I had misidentified him as Joseph Brown, who appears in the subscription list to Williams' book Souvenir of the Bristol Chess Club.
    • List of contributors page added.
  • 5 Jul. 2012 - Just a few additional results and biographical information from the book on A.B. Hodges by Hilbert and Lahde, from the Era, now up to 1851, and some corrected biographical information (particularly H.E. Bird's birth year!) from Harding's new book, Eminent Victorian Chess Players. I have also now made the player named E. Schneider in the Coburg 1904 Nebenturnier different from the player named E. Schneider at Dresden 1892, as it is not clear that they are the same (considering them the same had produced some improbably high ratings for some other players in these lesser events at Coburg 1904).
  • 19 Apr. 2012 - I have been attempting to make updates approximately every two months (and lately this has been roughly in time with FIDE rating updates). This update is coming a little early and the next one will probably be a little late. Changes:
    • I have now been through Hilbert's books on W.E. Napier and W.P. Shipley, mining them thoroughly for results and information on players, up until 1910. I have also started going through the biography of A.B. Hodges by Hilbert and Lahde. These have resulted in a number of additional events being rated, though for many, only partial information has been obtained. In a few cases, I have gone back to contemporary journals, such as the Brooklyn Daily Eagle, for further information on these events.
  • 29 Feb. 2012 - Additions and changes:
    • Mining of Bell's Life in London and La Régence up to 1850, the Chess Player's Chronicle up to 1851, and The Era up to 1844 has turned up some additional match results from the late 1840s up to 1851, and some more information on players.
    • Sarah (a.k.a. Batgirl) sent me a link to a Russian website with information on another early match by Petrov (versus Kopev in 1809), now included.
    • I have modified the results I've used for Petrov versus von Jaenisch in 1844. The result from Di Felice looks likely to be just a result of some recorded games, not a full match score, or else missing a statement about odds given. Typically, Petrov was giving von Jaenisch at least Pawn and Move odds, and was winning matches against him, so I have decided to just use a soft result of 1-0 for Petrov giving Pawn and Move odds. This slightly increases the assessment of his rating in 1844 and earlier.
    • I received a compilation of results of events in which T.Y. Stokoe participated, from his descendent, also named Thomas Stokoe. These include many minor Yorkshire events, but so far I have only included the first tournament result in 1870 (but with several classes playing in separate groups).
    • Results from the Victorian Tournament (Melbourne 1866) from the first issues of the Australasian.
  • 22 Dec. 2011 - Additions and changes:
    • Errors corrected in dates and identification of a few players, including Jan Vančura, Jan Kleczyński, Béla Bayer, Wilberforce Tribe, Hubert Greenwell, and less famous players named Marshall and Neumann.
    • Added a few British results from the 1850's to 1870's (references sent by Tim Harding).
    • Added a number of results from the Chess Player's Chronicle of 1855.
    • Added fairly complete results from Feenstra Kuiper, Gaige and Di Felice for the years 1909-1910 (though still not some events with only weaker players).
    • Mined van Winsen's book James Mason in America, covering the period 1867-1878.
    • Added additional results from OzBase and the Canadian Championships site to 1910.
    • Added a couple of 1830 match results from Forster's book on the Zurich Chess Club.
    • Updated the Rynd-Harvey match result with new information from David McAlister.
    • Mined the chess column of Bell's Life in London from 1845 to mid-1846.
  • 1 Nov. 2011 - Additions and changes:
    • Inclusion of additional results from Feenstra Kuiper, Gaige and Di Felice for the years 1905 - 1910. Now years up to 1908 are covered fairly completely from those sources at least. Matches and just a few strong tournaments from 1909 and 1910 are included so far.
    • Forster's biography of Burn now mined up to 1910.
    • Gilberg's book on the 5th American Chess Congress mined.
    • Hilbert's book Young Marshall mined. I also followed up some of his references to reports of New York events from the Brooklyn Daily Eagle.
    • A few casual matches of Steinitz' in 1882 included from The Steinitz Papers, by Landsberger.
    • New information on French players from Dominique Thimognier's Héritage des Echecs Français website. In particular, he has identified the two mid-19th century players named Laroche.
    • An additional match of J.A. Laroche against de Rivière in 1860 included from La Nouvelle Régence.
  • 24 Aug. 2011 - Two main changes:
    • Results extended to 1908. For the 19th century I have tried to include every rateable result I can find. I will not be able to cover 20th century events as thoroughly. For now, I have concentrated on results from Feenstra Kuiper, Gaige and Di Felice, including all results listed there for 1903 and 1904 and a selection of stronger events from 1905 to 1908. For 1903 to 1904, I also consulted the Canadian Championships website and the OzBase website as well as Forster's book on Amos Burn.
    • Deflation adjustment. There seemed to be increasing evidence of a slight deflation in the ratings over the years, though this is not certain and is really a matter of judgement. However, the trend seemed to be more noticeable in the years after 1900, so I have now set the self-game results to be 50.1% in favour of the player in the later year to correct for this (see the section on Long Term Consistency in the Explanations Page).
  • 28 June 2011 - I have now included results from the Illustrated London News up to 1878, and I have extracted results and information on players from Forster's biography of Burn up to 1902. These have produced a significant number of new results, particularly of county events in Britain. Some new results have been added to OzBase since the last time I checked it, and these have been included, too. I also received information from Tim Harding that allowed me to include some additional county and minor events in Britain, and more detailed information on some I already had partial results for, including Redcar 1865 and 1866, Birmingham 1877, Llandudno 1893 and others.

