Forecasting Ideal 3-and-D Players: Part One

We are down to 4 teams in the playoffs, which means 26 teams and their fan bases are anxiously looking forward to the draft. As most know, I’m a huge Oklahoma City Thunder fan. As I look at the many possible places for improvement, one that immediately pops out is our perimeter shooting. The Thunder had the sixth worst 3-pt shooting percentage in the league at 33.1%. However, drafting a player that is a pure sharpshooter isn’t enough; in today’s league, a wing player needs to be sound in both shooting and perimeter defense—a role often termed as a “3 and D” player. In this blog post, I have created two separate regression models to forecast a player’s catch-and-shoot ability and defensive rating. This first blog post will examine a player’s catch-and-shoot ability.

Data Collection:

In both regression models, the sample size chosen was completely random. I picked 49 players from the 2013 draft and onwards, looking at both their college and draft combine measurement stats. The table below shows the players that were studied:

Kentavious Caldwell-Pope	Allen Crabbe	Michael Carter-Williams	Tim Hardaway Jr	Victor Oladipo	CJ McCollum	Trey Burke
Otto Porter	Reggie Bullock	Andre Roberson	Tony Snell	Robert Covington	Zach Lavine	Jordan Clarkson
Aaron Gordon	Marcus Smart	Elfrid Payton	Doug McDermott	T.J. Warren	Joe Harris	Rodney Hood
Shabazz Napier	Devin Booker	Rondae Hollis-Jefferson	Pat Connaughton	Terry Rozier	Cameron Payne	TJ McConnell
Kelly Oubre	Delon Wright	Denzel Valentine	Ron Baker	Malcolm Brogdon	Tyler Ulis	Taurean Prince
Patrick McCaw	Josh Hart	Justin Jackson	Kyle Kuzma	Derrick White	Frank Jackson	Dillon Brooks
Donovan Mitchell	Jalen Brunson	Kevin  Huerter	Aaron Holiday	Allonzo Trier	Josh Okogie	Landry Shamet

 

The combine and college stats (sources are NBA.com/stats & sports-reference.com/cbb respectively) that I factored in as independent variables were as such:

• Body Fat (%)
• Hand Length (inches)
• Hand Width (inches)
• Standing Reach (inches)
• Weight (lbs)
• Lane Agility (seconds)
• Shuttle run (seconds)
• Height (inches)
• Wingspan (inches)
• Years in College
• College Average Career 3PT%
• College Average Career FT%
• College Average Career 3Par
• College Average Career TS%
• College Average Career eFG%

Finally, the dependent variable—the variable of interest—was a player’s average career catch-and-shoot 3PT% in the NBA, provided by NBA.com/stats.

Creating a Regression Model

I ran the regression for a first iteration. I set my alpha level for my model at 10%. What does this mean? It means any independent variable that returns a p-value less than 10% are deemed as significant variables for the model. It essentially means that if a variable (i.e. College Average Career FT%) returns a p-value of 5% then that variable is deemed as an important piece of the puzzle for forecasting a player’s catch-and-shoot ability in the NBA. If a variable exhibits an alpha level at or slightly above 10%, then that variable is subject for further examination rather than being immediately thrown out. If it makes logical sense to include that variable, then that variable is still included. For example, pretend that I’m forecasting Coachella’s attendance numbers and a variable (like weather) returns a p-value of 15%. Although it’s not below 10%, I may still try to include it because it makes intuitive sense that if the weather is bad, then attendance numbers may be dismal—conversely if the weather is great, then attendance numbers may be great.

OK! Enough nerd talk. After the first iteration, these variables were deemed as outside my alpha level of 10%:

• Body Fat (%)
• Hand Length (inches)
• Hand Width (inches)
• Standing Reach (inches)
• Weight (lb)
• Years in College
• College Career 3Par

The next step is to re-do the regression taking the above variables out. Once again, we determine if our variables in our second iteration meet the alpha level of 10%. These variables were ruled out after the second variable:

• Shuttle Run
• College Career TS%
• College Career eFG%

I then repeat the practice for a third iteration with the remaining variables. Here’s where it gets interesting. The variables height; wingspan; college career 3PT%; and college career FT% all meet the alpha level of 10%. The variable lane agility exhibits an alpha level of 15%. This model gives me a R-squared value of 41%. [A brief aside for those that are unfamiliar with “r-squared”:  to put it simply, r-squared is essentially how much of the puzzle has the model figured out. An r-squared of 41% means that the model has figured out 41% of the puzzle, while 59% is still waiting to be solved. An r-squared value of 41% isn’t great, but nonetheless, these variables are still significant pieces to the puzzle.]

Back to the lane agility variable. In one sense, it could make sense to include lane agility as lane agility is a combination of how a player changes direction; how fast that player changes directions; along with how fast the player is. This could indicate how a player moves without the ball as the player comes off screens. However, here I choose to not include it (shoot me an email if you want to understand my rationale for not including it).

