Stonebridge re-rated / re-measured

[mjf]

I played at Stonebridge today and was given a 'temporary' scorecard. They've re-measured and re-rated the course and the new cards aren't ready yet. The new card does have the new yardages and course rating / slope.

Get this... the course now measures a little shorter than it did originally and they've made several minor changes to make the course easier. So what do you suppose happened to the course rating?? It went UP of course. At least, it went up from the back tees.

New rating / slopes (MEN's):

Black: 71.3 / 133 (was 71.1 / 132)
Blue: 69.3 / 128 (was 68.8 / 129)
Gold: 68.7 / 126
White: 66.8 / 124 (was 66.2 / 123)
Red: 64.0 / 112 (was 63.8 / 109)


Rating 71.3 for a course that measures 6442 from the back. I wonder what the course would rate if they stretched it out another 500 yards?

[sensfan63]

I played at Stonebridge today and was given a 'temporary' scorecard. They've re-measured and re-rated the course and the new cards aren't ready yet. The new card does have the new yardages and course rating / slope.

Get this... the course now measures a little shorter than it did originally and they've made several minor changes to make the course easier. So what do you suppose happened to the course rating?? It went UP of course. At least, it went up from the back tees.

New rating / slopes (MEN's):

Black: 71.3 / 133 (was 71.1 / 132)
Blue: 69.3 / 128 (was 68.8 / 129)
Gold: 68.7 / 126
White: 66.8 / 124 (was 66.2 / 123)
Red: 64.0 / 112 (was 63.8 / 109)


Rating 71.3 for a course that measures 6442 from the back. I wonder what the course would rate if they stretched it out another 500 yards?

This is a microcosm of Ottawa golf, IMO. It's a pretty new course and yet somehow the back tees measure under 6500 yards. Just doesn't compute with today's game.

I find it...odd...that a new course doesn't open up that goes about 7300 or 7400 yards. They're all over Toronto, and it's not like these new courses in Ottawa lack the space to make them longer.

Reason # 17 why Ottawa will never host a Canadian Open...

[jvincent]

That is, as the saying goes, ridonculous!

There is no way that it plays harder from the back tees now. To me this just highlights the inconsistencies in the current rating system/methodology.

[bobblehead]

I usually play from the white tees and for sure i thought the slope would go down after they made those changes to the course. The course plays easier now that a few years ago.

[mjf]

I wonder if the methodology used to rate courses has changed in the last few years?

I suppose the change isn't that big... almost within the margin of error (if there was one).

[jvincent]

I wonder if the methodology used to rate courses has changed in the last few years?

As far as I know it hasn't. But IMNSHO, it is flawed because it has people in the loop.

I've described my "better" system before, but I'll do it again here.

Course ratings should all be done using computers and topographical maps. The process would be as follows.

For every club generate a probability distribution of where the shot will end up for your "reference" golfer. This is the hard part since the data gathering will be difficult. If it were me, I would just use the PGA tour shotlink data to create the model.

To rate a course, you overlay the statistical shot distribution onto each hole and generate a statistical score for the course. The statistical score takes into account the fact that some shots will end up in the water/trees/bunker as determined by the topography of the course and the shot distribution.

Since you can run this on a computer you can generate as many sample shots as you need.

As an example, SB #1 will have a certain percentage of shots that go into the water. The course rating program will "play" the hole for every sample shot based on where it ends. So some will be played from the fairway directly to the green, some will be played from the drop point of the lateral hazard, etc. for each sample, you generate a score and after a suitable number of samples, you get a score distribution.

Bingo! You now have a mathematical model for the course rating that factors topography, club selection, and shot distribution.

[Kilroy]

Or you could use hole by hole scores from actual golfers.

[sensfan63]

Or you could use hole by hole scores from actual golfers.

Wouldn't that be best for handicapping the holes of a specific golf course, but not necessarily rating it?

I mean, using the scores from the Citizen tournament, you'd think we have a collection of fantastically difficult courses in the Ottawa area. When, in reality, the concept of "handicaps" is simply indicating potential instead of true ability. So, that makes the courses not as hard as they appear via tournament scores.

[Ruskie]

One little problem with that stats approach: it doesn't take into account the mental side of the game. Little things like visual deception, tee box aiming, people aiming away from hazards, etc.

To use your sample #1 at SB, most people would aim away from the lake that they can see, only to increase the probability of going into the fesque on the left (effectively OB).

