Human Polls vs. Algorithms

[visitingcock]

Okay, Let’s talk Human Polls vs. Algorithms by comparing the NCAA Net rankings to the AP Poll as of today. Computer with algorithms vs. Humans with eyeballs. There are several glaring discrepancies.

First, let’s compare Oregon’s ranking of #4 by the computers to #13 by the humans. The Ducks started out in the first NET ranking as number 4. They lost three games to good teams (Stanford, Arizona, and UCLA) during January. The first two were closely fought games. The last was a 16 point loss where they scored only 41 points. They have beaten nobody in the top 25 of the human or computer rankings. The press dropped them significantly after their 6 point loss.

Let's compare NC State's ranking of #9 by the computers to number 2 by the humans. The Pack started out in the first ranking as number 7. They have lost no games and have a win over then-#1 (now#3/4 SCar).

Thus, we have a situation where three losses are considered better than one win. "Good losses are better than "Good wins" to a computer.
The human in me rejects that and I can see nothing to justify the Duck's higher ranking. Oh yeah, I have eyeballs. I saw the Ducks loss to Arizona. How are they ranked higher than Arizona by the algorithm?

Somebody else can pick out human problems if they want

NCAA NET

1	1	UConn	Big East	10-0
2	3	Stanford	Pac-12	12-2
3	2	South Carolina	SEC	12-1
4	4	Oregon	Pac-12	11-3
5	6	Baylor	Big 12	10-2
6	5	Louisville	ACC	14-0
7	8	UCLA	Pac-12	9-2
8	9	Indiana	Big Ten	9-3
9	7	NC State	ACC	11-0
10	10	Maryland	Big Ten	11-1
11	12	Ohio St.	Big Ten	8-1
12	11	Michigan	Big Ten	10-1
13	13	Georgia	SEC	12-2
14	18	Tennessee	SEC	10-3
15	16	Arizona	Pac-12	11-2
16	14	Texas A&M	SEC	14-1
17	15	South Fla.	AAC	10-1
18	19	Gonzaga	WCC	13-2
19	17	Kentucky	SEC	11-4
20	20	Rutgers	Big Ten	5-3
21	21	Georgia Tech	ACC	8-3
22	22	West Virginia	Big 12	11-2
23	23	Missouri St.	MVC	7-2
24	25	SFA	Southland	12-2
25	32	Bucknell	Patriot	6-0

AP Poll

1	Louisville (22)	14-0		1
2	North Carolina State (5)	11-0		2
3	UConn (1)	10-0		3
4	South Carolina (2)	12-1		4
5	UCLA	9-2		6
6	Stanford	12-2		5
7	Maryland	11-1		7
8	Texas A&M	14-1		8
9	Baylor	10-2		9
10	Arizona	11-2		10
11	Oregon	11-3		13
12	Michigan	10-1		11
13	South Florida	10-1		14
14	Ohio State	8-1		17
15	Kentucky	11-4		12
16	Indiana	9-3		16
17	DePaul	9-3		18
18	Gonzaga	13-2		20
19	Arkansas	11-5		15
20	Tennessee	10-3		25
21	Mississippi State	8-4		19
22	Georgia	12-2		22
23	Northwestern	8-3		21
24	West Virginia	11-2		NR
25	South Dakota State	13-2		NR

 

[Bama fan]

Okay, Let’s talk Human Polls vs. Algorithms by comparing the NCAA Net rankings to the AP Poll as of today. Computer with algorithms vs. Humans with eyeballs. There are several glaring discrepancies.

First, let’s compare Oregon’s ranking of #4 by the computers to #13 by the humans. The Ducks started out in the first NET ranking as number 4. They lost three games to good teams (Stanford, Arizona, and UCLA) during January. The first two were closely fought games. The last was a 16 point loss where they scored only 41 points. They have beaten nobody in the top 25 of the human or computer rankings. The press dropped them significantly after their 6 point loss.

