|
Post by The Bofa on the Sofa on Oct 29, 2014 11:20:11 GMT -5
Thanks for a very clear analysis, BOFA. Many of the issues that arise over ranking or comparing teams have their roots in a fundamental misunderstanding of the differences between RPI, Pablo, and RPI futures. In order to understand the differences, one must know what it is they actually do. I'm curious how you would describe Pablo? Is it a predictor, a probabilistic estimator (of something?). I understand the RPI future to be an estimate of what the final RPI will be. How would describe the function/purpose of RPI. I guess the way to think about it is a probability estimator for a hypothetical matchup of two teams at the current moment in time. Even for teams that have played and therefore have an outcome in the past, the rankings reflect the probability of the hypothetical matchup. Each week, a lot of those hypothetical matchups are played, although typically in a single event. Consequently, the outcomes are subject to the problems associated with interpreting a single event in light of a probability, which is a property of a population. However, we can just view the events as flipping weighted coins. Back to the issue of teams who have already played. The main difference between Pablo and something like RPI is that Pablo does not consider wins and losses to be absolute, and works on the premise that not all wins are equal in terms of determining win probabilities. This is shown clearly by the results I have shown above. So in the end, Pablo is saying, hey, I see that you won the match against team Y and scored 54.5% of the points. That result tells me that your chance of beating them again the next time you play is 76%, so I am going to separate you by TTT ratings points, which is the ratings difference that corresponds to 76% (the relationship between winning pct and rating difference is not linear). Do this for all of the matches that you've played, and all the teams to find a happy set of values that best describes all of you. In this approach, I am not concerned about "postdicting" wins and losses, in two respects. First, as noted, points is a better predictor of what's going to happen in those upcoming matches, and you can win and be outscored and vice versa. But second, and this is important, the whole premise of the probabilistic model is that not only can upsets happen, but they MUST happen. If there were no upsets, Pablo wouldn't work. Now, the likelyhood of an upset is not the same everywhere. Upsets are more likely when teams are closer (duh) and less likely when teams are more different. Put in terms of reflecting what happened, I think about it in terms of what does it mean for the future. I don't need a ranking system to tell me who won a match that has already been played. It has to be about predicting the outcome of a match that hasn't been played, since that applies to most of the pairings on the list. And in doing this, it applies just as much to to teams that have played previously as to those who haven't. For the teams that have played, we have to think about what that outcome is telling us, and that gets reflected in the rankings, so that outcome is absolutely reflected in the rankings, but the rankings are not meant to tell us about that outcome. I hope this helps and makes sense in some way.
|
|
|
Post by vbman100 on Oct 29, 2014 11:22:40 GMT -5
'Irrespective' and 'detailedly' Nice use of words Bofa. It is kind of interesting that there are predictors and algorithms that can be calculated, when all of these W/L results and points scored are determined by humans. Is this information broken down by conference as well? For example, what is the winning repeatability % in the Pac 12, B1G 14, WCC, Big West, as opposed to the MEAC, SWAC, Colonial, Patriot? I would also be interested to see 'Pablo winning repeatability' for Men's NCAA matches, since sometimes they play the same team on back to back nights. And just see the differences for the men's and women's game.
|
|
|
Post by The Bofa on the Sofa on Oct 29, 2014 11:27:28 GMT -5
Thanks for a very clear analysis, BOFA. Many of the issues that arise over ranking or comparing teams have their roots in a fundamental misunderstanding of the differences between RPI, Pablo, and RPI futures. In order to understand the differences, one must know what it is they actually do. I'm curious how you would describe Pablo? Is it a predictor, a probabilistic estimator (of something?). I understand the RPI future to be an estimate of what the final RPI will be. How would describe the function/purpose of RPI. As for RPI... RPI is just an attempt to put win/loss record into the context of schedule strength. I won't argue about whether it does that effectively or not, but the concept is right. W/L record is highly dependent on the quality of your competition.
