PGA Tour Success Metrics Analysis
Li Wang
2022-05-03
Business Problem:
What metric is the best predictor of success on the modern PGA Tour?
If I were a PGA Tour professional, what area of my game should I focus
on in order to make the most money possible, while shooting the lowest
scores possible?
Metrics:
The success metrics are defined as:
Earnings Per Event ($USD), a measure of how much money
a player made over the timeline of the analysis per tournament,
calculated by dividing the sum of their earnings by the number of events
they played (the higher, the better). Missed cuts and withdrawals,
both medical and non-medical, automatically translate to $0 of earnings
for the event.
Scoring Average, a measure of how many strokes it takes
a player to complete a round, calculated by the total number of strokes
a player makes divided by the total number of rounds they have played
(the lower, the better).
The input metrics are:
Driving Accuracy (%), a measure of how accurate a
player’s tee shots are, calculated by dividing the total number of times
a player’s tee shot finished in the fairway by the total number of tee
shots they hit (the higher, the better);
Driving Distance (yards), a measure of how far a
player’s average tee shot travels (the higher, the better);
Greens in Regulation (%), a measure of how accurate a
player’s approach shots are, calculated by dividing the total number of
times a player’s approach shot finishes on the putting green by the
total number of approach shots they hit (the higher, the better);
Putts Per Round, a measure of how many strokes a player
takes to put the ball in the hole once reaching the putting green,
calculated by dividing the total number of putts a player had by the
total number of holes the player played (the lower, the better);
Scrambling (%), a measure of how often a player gets
“up and down” (when a player puts the ball in the hole in two or fewer
strokes after their approach shot missed the putting green), calculated
by dividing the total instances of successfully getting up and down by
the total number of times a player missed the putting green on their
approach shot (the higher, the better);
Data Source:
The Official PGA Tour
website keeps raw data of all of the above metrics, from 1980 to the
present. However, I am only interested in the best predictor of success
on the “modern” PGA Tour; due to two reasons, tournaments conducted
prior to 2010 are not considered modern enough to be consistent with
those conducted in the 12 years since. Firstly, recent advances in
technology have allowed drivers to go further, irons to be more
accurate, and putters to be more consistent, which has effectively made
the game easier, and therefore the stats would heavily favor the
post-2010 data. Secondly, the sizes of the purses from 1980 to 2010
dramatically increased over the years due to more sponsors and more fan
attendance; while the rate of increase of modern-day purses since 2010
has been negligible in comparison, offering more consistency. As a
result, the data for this analysis includes only the years
2010-2021.
Note also that the 2021-2022 season is only halfway complete, so
data from 2022 will not be used.
Processing the Data:
Step 1: Obtained the appropriate data from the official
PGA Tour website, and used Excel to clean the data so that it is uniform
and properly formatted for the next step.
* Found and deleted duplicates across all tables;
* Used business sense to find and replace null values with the correct
values (i.e. in the “victories” field of the “money_made” table, if no
record was found for that field, it meant that the player did not win
any tournaments that year);
* Appropriately renamed field headers across all tables, to make them
consistent and more descriptive (i.e. renaming the “gir” field to
“greens_in_regulation” to make it easier to understand);
* Created two matching keys across all data tables for the future
joining of as many tables as necessary (the “player” and “year”
fields);
* Assigned corresponding year values to each record across all
tables;
* Exported each metric as an individual data table in respective .csv
files.
Step 2: Imported the data from the respective .csv
files into Google Big Query and used the sandbox SQL platform to
aggregate, filter, and sort the data for the next step.
* Created a new project titled “lw-capstone” inside of the SQL
workspace;
* Created seven data tables by importing the seven .csv files created
earlier;
* Inside a nested query, joined all seven tables together using LEFT
JOIN’s on a joint primary key created by “player” and “year” – the
reason for joining is to combine the data from all tables into one
table, the reason for using two keys jointly is to make sure all records
match player names as well as the corresponding years they played, and
the reason for using a LEFT JOIN rather than an INNER JOIN is that not
all players were on the PGA Tour for all 12 years, but I still want to
return their stats for the years in which they were active;
* Within the same nested query, returned and renamed only the fields
that I need;
* In the outer query, aggregated the data by created and returning
summarized values for all of the targeted metrics, grouped by player,
filtered by players who have averaged more 10 or more events per year
over the course of the 12 years (to ensure large enough of a sample
size), and sorted the data from most earnings per event to least
earnings per event;
* Exported the data into a .csv file called “pga_aggregate_stats.csv”
for further analysis.