A final regression iteration is run without include the lane agility variable. All of my variables now exhibit less than the 10% alpha level threshold and a r-squared value of 38.2%. Once again, further improvements to the model could be made in the future to raise the r-squared value. Nonetheless, we have our final model:

where:

Interpreting the Results:

Along with making sure that the variables meet the alpha level threshold, it’s good practice to rationalize why each variable may influence our dependent variable: catch-and-shoot 3PT%. Let’s run through each variable:

• Height: The sign in front of our height variable is positive—this means that a player is a better catch-and-shoot shooter if that player is taller. This makes logical sense. The taller a player is, the likelier that player is to see over the defender (and his hand) when shooting the ball.
• Wingspan: The sign is negative in front of our wingspan variable. This also makes sense. Every shooting coach mentions that the more compact the shooter’s form, the better. Longer wingspans make that much harder. Furthermore—and this is just pure subjective conjecture—the longer a player’s wingspan, the likelier it is for a player’s shooting arc to be flat.
• College Career 3PT% & College Career FT%: These obviously make sense to be positive.

Perhaps more interesting is that height and wingspan actually play a larger role than a player’s college shooting résumé. To examine, let’s create an “average” shooting guard: 6’6” with a 6’9” wingspan who shot an average of 35% from the 3’s and 70% from the free-throw line in college. Plugging in these variables, we obtain these results.

The forecast shows the player to have a catch-and-shoot 3PT% of 34%, but more importantly, observe the impact of a player’s physicals. The total magnitude of height and wingspan are 58.828 and -37.017, respectively, while the college shooting percentages are much lower than that in total magnitude.

Forecasting 2019 Shooter’s 3PT Catch-and-Shoot Ability:

Now, the fun part. The NBA combine results are slowly being shared with the public online. Using the above regression model, I was able to forecast some of the best 3PT catch-and-shoot shooters in this draft [note: combine results are incomplete, so an update to this post will be available later. Furthermore, some of these results are using unofficial height & wingspan data]. These are picks that are within the range for the Thunder’s first round selection. A full table of all available players will be in the subsequent update.

Player	College	Forecasted NBA Catch-and-Shoot 3PT %
1. Tyler Herro	University of Kentucky	42.29%
2. Cameron Johnson	University of North Carolina	40.64%
3. Ty Jerome	University of Virginia [Go Hoos!]	39.99%
4. Louis King	University of Oregon	37.84%
5. Coby White	University of North Carolina	37.74%
6. Admiral Schofield	University of Tennessee	36.92%
7. Nickeil Alexander-Walker	Virginia Tech	36.61%
8. Chuma Okeke	Auburn	36.35%
9. Keldon Johnson	University of Kentucky	35.77%
10. Matisse Thybulle	University of Washington	34.54%

 

Before jumping into any conclusions about drafting Tyler Herro, be sure to check out the second part of the blog which will forecast each player’s defensive rating!

Updated table of this year’s wing players based on combine results:

Player	College	Forecasted NBA Catch-and-Shoot 3PT%
Tyler Herro	University of Kentucky	42.3%
Cameron Johnson	University of North Carolina	40.6%
Ty Jerome	University of Virginia	40%
Kyle Guy	University of Virginia	38.9%
Jordan Poole	University of Michigan	38.8%
Dylan Windler	Belmont	38.7%
Ignas Brazdeikis	University of Michigan	38.6%
Zach Norvell Jr	Gonzaga	38.5%
Louis King	University of Oregon	37.8%
Coby White	University of North Carolina	37.7%
Jaylen Nowell	University of Washington	37.7%
Jordan Bone	University of Tennessee	37.4%
Admiral Schofield	University of Tennessee	36.9%
Nickeil Alexander-Walker	Virginia Tech	36.6%
Devon Dotson	Kansas	36.4%
Chuma Okeke	Auburn	36.4%
Jaylen Hands	UCLA	36.1%
Quinndary Weatherspoon	Mississippi State	35.9%
Keldon Johnson	University of Kentucky	35.8%
Miye Oni	Yale	35.5%
Cam Reddish	Duke	35.5%
Ky Bowman	Boston College	35.4%
Daquan Jeffries	Oral Roberts/University of Tulsa	35.4%
Jalen McDaniels	San Diego State	35.3%
Shamorie Ponds	St Johns	35.1%
Mariol Shayok	Iowa State	35.0%
Eric Paschall	Fordham/Villanova	35.0%
Carsen Edwards	Purdue	34.8%
Matisse Thybulle	University of Washington	34.5%
Grant Williams	University of Tennessee	34.5%
Jarrett Culver	Texas Tech	34.4%
Kris Wilkes	UCLA	34.2%
Terance Mann	Florida State	34.1%
Tremont Waters	LSU	34.1%
Oshea Brissett	Syracuse	34.0%
KZ Okpala	Stanford	33.9%
Rui Hachimura	Gonzaga	33.6%
Terance Davis	Ole Miss	33.6%
Reggie Perry	Mississippi State	33.3%
Jared Harper	Auburn	33.2%
Cody Martin	NC State/University of Nevada	32.5%
PJ Washington	University of Kentucky	32.5%
Kevin Porter Jr	USC	32.4%
Quentin Grimes	Kansas	32.2%
Luguentz Dort	Arizona State	31.9%
Romeo Langford	University of Indiana	31.3%
Brandon Clarke	Gonzaga	31.2%
Nassir Little	University of North Carolina	31.1%
Charles Matthews	University of Kentucky/Michigan	30.4%
Talen Horton-Tucker	Iowa State	28.2%

As always, I appreciate all of my readers and feel free to email me with any questions:

KangJ20@darden.virginia.edu