|
|
|
Post by The Bofa on the Sofa on Oct 29, 2014 11:44:01 GMT -5
'Irrespective' and 'detailedly' Nice use of words Bofa. It is kind of interesting that there are predictors and algorithms that can be calculated, when all of these W/L results and points scored are determined by humans. Is this information broken down by conference as well? For example, what is the winning repeatability % in the Pac 12, B1G 14, WCC, Big West, as opposed to the MEAC, SWAC, Colonial, Patriot? I would also be interested to see 'Pablo winning repeatability' for Men's NCAA matches, since sometimes they play the same team on back to back nights. And just see the differences for the men's and women's game. Yeah, I've never done this type of analysis on the men's side. I suspect that conceptually the concept would be the same, but that the actual values might change some. I could probably get the data from MHPR, I'm sure he's kept it all. There is not as much, of course. My full data set of match pairs is almost 25000 pairs of matches (5 years worth of data). However, in my women's database, which includes D1, d2, d3 and naia, there are in fact more than 200 pairs of matches played on the same day! When teams play on the same day, the same team wins both 90% of the time. However, that is due to the fact that the teams are very lopsided mostly due to schedule bias, I think. For example, in those matches, the average point pct in the first match is 61%, so of course the same team wins the second match. In fact, if you look at the table above, 90% is pretty much what you expect for when the team wins the first match with 61% of the points. It's actually pretty interesting, because the average time between matches in the database is 30 days, so you would expect that there would be some loss of predictability due to their being a month between matches, but the result with 0 days shows that there isn't much of a time effect on these probabilities. One of the first things I did with this database is to look at the effect of time on predictability, and one of the first things I learned is that I had way overstated the effect in Pablo. I have a D1 only dataset, but there aren't as many matches there, so it's harder to look into the nitty gritty. In the broad things I have compared, there is little difference between the full set and the d1 set, but there's not enough data to look at things as detailedly as I would like (I know you like that word; irrespective is a perfectly good descriptor, though). Once you get to individual conferences, there won't be near enough data to draw any conclusions.
|
|
bluepenquin
Hall of Fame
4-Time VolleyTalk Poster of the Year (2019, 2018, 2017, 2016), All-VolleyTalk 1st Team (2021, 2020, 2019, 2018, 2017, 2016)
Posts: 12,423
|
Post by bluepenquin on Oct 29, 2014 12:44:32 GMT -5
Thanks for a very clear analysis, BOFA. Many of the issues that arise over ranking or comparing teams have their roots in a fundamental misunderstanding of the differences between RPI, Pablo, and RPI futures. In order to understand the differences, one must know what it is they actually do. I'm curious how you would describe Pablo? Is it a predictor, a probabilistic estimator (of something?). I understand the RPI future to be an estimate of what the final RPI will be. How would describe the function/purpose of RPI. I guess the way to think about it is a probability estimator for a hypothetical matchup of two teams at the current moment in time. Even for teams that have played and therefore have an outcome in the past, the rankings reflect the probability of the hypothetical matchup. Each week, a lot of those hypothetical matchups are played, although typically in a single event. Consequently, the outcomes are subject to the problems associated with interpreting a single event in light of a probability, which is a property of a population. However, we can just view the events as flipping weighted coins. Back to the issue of teams who have already played. The main difference between Pablo and something like RPI is that Pablo does not consider wins and losses to be absolute, and works on the premise that not all wins are equal in terms of determining win probabilities. This is shown clearly by the results I have shown above. So in the end, Pablo is saying, hey, I see that you won the match against team Y and scored 54.5% of the points. That result tells me that your chance of beating them again the next time you play is 76%, so I am going to separate you by TTT ratings points, which is the ratings difference that corresponds to 76% (the relationship between winning pct and rating difference is not linear). Do this for all of the matches that you've played, and all the teams to find a happy set of values that best describes all of you. In this approach, I am not concerned about "postdicting" wins and losses, in two respects. First, as noted, points is a better predictor of what's going to happen in those upcoming matches, and you can win and be outscored and vice versa. But second, and this is important, the whole premise of the probabilistic model is that not only can upsets happen, but they MUST happen. If there were no upsets, Pablo wouldn't work. Now, the likelyhood of an upset is not the same everywhere. Upsets are more likely when teams are closer (duh) and less likely when teams are more different. Put in terms of reflecting what happened, I think about it in terms of what does it mean for the future. I don't need a ranking system to tell me who won a match that has already