Here’s the SQL query in its full form:
You can also access the code on GitHub HERE
Analyzing the Data:
Step 1: Launched R-Studio, installed and loaded the
“tidyverse”, “readr”, “ggplot2”, and “dplyr” packages used for data
analysis, imported the data from the previously generated .csv file
using the read.csv function, and saved the data as a dataframe called
“pga_tour_stats”:
setwd("~/Desktop/Learning/Google Data Analytics Certificate/Capstone Project")
pga_stats <- read.csv("pga_aggregate_stats.csv")
as_tibble(pga_stats)## # A tibble: 502 × 12
## player tot_events_played tot_earnings years_active events_per_year
## <chr> <int> <int> <int> <int>
## 1 Rory McIlroy 182 54764656 11 17
## 2 Jon Rahm 100 28771788 5 20
## 3 Dustin Johnson 239 67826516 12 20
## 4 Collin Morikawa 53 14065666 3 18
## 5 Justin Thomas 167 43915588 7 24
## 6 Brooks Koepka 139 35612106 7 20
## 7 Tiger Woods 114 27989167 10 11
## 8 Jordan Spieth 207 47750892 9 23
## 9 Patrick Cantlay 99 21979822 7 14
## 10 Bryson DeChambeau 117 25522685 5 23
## # … with 492 more rows, and 7 more variables: earnings_per_event <dbl>,
## # driving_accuracy_percent <dbl>, avg_driving_distance <dbl>,
## # gir_percent <dbl>, putts_per_round <dbl>, scrambling_percent <dbl>,
## # scoring_average <dbl>
Step 2: Calculated the correlation coefficients of all
five input metrics (Driving Accuracy, Driving Distance, Greens in
Regulation, Putts Per Round, and Scrambling) against the success metrics
(Earnings Per Event and Scoring Average).
The official definition of the correlation coefficient is: “The
correlation coefficient is a statistical measure of the strength of the
relationship between the relative movements of two variables. The values
range between -1.0 and 1.0. A calculated number greater than 1.0 or less
than -1.0 means that there was an error in the correlation measurement.”
-Investopedia
Correlation Coefficients of Earnings Per Event vs. the Input
Metrics:
as_tibble(earnings_correlations)## # A tibble: 1 × 5
## driving_accuracy_earnings driving_distance_earn… gir_earnings putting_earnings
## <dbl> <dbl> <dbl> <dbl>
## 1 0.00955 0.361 0.299 -0.295
## # … with 1 more variable: scrambling_earnings <dbl>As expected, all of the correlation coefficients were positive except
for Putting, because in each of the other cases, a higher number is seen
as better, but Putting is the only metric where less is better. It is
also nice to see that the coefficients were all between -1.0 and +1.0,
which means that there were no errors in calculation.
Based on these results, Driving Distance (0.3612) and Scrambling
(0.3344) are the metrics most closely correlated with earning more money
per event, while Greens in Regulation (0.2990) and Putting
(-0.2949) still had an impact but not as much, and Driving Accuracy
(0.0096) did not make much of a difference at all. Moving onto the other
success metric, Scoring Average, we get the following results:
Correlation Coefficients of Scoring Average vs. the Input
Metrics:
as_tibble(scoring_correlations)## # A tibble: 1 × 5
## driving_accurac… driving_distanc… gir_scoring putting_scoring scrambling_scor…
## <dbl> <dbl> <dbl> <dbl> <dbl>
## 1 -0.248 -0.279 -0.621 0.284 -0.639Just like above, at first glance, the results were as expected; all
of the correlation coefficients were negative except for Putting,
because the success metric of Scoring Average is now defined as
less-is-better. In addition, the coefficients were again all between
-1.0 and +1.0, meaning there were no errors in calculation.
Based on these results, Scrambling (-0.6388) and Greens in
Regulation (-0.6215) are the metrics most closely correlated with
shooting lower scores, while Putting (0.2842) and Driving
Distance (-0.2786) still had an impact but not as much. Once again,
Driving Accuracy (-0.2479) made the least difference, although its
effect on Scoring Average was much more significant than on Earnings Per
Event.
Step 3: Calculated two correlation coefficients between
metrics themselves (Driving Accuracy vs. Driving Distance, and Earnings
Per Event vs. Scoring Average):
as_tibble(other_correlations)## # A tibble: 1 × 2
## driving_correlations earnings_scoring
## <dbl> <dbl>
## 1 -0.580 -0.651
Both correlation coefficients showed statistically significant
relationships. In the case of Driving Accuracy vs. Driving
Distance (-0.5803), there was a noticeable negative correlation, which
makes sense because as a player hits the ball further, they are
naturally less accurate. Likewise, in the case of Earnings Per Event
vs. Scoring Average (-0.6509), there was also a noticeable negative
correlation, which also makes sense, because as a player scores lower in
tournaments, they will naturally finish higher in the events and make
more money. But with both, the correlation was not close enough
to -1.0 that they could be considered strictly correlated with each
other, because there are other variables that impact them (namely, the
other input metrics).
