Pool C

[KnightSlappy]

Batting average is winning percentage. It’s how many hits (wins) did I get without considering the quality of the pitcher (opponent).

OWP is like asking “how many hits (or wins, I’ve already lost the metaphor) would a neutral team get in a specific number of at-bats considering the quality of the pitcher.

If I’m facing each pitcher for one at bat, then all I care about is the rate at which he allows hits.

Verlander (.200)
Scherzer (.250)
Porcello (.300)

We’d expect to get a hit after facing these three pitchers 25% of the time. Even if Verlander has thrown to many more batters than Porcello has.

OWP, in it’s own way, is a measure of how many wins a neutral team would get given it’s schedule.

The number of batters he’s previously faced only matters to the degree that it truly reveals his true talent level. This is probably the sticking point -- that we don’t know a team’s true talent winning percentage, but we feel we know a 3-24 team is actually bad when a 1-8 team might have simply run into a string of bad luck.

The proper thing to do, though, when presented with “not enough data” is either to (1) scale down all numbers to the one with the fewest population size or (2) regress the numbers by adding in a number of games at a league-average rate (.500 in this case).

[Pat Coleman]

If I play a 20 game schedule, some games should count for more then 1/20th and some less than 1/20th? That is exactly what happens when the SOS calculation depends on the total number of games each individual opponent has played. And that is my issue.

Those aren't your games. They're your opponents' games.

Your games are measured in your winning percentage and your opponents' games are measured in the SOS.

This has been fun and you can continue to argue it if you like but you won't be arguing it with me. Wasted too much of my time today on something that is basically a Ctrl-Z of a mistake in the handbook.

[ziggy]

Sheesh - it only took all day to come up with this thought:

I think the basis to this is that the strength of a team's schedule percentage should be based on the fact that they play more games than another team who may have the same WP, but less games doesn't necessarily mean more strength.

I am not sure if that makes complete sense... but I don't know if you can say a team has an equal strength of schedule by playing a number of games less than another team.

Team A is 15-5 in region - their strength is thus the fact they have played 20 games and won 15 of them.
Team B is 9-3 in region - same percentage, but they don't appear to have as strong a schedule because they have played 8 fewer games and won 6 fewer.

So why should we be allowing the 9-3 team to have an same strength of schedule as a 15-5 team? They have played far less games and haven't proven their .750 WP is as legit as the team that has played those 20 games. However, in the former math... we considered them even.

But they haven't proven it isn't and therein lies the problem. With the eight extra games they could be 17-3 or they could be 9-11. The point is that we don't know, all we know is what they are and that is .750.

[ziggy]

If I play a 20 game schedule, some games should count for more then 1/20th and some less than 1/20th? That is exactly what happens when the SOS calculation depends on the total number of games each individual opponent has played. And that is my issue.

Those aren't your games. They're your opponents' games.

Your games are measured in your winning percentage and your opponents' games are measured in the SOS.

What I mean is 1/20th towards SOS.

[KnightSlappy]

If I play a 20 game schedule, some games should count for more then 1/20th and some less than 1/20th? That is exactly what happens when the SOS calculation depends on the total number of games each individual opponent has played. And that is my issue.

Those aren't your games. They're your opponents' games.

Your games are measured in your winning percentage and your opponents' games are measured in the SOS.

This has been fun and you can continue to argue it if you like but you won't be arguing it with me. Wasted too much of my time today on something that is basically a Ctrl-Z of a mistake in the handbook.

This is at least the fifth year in a row it's been written in the handbook as an average of the percentages.

[Pat Coleman]

It has been written that way, yet it has not been calculated that way, a point I have made at least twice previously today. I would guess it's probably because they borrowed the wording from D-I, which actually has an RPI, while D-III does not, and therefore does not average its SOSs.

[Hugenerd]

Right, but someone facing two batters back to back isn't going to be more likely to give up a hit to the person who had more at-bats.

Thing is, that just isn't what we're measuring here. We're measuring a team's batting average, akin to what sac said:

I'm not sure I see the horribleness, when you figure out a baseball teams batting avg you don't avg the percentages, nor do you average the percentages of a player to find the career average.

