Question about Fargo that only Mike Page has access to the data to answer

JC

JC

Coos Cues
We have many thousands of players now who are established with millions of games played.

So here is the question. How does a fargo rating translate to a percentile rating of all players?

Take every established player and list them from bottom to top. Then divide the list into ten equal parts with the same number of players on each list. Just like you are going to have a tournament with ten divisions and the entire pool world with fargo ratings is entering.

Where do the ten divisions break? Is it linear or a bell curve? Or something else? Are the break points stable at this point where it will not change much as more games are played and continuing to use the entire data base of established players?

As I said in the title probably only Mike Page can answer this.

I think the answer will be interesting because I'm a numbers/statistics freak.

JC
 
Jeff G. Martin

Jeff G. Martin

AzB Silver Member
Silver Member
Here is the last distribution of ratings I saw Mike post. Maybe there is a more recent one out there.

575579455c2545bb37698295680c293b.png
 
hang-the-9

hang-the-9

AzB Silver Member
Silver Member
Funny, I am almost exactly average of the players with over 800 games but 100 points over the average of the larger group of 50,000 players.

Makes sense, those that have more Fargo games in play in more events that would be tracked, those that play in more events likely play in more events because they are better players and like to compete and test their skill.
 
Bob Jewett

Bob Jewett

AZB Osmium Member
Staff member
Gold Member
Silver Member
Perhaps it was noted before, but although the distributions look similar to standard bell curves (Gaussian distributions) they are significantly different. In particular, for the top group with a mean value of 555 and a standard deviation (often called sigma) of 110, and 3350 players in the group, you would expect:

1675 above 555
531 above 665
76 above 775
4 above 885

Just looking at the curves it seems that they fall off faster on the high side than the low side.
 
M

Masayoshi

Fusenshou no Masa
Silver Member
Bob Jewett said:
Perhaps it was noted before, but although the distributions look similar to standard bell curves (Gaussian distributions) they are significantly different. In particular, for the top group with a mean value of 555 and a standard deviation (often called sigma) of 110, and 3350 players in the group, you would expect:

1675 above 555
531 above 665
76 above 775
4 above 885

Just looking at the curves it seems that they fall off faster on the high side than the low side.
Click to expand...

I think it would be more surprising if Fargo rate followed a normal distribution considering there are many orders of magnitude more people who have never put time into getting good at pool or who have never even picked up a cue than there are people who put in the time to get good and even fewer who put in the time to become a top pro.
 
JC

JC

Coos Cues
Jeff G. Martin said:
Here is the last distribution of ratings I saw Mike post. Maybe there is a more recent one out there.

575579455c2545bb37698295680c293b.png
Click to expand...

I hadn't seen that thank you.

My guess is that if the system itself is valid that this won't change much as more data is added. And the system is valid so there it is.

Although this graph is informative it doesn't really answer my question of translating the fargo data to a 1-10 rating by evenly breaking up the players by volume and noting the cutoff score for each break.

A little background on what got me thinking about this. My buddy was upset when he was put out of the last Western BCA 8 ball event after dogging an out hill hill. We were reflecting on the event and fargo afterward and counted up the players entered and looked at the fargo scores and there were almost 1000 participants and only 80 had a higher fargo than my buddies which was 605. So I told him that there were 90% of all the mob there that he played better than. But then I wondered if this was true for the larger pool world or if that our tournament was somehow different. Like the skill level was on average lower than the pool world at large.

JC
 
Last edited:
hang-the-9

hang-the-9

AzB Silver Member
Silver Member
garczar said:
What are the numbers on left side of graph? Yed, yellow, blue? Have no clue what i'm lookin at here.
Click to expand...

Left is number of people, bottom is Fargo rating, color graphs is based on how many games they have in the system. So at Fargo 500 the long red/pink line is those with 50-199 games, the yellow is 200-799, the blue is 800+

The mid point for everyone seems to be in the B range, for those that have a Fargo rating. I'm sure there are many players not in Fargo that would drop this lower if all the APA, TAP, non league bangers were entered.
 
Last edited:
G

garczar

AzB Silver Member
Silver Member
hang-the-9 said:
Left is number of people, bottom is Fargo rating, color graphs is based on how many games they have in the system. So at Fargo 500 the long red/pink line is those with 50-199 games, the yellow is 200-799, the blue is 800+

The mid point for everyone seems to be in the B range, for those that have a Fargo rating. I'm sure there are many players not in Fargo that would drop this lower if all the APA, TAP, non league bangers were entered.
Click to expand...
Cool. Thanx.
 
BRussell

BRussell

AzB Silver Member
Silver Member
Here are approximate percentiles based on assuming the red graph of 50 or more games is normal.

top 10% = 629
20% = 573
30% = 532
40% = 498
50% = 466
60% = 434
70% = 400
80% = 359
90% = 303

Given the skew of the red distribution, I might subtract a few points from the higher scores (e.g., 625 rather than 629 for the top 10%, 570 rather than 573) and add a few to the bottom (e.g., 305 for bottom 10% rather than 303).

