Every February, NBA All-Star voting produces a familiar two-act argument. Act one is the fan vote, which rewards popularity, market size, and accumulated star power. Act two is the analytical pushback that names the snubs and questions the inclusions.
Both acts are reasonable. They are answering different questions. The fan vote asks who fans want to watch. The analytics answer who actually delivered All-Star-level value. The two answers overlap most years but diverge in revealing ways at the margins — and the margins are where the conversation gets interesting.
The piece below is the working version of how to read All-Star voting through the analytics frame, what the data actually says about the snub and selection cases, and the short framework we apply before declaring any selection a mistake.
Quick read: All-Star voting through analytics in 60 seconds
- The voting structure: Fans vote 50%, players and media 25% each for starters; coaches pick reserves.
- What fans reward: Popularity, scoring volume, market visibility, accumulated reputation.
- What analytics rewards: Efficiency, on/off impact, role-adjusted production, defensive value.
- Where they agree: The top-15 players almost always (top stars are top stars by any frame).
- Where they diverge: Borderline cases — bench-heavy contributors, defensive specialists, high-usage role players on weak teams.
How the All-Star voting process actually works
The NBA’s All-Star starter selection breaks down by formula: 50% fan vote, 25% media vote, 25% player vote. Reserves are chosen entirely by the league’s head coaches. This split was designed in 2017 explicitly to soften the influence of pure popularity on the starting lineup, which had produced several lineups in the early 2010s that fans loved and analysts hated.
The fan-vote portion remains the largest single input. A player with significant social media following, a major-market team, or an accumulated reputation will almost always poll well in the fan vote regardless of his current season. The media and player votes correct for some of that drift but cannot override it entirely if the fan margin is large enough.
The coach-selected reserves often produce the more analytically defensible picks. Coaches see the league daily, talk to scouts, and tend to weight current-season performance more than reputation. This is where defensive specialists, two-way wings, and high-efficiency role players have a better chance of being recognized than they do in the fan vote. The official process is published on NBA.com; Basketball Reference archives historical selection data going back to the league’s first All-Star Game.
What analytics actually wants from a player evaluation
The analytical case for any player rests on a small handful of metrics rather than a single number. The table below maps the inputs that careful evaluators tend to weight and how each is captured.
| Metric family | What it captures | Public source |
|---|---|---|
| True shooting percentage | Scoring efficiency adjusted for shot type | NBA.com, Basketball Reference |
| Usage rate | Share of team possessions a player ends | NBA.com, Basketball Reference |
| On/off splits | How team performance shifts when the player plays vs sits | Cleaning the Glass, NBA.com |
| Lineup net rating | Five-man combination effectiveness | NBA.com, Cleaning the Glass |
| BPM / EPM / RAPM | All-in-one estimated impact per 100 possessions | Basketball Reference (BPM), DunksAndThrees (EPM) |
| Defensive metrics (D-EPM, defensive on/off) | Estimated defensive contribution | DunksAndThrees, Cleaning the Glass |
| Playmaking rate (assist %, potential assists) | Creation for teammates beyond raw assists | NBA.com tracking data |
A player who scores well on five or six of these is almost certainly an All-Star regardless of fan vote. A player who scores well on two and poorly on four is borderline. The analytical case is the intersection of multiple metrics agreeing, not any single number. The full vocabulary that sits around these metrics lives in our sports analytics field guide.
Where the fan vote and analytics diverge
The most informative All-Star voting conversations are the disagreements. Several patterns recur.
High-scoring stars on poor teams. A wing averaging 28 points per game on 53% true shooting for a 28-win team will poll well with fans and frustrate analysts. The fan case is the scoring. The analytical case is that the team is losing despite the scoring, often because the efficiency is below what his usage level should produce. The 28-point average earns the spotlight. The on/off split tells a quieter story.
Defensive specialists overshadowed by scorers. A wing who anchors a top-five defense, posts solid offensive numbers, and changes opponent shot selection rarely polls well with fans. The eye test for defensive impact is hard. The data captures it through defensive on/off, opponent-shot-quality differential, and team-level defensive rating with the player on the floor. The analytical case for these players is consistently stronger than the fan vote suggests.