We're not predicting how a team would do when playing Adrian (like we would predict this pitcher facing two batters), we're measuring how a group of teams performed (a group of batters). And if one team happened to play more measurable games than the other, so be it. It should count more.

If I play a 20 game schedule, some games should count for more then 1/20th and some less than 1/20th? That is exactly what happens when the SOS calculation depends on the total number of games each individual opponent has played. And that is my issue.

You have to remember the home and away weighting also, an away game is already weighted 2/3 higher than a home game, so your argument about each game counting toward 1/20th of the SOS doesnt make sense to begin with. 

I tend to like the new system better because it takes each of your opponents games as an individual event.  Therefore, for every game your opponent plays (that counts according to the NCAA criteria) you get either 1.25, 1.0, or 0.75 points (Home, Neutral, Away) towards your OWP statistic in either the Win or Loss column.  Then, you average over all events to give the OWP statistic.  In the previous method, whether your opponent played 4, 10 or 20 games, they were all treated the same.  As the number of events increase, the certainty of the statistic also increases.  Meaning that I have more confidence that a 12-12 team is a 0.500 team, than a 1-1 team, a 2-2 team, or even a 5-5 team, because we just dont have a lot of information on those teams yet.  The same analogy can be drawn to baseball as has already been discussed.  Do you have more confidence in someone who has gone 1-3 on the season to get a hit or someone who has gone 33-99?  The same thing is true here. When you have such small samples of data, your certainty in that team's WP is low. Thats why I have a problem with the previous batting average examples, you are never going to get 600 observations in basketball games for a single team in a season.  By averaging the WP for each opponent, you collapse the number of observations to the number of regional games played. Conversely, by doing it this way, you approximately square the number of observations.    As the number of events increases, your confidence in the true OWP of that team increases.  Scaling linearly with the number of events is the easiest way of doing this (which the NCAA has incorporated) and it could be debated whether it is the best way.  For example, in statistics, critical values for a t-stat are not linear, above 20 or 30 events you begin to approach the infinite observation t-stat.  However, for this purpose, I have absolutely no problem with what the NCAA is doing and think it is definitely an improvement on the alternative that is being debated.

[AO]

Right, but someone facing two batters back to back isn't going to be more likely to give up a hit to the person who had more at-bats.

Thing is, that just isn't what we're measuring here. We're measuring a team's batting average, akin to what sac said:

I'm not sure I see the horribleness, when you figure out a baseball teams batting avg you don't avg the percentages, nor do you average the percentages of a player to find the career average.

We're not predicting how a team would do when playing Adrian (like we would predict this pitcher facing two batters), we're measuring how a group of teams performed (a group of batters). And if one team happened to play more measurable games than the other, so be it. It should count more.

If I play a 20 game schedule, some games should count for more then 1/20th and some less than 1/20th? That is exactly what happens when the SOS calculation depends on the total number of games each individual opponent has played. And that is my issue.

You have to remember the home and away weighting also, an away game is already weighted 2/3 higher than a home game, so your argument about each game counting toward 1/20th of the SOS doesnt make sense to begin with. 

I tend to like the new system better because it takes each of your opponents games as an individual event.  Therefore, for every game your opponent plays (that counts according to the NCAA criteria) you get either 1.25, 1.0, or 0.75 points (Home, Neutral, Away) towards your OWP statistic in either the Win or Loss column.  Then, you average over all events to give the OWP statistic.  In the previous method, whether your opponent played 4, 10 or 20 games, they were all treated the same.  As the number of events increase, the certainty of the statistic also increases.  Meaning that I have more confidence that a 12-12 team is a 0.500 team, than a 1-1 team, a 2-2 team, or even a 5-5 team, because we just dont have a lot of information on those teams yet.  The same analogy can be drawn to baseball as has already been discussed.  Do you have more confidence in someone who has gone 1-3 on the season to get a hit or someone who has gone 33-99?  The same thing is true here. When you have such small samples of data, your certainty in that team's WP is low. Thats why I have a problem with the previous batting average examples, you are never going to get 600 observations in basketball games for a single team in a season.  By averaging the WP for each opponent, you collapse the number of observations to the number of regional games played. Conversely, by doing it this way, you approximately square the number of observations.    As the number of events increases, your confidence in the true OWP of that team increases.  Scaling linearly with the number of events is the easiest way of doing this (which the NCAA has incorporated) and it could be debated whether it is the best way.  For example, in statistics, critical values for a t-stat are not linear, above 20 or 30 events you begin to approach the infinite observation t-stat.  However, for this purpose, I have absolutely no problem with what the NCAA is doing and think it is definitely an improvement on the alternative that is being debated.