Another one would be SB #4. It looks like there's not a lot of room over the center bunkers, and before they put the red flag at the edge of the lake, it was very hard to convince yourself that you can go there, even after playing the hole countless times. So people aim left, bringing fesque and bunkers into play.

It would be very difficult, if not impossible, to account for these factors in a computer program.

[Kilroy]

Seems to me that the best way to "rate" anything would be to use actual real world statistics gathered over time. Since the handicapping system is there to equalize the lesser skilled players with the more skilled ones, then such statistics could provide clear indicators as to what courses and holes are tougher than others for players of all levels.

I am sure Dave Peltz or some other stats junkie would be able to crunch the numbers in a way that actually worked. I am not convinced the current system is the best we can do.

[jvincent]

Remember, the whole concept of a ratings system is about a "normalized" or reference golfer so all scores are compared to that score. In my system there would be no slope. The reference golfer, let's call him Mr. Shotlink, is the standard.

Ruskie's point about aim points is a good one. This is one area where a person could in theory have to make a call. The way to handle that would have the person running the program designate the aim point for each shot. Or you could have the program decide that for any tee shot the aim point is such that the area covered by the statistical distribution maximizes the "safe" landing area.

My whole point is that if you take out the "judgement" factor you get a better statistical model, which is what the rating is supposed to be.

[Ruskie]

Oh, but aim point is not the only "human" variable. You have deceptive changes in elevation, false fronts, deceptive green shapes and mounding, deceptive visual cues for the slope of the green, etc. And then there's the subconscious swing changes we make in response to visual cues. A hazard on the left will often result in a more pronounced push or slice to the right. Even things like walking up or down to the tee-box would make a difference.

The only way I can think of using a purely stats approach would be to create a collection of "template" holes, record how real golfers play on them, and apply it to other courses. That would be a monumental task though, given the variety of designs.
I wouldn't be surprised if they already use some form of that approach in the current system.

Remember, the whole concept of a ratings system is about a "normalized" or reference golfer so all scores are compared to that score. In my system there would be no slope. The reference golfer, let's call him Mr. Shotlink, is the standard.

Ruskie's point about aim points is a good one. This is one area where a person could in theory have to make a call. The way to handle that would have the person running the program designate the aim point for each shot. Or you could have the program decide that for any tee shot the aim point is such that the area covered by the statistical distribution maximizes the "safe" landing area.

My whole point is that if you take out the "judgement" factor you get a better statistical model, which is what the rating is supposed to be.

[jvincent]

If you believe that those visual cues will affect the resulting shot, then my solution to use the maximum statistically safe point as the aim point will take that into effect.

Remember, the reference number is based on a highly skilled golfer, not some average hacker like you or me.

If the aim point on a green with a false front is back middle, then statistically speaking some shots will miss short, some long, some left, etc.

Template holes won't work because the templates would need to be infinite.

[BC MIST]

Or you could use hole by hole scores from actual golfers.

Wouldn't that be best for handicapping the holes of a specific golf course, but not necessarily rating it?

Exactly.

Collecting scores IS the best way to handicap holes but one must use these scores in the proper manner. There are still many gofers who believe that the #1 handicap hole is the HARDEST hole and the #18 handicap hole is the easiest and so on. This is UNTRUE!!! The #1 handicap should be the hole on which the higher handicap golfer NEEDS a stroke.

To determine the ranking, holes are averaged for the low golfer and for the higher handicapped golfer and these numbers are then subtracted. The one with the greatest difference will be your # 1 handicap hole. For example: After analyzing over 230 scores from Greensmere's Premiere course, the #1 handicap hole should be #6 a short, 500 yard, par 5. The scoring average for the high handicap golfers, averaging 26, was 6.92 while for the low, averaging 3, was 4.86, creating a difference of over 2. It's an easy hole for the better players but a harder one for the weaker ones because it is longer and they have a greater chance of messing up the hole. It is a hole where they need a stroke. Courses that have handicapped the holes correctly will usually have the longer par 5's as their lowest ranked holes

A significant quote from the Handicap Manual: Difficulty in making par on a hole in NOT an effective indicator of a need for a stroke.

[Ruskie]

If you believe that those visual cues will affect the resulting shot, then my solution to use the maximum statistically safe point as the aim point will take that into effect.

Remember, the reference number is based on a highly skilled golfer, not some average hacker like you or me.

