Round Ratings

[chrishysell]

one of the things I do with our league is keep a scoring average for all players. It's almost meaningless in reality because the course we play on changes every 3 weeks. Some years I average 48 and this year I have the league high at 46.8. It's an odd stat to me because we have used more difficult layouts this year and also have had worse weather conditions. But my PDGA rating is as high as it has ever been. Perhaps i'm evolving.

 

[bhadella]

Very cool stuff. The round ratings also confirm that I play 30 points better on my practice rounds than my sanctioned rounds. Oh and that Winthrop Gold with the ropes up gave me a pretty epic beatdown with a 812 rated round.

 

[jeverett]

Very cool stuff. The round ratings also confirm that I play 30 points better on my practice rounds than my sanctioned rounds. Oh and that Winthrop Gold with the ropes up gave me a pretty epic beatdown with a 812 rated round.

Winthrop Gold is a good example of a course layout that the SSE formulas (and thus by extension the round ratings) are not coming out accurate for at all. The SSE for the Gold layout is estimated at a ~60.1 (Par of 68). So I definitely wouldn't trust the round rating for that at all. I strongly suspect that your 'epic beatdown' would rate better than the DGCR rating suggests.

 

[mashnut]

Winthrop is also an outlier in terms of how much it separates the top players from those of us with lesser skill levels (similar to Fountain Hills used for the Memorial). Big distance and lots of OB tends to mean the best players have a chance to shoot really well and the worst players are hit really hard by the OB penalties.

 

[New013]

yea winthrop is a case where you probably don't list it as lightly wooded

 

[mashnut]

Except that it's more important to have accurate course information than to get accurate round rating estimates. I would be annoyed if I showed up at Winthrop expecting a heavily wooded course, it looks and feels like a lightly wooded course there's just a lot of OB.

 

[jeverett]

Winthrop is also an outlier in terms of how much it separates the top players from those of us with lesser skill levels (similar to Fountain Hills used for the Memorial). Big distance and lots of OB tends to mean the best players have a chance to shoot really well and the worst players are hit really hard by the OB penalties.

Yeah, I plan on using the round score/player rating data for Winthrop this year to examine the possibility of designing a non-linear formula to model the relationship between player (initial) rating and round score. I saw a little bit of an unusual shape to the scoring spread when I looked at the Memorial data, but still not enough data to really suggest that the PDGA method of determining SSA using linear regression is less accurate (than anything else). For reference, here are scatterplots of the first two rounds (MPO plus FPO) of the Memorial, with the best-fit line added:

Note the grouping for players rated above 1020.. to me, from that data the assumption of a linear relationship between player initial rating and round score is a little suspect for that particular layout.

 

[Menacewarf]

Ratings are calculated immediately after every score is entered. You didn't choose a tee in any of your rounds Menacewarf so those rounds are thrown out.

Gotcha, thanks.

 

[mashnut]

Yeah, I plan on using the round score/player rating data for Winthrop this year to examine the possibility of designing a non-linear formula to model the relationship between player (initial) rating and round score. I saw a little bit of an unusual shape to the scoring spread when I looked at the Memorial data, but still not enough data to really suggest that the PDGA method of determining SSA using linear regression is less accurate (than anything else). For reference, here are scatterplots of the first two rounds (MPO plus FPO) of the Memorial, with the best-fit line added:


Note the grouping for players rated above 1020.. to me, from that data the assumption of a linear relationship between player initial rating and round score is a little suspect for that particular layout.

Cool, thanks for posting those. Just by eyeballing those, I see that if you just look at players over 1000 you'd get a much steeper line, and overall a curve that mostly followed those best fit lines but was steeper at the high end would fit that data a lot better. That also jives with so many of the highest rated rounds coming out of the Memorial.

 

[vonDrehle]

Except that it's more important to have accurate course information than to get accurate round rating estimates. I would be annoyed if I showed up at Winthrop expecting a heavily wooded course, it looks and feels like a lightly wooded course there's just a lot of OB.

The thing is 95% of the time there isn't a lot of OB because the ropes aren't down so it is just a hard course to try and get a good estimate with.

 

[bhadella]

Winthrop Gold is a good example of a course layout that the SSE formulas (and thus by extension the round ratings) are not coming out accurate for at all. The SSE for the Gold layout is estimated at a ~60.1 (Par of 68). So I definitely wouldn't trust the round rating for that at all. I strongly suspect that your 'epic beatdown' would rate better than the DGCR rating suggests.

Just looked at least years Performance Flight results and my 84 would have been a 872, which makes me feel marginally better.

 

[jeverett]

Cool, thanks for posting those. Just by eyeballing those, I see that if you just look at players over 1000 you'd get a much steeper line, and overall a curve that mostly followed those best fit lines but was steeper at the high end would fit that data a lot better. That also jives with so many of the highest rated rounds coming out of the Memorial.