it's not merely a "measure of confidence" of the OWP, it is a penalty.  If you play Nebraska Wesleyan and have only 5 non-conference games, you likely won't make up the difference.

[Dave 'd-mac' McHugh]

In other words, the old way was based on the assumption a 9-3 team was going to like a 15-5 team should the 9-3 team have played 20 games. You can't base SOS numbers on assumptions... they should be based on hard numbers.

And if you decide to play Nebraska Wesleyan or the like, that is a decision that coach has made. It isn't like we have teams who all of the sudden change their minds and only play half their games in region. NW and others have long stranding track records or scenarios which everyone knows... and rewarding a team for playing a 3-3 NW the same as playing a 10-10 team doesn't make any sense.

If all games count when the committee looks at selecting Pool Bs and Cs and bracketing the teams, then the SOS should count all games and not the average.

[Ryan Scott (Hoops Fan)]

Right, but someone facing two batters back to back isn't going to be more likely to give up a hit to the person who had more at-bats.

Thing is, that just isn't what we're measuring here. We're measuring a team's batting average, akin to what sac said:

I'm not sure I see the horribleness, when you figure out a baseball teams batting avg you don't avg the percentages, nor do you average the percentages of a player to find the career average.

We're not predicting how a team would do when playing Adrian (like we would predict this pitcher facing two batters), we're measuring how a group of teams performed (a group of batters). And if one team happened to play more measurable games than the other, so be it. It should count more.

If I play a 20 game schedule, some games should count for more then 1/20th and some less than 1/20th? That is exactly what happens when the SOS calculation depends on the total number of games each individual opponent has played. And that is my issue.

You have to remember the home and away weighting also, an away game is already weighted 2/3 higher than a home game, so your argument about each game counting toward 1/20th of the SOS doesnt make sense to begin with. 

I tend to like the new system better because it takes each of your opponents games as an individual event.  Therefore, for every game your opponent plays (that counts according to the NCAA criteria) you get either 1.25, 1.0, or 0.75 points (Home, Neutral, Away) towards your OWP statistic in either the Win or Loss column.  Then, you average over all events to give the OWP statistic.  In the previous method, whether your opponent played 4, 10 or 20 games, they were all treated the same.  As the number of events increase, the certainty of the statistic also increases.  Meaning that I have more confidence that a 12-12 team is a 0.500 team, than a 1-1 team, a 2-2 team, or even a 5-5 team, because we just dont have a lot of information on those teams yet.  The same analogy can be drawn to baseball as has already been discussed.  Do you have more confidence in someone who has gone 1-3 on the season to get a hit or someone who has gone 33-99?  The same thing is true here. When you have such small samples of data, your certainty in that team's WP is low. Thats why I have a problem with the previous batting average examples, you are never going to get 600 observations in basketball games for a single team in a season.  By averaging the WP for each opponent, you collapse the number of observations to the number of regional games played. Conversely, by doing it this way, you approximately square the number of observations.    As the number of events increases, your confidence in the true OWP of that team increases.  Scaling linearly with the number of events is the easiest way of doing this (which the NCAA has incorporated) and it could be debated whether it is the best way.  For example, in statistics, critical values for a t-stat are not linear, above 20 or 30 events you begin to approach the infinite observation t-stat.  However, for this purpose, I have absolutely no problem with what the NCAA is doing and think it is definitely an improvement on the alternative that is being debated.

it's not merely a "measure of confidence" of the OWP, it is a penalty.  If you play Nebraska Wesleyan and have only 5 non-conference games, you likely won't make up the difference.