been played. It has to be about predicting the outcome of a match that hasn't been played, since that applies to most of the pairings on the list. And in doing this, it applies just as much to to teams that have played previously as to those who haven't. For the teams that have played, we have to think about what that outcome is telling us, and that gets reflected in the rankings, so that outcome is absolutely reflected in the rankings, but the rankings are not meant to tell us about that outcome. I hope this helps and makes sense in some way. I think most people have been made aware of how RPI can 'mess up' when comparing two teams where the team that lost scored more points - leadiing to a difference between Pablo and RPI. But I think baried in there is something much more fundemental to how Pablo is superior to RPI. RPI can treat any win against a 'bad' opponent as a bad win. However, Pablo treats any win or loss (or more specifically % of points scored) in the context of the team they played. No matter how bad Pittsburgh beats Bowling Green - they will end up worse in their RPI for having played the game. Pablo would look at the games scores (-19,-7,-14) and treat that different then if Pittsburgh had struggled some in winning. Put another way - Pittsburgh and Purdue where already limited in RPI by what was on their schedule, while Pablo allows for them to improve or fail based on how they play against some of their poor opponents.
|
|
|
Post by mikegarrison on Oct 29, 2014 13:03:26 GMT -5
I guess the way to think about it is a probability estimator for a hypothetical matchup of two teams at the current moment in time. Even for teams that have played and therefore have an outcome in the past, the rankings reflect the probability of the hypothetical matchup. Each week, a lot of those hypothetical matchups are played, although typically in a single event. Consequently, the outcomes are subject to the problems associated with interpreting a single event in light of a probability, which is a property of a population. However, we can just view the events as flipping weighted coins. Back to the issue of teams who have already played. The main difference between Pablo and something like RPI is that Pablo does not consider wins and losses to be absolute, and works on the premise that not all wins are equal in terms of determining win probabilities. This is shown clearly by the results I have shown above. So in the end, Pablo is saying, hey, I see that you won the match against team Y and scored 54.5% of the points. That result tells me that your chance of beating them again the next time you play is 76%, so I am going to separate you by TTT ratings points, which is the ratings difference that corresponds to 76% (the relationship between winning pct and rating difference is not linear). Do this for all of the matches that you've played, and all the teams to find a happy set of values that best describes all of you. In this approach, I am not concerned about "postdicting" wins and losses, in two respects. First, as noted, points is a better predictor of what's going to happen in those upcoming matches, and you can win and be outscored and vice versa. But second, and this is important, the whole premise of the probabilistic model is that not only can upsets happen, but they MUST happen. If there were no upsets, Pablo wouldn't work. Now, the likelyhood of an upset is not the same everywhere. Upsets are more likely when teams are closer (duh) and less likely when teams are more different. Put in terms of reflecting what happened, I think about it in terms of what does it mean for the future. I don't need a ranking system to tell me who won a match that has already been played. It has to be about predicting the outcome of a match that hasn't been played, since that applies to most of the pairings on the list. And in doing this, it applies just as much to to teams that have played previously as to those who haven't. For the teams that have played, we have to think about what that outcome is telling us, and that gets reflected in the rankings, so that outcome is absolutely reflected in the rankings, but the rankings are not meant to tell us about that outcome. I hope this helps and makes sense in some way. I think most people have been made aware of how RPI can 'mess up' when comparing two teams where the team that lost scored more points - leadiing to a difference between Pablo and RPI. But I think baried in there is something much more fundemental to how Pablo is superior to RPI. RPI can treat any win against a 'bad' opponent as a bad win. However, Pablo treats any win or loss (or more specifically % of points scored) in the context of the team they played. No matter how bad Pittsburgh beats Bowling Green - they will end up worse in their RPI for having played the game. Pablo would look at the games scores (-19,-7,-14) and treat that different then if Pittsburgh had struggled some in winning. Put another way - Pittsburgh and Purdue where already limited in RPI by what was on their schedule, while Pablo allows for them to improve or fail based on how they play against some of their poor opponents. RPI doesn't know anything about any win (except maybe for the secret adjustments?). It knows records, not wins. It doesn't care who you win or lose to, only what your overall record was and what their overall records were. You don't get more or less credit with RPI for beating good or bad teams. You do get more or less credit in RPI for playing good or bad teams.