Data Visualizations:
Visually, the relationship between these variables can be plotted
against each other like this:
Correlation Coefficients of Earnings Per Event vs. the Input Metrics:
Correlation Coefficients of Scoring Average vs. the Input Metrics:
Correlation Coefficients Between Metrics Themselves:
Conclusions and Recommendations
Before tying these numbers back to the business problem at hand (what
is the best predictor of success on the PGA Tour?), let’s first recap
the findings of the analysis:
* Driving Distance (cor: 0.3612) and Scrambling (cor: 0.3344) are the
metrics most closely correlated with earning more money per event, while
Greens in Regulation (cor: 0.2990) and Putting (cor: -0.2949) still had
an impact but not as much, and Driving Accuracy (cor: 0.0096) does not
make much of a difference at all;
* Scrambling (cor: -0.6388) and Greens in Regulation (cor: -0.6215) are
the metrics most closely correlated with shooting lower scores, while
Putting (0.2842) and Driving Distance (cor: -0.2786) still had an impact
but not as much. Once again, Driving Accuracy (cor: -0.2479) made the
least difference;
* As Driving Distance increases, Driving Accuracy decreases (cor:
-0.5803), and as Scoring Average decreases, Earnings Per Event increases
(cor: -0.6509).
Based on the above, I gave each of the input metrics a score of 1-5 for
each category, with 5 being the best:
* Earnings Per Event: (5) Driving Distance, (4) Scrambling, (3) Greens
in Regulation, (2) Putting, (1) Driving Accuracy;
* Scoring Average: (5) Scrambling, (4) Greens in Regulation, (3)
Putting, (2) Driving Distance, (1) Driving Accuracy;
It can be assumed that the average PGA Tour player cares far more about
money than about his average score, because at the end of the day, they
are competing for their livelihoods. However, we should not disregard
the significant correlation between lower average scores and higher
amounts of money earned, which means that Scoring Average contributes to
overall success in other ways, and should still be factored into the end
decision. Therefore, I will weight these scores in a 75%-25% split, in
favor of Earnings Per Event.
Taking into account the above weighting system, and combining it with
the assigned input metric scores, I calculated the total scores for each
input metric:
earnings_coeff <- 0.75
scoring_coeff <- 0.25
input_scores <- data.frame(input_metrics = c("driving_distance", "scrambling", "greens in reg",
"putting", "driving_accuracy"),
earnings_scores = c(5, 4, 3, 2, 1),
scoring_scores = c(2, 5, 4, 3, 1))
input_scores <- input_scores %>%
mutate(combined_score = (earnings_coeff * earnings_scores) +
(scoring_coeff * scoring_scores))
arrange(input_scores, desc(combined_score))## input_metrics earnings_scores scoring_scores combined_score
## 1 driving_distance 5 2 4.25
## 2 scrambling 4 5 4.25
## 3 greens in reg 3 4 3.25
## 4 putting 2 3 2.25
## 5 driving_accuracy 1 1 1.00
Once the input metrics have all been assigned scores, two of the metrics
have distanced themselves from the others and are tied for most
important: Driving Distance and Scrambling. Business
intuition confirms that this indeed makes sense; the “modern” form of
golf is widely regarded as favoring “bomb-and-gouge” players, or players
that drive the ball a long ways (the “bomb”), and because distance is
inversely correlated with accuracy, these same players will ultimately
need to play more of their approach shots from the rough (the “gouge”),
thus placing a high emphasis on their short games and their ability to
scramble!
In conclusion, while golf is a multi-faceted sport with many
variables at play, at the highest levels of golf on the modern PGA Tour,
in order to have greater chance of success, a player should focus on
increasing their Driving Distance, as well as their Scrambling
Percentage, for the highest return on the investment of their training
time.
Future Considerations
In the future, I would love to expand on this current project, and have
the following modifications / improvements in mind:
* I would conduct a separate analysis on PGA metrics from the years 1980
to 2010, to see how the game has evolved over time;
* I would dive deeper into Driving Distance and Scrambling to see if
there are particular aspects of these metrics that matter more than
others (i.e. “number of drives per season over 300 yards,” or “bunker
scrambling percentage” vs. “rough scrambling percentage”);
* I would look at other success metrics, such as the number of
tournament victories a player has, although I hypothesize that the
results would be similar, since the players with making the most money
and having the lowest scoring averages are probably also the ones
winning the most;
* I would take a handful of the most successful golfers in the modern
era (i.e. Tiger Woods, Phil Mickelson, Rory McIlroy, Dustin Johnson,
etc.) and dive deeper into what made them so successful;
* I would be interested in seeing how these results are similar or
different on other major worldwide professional golf tours such as the
DP World Tour (formerly the European Tour); the PGA Tour competes
primarily in the United States, and maybe the courses that players play
on elsewhere would not reward the “bomb-and-gouge” playing style in the
same way;
* I would be interested in diving deeper into any outliers found by this
study (i.e. players who are low on Driving Distance, but have still made
a lot of money per event and shoot low scores) to find out how and
why;
* I would be interested in seeing which input metrics correlate the most
strongly with Scoring Average for the average, 15-handicap golfer.