    I had already included Cochrane's result against Mouret (playing as the Turk) in 1820, because it was referred to as a result of all games played, though Hunneman's book, from which these games were taken only claims that they are a selection of 50 games played by the Turk. However, it does seem likely that Hunneman was recording all the games by the Turk against notable players during the period he was recording games, so I have included all the results for opponents of the Turk for which at least 2 games are given.

    As usual, a few other errors have been corrected (e.g. the Blackburne-Zukertort match given by Di Felice as +2-1 in 1876, which seems to have been, in fact, +1-1 in 1877; the identity of the 'Hodges' who played at Cheltenham 1876, given incorrectly as 'A. Hodges' by Di Felice; and Popert's initials, given incorrectly by William Greenwood Walker and almost everyone since).

  • 31 Mar. 2011 - I have extended my mining of results from the Illustrated London News up to June 1874, with a bit of additional information from other contemporary sources.
  • 25 Jan. 2011 - New results and player information mostly from going through the Illustrated London News (now up to 1868) and Bell's Life in London (now up to 1844). The Illustrated London News, in particular, produced results of a number of county tournaments and inter-club matches in England in the 1850s and 1860s, which included a few relatively strong players like John Watkinson and James Kipping. Some correspondence with Tim Harding has led to more complete results for some of Gunsberg's events, such as the Simpson's Divan tournament of 1888 and the 2nd BCA congress, London 1886, as well as for Dublin 1868. I found a further source that enabled me to rate the handicap tournament in Dundee 1867, which I had left out for lack of full information. I have also now included some previously missing Canadian Championships from the late nineteenth century from Stephen Wright's Canadian Championships website.
  • 14 Nov. 2010 - A few minor changes:
    • One of the unexpected consequences of switching my programming language from SPlus to R was a different alphabetical ordering of player names on my Players page, as a result of R's different default handling of character comparisons. I've now coaxed it to put the ordering back to the way I had it before with initial 'de', 'la', etc., ignored.
    • I have decided to remove the few matches involving the versions of chess known as 'the game of pawns'. These were a strange species of odds game, in which one player gives up the queen (or queen's rook) in exchange for being given some number of additional pawns at the start of the game. I have some results of this kind of game for Deschapelles against de la Bourdonnais, de Saint-Amant and Boissy d'Anglas as well as Kieseritzky against Dumoncheau. The reason for including these in the first place was that in that era when these games were played a great deal in some circles, a well-defined system of equivalences had been worked out with the more standard types of odds, so we had a way of gauging the rating effect. On the other hand, this really is quite a different version of the game, and might be compared to rating FischerRandom Chess with regular chess nowadays. Some players (like Deschapelles) were more practiced at these strange odds games than others and this may have biased the results. So now I have removed them from my rating calculations, which has lowered the estimates of Deschapelles strength a bit.
    • I have continued to go through Bell's Life in London (now up to the end of 1843), mining for results and information on players. This has resulted in a few new match results and new information about others.
  • 20 Aug. 2010 - I have continued to add material from early sources available to me, covering most of them thoroughly up to 1849. I have also recently discovered that I have online access to Bell's Life in London, an important early source, and have so far mined it for information from its beginning in 1822 up to the end of 1837.

