Okay, last season I tried with the main stats to try and gauge the standards of our players - found the whole process too tedious to maintain and didn't get close to anywhere the level of accuracy I'd hoped for - although it did tell us that our rockets wasn't too bad. Just thought for this year, I will try a new approach - something close to the Elo system employed in the Chess world - which ranks its players according to the strength of their plays against each other. What I like is that - it is a self-correcting system (i.e. the more we use it, the closer we get towards determining the relative strength of the person/team being assessed). Please note it is a relative ranking system - so it only make sense within defined constraints. Criticism, suggestion and feedback are most welcomed. So here goes.... Starting premise - in NBA we have 30 teams, who plays each other over the season in a 82 games schedule. So we begin by defining an arbitary range of scores beginning with the lowest ranked team all the way to the top team. In an ideal world, all 30 teams would have sorted themselves out ranking from 1st to 30th. And in that same ideal world, the rank score differentials between each ranked team would be a consistent gap. We could choose any value to define the gap - but for my purpose I like the number 82. So we will score the lowest ranked team with a score of 82 (no team will ever score below this score) and by adding 82 consistently upwards, we will end up with a ceiling (1st ranked team) holding the max score of 2460. i.e. so the defined limits for our ranking system is a score range of 82 (min) up to 2460 (max). For interest sake - an average team should theoretically hold on to a score value of 1271 (i.e. a team that wins 41 out of 82 games). Using this range of scoring, by looking at last season win/loss values of our NBA teams - we start with a baseline score of them (from which we will use to calculate changes in rankings hereforth). (Calculated using W/82 * 2460 + (82-W) rounded to the nearest 1 since I hate decimals ;-) ). So to start our new 2013/14 system - here are our baseline scores (in order of highest to lowest): Heats - 1996 Thunder - 1822 Spurs - 1764 Nuggets - 1735 Clippers - 1706 Grizzlies - 1706 Knicks - 1648 Pacers - 1536 Nets - 1536 Warriors - 1445 Bulls - 1387 Lakers - 1387 Rockets - 1387 Hawks - 1358 Jazz - 1329 Celtics - 1271 Mavericks - 1271 Bucks - 1188 76ers - 1068 Raptors - 1068 Trailblazers - 1039 Timberwolves - 981 Wizards - 923 Pistons - 923 Kings - 894 Pelicans - 865 Suns - 807 Cavaliers - 778 Bobcats - 691 Magic - 662 Note that this is starting estimation of rank based on last year's scores - which is obviously not going to be accurate since there have been trades and what nots but it would be a useful point to start (we gotta start somewhere). Like I mentioned the system I will be using is self-correcting, so as the season progresses, I hope to see a better fit in terms of the teams' rank scores to their actual expected performance. But this baseline has its uses as well. We all know our Rockets should in theory have improved over last year - but how much improved? No one can really say - but with this rank scores - we can compare the baseline score to say the mid-season score (which should already show us how much "improved" our team has become). Now, I need to mention that the rank scores will change as the teams start to play each other and here's how it will be calculated. Let's take for example the 1st game coming up on Oct 29, Magic (662) vs Pacers (1536). Here's what the scores mean - the relative scores between Magic and Pacers simply translate to Magic having a 30% chance of winning versus Pacers having a 70% chance of winning. As I have decided on a 82-point system - this just means that for an evenly matched team (i.e. if they have the same rank scores, both sides would have a 50/50 chance of winning - the winner would get 41 points and loser lose 41 points) but for uneven odds like 30/70 - the +/- between lower rank team versus higher rank team would work out to be of the ratio of around 25/57. So what does this mean? It simply means that if Pacers win (which is expected), they get only a 25 points increase and Magic will get a 25 points decrease. On the other hand if Magic was to win, Magic will get a 57 points increase and Pacers would suffer a 57 points decrease (i.e. a differential of 104 points!). Yes, I can already hear some of you screaming isn't this unfair to the higher ranked team - why yes in fact it is, but it also means that that team might have been ranked wrongly too high so will be brought back to their norm faster. So enough about other teams - how about the situation for our Rockets and our first game against Bobcats on Oct 30? Here is what's at risk - Rockets (1387) vs Bobcats (691) - translates to a 67:33 ratio in our favor - the points at risk would be 27/55 i.e. if Rockets win we get +27 and Bobcats -27 but if Rockets lose it would be Rockets -55 and Bobcats +55. I will try and update as the season progress (and when I have the time to do so - please bear with me as this works in progress). Now to another interesting thing - I remember last season there were many post lamenting about Rockets being placed in the toughest division - but just how tough it was no one could really nail a figure to it - here's what my method has come up with as a means to show how tough the various divisions were: The