If the aim point on a green with a false front is back middle, then statistically speaking some shots will miss short, some long, some left, etc.

Template holes won't work because the templates would need to be infinite.

Do you really believe that even a scratch golfer would have the exact same shot dispersion when shooting to a flat unprotected green as opposed to a well bunkered blind green with a false front and lots of mounding around it? Even if the flag is in the middle and the golfer uses it as the aim point? I would think not. Then again, you're much closer to scratch than me, so maybe I'm missing something

Here's an example of a hole that plays a lot harder than its design would suggest: #17 at TPC Sawgrass. Remove the water from that hole, and 90% of the pros would probably be on the green, every time. Why do they miss it so often then? And how can you account for that with a statistical model?

[jvincent]

I think you are falling into the trap that the current system has in it, i.e. too much reliance on psyche factor, etc.

#17 at Sawgrass is hard because the wind swirls and the penalty for a miss is extreme. It doesn't change the fact that the pros are hitting a PW and the shot distribution for a PW is, statistically speaking, a constant.

If I take "blind" Shotlink data for EVERY club that the pros hit, then all of the psyche out factors of #17 and #12 at Augusta, uphill, downhill, well bunkered, etc. will be accounted for in the shot distribution.

When you apply the model to a given hole then you get a range of scores for that hole. Based on the statistical model you will get some aces, birdies, pars, bogies, doubles, etc. But, if the model says that within one standard deviation the scoring average for the hole is 3.12578 then that is what you would use for the course rating.

The other thing to remember is that for a hole of 150 yards, Mr Shotlink can probably choose 2 or 3 different clubs to play that hole so you need to run the model using all of those data points.

This is one obvious thing that needs to be accounted for in generating the shot distribution. You need to have data that maps to an intended result. By that I mean, if your "target distance" is 150 yards, you need to generate the distribution for someone trying to hit a 150 yard shot, potentially with different clubs. I believe that they have all of this data in Shotlink, it's just a matter of mining it properly.

As I said earlier, generating the reference model would be the hard part.

From a purely geekazoid perspective, this would be a very cool project to run. I am waiting for the call from the USGA to start the project.

[Ruskie]

By "blind" ShotLink data, do you mean "take all 150yard shots, no matter the hole, and average them" ?

[jvincent]

Correct, but you don't average them, you generate a statistical distribution for them. If you imagine the distribution in 3D it will look like a rounded pyramid with the highest probability being at the center.

[Ruskie]

Ok, that's what I thought you meant.

So, let's say you have two par 3's, both 150 yards. One with a flat, open, "easy" green. The other with a bunker left of the green, a ravine in front making it seem to play downhill and a lake in the back. Would you agree that shots to the second green would tend to come up short and right, even from a scratch golfer?

If you now throw all the shots from both holes into the same pile and crunch the numbers of them you'll end up with a distribution centered somewhere in between the two groups, a little short and right.

If you now apply this distribution to either hole, you'd get incorrect predictions. For the "easy" hole, your prediction would be worse than reality. For the "hard" hole, it would be better than reality.

Am I missing something?

[jvincent]

The thing that the model depends on is that Mr Shotlink is trying to hit an exactly 150 yard shot. The model has to take into account all shots of exactly 150 yards so it averages out any effects due to any individual hole.

Now, when you apply the model to the two different holes, some of the "misses" on the harder hole will be in the water, ravine, etc, which will make the average score higher.

[Ruskie]

And how would you determine that Mr.Shotlink wanted to hit exactly 150 yards? The only data available would probably be distance to the flag and where the shot ended up.

[jvincent]

The shotlink data presumably has the distance to the pin as part of the database so you'd have to use that as an assumption.

Generating the Mr. Shotlink distribution data is the hardest part of this exercise. Once you have that you can generate ratings for the courses and see how it correlates to the real world.

I contend that this will be more consistent than using teams to rate courses.

[BC MIST]

As far as I know it hasn't. But IMNSHO, it is flawed because it has people in the loop. Course ratings should all be done using computers and topographical maps. The process would be as follows. For every club generate a probability distribution of where the shot will end up for your "reference" golfer. This is the hard part since the data gathering will be difficult. If it were me, I would just use the PGA tour shotlink data to create the model. To rate a course, you overlay the statistical shot distribution onto each hole and generate a statistical score for the course. The statistical score takes into account the fact that some shots will end up in the water/trees/bunker as determined by the topography of the course and the shot distribution.
As an example, SB #1 will have a certain percentage of shots that go into the water. The course rating program will "play" the hole for every sample shot based on where it ends. So some will be played from the fairway directly to the green, some will be played from the drop point of the lateral hazard, etc. for each sample, you generate a score and after a suitable number of samples, you get a score distribution.