    The entire Morphy website of Sarah (a.k.a. Batgirl) and her Chess Journal have been removed from the web. I cite both of them heavily in my player and event references here, so, with Batgirl's cooperation, I have installed the majority of that material on my own webspace at www.edochess.ca/batgirl/ (for the Morphy site) and www.edochess.ca/batgirl/archives.html (for Sarah's Chess Journal).

    Also, I've transferred all my software to R from an old version of SPlus. This has no visible effect on the website but has enabled me to calculate the ratings and produce the web pages far more efficiently. I worried about doing this for a long time but it turned out to be fairly easy and I'm thrilled at the increase in processing speed and additional automation that it made possible.

    Graphs of tournaments with more than 20 players now include the highest rated 20 players.

  • 8 June 2010 - A few minor enhancements:
    • Errors of previous version corrected: B.V. Ljubimov and Boris Vasilievich Liubimov merged; Buckle-Tuckett 1842 score corrected; German results of Roberts disassociated from Albert Roberts; Manhattan Club tournament, 1886, handicaps accounted for, etc.
    • A few additional results found or corrected by mining early chess sources up to 1845 (books by W.G. Walker, G. Walker and E. Williams, the chess journals Le Palamède and the Chess Player's Chronicle, and the chess column of the Illustrated London News). Also, some results added from reminiscences of John Donaldson published in the Illustrated London News between 1878 and 1880. It is interesting to note that as I have been adding additional results and correcting other results from the first half of the nineteenth century, the ratings of these early players have been increasing over previous versions of the Edo ratings. They are still very uncertain (as shown by the rating deviations) but Deschapelles and de la Bourdonnais, in particular, now appear very highly rated and comparable in strength to many top players later in the century.
    • A website (one of Spinrad's sources) by Marco A. Alberti turned up some additional information, for example on the odds that appear to have been given in the match de la Bourdonnais-Church, 1837.
    • Small tweaks to the graphs of top players by decade, 30-year span, and the whole period.
  • 23 Jan. 2010 - Here at last is a major update. This has taken a rather long time (my spare time being really very spare), but it was difficult to do piecemeal. My original intention was to add basic biographic information on players and to document my references for this information as well as the match and tournament results I had already obtained, and cross-check everything as I went. However, I had already started collecting information from other sources, and in the process of trying to verify results and player identifications, I inevitably started including additional information. It also seemed essential to try to date and locate events and document this information as well. I have erred on the side of inclusiveness in citing sources, despite their variability in quality, except in the case of very well-known players for whom there are too many references. I do not necessarily endorse or recommend any of the sources I've cited; rather I have been interested in reporting discrepancies in the information given.
    A number of other modifications to the Edo ratings have also been made. I summarize the main changes below:
    • Addition of biographical information (full name, alternate names, birth and death dates) where available on all player pages, as well as all references used for the information on each player from books, periodicals and web pages. Generally, only sources where a first name or initial or name variant is given, or biographic information, are cited, not just the occurrence of a surname. The single most important source here is Gaige's Chess Personalia. Discrepancies in information and other comments are included in 'Notes' on these pages where needed.
    • When the birth year is known, approximate ages (years since birth year) are marked in decades at the top of the rating graph for a player. (Note that I only give a graph for players with ratings in more than 2 years.)
    • Addition of event lists on the pages for each year, with links to the new pages for each individual event.
    • Addition of pages for each event, with links from the tournament and match data at the bottom of the player pages and from the event lists on the year pages. These event pages list results in the form of scores (and number of games) for each player involved, rather than reproducing entire crosstables. For matches that occurred as part of a tournament (knockout matches or tie-break matches), scores are given for each pairing. Also, event information (name, place, start and end dates) are given where available, along with notes covering especially any discrepancies in information among sources, or any uncertainties about the information. Finally, all references used for information on the event are indicated, from books, periodicals and web sites.
      'Matches' are here defined as two-player events, while 'tournaments' are defined as events involving structured play between more than two players (to determine one or more prize-winners). Thus, inter-club matches are counted as 'tournaments', as is the triangular contest in Paris, 1821, while a series of games between a pair of players at unstructured events count as matches (like at the gathering at Blue Lick, Kentucky in 1847, and early meetings of the Chess Associations, under various names, in Yorkshire from the 1840s).
    • Rating graphs are included for each event with more than one rated player (limited to the highest-rated 16 players where more than 16 were involved).
    • Event results are now classified as 'formal', 'casual', or 'soft'. These are only guesses in many cases and should not be taken as being definitive. All results have in any case been used equivalently in the rating calculations (except in the case of Adolf Anderssen). 'Soft' results are numerical results that I have essentially made up to match as closely as possible what information is known about an event, without giving it too much weight, in cases where it would be a serious mis-representation not to include the event at all.