|
|
bluepenquin
Hall of Fame
4-Time VolleyTalk Poster of the Year (2019, 2018, 2017, 2016), All-VolleyTalk 1st Team (2021, 2020, 2019, 2018, 2017, 2016)
Posts: 12,423
|
Post by bluepenquin on Oct 29, 2014 14:08:34 GMT -5
I think most people have been made aware of how RPI can 'mess up' when comparing two teams where the team that lost scored more points - leadiing to a difference between Pablo and RPI. But I think baried in there is something much more fundemental to how Pablo is superior to RPI. RPI can treat any win against a 'bad' opponent as a bad win. However, Pablo treats any win or loss (or more specifically % of points scored) in the context of the team they played. No matter how bad Pittsburgh beats Bowling Green - they will end up worse in their RPI for having played the game. Pablo would look at the games scores (-19,-7,-14) and treat that different then if Pittsburgh had struggled some in winning. Put another way - Pittsburgh and Purdue where already limited in RPI by what was on their schedule, while Pablo allows for them to improve or fail based on how they play against some of their poor opponents. RPI doesn't know anything about any win (except maybe for the secret adjustments?). It knows records, not wins. It doesn't care who you win or lose to, only what your overall record was and what their overall records were. You don't get more or less credit with RPI for beating good or bad teams. You do get more or less credit in RPI for playing good or bad teams. Correct. What I am trying to point out is that even winning a match against a team with a bad record will be worse for your RPI then not having played that game. I referred this to a 'bad' win. This is also what makes RPI so darn predictable. And is way different than how Pablo would treat matches between uneven opponents.
|
|
|
Post by alpacaone on Oct 30, 2014 8:24:58 GMT -5
I'll be brief, the data is incomplete; if you are using total points as a factor and not all the participants value total points, it's as if you were doing a study on the health benefits of aspirin and applying results if people were even using aspirin. How many coaches play the "I need to win total points" card? How many forfeit sets not caring too much if the team uses their energy battling for every point in a lobsided set? How many make a change or try something different against a clearly weaker opponent? How many times are bench players subbed in to finish things off? Clearly all these effect your total point analysis. I feel it is also clear that these are reasons why this rating would not be very good for seedings. NCAA is much more about sportsmanship about giving many more players opportunities to play than other legues.
One more thing than you can have the last word; it is your baby and I value it. Unless we are going to consede that wins and losses have no relevance, I don't think I with straight face could propose this ranking as a seeded criteria. I could not imagine any committee putting Stanford behind PSU for any reason, Stanford defeated PSU and both of the sides it lost to. At this point perhaps an arguement could be made for North Carolina being seeded ahead of FSU if not for a head to head loss, and how could one even imagine Illinois with 5 losses, one a shared 3-1 loss be somewhere in between? I understand your reasoning, but Some of it is based on objectives that aren't in place, and if they were then how much would valuing things like total points scored against even much weaker teams, or value on winning total points change the student athlete game?
