Do Early and Late Tee Times Affect Pro Golfers? A Comprehensive Analysis
I played an early bird round two weeks ago and shot a shocking five strokes higher than what I’m used to at my home course. While I managed to three-putt three times and hit my fair share of poor shots, as every golfer knows, it’s never that. Sometimes it’s the wind, other times it’s the course conditions. Ingeniously, I managed to create a new excuse: the early wake-up call. I felt as if I hadn’t actually woken up by the fourth hole, and I even managed to take a quick five-minute nap at the turn. My poor play had to be a result of an unusually dawning round of golf.
So, is there any truth to my hypothesis?
Using datagolf’s comprehensive strokes gained archive, I collected data from every PGA Tour tournament from the 2017 SBS Tournament of Champions to the 2023 John Deere Classics.
Note, due to lack of Strokes gained categories for certain events, data from Corales Puntacana, Puerto Rico, Hero World Challenge, WWT Championship at Mayakoba, Butterfield Bermuda Championship, (2017, 2018, 2019) ZOZO Championship, (2017, 2018, 2019) CJ Cup, The Open, (2017, 2018, 2019, 2020) Masters, (2017, 2018) US Open, WGC HSBC Championship, and 2017 CIMB Classic were omitted. 2020 Olympics data was utilized.
The data was filtered for only rounds 1 and 2 since better players tee off at later tee times after the cut (rounds 3, 4) for a given week. Then, all tee times were divided into two subgroups, early and late, below and above the median of tee times for a given event and round, respectively.
Strokes gained (SG), a metric created by Columbia professor and “Godfather of Golf Analytics” Mark Broadie, is a way of measuring a player’s performance relative to the field. It subtracts the player’s score of a given round from the average score of players in the field for a given round—thus it interprets the player’s strokes above the field for that round. A positive value represents strokes better than the field, so if Tiger Wood has 3.5 total strokes gained, he scored 3.5 strokes lower than the average score of the field.
Note, players with less than 30 rounds within the timeframe, essentially 15 events (the tour membership yearly minimum event requirement) were omitted from the data.
Then, I determined the difference between a player’s average SG Total if the player teed off early and the player’s average SG Total if the player teed off late. The distribution of differences between the 420 players who fit the criteria is shown below. According to the central limit theorem, with a sample size as large as ours, we would assume this distribution to be approximately normal.
Interestingly, the center of the distribution is situated less than 0, meaning that in our sample of 420 PGA tour players, on average, players who tee off later gain fewer strokes on average than players who tee off earlier. Interesting, but this is not enough to make a conclusion.
We need to run a hypothesis test. Assuming there was absolutely no difference in performance between a player teeing off early or late, we would expect the true difference to be 0. And our paired t-test indicates contrary results.
Notice that the title of the summary states “One Sample t-test,” but in essence, it is a paired test since we paired the average SG Total for early and late tee times for every player.
The mean of the sample situates at -0.286 and our p-value is extremely small. This means that if we assume that a true difference in performance between early and late is 0, there is almost a 0% chance of observing this difference of -0.286 SG. We have convincing evidence to suggest that players perform better earlier in the day than later in the day. This is the opposite conclusion that I expected to get after my disastrously early round of golf.
But can we get more specific? What if we subdivide the tee times into three categories, early, mid, and late, perhaps we could get a clearer picture.
In order to compare these three groups, I would have to run an Analysis of Variance (ANOVA), since we are comparing three groups. After subsetting the data, I checked for the three conditions to run an ANOVA.
Independence — we can safely assume that the occurrence of a player’s average Strokes Gained for an early group does not affect the player’s Strokes Gained for a mid or late group.
Roughly equal variance across groups — the variances of early, mid, and late draws were 5.7, 7.3, and 7.75, respectively. All variances were within a factor of 2, so this condition is met.
Normality — I created a Normality Probability Plot (Q-Q Plot) for each of the three groups.
For all three normality probability plots, while there is some departure from normality in extreme values, generally, the distributions of wave groups trend along the expected normality line, and thus we assume normality. We will set our alpha to 0.05.
F Test results
So, an ANOVA was conducted and the resulting f-test p-value was very small, suggesting that at least one of the groups has a different true mean average Strokes Gained, which is expected.
A following pairwise test was performed between the three groups, and the results show an interesting (and ambiguous) picture. We used a Holm-Bonferri method to account for the increase in a type I error when multiple pairwise tests are performed. As a result, the p-values from the following pairwise test can be compared against our original alpha value = 0.05.
The table suggests that the p-value between the early and late groups has a significant difference, and the middle and early groups have a significant difference—however, the late and middle groups have a p-value just above the cutoff of alpha value = 0.05. This is an ambiguous result as the p-value is small, but not small enough for us to consider it significant.
So, what is clear is that the mean average Strokes Gained for the early groups have a significant difference between the middle and late groups.
Given the results of the pairwise and the T-test performed earlier, I wanted to see if the best players fair better to the effects of early and late draws. But this analysis came up empty-handed.
Using data wrangling, I found the average SG total for each of the 420 players during the given 2017 to 2023 timeframe. I ran a plot between this, and the average strokes gained total difference. Surprisingly, there was little correlation (-0.08).
So while it’s not clear that better players are better at withstanding the effects of the late draw, it is clear that the draw has an effect on a player’s performance.
So could one assume that the time of a wake-up call (while not in the direction I expected) has an effect on a player’s performance?
Well not exactly. Unfortunately, with the nature of how professional golf works, we cannot be sure that the course plays in the same condition from early and late tee times. For example, as an HSAC analysis suggested, in which my article was heavily influenced, course conditions could become more difficult throughout the day as the greens dry out or heat takes over. Interestingly, the HSAC analysis did not take into account the early-morning effect of which I intended to analyze. Nonetheless, due to the nature of observational studies, we cannot be sure of the causation behind this early and late draw effect, only that there is one. As far as I am concerned, I am booking later tee times from now on, regardless of correlation or causation.
Sources:
datagolf.com
https://harvardsportsanalysis.org/2014/09/uspga-golf-the-impact-of-tee-times/
https://thegolfnewsnet.com/golfnewsnetteam/2022/01/14/why-pga-tour-players-have-late-and-early-tee-times-during-tournaments-124922/