But isn't that the point?  They've been trying for years to force teams, even geographically isolated ones, to play d3 teams.  This is simply incentivizing something they've been trying to incentivize.  If NebWes wants d3 games against good teams, they're going to have to play more d3 games.

[sac]

Everything would be better if the NCAA would just adopt the D-III Championship Belt^TM.

problem solved

[Hugenerd]

Right, but someone facing two batters back to back isn't going to be more likely to give up a hit to the person who had more at-bats.

Thing is, that just isn't what we're measuring here. We're measuring a team's batting average, akin to what sac said:

I'm not sure I see the horribleness, when you figure out a baseball teams batting avg you don't avg the percentages, nor do you average the percentages of a player to find the career average.

We're not predicting how a team would do when playing Adrian (like we would predict this pitcher facing two batters), we're measuring how a group of teams performed (a group of batters). And if one team happened to play more measurable games than the other, so be it. It should count more.

If I play a 20 game schedule, some games should count for more then 1/20th and some less than 1/20th? That is exactly what happens when the SOS calculation depends on the total number of games each individual opponent has played. And that is my issue.

You have to remember the home and away weighting also, an away game is already weighted 2/3 higher than a home game, so your argument about each game counting toward 1/20th of the SOS doesnt make sense to begin with. 

I tend to like the new system better because it takes each of your opponents games as an individual event.  Therefore, for every game your opponent plays (that counts according to the NCAA criteria) you get either 1.25, 1.0, or 0.75 points (Home, Neutral, Away) towards your OWP statistic in either the Win or Loss column.  Then, you average over all events to give the OWP statistic.  In the previous method, whether your opponent played 4, 10 or 20 games, they were all treated the same.  As the number of events increase, the certainty of the statistic also increases.  Meaning that I have more confidence that a 12-12 team is a 0.500 team, than a 1-1 team, a 2-2 team, or even a 5-5 team, because we just dont have a lot of information on those teams yet.  The same analogy can be drawn to baseball as has already been discussed.  Do you have more confidence in someone who has gone 1-3 on the season to get a hit or someone who has gone 33-99?  The same thing is true here. When you have such small samples of data, your certainty in that team's WP is low. Thats why I have a problem with the previous batting average examples, you are never going to get 600 observations in basketball games for a single team in a season.  By averaging the WP for each opponent, you collapse the number of observations to the number of regional games played. Conversely, by doing it this way, you approximately square the number of observations.    As the number of events increases, your confidence in the true OWP of that team increases.  Scaling linearly with the number of events is the easiest way of doing this (which the NCAA has incorporated) and it could be debated whether it is the best way.  For example, in statistics, critical values for a t-stat are not linear, above 20 or 30 events you begin to approach the infinite observation t-stat.  However, for this purpose, I have absolutely no problem with what the NCAA is doing and think it is definitely an improvement on the alternative that is being debated.

it's not merely a "measure of confidence" of the OWP, it is a penalty.  If you play Nebraska Wesleyan and have only 5 non-conference games, you likely won't make up the difference.

And if you play a non-region game, you get 0 games credit. That seems like a pretty steep penalty, but that's the way it is.

[Greek Tragedy]

LAST WEEK'S RESULTS (records not updated)