Let's compare NC State's ranking of #9 by the computers to number 2 by the humans. The Pack started out in the first ranking as number 7. They have lost no games and have a win over then-#1 (now#3/4 SCar).

Thus, we have a situation where three losses are considered better than one win. "Good losses are better than "Good wins" to a computer.
The human in me rejects that and I can see nothing to justify the Duck's higher ranking. Oh yeah, I have eyeballs. I saw the Ducks loss to Arizona. How are they ranked higher than Arizona by the algorithm?

Somebody else can pick out human problems if they want

NCAA NET

1	1	UConn	Big East	10-0
2	3	Stanford	Pac-12	12-2
3	2	South Carolina	SEC	12-1
4	4	Oregon	Pac-12	11-3
5	6	Baylor	Big 12	10-2
6	5	Louisville	ACC	14-0
7	8	UCLA	Pac-12	9-2
8	9	Indiana	Big Ten	9-3
9	7	NC State	ACC	11-0
10	10	Maryland	Big Ten	11-1
11	12	Ohio St.	Big Ten	8-1
12	11	Michigan	Big Ten	10-1
13	13	Georgia	SEC	12-2
14	18	Tennessee	SEC	10-3
15	16	Arizona	Pac-12	11-2
16	14	Texas A&M	SEC	14-1
17	15	South Fla.	AAC	10-1
18	19	Gonzaga	WCC	13-2
19	17	Kentucky	SEC	11-4
20	20	Rutgers	Big Ten	5-3
21	21	Georgia Tech	ACC	8-3
22	22	West Virginia	Big 12	11-2
23	23	Missouri St.	MVC	7-2
24	25	SFA	Southland	12-2
25	32	Bucknell	Patriot	6-0

AP Poll

1	Louisville (22)	14-0		1
2	North Carolina State (5)	11-0		2
3	UConn (1)	10-0		3
4	South Carolina (2)	12-1		4
5	UCLA	9-2		6
6	Stanford	12-2		5
7	Maryland	11-1		7
8	Texas A&M	14-1		8
9	Baylor	10-2		9
10	Arizona	11-2		10
11	Oregon	11-3		13
12	Michigan	10-1		11
13	South Florida	10-1		14
14	Ohio State	8-1		17
15	Kentucky	11-4		12
16	Indiana	9-3		16
17	DePaul	9-3		18
18	Gonzaga	13-2		20
19	Arkansas	11-5		15
20	Tennessee	10-3		25
21	Mississippi State	8-4		19
22	Georgia	12-2		22
23	Northwestern	8-3		21
24	West Virginia	11-2		NR
25	South Dakota State	13-2		NR

I see your point here, but at the risk of stating the obvious, the algorithms are created by humans and thus depend upon the quality of information and analysis of humans too. The "human only "polls have to contend with the obvious bias of the voters. But the algorithm based evaluations have bias built in that even their creators often do not even recognize. I prefer the "eye test" opportunity that coaches polls provide, but I guess that is my own bias showing. The manipulation of statistics is inherently problematic and I would guess that the most critical thinking isn't always provided in these data driven polls.

 

[Kaizen]

This is Massey's poll (they use computer poll) very close to Net poll.

Massey Ratings - College Basketball Women's : NCAA D1 Ratings

Computer ratings and rankings for CBW (College Basketball Women's), with links to team predictions, scores, and schedules.

www.masseyratings.com

 

[visitingcock]

My preference would be something like the old BCS system that combined computers and polls.

not that it matters too much in this Covid year where geography is being taken out of the tourney. Not much difference between 1st and second seeds.

stiil I would hate to be a team with some good wins that falls to 65 behind a team with good losses

 

[nwhoopfan]

Oregon at #4 is a headscratcher. Wins over Washington St. x2 is their only thing remotely close to a quality win. 0-3 against Stanford, UCLA and Arizona. I'm just not seeing it. I guess their margin of victory against some cream puffs must be propping them up.