    • Addition of bibliographic lists (books, periodicals and main web sites).
    • A complete revamping of results, with improvements in accuracy as far as is possible with the sources I've currently searched. Many errors have been corrected and a significant number of new results have been entered into the data set for rating calculations.
    • I have begun to do more historical work to make the often tricky identification of players more reliable, and to accurately locate events (matches and tournaments) in time (and place), as well as to pin down accurate results of events. I have begun to look at primary source material, rather than relying exclusively on secondary sources that are of varying accuracy. So far, I have mainly been using such nineteenth century sources as are available to me for making sporadic checks of results of events I already knew about, though I have occasionally picked up other events.
    • Results from a number of additional sources have been included. In particular, I have made a thorough search through Fiske's book of The First American Chess Congress, Reichhelm's on Chess in Philadelphia, Lawson's Paul Morphy, Hilbert's Essays in American Chess History and The Tragic Life and Short Chess Career of James A. Leonard, 1841-1862, Winter's four books from his Chess Notes column, Chess Explorations, Chess Facts and Fables, A Chess Omnibus and Kings, Commoners, and Knaves, and Cornil's collection of excerpts from nineteenth century sources covering goings-on at the Café de la Régence. A number of additional web sites have proven very useful, including Paul Dunn's Ozbase site, and David Cohen's Canadian Chess site. Actively updated websites, like ChessCafe and Chess Notes, have been mined for data up to the end of 2007 so far, with only occasional bits and pieces since then. I have also received some valuable information on Canadian events and players via correspondence with David Cohen and Erik Malmsten.
    • The Explanations page has been updated to reflect the latest round of changes. Rather than rewriting the text, I have added the updates in red and dated them, except in the 'Results' and 'Future Enhancements' sections, which have been rewritten.
    • I have added comments on two new rating systems that are similar to Edo ratings, 'TrueSkill Through Time' and 'Whole History Ratings', on the Explanations page.
    • I have now included events involving more varied odds, based partly on 19th-century assessments of the effects of giving these odds, and partly on (educated, I hope) guesses. See the changes on the Explanations page for details. Of particular note are matches of Deschapelles involving the 'game of pawns' as it was called, in which one player gave up his queen in exchange for some number of additional pawns. Eight pawns for the queen was considered to give an even game, and other numbers of pawns were given equivalencies in effect to more traditional odds games. These assessments allow such contests to be rated (with an admittedly very rough idea of the rating effect of the handicap), and Deschapelles phenomenal results against de la Bourdonnais in 1836 and de Saint-Amant in 1842 at odds of this type considerably increase his resulting ratings. I would like to point out that these high ratings correspond to Deschapelles' enormous reputation at the time.
    • I have also now rated more tournaments with only partial information, where information about relative playing strengths can still be extracted from them in a mathematically correct way. See the new information on the Explanations page for details.
    • After running the entire rating calculation on my updated data set, I discovered a few errors that will have to wait for the next update. The errors are corrected in the reported data, but are not yet reflected in the ratings. In particular: a result of Valson in 1849 was mistakenly attributed to Davessac; a similar confusion occurred between Sam Loyd and Thomas Loyd; the Buckle-Tuckett match of 1842 was scored 9-7 when it should have been 7-9; B.V. Ljubimov presumably should have been recognized as identical to Boris Vasilievich Liubimov; although Di Felice identifies the player named Roberts at Edinburgh 1892 as Albert Roberts, who was active in Philadelphia in the 1870s, I doubt that these are the same player (without more supporting evidence), and should have separated them.
    • I have just learned through Edward Winter (Chess Note 6450) that the Manhattan Chess Club tournament of 1886 was a handicap event, though neither Gaige nor Di Felice give any such indication with their crosstable, so my ratings are distorted by not taking the handicaps into account. In particular, Hyde and Hartshorne appear very highly rated, but should not be since they received heavy odds from the stronger players in this event. This will also be corrected in the next update.
  • 30 Aug. 2006 - 3 main enhancements:
    • New algorithm: block-structuring of data and efficient exploitation of block structure in calculating variances (which in principle involves inverting a huge matrix). The effects of this change in procedure are entirely invisible: The results are not affected.
    • More data: I have included additional results, particularly some that were added to Spinrad's list for the years 1836 - 1863 after its initial compilation. I have also extended my results database up to the year 1902.
    • Better handling of odds games. I have changed some of my 'educated guesses' at rating effects of giving various odds. I have left the rating disadvantage corresponding to odds of the move at 28 and to odds of pawn and move at 174. Odds of pawn and two moves are set at 226 and knight odds at 347. Previously, I did not include games at any other odds because there were too few to gauge the appropriate rating from the data. However, in some cases it distorts the results more seriously if other odds matches are left out, than if they are included with admittedly speculative rating effects. So I have guessed at a rook odds effect of 500 rating points and an exchange-odds effect of 300 rating points.