ATL	TEAM	CONF.	IN-REGION	OVERALL	SOS	SCHEDULE
1	Ramapo	NJAC	18-1	20-2	0.500	LOST at Rutgers-Newark 72-69; vs Richard Stockton 2/9 (rescheduled 2/11)
2	Old Westbury	SKY	18-1	19-3	0.517	WON at NYU-Poly 93-81; BEAT SUNY-Maritime 82-62
3	SUNY-Purchase	SKY	16-4	16-4	0.522	LOST at Trinity (Conn) 68-66; BEAT Farmingdale St. 84-65
4	Ricard Stockton	NJAC	16-5	16-5	0.519	BEAT Rutgers-Camden 60-45; at Ramapo 2/9 (rescheduled 2/11)
5	Rutgers-Newark	NJAC	15-6	16-6	0.525	BEAT Ramapo 72-69;WON at Rowan 73-69
EAST	TEAM	CONF.	IN-REGION	OVERALL	SOS	SCHEDULE
1	Rochester	UAA	18-1	19-1	0.580	WON at Chicago 68-57; LOST at Washington U. 72-53
2	Cortland St.	SUNYAC	17-2	17-3	0.514	BEAT Oneonta St 84-56; BEAT Oswego St. 78-46
3	Stevens	E8	15-3	17-3	0.540	WON at Baruch 70-67; BEAT Elmira 59-46; BEAT Ithaca 87-65
4	NYU	UAA	13-7	13-7	0.612	LOST at Case Western Reserve 71-61;LOST at Carnegie Mellon 56-55
5	Hobart	LL	12-6	13-6	0.596	BEAT Union 85-82; WON at Vassar 61-53; WON at Bard 76-59
6	SUNY-Geneseo	SUNYAC	14-6	14-7	0.534	LOST at Potsdam St. 75-61; LOST at Plattsburgh St. 84-62
GT LK	TEAM	CONF.	IN-REGION	OVERALL	SOS	SCHEDULE
1	Wooster	NCAC	17-3	17-3	0.552	BEAT Ohio Wesleyan 74-67 OT; BEAT Wittenberg 75-71
2	Ohio Wesleyan	NCAC	16-3	16-4	0.541	LOST at Wooster 74-67 OT; BEAT Hiram 63-61
3	Thomas More	PrAC	17-2	18-3	0.475	WON at Westminster (Pa) 83-77; BEAT Thiel 72-62
4	Calvin	MIAA	15-0	19-2	0.414	WON at Alma 90-66; LOST at Hope 73-70
5	St. Vincent	PrAC	14-3	16-5	0.493	WON at Geneva 74-66; WON at Bethany 69-66
6	Marietta	OAC	16-5	16-5	0.488	LOST to Capital 70-67; BEAT Baldwin-Wallace 72-70
MID-ATL	TEAM	CONF.	IN-REGION	OVERALL	SOS	SCHEDULE
1	Catholic	LAND	16-2	19-2	0.534	WON at Susquehanna 74-53; LOST at Scanton 82-74
2	Albright	MACC	18-3	18-3	0.554	LOST at Hood 70-60; LOST to Stevenson 82-81
3	St. Mary's (Md.)	CAC	14-2	19-2	0.546	BEAT Frostburg St. 73-53; WON at Marymount 74-64
4	Alvernia	MACC	16-4	16-4	0.562	BEAT York (Pa) 66-55; WON at Arcadia 64-47; BEAT Widener 75-59
5	Wesley	CAC	14-2	17-5	0.509	BEAT Mary Washington 67-52; WON at York (Pa) 69-61
6	Scranton	LAND	15-6	15-6	0.539	WON at Drew 77-68; BEAT Catholic 82-74
7	Arcadia	MACC	13-6	13-8	0.579	LOST to Alvernia 64-47; LOST to Lycoming 82-76
8	Cabrini	CSAC	14-4	16-5	0.490	WON at Cairn 99-76; BEAT Gwynedd-Mercy 80-68; BEAT Rosemont 96-66
9	F & M	CC	13-4	16-5	0.512	WON at Gettysburg 56-49; LOST at Swarthmore 73-63
MW	TEAM	CONF.	IN-REGION	OVERALL	SOS	SCHEDULE
1	Illinois Wesleyan	CCIW	15-3	18-3	0.560	BEAT North Park 88-44; WON at Carthage 78-59
2	Transylvania	HCAC	15-4	16-5	0.565	WON at Mt. St. Joseph 65-39; WON at Manchester 91-47
3	Wheaton (IL)	CCIW	13-5	16-5	0.577	BEAT Augustana 58-57; WON at Milikin 62-52
4	Washington U.	UAA	15-4	16-4	0.544	BEAT Emory 68-65; BEAT Rochester 72-53
5	NCC	CCIW	16-3	18-3	0.525	WON at Millikin 59-58; WON at Augustana 76-62
6	Rose-Hulman	HCAC	18-2	19-2	0.502	BEAT Franklin 72-63; WON at Defiance 63-62