|
|
|
Post by mikegarrison on Oct 30, 2014 9:15:22 GMT -5
I'll be brief, the data is incomplete; if you are using total points as a factor and not all the participants value total points, it's as if you were doing a study on the health benefits of aspirin and applying results if people were even using aspirin. How many coaches play the "I need to win total points" card? How many forfeit sets not caring too much if the team uses their energy battling for every point in a lobsided set? How many make a change or try something different against a clearly weaker opponent? How many times are bench players subbed in to finish things off? Clearly all these effect your total point analysis. I feel it is also clear that these are reasons why this rating would not be very good for seedings. NCAA is much more about sportsmanship about giving many more players opportunities to play than other legues. One more thing than you can have the last word; it is your baby and I value it. Unless we are going to consede that wins and losses have no relevance, I don't think I with straight face could propose this ranking as a seeded criteria. I could not imagine any committee putting Stanford behind PSU for any reason, Stanford defeated PSU and both of the sides it lost to. At this point perhaps an arguement could be made for North Carolina being seeded ahead of FSU if not for a head to head loss, and how could one even imagine Illinois with 5 losses, one a shared 3-1 loss be somewhere in between? I understand your reasoning, but Some of it is based on objectives that aren't in place, and if they were then how much would valuing things like total points scored against even much weaker teams, or value on winning total points change the student athlete game? Sorry, but this just makes no sense. Is there any coach in the world who does not want to score a point during any given rally? No? So if a coach plays to win every individual point, how can you say they aren't playing to maximize their total points? It's clear to any of us who have been using this for a while that pablo really does work as advertised. It finds the teams more likely to win in upcoming matches. Whether you want to seed the tournament based on "the best teams" or based on "the teams that have had the best results" is up to you. It's clear that you prefer the latter.
|
|
|
Post by s0uthie on Oct 30, 2014 10:03:53 GMT -5
I'll be brief, the data is incomplete; if you are using total points as a factor and not all the participants value total points, it's as if you were doing a study on the health benefits of aspirin and applying results if people were even using aspirin. How many coaches play the "I need to win total points" card? How many forfeit sets not caring too much if the team uses their energy battling for every point in a lobsided set? How many make a change or try something different against a clearly weaker opponent? How many times are bench players subbed in to finish things off? Clearly all these effect your total point analysis. I feel it is also clear that these are reasons why this rating would not be very good for seedings. NCAA is much more about sportsmanship about giving many more players opportunities to play than other legues. One more thing than you can have the last word; it is your baby and I value it. Unless we are going to consede that wins and losses have no relevance, I don't think I with straight face could propose this ranking as a seeded criteria. I could not imagine any committee putting Stanford behind PSU for any reason, Stanford defeated PSU and both of the sides it lost to. At this point perhaps an arguement could be made for North Carolina being seeded ahead of FSU if not for a head to head loss, and how could one even imagine Illinois with 5 losses, one a shared 3-1 loss be somewhere in between? I understand your reasoning, but Some of it is based on objectives that aren't in place, and if they were then how much would valuing things like total points scored against even much weaker teams, or value on winning total points change the student athlete game? Sorry, but this just makes no sense. Is there any coach in the world who does not want to score a point during any given rally? No? So if a coach plays to win every individual point, how can you say they aren't playing to maximize their total points? It's clear to any of us who have been using this for a while that pablo really does work as advertised. It finds the teams more likely to win in upcoming matches. Whether you want to seed the tournament based on "the best teams" or based on "the teams that have had the best results" is up to you. It's clear that you prefer the latter. He is talking about coaches who may ease up on a bad team by playing their whole roster. He's talking about coaches who may experiment with lineups or let the team "play through" vs. taking a timeout. Many (smart) coaches are scheduling to give themselves the best chance at a high RPI. If the goal were a higher Pablo rating, perhaps that would influence certain in-game decisions. In the same way that the original BCS formula partially used a margin of victory measurement which led to some running up the score by power teams, you would probably see a bit less "sportsmanship" in volleyball if you incentivize point-scoring. None of this changes the valid points other people have made: -Point scoring is far and away the best predictive measure of future success. -Ratings are used to assign a likelihood of an event occurring. -Rankings are a reflection of what actually occurred, regardless of the odds for it to happen at the time or again in the future. (Upsets happen!) -RPI, though flawed, is the law of the land and good coaches schedule accordingly. If you're unhappy with your team's RPI, schedule (and hopefully beat) better RPI teams in non-conference play.