averaged rank scores of teams in Atlantic Div was 1318 Central Div was 1196 Southeast Div was 1126 Northwest Div was 1381 Pacific Div was 1248 Southwest Div was 1399 Yes, we now have a value-method in comparison with the other divisions to show that indeed SW was really the toughest of the lot - by a whooping 273 differential to the lowest div (SE) - so what does 273 means - if we divide this by 82 - it means we are about 3 ranks more difficult than them (please note not 3 times, but 3 ranking positions - my own interpretation just means that in general we play opponents in SW who should be ranked 3 ranks higher than those in SE on average). Even the comparison of the Western conference vs Eastern conference shows a pretty significant difference: The averaged rank score for the West is 1343 vs East 1214 - a differential of 129 (or about 1.57 rank difference, if you like we could round it up to 2 ranks). Now note that we are dealing with averages here - but if I were to try and convert say to a percentage difference in difficulty .... a SW team in general should have a 55% chance of winning against a SE team (45%) - a difference of 10%. Likewise a WCF team has a 53% chance of winning against the ECF (47%) - a differential of 6%. So it may mean that it is 5% more difficult for us to play against an SW team compared to an SE and 3% more difficult for us to play against a WCF team as compared to an ECF team. So what's the aim for our rockets this year? I think people have been thinking we ought to take the SW division crown - but what rank score should we achieve to get us there? By my own estimates - I think a rank score of about 1745 should bring us there (but take note last year Spurs got a rank score of 1764). To me - if our Rockets can achieve or surpass a rank score of 1745 (from our current of 1387) I would be happy (by the way this represent something like a jump of at least 4 rank levels)..... Have fun - questions are welcome.... Go rockets!
Unstable I think this is a great idea. Play around with it and then use a single thread to update us at reasonable intervals (not daily). I will be interested in seeing the progress. You might also consider playing around with folding in other metrics like RAPM, Wins Produced, Hollinger's Power Rankings, etc. At the beginning of the season of course, before the data obtained gathers its own significant "weight" so to speak.
Thanks for the suggestions, I don't think folks here would be interested in the rank values of the other teams so am only likely to do so after each rockets game (hopefully). Will try this at the team level before exploring the options on the player level, I think hollingers method although not ideal is pretty tough to displace. The method I am employing is just relative strength estimation so would require direct competition as a means to gauge - currently at the player would be challenging to do, as the way basketball games go, it's not easy to gauge a 1 on 1 player vs player performance but I may muck around using last season's stats and see what I can come up with....
Dude you even added an extra letter to TLDR, not to mention the two exclamation points. You definitely had time to read this interesting post. I'm kind of a nerdy fan of this kind of thing. I appreciate that the OP spent a lot of time on this. Wouldn't mind seeing where this goes vs. real results next season.
As a side interest - I had completed calculating some values based on Players' rankings and tried a comparison i.e. averaging player rank points and comparing against team rank points to see if there differences mean anything. For example for Rockets - team rank score (based on last season stats) - came up to be 1387 Our new player averages (again based on last year's stats but adding Dwight to the mix) - came up to 1595 (Lin (1428) + Harden (1797) + Parsons (1010) + Asik (1485) + Howard (2256)/5). This is in fact short of the 1745 target if we want to get the Div Championship. So what does this mean? (i.e. how can I make use of this set of calculations I have churned out)? Theoretically - this means if we are aiming to change Asik to a PF - we need to aim for someone with a score of 2234 and above - forwards scoring this much are: Tim Duncan (2257) and Dwight Howard (2256) - but we already have Howard. So in theory if we want to be Championship contender we should get Duncan to replace Asik. (but of course there is a risk that Duncan's really over his sell-by date). On the other hand, if we are only replacing Parsons - we would only need to look for a Forward scoring at 1759 and above. There are 9 others (not counting Howard): Tim Duncan S Ibaka A. Bynum L. Sanders R. Hibbert J Noah K Durant B. Lopez J. Smith strange eh.... yeah I know it looks weird....
I get the premise .... But I don't see the starting point as being accurate as there are some obvious GM's (2400+ is a GM under the USCF ranking) among the 30 teams, Miami & OKC for sure .... and some others that should probably be ranked much lower. I think this could be solved by calculating last years wins / losses with the same system you plan to use from this point forward. Also players defeating an opponent with a gap of 400 or greater receive a 0 (zero) increase tho this may overcomplicate your system .... I haven't played tournament chess in over two decades but had a 2200+ ranking at one time & a 2360 postal ranking. My brain hurts just thinking about the complexity now ....
