Draft by Numbers: Ranking the WNBA Draft Prospects Statistically
Life comes at you fast in women’s basketball. Just 11 days after Arike Ogunbowale catapulted herself into the national consciousness and led Notre Dame to the NCAA Championship, the WNBA will conduct the 2018 Draft.
In my role as founder of Her Hoop Stats, my goal has been to unlock more and better information about the women’s game. Our focus since launch has been the 2017-18 NCAA season, so 11 days isn’t enough time to introduce an important feature to our site: a WNBA Draft Model. Fortunately, Jesse Fischer, Senior Software Engineer at Amazon by day and tothemean.com draft modeling expert by night, is here for us.
More from NCAA
- Your Day in Women’s Basketball, April 6: Stanford defeats Arizona in a tightly contested matchup to win the national title
- Your Day in Women’s Basketball, March 30: UConn and Baylor deliver a classic battle of storied programs
- Your Day in Women’s Basketball, March 26: Louisville and Texas A&M survive and advance
- Your Day in Women’s Basketball, March 23: Highlights from the first round of the NCAA Tournament
- Your Day in Women’s Basketball, March 16: Tournament bracket released
What is a draft model? It’s an algorithmic assessment of players based solely on a set of data about them. While it’s automatically calculated by a computer, there is no one right way to build a model. As a result, the results are not an objective truth as they reflect the choices made by the analyst in the design process. Among the decisions that Jesse faced:
- Should the model consider how the mock drafts rank players?
- Should players be rated based on their peak performance, the value they deliver over their career, the value they deliver on their rookie contract, or the value they deliver before they are eligible to be an unrestricted free agent?
- Should we consider the statistics of the player’s team?
- How many years of data should be considered?
It is completely reasonable to have different answers to questions such as these. In Jesse’s case, his PLUM model detailed on tothemean.com is stats-only and does not incorporate the mock drafts. It uses the player’s box score data to predict their WNBA performance after adjusting for pace, team and opponent strength, and the player’s position, height, and year in school.
Ideally, we would be able to combine the PLUM results with the results of other draft models that made different choices or use different data such as rank in high school class to produce an even better prediction. This machine learning technique has the relatively fancy name “ensembling”. However, it is the same principle as having many voters in the AP poll who weigh the available facts differently to create a better overall ranking of teams than any single voter could produce.
Unfortunately, I’m not aware of any other WNBA draft models at the moment. Please let us know in the comments if we missed any. Next year, we plan to have a WNBA model from Her Hoop Stats and hopefully a number of others will be developed as well. One of the best aspects of tothemean.com is their NBA Draft Tool showing the rating and ranking from numerous analytical models as well as multiple mock drafts. As always, there is no reason we shouldn’t have the same level of analysis for the women’s game.
So, with those caveats, how does the PLUM model rate prospects in the 2018 WNBA Draft? Check out the table below for the high-level results. For a more in-depth look, read Jesse’s 2018 draft analysis blog post on tothemean.com.
Don’t forget, no model (or scout, GM, team or journalist) will be perfect. Also, what’s most interesting is not that (spoiler alert) A’ja Wilson ranks number 1. You don’t need stats to say that A’ja Wilson and LeBron James are great at basketball. The real value comes from understanding the discrepancies between the draft model and scouting ranks. You should hope your favorite team’s GM is looking at is why there is a difference between the draft model and scouting and be sure to be comfortable about why they trust one over the other.