Bingo! You now have a mathematical model for the course rating that factors topography, club selection, and shot distribution.

Your system is interesting and would certainly add to the accuracy of those rating factors involved. Is there a need to include separately, other factors such as prevailing winds, speed of greens, lushness of fairways and so on, that ARE included in the current system, or are all part your shot distribution factor?

Ratings for some courses have recently changed, ours at Greensmere for example, because the team of 8 raters are doing their work more "by the book," than was done when the courses were originally rated, years ago. I always felt that our Premiere course had a rating that was too high and the new, lower rating confirmed that.

[Ruskie]

The shotlink data presumably has the distance to the pin as part of the database so you'd have to use that as an assumption.

Generating the Mr. Shotlink distribution data is the hardest part of this exercise. Once you have that you can generate ratings for the courses and see how it correlates to the real world.

I contend that this will be more consistent than using teams to rate courses.

I think I understand what you're saying now, but I still think there's no way to build an even remotely correct statistical model based solely on shot data. You would have to take into account course features and environmental conditions, and those kinds of things are very difficult to quantify. Which is where the rating teams come in.

[jvincent]

Addressing two replies at once here.

Day to day variances in environmental conditions are not currently factored into the rating system so there is no change there.

Case in point, played the Marshes yesterday in 30K+ winds on firmer and faster than normal greens. Compare those conditions to a calm day after some rain where the greens are slow and soft. Or during a rainstorm. The effective rating for the course on each day would be quite different.

It would be possible to factor things like rough length or fairway firmness into the model by adjusting the "effect" of the course topography, but then again you run into the situation where sometimes the rough is short, sometimes it isn't.

Prevailing wind conditions would affect the shot distribution model. That is very easy to do. Of course when the wind changes the course rating is completely invalid.

[Ruskie]

You can account for some of these things for your target course, but you can't possibly account for them in the input data from Shotlink. You can't tell if a particular shot in Shotlink was executed on a windy day, when the rough was long and greens hard, or on a calm day with short rough and soft greens. You also can't tell things like which way the fairway or the green funnel the shots and the effects of these variables on the thinking process of the golfer, the type of shot he was trying to hit. You won't even know where the recorded shots landed and weather they, say, ran up onto the green or hop-and-stopped on it.

Not knowing all these and more variables, you can't just throw all these shots into the same pot and draw a model out of it. It'd be like calculating the odds of winning at all the casino games combined, and then using those odds to try to predict winning percentages for a new game.

[jvincent]

Strictly speaking the shotlink data accounts for all of those conditions in the probability distribution. That's the beauty of this system. Making up numbers, 75% of all 150 yard shots will land within a 10 yard circle of the target. The remaining 25 percent will be distributed short/long/left/right because of wind, hitting fat, thin, etc. That is the whole point of using a probability distribution. Once the sample size is sufficiently large you lessen the impact any one effect.

They do track data from the fairway/tee/rough independently so while you can't gauge severity of those things you do have a first order differentiator.

The current rating system is even more flawed. The raters apply a subjective criteria for how difficult a shot is. That is going to be much more error prone because they will bring their personal bias into it.

[BC MIST]

The current rating system is even more flawed. The raters apply a subjective criteria for how difficult a shot is. That is going to be much more error prone because they will bring their personal bias into it.

What is the subjective part of the current "flawed" rating system? We know that 90% of a rating is based on measured distance with the other 10% based on all the others factors combined, each starting with a standard, then increasing or decreasing that standard based on further measurements. How does "personal bias" leak into cold hard, measurements?

There is no doubt that the rating system could be improved, but if all courses were rated based on all the criteria taught at the certification courses, there should, at least, be consistency from course to course. The current ratings teams do a lot of measuring, whereas, those from the past, did a lot less.

[jvincent]

Yeah, my personal bias comment is probably overstated since length is the predominant factor.

I guess my point is really that a formula which places such and emphasis on length and then gets modified based on subjective criteria is the problem.

My proposal takes length into account, but also, IMHO, more accurately reflects the impact of not hitting your target.