7	Augustana	CCIW	15-5	16-5	0.549	LOST at Wheaton 58-57; LOST to North Central 76-62
8	St. Nobert	MWC	15-4	15-4	0.525	BEAT Carroll 63-50; LOST at Grinnell 104-99
NE	TEAM	CONF.	IN-REGION	OVERALL	SOS	SCHEDULE
1	WPI	NEWMAC	21-0	21-0	0.548	BEAT Wheaton (Mass) 86-59; LOST at Springfield 66-60
2	Amherst	NESCAC	20-2	20-2	0.553	WON at Williams 65-48; at Middlebury 2/9 (rescheduled 2/12)
3	Williams	NESCAC	18-2	20-2	0.553	LOST to Amherst 65-48; vs. Trinity (Conn) 2/9 (rescheduled 2/12)
4	Middlebury	NESCAC	16-1	19-1	0.527	BEAT Lyndon State 89-59; BEAT Trinity (Conn) 66-59; vs. Amherst 2/9 (rescheduled 2/12
5	RIC	LEC	18-3	18-3	0.544	BEAT Mass-Dartmouth 71-60; BEAT Southern Maine 64-46
6	Brandeis	UAA	15-5	15-5	0.588	WON at Carnegie Mellon 73-68; LOST at Case Western Reserve 56-47
7	MIT	NEWMAC	15-4	16-4	0.553	WON at Babson 69-53; vs. Clark 2/9 (rescheduled 2/11)
8	Curry	CCC	15-6	15-6	0.555	BEAT Western New England 71-60; at Endicott 2/9 (rescheduled 2/14)
9	Westfield St.	MASCAC	15-4	17-4	0.517	WON at Massachusetts College 71-69; BEAT Western Connecticut 72-59
10	Springfield	NEWMAC	15-7	15-7	0.559	BEAT WPI 66-60
11	Eastern Conn.	LEC	14-4	14-7	0.517	BEAT Western Conn. 64-51; at Mass-Dartmouth 2/9 (rescheduled 2/11)
12	Albertus Mag.	GNAC	20-2	20-3	0.454	vs. St. Joseph's (Maine) 2/9 (rescheduled 2/14)
SOUTH	TEAM	CONF.	IN-REGION	OVERALL	SOS	SCHEDULE
1	Hamp-Syd	ODAC	15-2	19-2	0.529	WON at Randolph-Macon 66-47; BEAT Bridgewater 95-49
2	MHB	ASC	18-3	18-3	0.552	BEAT Schreiner 79-65; BEAT Texas Lutheran 74-54
3	Chris. Newport	USAC	14-3	15-3	0.537	LOST to Virg. Wesleyan 79-66; LOST at LaGrange 73-68; at Piedmont (cancelled)
4	Concordia (TX)	ASC	14-4	16-5	0.536	BEAT Texas Lutheran 92-86 ; BEAT Schreiner 115-92
5	Emory	UAA	14-5	14-5	0.561	LOST at Washington U. 68-65; WON at Chicago 82-59
6	Virginia Wesleyan	ODAC	11-5	14-6	0.510	WON at Christ New 79-66; WON at East Menn 70-65; WON at Guilford 70-62
7	Lynchburg	ODAC	12-5	16-5	0.502	BEAT Roanoke 63-61; WON at Emory and Henry 91-79
8	Guilford	ODAC	13-5	16-5	0.530	LOST at Randolph 72-70; LOST to Virginia Wesleyan 70-62
WEST	TEAM	CONF.	IN-REGION	OVERALL	SOS	SCHEDULE
1	St. Thomas	MIAC	20-1	20-1	0.545	BEAT Bethel 75-64; WON at Gustavus Adolphus 68-63; WON at St. John's 93-68
2	Whitworth	NWC	19-1	20-1	0.546	LOST to Whitman 93-90; LOST to George Fox 89-81
3	Stevens Point	WIAC	18-4	18-4	0.593	BEAT Oshkosh 62-52
4	Whitewater	WIAC	16-4	17-4	0.606	BEAT Platteville 64-50; BEAT Superior 84-53
5	Stout	WIAC	16-4	17-4	0.538	LOST at La Crosse 62-48; BEAT at Eau Claire 70-54
6	Buena Vista	IIAC	15-5	16-5	0.557	LOST to Dubuque 73-58; BEAT Luther 70-41
7	Augsburg	MIAC	15-5	15-5	0.526	BEAT St. John's 106-87; LOST at Carleton 63-58; BEAT St. Olaf 82-71
8	Luther	IIAC	14-4	15-6	0.497	BEAT Central 72-66 OT ; LOST at Buena Vista 70-41
9	Concord. Moorhead	MIAC	15-6	15-7	0.509	BEAT St. John's 74-67; WON at Hamline 90-78