 

[MD#12Fan]

Really good post! Over my head!
I've always wondered at what point a team establishes its credentials by which wins and losses are evaluated. Like the SEC in football. Is Miss St really that much better than Wake Forest or say Washington St? When is that initial judgement made. After a couple of meaningless "warmup games", the teams start playing each other and chi-ching establishing credentials from playing other teams in the league. Recently the Pac 10 in football has been virtually slighted. With their style of play being a bit different, who's to say that the SEC is that much better than the Pac !0. Some person has to enter into the data base some initial information, then decisions can be made based on that. What is a strength of schedule if nobody knows how good anyone is.

 

[SVCBeercats]

Somebody else can pick out human problems if they want

Human Problems:

• Can't possibly watch every game of the top 30-40 teams let alone analyze the games and effectively compare all the teams' performances,
• They have built in biases favoring certain conferences over other conferences and teams they like and teams they hate,
• They lack the knowledge and skill to objectively rank the teams,
• They believe they know more than they do,
• Their opinions are just that - opinions,
• They have one infallible method for correctly ranking thr teams, the NCAA Tourney.

 

[diggerfoot]

The two biggest factors that will increase the accuracy or algorithms are: 1) increased number of games and 2) increased mixing between conferences. Both are compromised this season. Right now in this season the eyeball test may be more accurate. Towards the end of a normal season the algorithms would be more accurate.

 

[Glenn]

I still prefer the algorithms. That said, the Massey ratings are hurt by the lack of interconference play this year--it relies on having those matchups as diggerfoot noted. But humans also do not have those matchups and so are relying on biases more than normal. And the human "eye" test is full of biases when they watch games--and we know that the many games are not even watched. NC State is a mystery why they are ranked so low by the algorithms. I think Massey dings them because of all of their close wins. I think the NET does not like their "softer" schedule.

 

[Alydar]

I see your point here, but at the risk of stating the obvious, the algorithms are created by humans and thus depend upon the quality of information and analysis of humans too. The "human only "polls have to contend with the obvious bias of the voters. But the algorithm based evaluations have bias built in that even their creators often do not even recognize. I prefer the "eye test" opportunity that coaches polls provide, but I guess that is my own bias showing. The manipulation of statistics is inherently problematic and I would guess that the most critical thinking isn't always provided in these data driven polls.

Using "eye test" presumes that these coaches, or the persons delegated to do it, watch maybe 15 or 20 games a week to stay up on the teams. That's a full time job and I doubt the coaches spend more than an hour or two on it.

 

[Bama fan]

Using "eye test" presumes that these coaches, or the persons delegated to do it, watch maybe 15 or 20 games a week to stay up on the teams. That's a full time job and I doubt the coaches spend more than an hour or two on it.

I am sure you are right about that, but to me it is better than stat crunching programs.

 

[willtalk]

The problem is that very few interconference games were played this season. In the past they helped provided a gage as to the relative strength of each conference. Now the only gage is how well the conference teams do against each other. How can you gage a conferences strength if the teams only play each other. Then you have to also add the problems caused by quarantines , cancelled and postponed games. I doubt anything will be answered if and when the tournament is ever played.

 

[Jim]

Guys, the computer algorithms were programmed by humans, so this entire thread is about the opinions of one set of humans versus another set of humans. The two groups just use different tools at their disposal and have different biases.

 

[MooseJaw]

Using "eye test" presumes that these coaches, or the persons delegated to do it, watch maybe 15 or 20 games a week to stay up on the teams. That's a full time job and I doubt the coaches spend more than an hour or two on it.

Agree, but how many games do the computers watch. There are factors about games and teams and even players that must be seen by the human eye and evaluated by the fastest computer known, the human brain. Both are fallible that being said there is no way get a computer to explain it's choices.

 

[visitingcock]

The problem is that very few interconference games were played this season. In the past they helped provided a gage as to the relative strength of each conference. Now the only gage is how well the conference teams do against each other. How can you gage a conferences strength if the teams only play each other. Then you have to also add the problems caused by quarantines , cancelled and postponed games. I doubt anything will be answered if and when the tournament is ever played.