      I had been handling odds games by adjusting scores of individual matches or tournaments in an attempt to compensate for a supposed rating disadvantage of giving odds. This worked well for long matches such as the 69-game odds match between Dubois and Wyvill in 1846. But for short matches or knockout tournaments, the approach was not very successful at achieving this compensation. In the extreme case of a 100% or 0% score, the odds adjustment made no compensation at all, because the indicated rating difference in such a case is infinite (i.e. the larger the rating difference, the larger the probability of a 100% - 0% score) with or without the odds. This applied in many results in knockout tournaments, for example, where only one or two games were played between a pair of players. In matches where the score was close to 100% - 0%, the compensation for the odds obtained by adjusting that score alone tends to be insufficient. The more even the result, the better the odds adjustment.

      The improvement I made was to include an odds-giving player as a separate entity in the huge results crosstable, so, for example, there would be a player 'Staunton-1845 giving pawn and two moves' who is distinct from 'Staunton-1845' (not giving odds). The rating we want is the one for the 'real' Staunton, but the matches with Williams, Spreckley and Mongredien in 1845 were with the odds-giving fellow, a sort of 'ghost' Staunton who will be rated lower than the 'real' Staunton by an amount corresponding to the rating adjustment for pawn and two move odds. To fix the relationship between the odds-giving Staunton and the 'real' Staunton, I simply posit 100 games between them with a score that corresponds to the desired rating difference, which will force them to come out of the rating iteration with approximately that desired difference. (Only approximately, because the real games that are played will pull at the two players' ratings - ghost and real - to some extent, but the 100 games should be enough to set the rating difference fairly firmly.) All odds effects are thus fully accounted for, and integrated with the network of other results.

      A particularly problematic example is that of the Wayte vs. Lowenthal matches of 1852. Lowenthal gave Wayte odds of pawn and two moves but lost 6-0. Then, giving odds of the exchange, he lost only 3.5-1.5. By itself, the 6-0 match would have avoided adjustment by my old method, and thus Wayte's rating would be driven well above Lowenthal's. I didn't include exchange odds matches previously, so I also left out the pawn and two move odds match, since by itself it would have given a very misleading picture. With the new system, both can be included.

      The net effect of handling all odds games properly is quite significant. It is not primarily the small changes in my built-in rating differences corresponding to various odds that have caused these significant rating changes. Rather, the main difference is that the rating effect is now fully accounted for by the procedure. The main odds-giving players of the early period have benefitted from this change (or rather, the rating procedure now gives them due credit). In particular, Staunton has higher ratings than in previous versions of the Edo ratings, and now looks more like the dominant player his reputation suggests, though von der Lasa's ratings in the same period are also higher. The most dramatic change, however, is in Morphy's rating, which is not surprising since he played so many games giving odds of one kind or another. He now appears as the strongest player of the nineteenth century, exceeding (just) even Steinitz' peak.