If you want to look at established players (200+ games) only, using the yellow distribution you get these percentiles:
top 10% = 651
20% = 599
30% = 562
40% = 530
50% = 501
60% = 472
70% = 440
80% = 403
90% = 351
 
Last edited:
JC

JC

Coos Cues
BRussell said:
Here are approximate percentiles based on assuming the red graph of 50 or more games is normal.

top 10% = 629
20% = 573
30% = 532
40% = 498
50% = 466
60% = 434
70% = 400
80% = 359
90% = 303

Given the skew of the red distribution, I might subtract a few points from the higher scores (e.g., 625 rather than 629 for the top 10%, 570 rather than 573) and add a few to the bottom (e.g., 305 for bottom 10% rather than 303).

If you want to look at established players (200+ games) only, using the yellow distribution you get these percentiles:
top 10% = 651
20% = 599
30% = 562
40% = 530
50% = 501
60% = 472
70% = 440
80% = 403
90% = 351
Click to expand...

I'm not sure how you ascertained that from the data on that graph. It must be an estimate. I'm hoping for exact numbers, which is of course possible.

JC
 
BRussell

BRussell

AzB Silver Member
Silver Member
JC said:
I'm not sure how you ascertained that from the data on that graph. It must be an estimate. I'm hoping for exact numbers, which is of course possible.

JC
Click to expand...

It's just assuming a normal distribution, which the red and yellow are pretty close to, and then converting to percentiles with rawScore = (z)sd + mean. It's what you do to calculate SAT or birth weight or other percentiles from standardized scores.

Without making Mike Page go in and do it with all the raw data, this will be within several points of the raw percentiles, and this method is probably better than just dividing up the raw scores because it's based on the theoretical distribution rather than the weird patterns that might be in the raw scores.
 
iusedtoberich

iusedtoberich

AzB Silver Member
Silver Member
Who are the players in the high 700's to 800's that have less than 200 games? I know Josh Brothers from my local area is one in the 750's or so.
 
Bob Jewett

Bob Jewett

AZB Osmium Member
Staff member
Gold Member
Silver Member
iusedtoberich said:
Who are the players in the high 700's to 800's that have less than 200 games? I know Josh Brothers from my local area is one in the 750's or so.
Click to expand...
If I had to guess, I'd say they are non-American players who have had a few good matches in international tournaments.
 
Bob Jewett

Bob Jewett

AZB Osmium Member
Staff member
Gold Member
Silver Member
JC said:
I'm not sure how you ascertained that from the data on that graph. It must be an estimate. I'm hoping for exact numbers, which is of course possible.
Click to expand...
The measurement is uncertain since it is based on playing statistics, so values derived from the data will not be exact/accurate. The numbers will vary as new games are added.

If you want an estimate from the data shown, just put a number on each bar and add them up. For example, the left-most red bar (above "140") is 223 players. So you know that 223 players are below 155 in rating out of 50,000 players.
 
G

garczar

AzB Silver Member
Silver Member
You guys really need to get out more. ;) Seriously, what would a higher level of exactitude(damn i love that word) in this pile of data really tell you? For me, i play and my results go in a computer and out pops my FR. Whatever floats-ur-boat i guess. As ColinK. might say, " JustKeepCrunchin".
 
JC

JC

Coos Cues
Bob Jewett said:
The measurement is uncertain since it is based on playing statistics, so values derived from the data will not be exact/accurate. The numbers will vary as new games are added.

If you want an estimate from the data shown, just put a number on each bar and add them up. For example, the left-most red bar (above "140") is 223 players. So you know that 223 players are below 155 in rating out of 50,000 players.
Click to expand...

I understand that the data is dynamic but what I'm wondering is if the percentiles are static as one player goes up and another goes down is there a predictable score that would put a player at 60th percentile (example) that will remain stable as the total data increases. I think there may be.

How do you know that the red bar above 140 is exactly 223 players BTW? I don't see that indicated anywhere. I must be missing it.

Thanks,

JC
 
JC

JC

Coos Cues
garczar said:
You guys really need to get out more. ;) Seriously, what would a higher level of exactitude(damn i love that word) in this pile of data really tell you? For me, i play and my results go in a computer and out pops my FR. Whatever floats-ur-boat i guess. As ColinK. might say, " JustKeepCrunchin".
Click to expand...

There are two kinds of people. Those who like numbers and those who don't.

Getting out more has nothing to do with it.:)

JC
 
A

AtLarge

AzB Gold Member
Gold Member
Silver Member
Bob Jewett said:
I pulled it out of ..... the air.:wink:
Click to expand...

No, no. It was highly scientific. I enlarged the chart as much as I could on my screen. The distance from 0 to 500 on the left axis was then 83 mm on my screen. The height of the first pink bar was 37 mm. (37/83) x 500 = 223.

[Maybe that's "the air."]
 
You must log in or register to reply here.
Top