[ziggy]

In other words, the old way was based on the assumption a 9-3 team was going to like a 15-5 team should the 9-3 team have played 20 games. You can't base SOS numbers on assumptions... they should be based on hard numbers.

And if you decide to play Nebraska Wesleyan or the like, that is a decision that coach has made. It isn't like we have teams who all of the sudden change their minds and only play half their games in region. NW and others have long stranding track records or scenarios which everyone knows... and rewarding a team for playing a 3-3 NW the same as playing a 10-10 team doesn't make any sense.

If all games count when the committee looks at selecting Pool Bs and Cs and bracketing the teams, then the SOS should count all games and not the average.

Right, and the hard numbers say a 9-3 team is a .750 team, just like a 15-5 team. Both teams have achieved equally in their opportunities.

[ronk]

Right, but someone facing two batters back to back isn't going to be more likely to give up a hit to the person who had more at-bats.

Thing is, that just isn't what we're measuring here. We're measuring a team's batting average, akin to what sac said:

I'm not sure I see the horribleness, when you figure out a baseball teams batting avg you don't avg the percentages, nor do you average the percentages of a player to find the career average.

We're not predicting how a team would do when playing Adrian (like we would predict this pitcher facing two batters), we're measuring how a group of teams performed (a group of batters). And if one team happened to play more measurable games than the other, so be it. It should count more.

If I play a 20 game schedule, some games should count for more then 1/20th and some less than 1/20th? That is exactly what happens when the SOS calculation depends on the total number of games each individual opponent has played. And that is my issue.

You have to remember the home and away weighting also, an away game is already weighted 2/3 higher than a home game, so your argument about each game counting toward 1/20th of the SOS doesnt make sense to begin with. 

I tend to like the new system better because it takes each of your opponents games as an individual event.  Therefore, for every game your opponent plays (that counts according to the NCAA criteria) you get either 1.25, 1.0, or 0.75 points (Home, Neutral, Away) towards your OWP statistic in either the Win or Loss column.  Then, you average over all events to give the OWP statistic.  In the previous method, whether your opponent played 4, 10 or 20 games, they were all treated the same.  As the number of events increase, the certainty of the statistic also increases.  Meaning that I have more confidence that a 12-12 team is a 0.500 team, than a 1-1 team, a 2-2 team, or even a 5-5 team, because we just dont have a lot of information on those teams yet.  The same analogy can be drawn to baseball as has already been discussed.  Do you have more confidence in someone who has gone 1-3 on the season to get a hit or someone who has gone 33-99?  The same thing is true here. When you have such small samples of data, your certainty in that team's WP is low. Thats why I have a problem with the previous batting average examples, you are never going to get 600 observations in basketball games for a single team in a season.  By averaging the WP for each opponent, you collapse the number of observations to the number of regional games played. Conversely, by doing it this way, you approximately square the number of observations.    As the number of events increases, your confidence in the true OWP of that team increases.  Scaling linearly with the number of events is the easiest way of doing this (which the NCAA has incorporated) and it could be debated whether it is the best way.  For example, in statistics, critical values for a t-stat are not linear, above 20 or 30 events you begin to approach the infinite observation t-stat.  However, for this purpose, I have absolutely no problem with what the NCAA is doing and think it is definitely an improvement on the alternative that is being debated.

I doubt these points are being expressed over on the D1 message boards.