Our Blog


Course History at Specific Courses

In my first two articles on course history, I outlined some principles for quantifying course history and proved that course history is predictive of future performance. In this article, we’ll push the envelope of how specific we can get with course history and speculate a little more about what course history might really be.

All courses have their individual tendencies, so it stands to reason that course history might matter more or less at different courses. Maybe on a ball-striker course, understanding how to navigate the fairways is a measure of strategy and eye fit, which might make course history more important there than a bomber’s course, where raw distance is more important than strategy. Do these courses exist? — where course history is more or less important than others? And if so, can we reliably predict which courses where history will matter more or less?

To answer this question, I looked at my historical data set after I applied generic, best-fit adjustments for course history. I plotted each golfer’s quantified course history going into the tournament versus how they did relative to their DraftKings scoring expectations. Here’s a random sample of six such tournaments in 2016:

courses1

courses2

courses3

Some courses, you can see a decent downward slope correlation, indicating course history might be overrated. Other courses look really noisy, indicating there’s no pattern to be found. It’s basically all over the map. That is a finding in itself for me, because it bounds the degree to which we can say course history matters more or less than our generic rules at a given course. Given the noise level shown in the plots above, it’s a lot less than I might have thought.

These results are consistent with the research I did into specialists. The maxim from that series holds: The more specific you try to get with your adjustments, the weaker your conclusions have to be due to all the noise. The concept of course history is already a pretty specific adjustment, and you have to be very careful about how you calculate it in order for it to stick.

On the other hand, the fact that generic rules apply pretty well could just as easily be an exploitable edge. If you polled most people about where course history matters the most, Augusta National would probably be up there given how often you hear about its impossible greens and the value of experience. You might conflate that with course history being extra important. However, at least from this year alone, generic rules on course history applied just as well. All of those things could be true, and we could still be picking up people like Jordan Spieth who excel there. If those things matter more at a given course, that should be reflected in how much a golfer overperforms relative to the field at a course like Augusta.

I think this is about the limit to which I am willing to incorporate course history. Generic rules do just fine 80% of the time, and if there’s some super-small sliver of edge in understanding where course history might matter more, it may be beyond the realm of quantitative analysis. I’ll conclude this series next week on future course history topics to explore.

In my first two articles on course history, I outlined some principles for quantifying course history and proved that course history is predictive of future performance. In this article, we’ll push the envelope of how specific we can get with course history and speculate a little more about what course history might really be.

All courses have their individual tendencies, so it stands to reason that course history might matter more or less at different courses. Maybe on a ball-striker course, understanding how to navigate the fairways is a measure of strategy and eye fit, which might make course history more important there than a bomber’s course, where raw distance is more important than strategy. Do these courses exist? — where course history is more or less important than others? And if so, can we reliably predict which courses where history will matter more or less?

To answer this question, I looked at my historical data set after I applied generic, best-fit adjustments for course history. I plotted each golfer’s quantified course history going into the tournament versus how they did relative to their DraftKings scoring expectations. Here’s a random sample of six such tournaments in 2016:

courses1

courses2

courses3

Some courses, you can see a decent downward slope correlation, indicating course history might be overrated. Other courses look really noisy, indicating there’s no pattern to be found. It’s basically all over the map. That is a finding in itself for me, because it bounds the degree to which we can say course history matters more or less than our generic rules at a given course. Given the noise level shown in the plots above, it’s a lot less than I might have thought.

These results are consistent with the research I did into specialists. The maxim from that series holds: The more specific you try to get with your adjustments, the weaker your conclusions have to be due to all the noise. The concept of course history is already a pretty specific adjustment, and you have to be very careful about how you calculate it in order for it to stick.

On the other hand, the fact that generic rules apply pretty well could just as easily be an exploitable edge. If you polled most people about where course history matters the most, Augusta National would probably be up there given how often you hear about its impossible greens and the value of experience. You might conflate that with course history being extra important. However, at least from this year alone, generic rules on course history applied just as well. All of those things could be true, and we could still be picking up people like Jordan Spieth who excel there. If those things matter more at a given course, that should be reflected in how much a golfer overperforms relative to the field at a course like Augusta.

I think this is about the limit to which I am willing to incorporate course history. Generic rules do just fine 80% of the time, and if there’s some super-small sliver of edge in understanding where course history might matter more, it may be beyond the realm of quantitative analysis. I’ll conclude this series next week on future course history topics to explore.