Hi Corrosion, thanks for your points, my system is not using the same yardstick (or ruler) as USCF. As for now the working range of ranking scores runs from 82 (lowest) all the way to 2460 (highest) - so if we are looking for GM equivalent levels, we will probably want to consider top 10% (i.e. top 3 teams of a 30 teams system) - so an equivalent value for GM-like teams would be any team that scores 2296 and above. Currently it doesn't seem that any team has hit that mark yet - but that's probably because I was just using last year's win/loss ratio to calculate a baseline - as the season progress, the GM-like teams will gain enough points based on head-to-head performances to likely breach that mark. Using the current yardstick - if we split the league into 3 blocks of 10 teams, the top tier teams should hover 1722 and above and bottom tier teams should be 820 and below. It would be interesting to see if we do actually get a Normal distribution at all for NBA, but the way NBA seem to be working we might see a compressed ND i.e. the bulk of the teams (i.e. 80% scrunched in the mid-tier with only 3 (or maybe 4) teams in the top/bottom tiers.
may be a dumb question but would it make it more interesting if every team started with the same score of 1271 (the score every team should theoretically have) just wondering correct me if im wrong
If we are to receive updates on this ranking, I think we need to give it a name to refer to. You said it is similar to the ELO system in chess. Can we just call this the ELO rankings? What does ELO stand for btw? I'm not a chess player. I'm much more into Catan.
Sorry I should have added to my previous post. I think this is a good point. From season to season, everyone begins on an even playing field starting with a 0-0 record. While I understand the need to reward teams who upset the favourites, you technically cannot consider anyone a favourite at the beginning of the season. Yes, I understand that is not the case in reality where Heat are the favourite, but I just don't think it seems right to start this ranking this way. This ELO ranking needs to be self correcting as of the start of this season. I think the ranking will take form after only a month.
No. You are not wrong, and yes we could have started every team on the same footing every season. The system will correct itself over the games and by the end of the season we will get the relative ranking sorted out. As to the reasons why I have chosen to make use of last season's stats to make a baseline - (1) it's to try and get the teams as close to their potential rank as possible. i.e. by starting everyone in the middle - it will take a longer time for the champion team to get to the top of the pile (and the bottom team to get to the bottom of the pile). By using an estimated rank value, we are trying to shorten the self-correcting process. (2) as a means to gauge if the teams have improved or regressed in performance (in comparison to last season).
Elo is someone's name - it was the person who devised the ranking method for the Chess World. Not so sure, that I want to give it a name as yet - since this is still works in progress. I may tweak it mid-season when I gauge how far off we may have varied from expectations. Maybe we can call this RRS (relative rank scoring)...
If the playoffs started tomorrow, the Rockets would lose to the following teams in a 7 game series. Spurs, Grizzlies, Clippers, OKC, Heat, Pacers and Bulls. Teams that would be a toss up...Golden State, Brooklyn, Dallas, Minnesota.
how can you possibly say this. we don't even know what dwight will do on this team,heck we still have 2 positions being debated on who will start for the starting 5.. very ignorant statement
As per suggested by one fellow member, I was going through the data of last season, and realized that it would take me a fair bit of time to compute the rank values, so please bear with me while I work things out. But while i was working on this, I realized that I needed to establish some kind of points assignment rules and controls, here's a summary of what I plan to use as guides for computation. 1. Max rank value for team/player would be 2460 2. Minimum rank value would be 82 3. Provisional rank value would be 1271 4. There are 3 results when teams meet I.e. win, lose or draw (I will not be counting ot results) 5. Rv adjustment computation will be dependent on the difference in rv between them and outcome of their match in regular time. Team with higher rv will only gain point if they win (but they may not gain any points if the rv difference becomes higher than about 5 grade difference which is about 410 points - minimum gain points 1 max 82 per match. For draw results max gain would be capped at 41 points. There is a graded difference in addition an subtraction which is a tad complicated to explain using text.) I hope to post something up when I get to around the mark where all teams have completed at least 10 matches. I am currently up to about 85 games, sorry I have my day job to take care of, so can only do this on the little spare time I have. At current rate I am not sure I can complete this before our regular season starts, but I will give it my best shot.... Thanks for reading and your interest
Just a sampling of the progress of work with last season's data (after having gone through about 100 matches - about at least 5 games per team except for NYK) - but we can start to see some of the teams already pulling away from the norm.... This is on the basis that all teams started with the same provisional RV of 1271 Team (games played) - RV ATL (6)- 1293 BKN (5) - 1293 BOS (7) - 1353 CHA (5) - 1230 CHI (7) - 1294 CLE (7) - 1200 DAL (8) - 1250 DEN (8) - 1221 DET (8) - 1073 GSW (7) - 1270 HOU (7) - 1207 IND (7) - 1237 LAC (7) - 1363 LAL (7) - 1284 MEM (6) - 1420 MIA (8) - 1354 MIL (6) - 1317 MIN (7) - 1336 NOH (5) - 1308 NYK (4) - 1419 OKC (8) - 1365 ORL (6) - 1175 PHI (7) - 1295 PHX (8) - 1288 POR (7) - 1193 SAC (7) - 1192 SAS (7) - 1409 TOR (7) - 1232 UTA (8) - 1238 WAS (5) - 1120