|
|
|
Post by n00b on Oct 30, 2014 10:12:37 GMT -5
I'll be brief, the data is incomplete; if you are using total points as a factor and not all the participants value total points, it's as if you were doing a study on the health benefits of aspirin and applying results if people were even using aspirin. How many coaches play the "I need to win total points" card? How many forfeit sets not caring too much if the team uses their energy battling for every point in a lobsided set? How many make a change or try something different against a clearly weaker opponent? How many times are bench players subbed in to finish things off? Clearly all these effect your total point analysis. I feel it is also clear that these are reasons why this rating would not be very good for seedings. NCAA is much more about sportsmanship about giving many more players opportunities to play than other legues. One more thing than you can have the last word; it is your baby and I value it. Unless we are going to consede that wins and losses have no relevance, I don't think I with straight face could propose this ranking as a seeded criteria. I could not imagine any committee putting Stanford behind PSU for any reason, Stanford defeated PSU and both of the sides it lost to. At this point perhaps an arguement could be made for North Carolina being seeded ahead of FSU if not for a head to head loss, and how could one even imagine Illinois with 5 losses, one a shared 3-1 loss be somewhere in between? I understand your reasoning, but Some of it is based on objectives that aren't in place, and if they were then how much would valuing things like total points scored against even much weaker teams, or value on winning total points change the student athlete game? Sorry, but this just makes no sense. Is there any coach in the world who does not want to score a point during any given rally? No? So if a coach plays to win every individual point, how can you say they aren't playing to maximize their total points? It's clear to any of us who have been using this for a while that pablo really does work as advertised. It finds the teams more likely to win in upcoming matches. Whether you want to seed the tournament based on "the best teams" or based on "the teams that have had the best results" is up to you. It's clear that you prefer the latter. I disagree, I think his post makes perfect sense. Coaches start seniors on senior day. Coaches often start non-starters against teams they know they are going to beat leading to closer set scores. Coaches put in backups late in sets to get them playing time late in sets that they are going to win. Do they still want to win those points? Of course. But they aren't doing everything they can to win those points. They're conceding some points in order to make they're bench players happy. Nearly every team does this and it likely only has an extremely small impact on actual Pablo ratings, so in no way am I implying these things make Pablo less accurate. HOWEVER, if the committee announced that they were going to choose the 32 At Large teams based strictly on Pablo, coaches would undoubtedly coach differently. If you know that point differential could determine your inclusion or not, coaches of bubble teams will not play backups and will run up the score. It's why the NCAA selection committee doesn't like rankings with point differentials and it's why the BCS didn't allow computer rankings with point differential included. Running up the score goes against the NCAA's goals. I like Pablo a lot and am very grateful for it. That doesn't mean I think it should be used for tournament selection. (I could be more convinced that it should be used for seeding because if Penn State gets underseeded, it doesn't really punish PSU much, but it REALLY punishes the top seed whose bracket they get put in)
|
|
|
Post by The Bofa on the Sofa on Oct 30, 2014 10:17:43 GMT -5
Unless we are going to consede that wins and losses have no relevance, I don't think I with straight face could propose this ranking as a seeded criteria. If there were no relationship between wins and losses and Pablo, no one would pay any attention to it. But there is. And with good reason: there is a very strong relationship between scoring points and winning matches. In fact, far and away, the team that scores more points in a match wins. And those matches where they don't? It's usually close enough that everyone can recognize that, yeah, this could have gone either way. In fact, how good is Pablo at reflecting who won and lost? <b>JUST AS GOOD AS RPI!!!!</b> That's right, even though Pablo puts very little premium on winning, and focuses on those factors that are predictive, if you look at how well Pablo and RPi reflect matches that have already taken place, you discover that there is in fact very little difference in their overall results. The main difference is that Pablo does a lot better at inter-regional matches and doesn't systematically overrate teams from the south and east and underrate teams from the west. So the dirty little secret, which isn't really a secret to anyone paying attention, is that while the stated motivation of RPI is to solely evaluate what a team has done, it doesn't even do that better than Pablo. Despite the fact that Pablo cares very little about who won, it does just was well (maybe a tad better, but I won't claim it; certainly not worse) as RPI at reflecting who even won the matches that already been played. Something like Massey rankings, which are based on W/L records, do even better in that regard. And whether you would propose it as a selection/seeding criteria is pretty irrelevant, considering that this is what the coaches, led by the AVCA, tried to do. In all seriousness, and with a complete straight face. They put a lot of time and resources into it, and had a lot of discussions with the NCAA and the selection committee. It was serious, and, unfortunately, it got nowhere.