PLUM MODEL RESULTS
Player PLUM Rank Mock Draft Rank A'ja Wilson 1 1 Azura Stevens 2 4 Victoria Vivians 3 8 Gabby Williams 4 6 Jordin Canada 5 5 Shakayla Thomas 6 19 Ariel Atkins 7 15 Kaylee Jensen 8 40 Myisha Hines-Allen 9 18 Maria Vadeeva 10* 10 Monique Billings 11 11 Kelsey Mitchell 12 2 Vionise Pierre-Louis 13 23 (tie) Marie Gülich 14 14 Kristy Wallace 15 NR Katelynn Flaherty 16 23 (tie) Stephanie Mavunga 17 12 Dekeiya Cohen 18 NR Loryn Goodwin 19 22 Kia Nurse 20 9 Brooke McCarty 21 20 Diamond DeShields 22 3 Tyler Scaife 23 16 Lexie Brown 24 7 Mercedes Russell 25 17 Morgan William 26 30 Brittany McPhee 27 NR Khaalia Hillsman 28 NR Tinara Moore 29 NR Teana Muldrow 30 34 Jill Barta 31 21 Charise Beynon 32 NR Cierra Dillard 33 NR Jaime Nared 34 13 Natalie Butler 35 25 Maddie Manning 36 NR Notes: * Maria Vadeeva played in Russia rather than the NCAA; she is assigned PLUM rank 10 reflecting her mock draft consensus * Mock Draft Ranks reflects the average of the rank in the High Post Hoops v6 Mock Draft and Aneela's Final Mock Draft * NR reflects Not Ranked in either Mock Draft * All players not ranked in one of the Mock Drafts are considered to be ranked 37th in that Mock Draft * Diamond DeShields' projection is based on her 2016-17 NCAA stats
Comments
You probably have many questions and comments about the PLUM rankings for 2018. “That’s crazy, Player X is way too low!” “No way Player Y should be a 1st round pick!” “Did Jesse and Aaron watch any women’s games this season?” We certainly did watch a bit, but nowhere near as much as WNBA front offices. Of course, our personal observations from watching aren’t reflected in these results. There’s no doubt when we look back in perhaps four years PLUM will not have ranked the players perfectly. Neither will the WNBA front offices. Jesse’s original PLUM post and 2018 article includes some insight on how the model has performed historically that show the results seem reasonable. A future project is to compare the PLUM results to how well the league has ranked prospects via the actual draft picks.
Smart teams will incorporate these results into their ranking but no team should rely on them exclusively. The best teams (and models) combine the input from scouting, which includes what happens on and off the floor, with statistical models of on-court performance to form their final board. One way to think about it is rather than taking Shakayla Thomas sixth, teams should think about where she is on their “scouting board” and decide if she should perhaps be ranked higher overall.
Given the length of this post, we could dive into the details on many of these prospects. Here are a few quick comments on the PLUM rankings we found most notable:
- Kaylee Jensen probably benefits the most from being viewed through a stats-only lens. Keep in mind, on a good Oklahoma State team she ranked in the top 2% of the NCAA in defensive rebound rate and top 5% in both offensive rebound rate and shooting efficiency. She was also in the top 10% in usage and had a low turnover rate
- Kristy Wallace dropped in the mock draft because of her injury in late February. She had a great season before the ACL tear. The model isn’t factoring her injury into the prediction and is effectively assuming a full recovery. She could be a steal for a patient team that doesn’t need her contribution immediately. After all, you wouldn’t have expected her rookie year to be the best of her career. Wallace was in the top 5% of the country in shooting efficiency averaging 1.21 points per scoring attempt this year and top 10% in assist rate with an assist on 26% of her teammates’ baskets when she was on the floor.
- PLUM projects Kelsey Mitchell far below the mock draft consensus, including others we didn’t have time to include such as Mechelle Voepel’s espnW mock. Why so low? Even though Mitchell is 3rd in the country in scoring, Ariel Atkins ranks ahead of her in PLUM mostly due to her height advantage and scoring much more efficiently albeit on far lower volume. Mitchell is also far lower in steal rate than Atkins and Jordin Canada, a factor that appears to be important in the NBA. The good news for Mitchell in Jesse’s work is that he has an alternate “All-Star” model that ranks her 4th among prospects with a 60% chance of being an All-Star.
- Diamond DeShields also rates far below the consensus. She ranks highly in per game stats since she played relatively heavy minutes and was a high volume player. However, the only rate stat besides usage rate (shooting volume) that DeShields ranked in the top 10% of the NCAA in 2016-17 was assist rate. Her PLUM rating based on her freshman 2013-14 season was actually higher but would still place her just 15th in the rankings.
If you have feedback or more questions, leave us a comment below or reach out on Twitter to Jesse and Aaron and we’ll try to respond to as many as we can.