I agree with you and concede that this is the worst of all years to rely on computers because their data is limited by the scarcity of interconference games. This is especially true when ranking Oregon and the rest of the PAC that has no games against worthy nonconference foes.

AND I am not certain how far the reliance on NET rankings goes. Is it set in stone that the NET will make the determination on all non-conference champions or is it merely the guide to be used by a selection "committee?" If it rules in selection of the 64th team in, we should probably wait in using it until things return to normalcy.

 

[visitingcock]

Using "eye test" presumes that these coaches, or the persons delegated to do it, watch maybe 15 or 20 games a week to stay up on the teams. That's a full time job and I doubt the coaches spend more than an hour or two on it.

I doubt that coaches watch 10 games per week with the exception of scouting film re their opponents. They know alot about the teams they play during the next week or two but little about teams that are not on their schedule. Maybe they watch a few games between highly ranked teams that they are going to play. Dawn probably watched UConn v. Tennessee. I doubt very seriously that she took the time to watch Stanford vs. Colorado.

AP voters are different - though less knowledgeable. They have plenty of time to watch games - if they choose to.

 

[visitingcock]

FWIW

5th (NET) ranked Maryland lost last night to 11th ranked Ohio State.

The game has not affected either team's NET rank. They are exactly ranked where they were yesterday. A human would probably raise the winner and lowered the loser.

 

[Littlemin]

ALL polls are subjective because there is a human factor, directly or indirectly in all of them. IMO, they provide fodder for bar talk, nothing more and nothing less. And let me add, there are only a few few things concerning sports more fun but less meaningless than bar talk.

 

[Alydar]

Agree, but how many games do the computers watch. There are factors about games and teams and even players that must be seen by the human eye and evaluated by the fastest computer known, the human brain. Both are fallible that being said there is no way get a computer to explain it's choices.

If you want to evaluate one game then eyeballing is better than analytics, but if you are evaluating a season then I'll take analytics. Eyeballs often see what they expect to see rather than what is in front of them. If you think that S Car is the best team then you look for confirmation of that and if you think they aren't then you look for things to support that POV. Creme is a prime example. In his last Bracket update he said that S Car was the top seed and he "had the proof" of why SC was the No 1 seed. He had no proof, he had some evidence but in his own mind evidence supporting his POV going in was "proof". Actually I liked Creme's picks better when he did his own work. The last few years he has met with committee months before the tournament and a few times after that. And his bracket was virtually identical to the committee's, giving both a measure of legitimacy (if you think they arrived at their conclusions independently). But they were never independent. At least not in recent years.

 

[Alydar]

ALL polls are subjective because there is a human factor, directly or indirectly in all of them. IMO, they provide fodder for bar talk, nothing more and nothing less. And let me add, there are only a few few things concerning sports more fun but less meaningless than bar talk.

Doesn't "less meaningless" mean "more meaning"? And I agree that disagreements over sports is a great bar pastime.

 

[Alydar]

I doubt that coaches watch 10 games per week with the exception of scouting film re their opponents. They know alot about the teams they play during the next week or two but little about teams that are not on their schedule. Maybe they watch a few games between highly ranked teams that they are going to play. Dawn probably watched UConn v. Tennessee. I doubt very seriously that she took the time to watch Stanford vs. Colorado.

AP voters are different - though less knowledgeable. They have plenty of time to watch games - if they choose to.

AP voters also have a bias. If they cover the southeast it's natural that they will promote schools that their readers support, not exclusively but enough to sway the vote. Now if the writers are scattered around the country then the bias balances out to an extent but writers who depend on regional readership for a living are going to lean toward that readership.

 

[Littlemin]

Doesn't "less meaningless" mean "more meaning"? And I agree that disagreements over sports is a great bar pastime.

No it doesn’t, not in the context I used it.