  • 15 Apr. 2006 - layout rearrangement
  • 12 Apr. 2006 - Edo version 2: Major revision of the whole project.
    • I have improved the Edo method, particularly in the way it adjusts for the supposed underlying rating distribution. The reference games I used in the previous version were too ad hoc and despite many attempts to tune them to produce reasonable results, there were always instances where they made obvious distortions to the ratings. I now use a different type of adjustment, with a better theoretical basis and no need for tuning. First, as before, I calculate the 'raw' ratings based on competition scores using the Bradley-Terry method (with the usual self-matches between the same player in consecutive years). This produces a rating estimate along with a rating deviation, which together characterize a distribution of ratings that could have produced the observed results (with the most likely being the mean). But this estimate needs to be adjusted to account for our prior knowledge of a universal distribution of ratings of all chess players. This is done by finding the maximum of the combined rating distribution formed by the product of the score-based individual distribution (the 'raw' rating estimate) and the underlying universal one. This also provides an adjustment to the rating deviation.
    • The rating deviations in the old version were all too large, by about 1.4 times. This resulted from my naive calculation of the standard error of the mean from the Bradley-Terry method, without considering the effect of drawn games. In effect, the possibility of draws in chess means that a single game gives finer information than a binary decision and therefore less uncertainty. These are now correctly calculated, and lowered still further (if only slightly in most instances) by the adjustment described above.
    • The project now covers the entire nineteenth century up to and including the year 1900. Early results are still very sparse, especially before 1840, with only one result before 1821.
    • As well as adding on results for the last 5 years of the nineteenth century, I have included additional results, mostly from Gino Di Felice's Chess Results: 1747-1900. Partly with the help of this book and also with the help of Jeremy Gaige's Chess Personalia, I have also made a further effort to identify players. I give initials as well as last names wherever possible to distinguish players with the same last name. There is still the likelihood that some errors have been made, either because two players have been confused as one, or because one player has been confused as two (or more). This is a surprisingly difficult problem when we are dealing with more obscure players from the nineteenth century. For example, I had decided that the Carlo Schultz who played at Livorno 1878 was not likely the same as the Carl Schultz who played at Dusseldorf 1863 or the Schultz who played at Berlin 1881 or later tournaments in Germany. According to Di Felice, the first two are the same person, but not the third, though he lists this Carl Schultz as a participant in 3 other German tournaments between 1894 and 1900. Though Di Felice is not always completely reliable, I have trusted his identification of players here and in other places.
    • Comparitive ratings by other rating systems are no longer included.
  • 9 May 2005 - Extended results to 1895. Added a few early results, such as the Lewis-Cochrane match of 1821, as reported by Murray in A History of Chess. Re-evaluated the identities of some players, such as Carlo Schultz (who played at Livorno 1878) who I now think is unlikely to be the same person as Carl Schultz, who played at Dusseldorf 1863, or the Schultz who played at Berlin 1881. Adjusted the explanations page, especially the 'interesting examples' to reflect the new results. Added comments on comparison with the new Chessmetrics ratings to the explanations page.
  • 6 Apr. 2005 - Added "World champions by Edo rating" list and "Top 40 peak Edo ratings" list.
  • 27 Mar. 2005 - Added results up to and including 1890, and modified the explanations page to reflect new results. I also included an additional early result here and there, like the match between Cochrane and Mouret (playing as the Turk) in the early 1820's (I've listed it as occurring in 1821, though this is uncertain). Also, I attempted to reconcile player names, to avoid the two types of error that can result from incorrect identification: the same player listed as two or more players, or different players confused as one. The latter is probably the more serious. This can be very difficult for some less well-known players. Which results listed for "Bauer" for example belong to J.D.Bauer and which to W. Bauer, both active in the 1880's? I went through references with a reasonable amount of care and sorted out a number of problem cases, but an error-free accounting of which results belong to which players would be a historical research project requiring a huge amount more effort and time. Players for which I know first names or initials are now indicated with initials. This is extremely helpful in sorting out who is who, but many sources (including Gaige in vol.1) only give surnames. Many potential errors remain. For example, a "Lowe" played in a tournament in London in 1886. Is this the E. Lowe who was active in London from 1847 to 1858? It's possible (barely) but I don't know. In unlikely cases like this, I've listed them as two separate players.
  • 29 Jan. 2005 - Added results up to and including 1885 (which changes all the ratings somewhat, of course, given the nature of the method), and modified the 'interesting cases' on the explanations page in the light of the new ratings. Added this update page.
  • 12 Jan. 2005 - Added tournament and match results to player pages, so that you can see what data is going into the rating system. Remember that tournament scores may not always agree with published results, because I've endeavoured to exclude all defaulted games and to include draws that were not counted towards scores in some early tournaments. Also, made some minor formatting corrections and fixed up an occasional data error.
  • Sept. 2004 - Web site created with years 1809-1883.

Back to main Edo rating page