I've been hunting for a potential cause for that slight curvature, or at least something to help determine how best to model it. For example, here's a neat chart of Steve West's data from the MN Indoor Putting Championship:

Player rating is on the x axis, and percentage of putts made at 30ft. is on the y. I was hoping it might help, but the curve is actually the wrong way. i.e. instead of players getting better at putting as rating increases in a linear fashion, the curve suggests that players actually have slight diminishing returns on how much improvement they see in their putting (at 30ft.) as rating increases. So if it's not simply a factor of putting accuracy, it must be something else. I'll keep thinking on it.

 

[jkdisc]

i never knew what my rating would be, now i do. pretty cool, nice job men.

 

[jtencer]

Never joined the PDGA, but from playing with rated players I always figured I was around 900. My DGCR rating of 926 is consistent with my expectations. It does look like there are some discrepancies with different courses (some courses I seem to shoot 800s but others I consistently shoot over 950), but they appear to average out. It might just be a case of my game being a better fit for some courses rather than others.

Any chance we could see some site statistics for the DGCR player ratings? I think it would be interesting to look at a PDF of the player ratings for members on this site.

 

[mashnut]

I've been hunting for a potential cause for that slight curvature, or at least something to help determine how best to model it. For example, here's a neat chart of Steve West's data from the MN Indoor Putting Championship:



Player rating is on the x axis, and percentage of putts made at 30ft. is on the y. I was hoping it might help, but the curve is actually the wrong way. i.e. instead of players getting better at putting as rating increases in a linear fashion, the curve suggests that players actually have slight diminishing returns on how much improvement they see in their putting (at 30ft.) as rating increases. So if it's not simply a factor of putting accuracy, it must be something else. I'll keep thinking on it.

I think that probably contributes to why many courses do have a more linear distribution even at the top end. I would be willing to bet that OB strokes strongly contribute to what we were talking about earlier. The punishment at the USDGC and the Memorial doesn't affect players linearly, the best players are able to take a lot of it out of play while the lower level players end up penalized a lot.

 

[Dave242]

Never joined the PDGA, but from playing with rated players I always figured I was around 900. My DGCR rating of 926 is consistent with my expectations. It does look like there are some discrepancies with different courses (some courses I seem to shoot 800s but others I consistently shoot over 950), but they appear to average out. It might just be a case of my game being a better fit for some courses rather than others.

This is something I am very curious about. We all realize that SSE is a very imperfect estimate. But if we correctly chose things when setting up the formulas, some courses should be off in one direction and an equal amount should be off in the opposite direction. So, if a player has enough rounds spread evenly over enough courses will the DGCR Player Rating match the PDGA's? Of course, if people fudge their numbers here or do not include their bad rounds that skews things too (I anticipate this to happen - heck, I probably wouldn't include my bad rounds if I tracked my scores here).

Any chance we could see some site statistics for the DGCR player ratings? I think it would be interesting to look at a PDF of the player ratings for members on this site.

I am very interested in doing this sort of thing. If/when DGCR ratings are included on this page, I will probably do a whole bunch of number crunching.

 

[notroman]

I don't enter too many rounds here just cause it's usually time consuming and I play too much, but I did notice I have a very high number of 1000+ rated rounds among the few that I do have entered. I know some might be close to that, but I doubt I crack 1000 that often yet. One of them is showing up as 1070 - I think it was 1050-ish tops, but I doubt I was pushing 1070.

I also notice some courses are more likely to spit out a higher rating while others seem a bit low. Community Park in Channahon usually has SSA around 46-48 in tournaments but my round of 50 locked in a rating of 923 here. On the other hand it seems I have no problems shooting 1000+ at Sinnissippi and at Treehouz with several mediocre rounds registering over 1000 here.

 

[3j0hn]

I think that probably contributes to why many courses do have a more linear distribution even at the top end. I would be willing to bet that OB strokes strongly contribute to what we were talking about earlier. The punishment at the USDGC and the Memorial doesn't affect players linearly, the best players are able to take a lot of it out of play while the lower level players end up penalized a lot.

I wonder if it looks more linear if you pull the penalty strokes out of the scores. I am not sure if it would make a big difference, but (if it weren't a book keeping nightmare for TDs) I do think that SSA calculation should be done using scores with all penalties removed, but then round ratings given using the score with the penalties added back.

 

[jtencer]

Another thing that might be interesting would be comparing the distribution of ratings of players on DGCR with the calculated SSE for individual courses along with the actual rounds recorded on that particular course.

The distribution of player ratings along with the SSE would give you one distribution of potential scores. The actual rounds played would give you another.

I'm just brainstorming, but I might expect that the KL divergence between those two distributions would represent a measure of how accurate the SSE is for that given course.


Edit: Basically, I think that the scorebook provides us with a wealth of data about the relative difficulties of these different courses and that it would be interesting to make more use of it on the aggregate.

 

[Dave242]

That would be some interesting stuff.....but it would be an absolute nightmare to collect the data (unless timg did did a data dump for you).

the KL divergence between those two distributions would represent a measure of how accurate the SSE is for that given course.

Why not just compare SSE to SSA to get to that understanding? Well....I guess not all courses have an SSA established. Do I answer my own question?