|
|
|
Post by mikegarrison on Oct 30, 2014 10:18:49 GMT -5
In the same way that the original BCS formula partially used a margin of victory measurement which led to some running up the score by power teams, you would probably see a bit less "sportsmanship" in volleyball if you incentivize point-scoring. But volleyball is not the same as these other sports. You can't "run up the score." You have to play every set to 25. Are you saying there are coaches who advocate throwing a certain number of points in a match in the name of "sportsmanship"? Anyway, Washington has experimented with lineups this year, and also has pulled starters out of matches where the opponent was clearly overmatched. Hasn't stopped them from ending up number 1 in pablo this week. People have been raising these concerns for a long time about pablo, but there has never been any actual evidence that the concerns are valid.
|
|
|
Post by BeachbytheBay on Oct 30, 2014 10:19:56 GMT -5
I'll be brief, the data is incomplete; if you are using total points as a factor and not all the participants value total points, it's as if you were doing a study on the health benefits of aspirin and applying results if people were even using aspirin. How many coaches play the "I need to win total points" card? How many forfeit sets not caring too much if the team uses their energy battling for every point in a lobsided set? How many make a change or try something different against a clearly weaker opponent? How many times are bench players subbed in to finish things off? Clearly all these effect your total point analysis. I feel it is also clear that these are reasons why this rating would not be very good for seedings. NCAA is much more about sportsmanship about giving many more players opportunities to play than other legues. One more thing than you can have the last word; it is your baby and I value it. Unless we are going to consede that wins and losses have no relevance, I don't think I with straight face could propose this ranking as a seeded criteria. I could not imagine any committee putting Stanford behind PSU for any reason, Stanford defeated PSU and both of the sides it lost to. At this point perhaps an arguement could be made for North Carolina being seeded ahead of FSU if not for a head to head loss, and how could one even imagine Illinois with 5 losses, one a shared 3-1 loss be somewhere in between? I understand your reasoning, but Some of it is based on objectives that aren't in place, and if they were then how much would valuing things like total points scored against even much weaker teams, or value on winning total points change the student athlete game? i think the discussion gets too complicated somethimes Pablo is concerned about wins and losses, however in layman's terms, Pablo uses points won (factored as part of the Pablo formula) because points won is statistically the best predictor of future points won, and points wons correlates extremely high with wins. notwithstanding that sure coaches don't always get concerned about points one or on occasion sub to get players experience, etc. - but those are instances that by and large are exceptions to the rule RPI, by virtue of ingnoring points, doesn't account as well as it could for the quality of opponents - hence it ends up with bias (and the bias is in layman's view, is a bias that goes generally in favor of east/southeast and bias against the west) - so perceived 'bad' western teams are likely to score points in defeat that they don't get credit for in RPI and using RPI for BOTH initial ranking and for determining 'quality' wins-losses COMPOUNDS it's flaws.
|
|
|
Post by The Bofa on the Sofa on Oct 30, 2014 10:26:25 GMT -5
Coaches start seniors on senior day. Coaches often start non-starters against teams they know they are going to beat leading to closer set scores. Coaches put in backups late in sets to get them playing time late in sets that they are going to win. Do they still want to win those points? Of course. But they aren't doing everything they can to win those points. They're conceding some points in order to make they're bench players happy. And despite all this, the relationship between point percentage in the first match and the winning percentage in the second match is indistinguishable from what Pablo expects it to be. So either the stuff you claim is happening isn't happening, or it's not significant. The only place where we see deviation from the model is if we look at blowouts, but there the issue is not that teams do better than expected based on points, which would suggest they "let up" in blowouts, but that teams don't do as good as expected after blowouts, which is exactly opposite of what you'd expect if teams were letting off the gas. If anything, it is like the other team gives up.
|
|