Using Data To Predict The 2019–2020 NBA Most Improved Player
Generally, I’ve found people don’t put too much thought into predicting the winner of the Most Improved Player award. It’s one of the least-discussed awards in preseason predictions, and gets left off of many lists entirely. When predictions are made, they’re often based more on gut feeling than anything else. And I don’t blame them. The award is not as obvious or easy to pick as MVP, not as flashy as All-NBA, not as important as DPOY.
In addition, MIP seems to be a narrative-based award, with an improvement in points being more important than anything else. You wouldn’t know that unless you looked at the numbers. So I did.
I wrote a similar article back in 2017, with only cursory look at the data and limited knowledge of statistics. This year, brings a closer look at the data and a slightly stronger knowledge of statistics.
My model takes into account both a player’s prior season stats and “jumps” in stat categories from 2 seasons prior to winning MIP to the season prior to winning the award. Most players to win Most Improved do so after making small improvements year-to-year before that breakout season. Typically, a huge jump in production doesn’t come out of nowhere. I calculated MIP score using the distribution of values for various stat categories for MIP winners since 2000. If players fell more squarely in the middle of the distribution, they scored well. In addition, I saw which of these stat categories most strongly predicted MIP success and weighted these stats accordingly.
TL;DR For this year’s candidates, I used what their stats looked like compared to the stats of past winners, then weighted by importance.
90% of athletes to win Most Improved Player in the past decade scored above 54 on my model, 70% above 57. For reference, out of all players I considered eligible for this year, only about 15% scored above 54 and 10% above 57.
Eligibility:
Players must have scored between 4.8 and 19.6 ppg in 2019, and be entering year 3–6 of their career. 95% of non-sophomore MIP in the last 20 years meet this criteria. Sophomores were not graded using my model for two reasons: 1) there were only two sophomore MIP winners in the past 20 years and 2) these players could not experience jumps from two years prior to the year prior (as they had not been in the NBA two years prior). Once I got my results, I cut them down to the 7 highest-scoring player that were likely to start on their team.
Below is how all players to win Most Improved since 2000 scored on my model. You can see how, since 2009, the model has predicted the award with greater certainty.
MIP is very much a narrative based award. That’s why already-great players never win the award, even if they improve more than those who do (e.g. Steph Curry going from MVP to Unanimous MVP). Counting stats definitely matter the most for winning MIP, especially points per game. Here’s how much weight each stat warranted.
Honorable Mentions:
These players all scored highly in the model, but there’s varying degrees of certainty as to whether each will earn a starting role this upcoming season.
Kyle Kuzma
Score: 54.11
Kuzma’s job description reads just one word: Scorer. This year, he’ll be surrounded by two of the league’s best
Kris Dunn
Score: 57.39
Dunn’s PPG decreased last year
Bryn Forbes
Score: 57.46
Forbes lacks the physical tools held by many recent award winners
Terrance Ferguson
Score: 57.99
Ferguson’s prior-year USG% (10.4) is significantly lower than any player to win it in the last 20 years.
7: Shai Gilgeous-Alexander
Score: N/A
Sophomores are often seen as some of the players most likely to break out after getting a year of experience on an NBA roster. The same is true of players who change teams, the perception being that a fresh system and new group of teammates could better suit their talents. The reality is that while these perceptions can become reality, it often doesn’t lead to a Most Improved Player award. For sophomores, I would hypothesize it’s because as second-year players they are expected to make improvements, and so even somewhat large strides go under the radar. I think that to a lesser extent the same is true for players that change teams, or maybe there just aren’t as many team-changes happening for younger players. Either way, only 2 of the last 20 MIP award winners have been second-year players, only and only 3/20 were on new teams.
However, it still happens.
But I couldn’t use my model to evaluate these athletes (as the model takes into account prior-year jumps in stats, which can’t happen for sophomores). Instead, I looked at the 17 second-year players who have come in the top 3 in MIP voting in the past 20 years, and found those who most matched their profiles. Out came Shai Gilgeous-Alexander.
Honorable Mention: Wendell Carter Jr.
6: Caris LeVert
Score: 56.54
LeVert can ball. ATS.com puts him tied for 10th in MIP odds. Many people see him as a potential burgeoning star already. That’s all I have to say on the matter.
5: Clint Capela
Score: 60.18
This might seem far-fetched, but Gobert came close to winning the award multiple times, and there’s always the potential that Capela made strides on aspects of his game we haven’t heard about. Maybe he simply makes a leap in things he’s already great at. Maybe he starts shooting threes (probably not).
4: Joe Harris
Score: 62.97
Harris is the 2nd Nets player in my top 6. In fact, there were three Nets players that made the MIP cutoff: LeVert (56.54), Allen (59.04), Harris (62.97). Allen (against many Nets fans wishes) probably won’t start this season after their acquisition of DeAndre Jordan.
Even in addition to the Nets’ two superstars, they appear to have quite the young core. It will be interesting to see which players continue to come into their own as Durant sits out the year.
Despite scoring highest in my model, Harris only ended up 4th in my personal rankings. He’s not exactly the type of player you’d expect to win the award, but perhaps neither was Pascal Siakam or Ryan Anderson. The numbers are enough to earn him a spot on the list.
3: Myles Turner
Score: 61.35
When I did these rankings two years ago, Turner was my pick to win the award. While he didn’t end up succeeding,(and neither did anyone else on my list — the award went to model outlier Victor Oladipo) he has since made strides as a player, improving his three-ball and leading the league in blocks last season.
2: Jaylen Brown
Score: 58.25
Of Boston’s young wings, it’s Tatum, not Brown, that is the popular choice to take home the Most Improved Player award this year. In fact, Tatum is the top choice out of the entire field — he was given the best odds of winning the award in 2020 by ATS.com. It’s possible, however, that Tatum may have actually scored too much (15.7 PPG) to win the award last year. With Kyrie and Horford gone and Kemba as the only true star on the team, Brown may be primed to take the leap.
Maybe he’ll deserve the $115 million he just got. Siakam just got $130 million after winning the award.
1: Luke Kennard
Score: 59.09
There’s no guarantee that Kennard starts this season, but it felt wrong to put him in honorable mentions. He had the fourth-highest score of any player on the list, and is in my eyes best equipped to win the award from a skill and narrative standpoint. He certainly has the talent necessary to win the award, but goes under-the-radar. That’s a good thing. A large improvement this year would give him the narrative boost necessary to take home the hardware.
Though it’s not something I considered when creating these rankings, it’s probably worth noting that since 2012, every player to win the award has made the playoffs. Can Detroit crack the postseason? 538 gives them a 45% chance of doing so. If you believe making the playoffs is important to winning the award, perhaps Brown would be a better bet.
Gaps in the article:
I did not calculate scores in comparison to players that did not win MIP from 2000–2019, I only used players that did win the award as my data set. In other words, I made a coverage error — I only did sampling on the dependent variable. In addition, I assumed the distribution for each stat was normal and came from a normal population, I did not know how to do any differently at the time. Next year, I’ll take these factors into account and probably change the statistics I base this on. This model is significantly more accurate than my last (which wasn’t even really a model), and next year’s model should be an improvement on this year’s as I incorporate these and other changes.
For now, however, here’s who I got.
