By: Edward Egros

Predicting Putts on the PGA Tour

Pasted Graphic 1
If you go to your favorite search engine looking for studies on predicting the likelihood a PGA Tour player sinks a given putt, you will find enough research to suggest there is no other dimension for discovery. Distance and slope largely affect the odds a golf ball finds the bottom of the cup. Golf tournaments have even begun broadcasting both this research and these resulting probabilities as part of its televised coverage.

What you don't see in a lot of research or in broadcasts is how these odds can be adjusted based upon the golfer himself. Suffice to say, some players are better putters than others.
While statistics like Strokes Gained: Putting can be volatile, the results from specific golfers can be used to calculate a more accurate probability that any given putt will be successful.

With the help of students at Southern Methodist University, we put together a model for predicting the probability of a successful putt. Using data from 2014 to the 2018 seasons, we first used the following variables:

  • Distance (in inches)
  • Distance^2 (the squared term for distance)
  • Resulting Score (e.g. birdie, par, bogey, etc.)
  • Hole sequence (i.e. the player's progress throughout his round)
  • Position on the Leaderboard
  • Slope (uphill, downhill or level)
  • Interaction terms between slope and distance, as well as slope and distance^2
  • Par value
  • Specific golfers (Binary variables that equal 1 if the putt will be taken by a specific golfer, otherwise 0)

Pertaining to that last set of variables, for this study we took 20 of the more popular golfers on the PGA Tour with varying putting prowess:

  • Adam Scott
  • Brooks Koepka
  • Bryson DeChambeau
  • Bubba Watson
  • Dustin Johnson
  • Henrik Stenson
  • Jason Day
  • Jon Rahm
  • Jordan Spieth
  • Justin Rose
  • Justin Thomas
  • Patrick Reed
  • Phil Mickelson
  • Rickie Fowler
  • Sergio Garcia
  • Tiger Woods
  • Tony Finau
  • Webb Simpson
  • Xander Schauffele
  • Zach Johnson

Because of the pressure of winning a golf tournament and because the sample after the second-round cut only features players who have played well enough to post low scores, we constructed two model: one to represent the first three rounds of a Tour event and one for the final round. These models are
logistic regressions, with the coefficients for each variable representing an odds ratio. Before analyzing specific golfers, here are the results for the other variables:

Pasted Graphic

Even though some of these coefficients are close to one and perhaps difficult to analyze, all are statistically significant and logically intuitive. For instance, as distance increases, the putt becomes tougher. A downhill putt is tougher than an uphill putt. Bogey putts tend to be easier to make than pars and birdies (perhaps because they tend to be the second putt in a sequence where a golfer has learned how to read the green better).

As for the golfers themselves:



Pasted Graphic 1

These bar charts represent who tends to be better putters, and there are many observations available. In no particular order, Henrik Stenson is one of the more exceptional golfers with the flat stick, though there is a slight drop-off as the tournament concludes. Rickie Fowler performs consistently well, especially when he is in contention. Bryson DeChambeau has putted well in his victories, but it was a skill he had to work on during his professional career. Tiger Woods has a natural uptick when he is competing for a win. Lastly, when Jason Day or Jordan Spieth have momentum heading into the final round, they exhibit solid putting abilities.

There are many things that could improve this study. Our data did not include weather information, which affects greens (dry greens are faster and tougher, wet greens are slower and easier). There also isn't a way to analyze newer golfers who have not played in many PGA Tour events. Perhaps there is a way to convert other data and translate it to PGA Tour levels, but that is not at our disposal right now. Still, the results here are promising and should help refine the probability of putts when dealing with specific golfers.

Special thanks to
ShotLink for providing the data. Also, we put together a GitHub page with academic presentations, R code and other valuable information. For that, click here.