Bench-heavy contributors on contenders. A sixth man putting up 18 points on 60% true shooting in 28 minutes off the bench will not start an All-Star game by fan vote, but his role-adjusted impact often outpaces several starters by efficiency-at-usage and on/off measures. The framework on how role context affects evaluation lives in our context problem piece.
Rising stars vs accumulated reputation. A breakout third-year player having his best season often loses the fan vote to a veteran on his last All-Star bid. The fan vote rewards accumulated identity. The analytics reward this-season production. Both perspectives have merit. They are answering different questions.
A framework for evaluating All-Star selections
The table below is the workflow we run when reading any All-Star roster against the analytics. The job is not to call selections right or wrong but to identify where the process produced an inclusion or exclusion that the data sees differently.
| Question to ask | What it reveals | What to write |
|---|---|---|
| Is the player top-25 in any all-in-one metric? | Whether the analytical case is sound | Top-25 = defensible; outside top-40 = the selection is fan-driven |
| How does on/off split look? | Whether the player moves the needle when on the floor | Above +5 = real impact; below +2 = role-replaceable |
| Is efficiency-at-usage above league average? | Whether scoring volume is sustainable or empty calories | High usage + high TS% = elite; high usage + low TS% = empty |
| Does defensive on/off support the selection? | Whether defensive case complements the offensive one | Two-way candidates earn more analytical weight |
| How does the player rank within position? | Whether selection is fair given position scarcity | Position-adjusted ranking gives the fairer comparison |
| What is the team’s net rating with him vs without? | Whether his presence shifts team competitiveness | Large differential = analytical lock; small = borderline |
| Is the case being built on small samples? | Whether a hot stretch is being mistaken for a season | See our framework on small samples |
The framework’s job is to surface where the selection and the data agree, where they diverge, and which version of the case has more weight. Neither the fan vote nor the analytics should win every argument. The interesting cases are the ones where the disagreement reveals what each frame undervalues.
Where analytics still struggles with All-Star evaluation
The data does not capture everything. Three categories remain harder to quantify than the public conversation often admits.
Leadership and locker-room presence. Coaches consistently cite these factors in interviews. The analytical community has not produced a credible public metric for them. Until that gap closes, the eye test and the coach’s perspective carry weight in this dimension that the data cannot match. The companion read on what makes a metric worth quoting at all lives in our durability piece.
Late-game performance under pressure. Clutch metrics exist but are notoriously noisy. A player who posts elite numbers in clutch minutes across a single season is most often regressing toward his career baseline the following year. The eye-test version of clutch reads more reliably than the small-sample data version.
Two-way mid-season acquisitions. A player traded in February who immediately produces for a contender has been measured in two contexts during the same season. His analytical case is messy because the data has not had time to stabilize in the new role. All-Star reserves of this profile are some of the hardest selections to evaluate fairly.
Frequently asked questions
Why does the fan vote still matter if the data has improved?
Because the All-Star Game is a fan event as much as a competitive one. The league’s commercial interest in promoting visible stars is real and not unreasonable. The 2017 reforms — splitting the starter vote into fan, media, and player buckets — were a compromise between competitive accuracy and fan engagement. Both sides have legitimate stakes.
Which advanced metric best predicts All-Star selection?
Box Plus/Minus (BPM) has historically had the highest correlation with All-Star selections among public metrics, partly because it weights the box-score inputs that voters also see. Estimated Plus-Minus (EPM) and on/off splits diverge more from the actual selections, often surfacing cases where the analytical and voting cases differ most.
How often does the analytics community get the snub debate right?
Mixed record. Snub candidates frequently end up making the team the following year, which suggests the analytical case was directionally correct. Some snub candidates fade and never make a team, which suggests the original case was small-sample-driven. The honest version is that the analytics are usually directionally right and occasionally precisely wrong.
Has the rise of analytics changed coach-selected reserves more than fan-vote starters?
Yes, clearly. Coach-selected rosters since 2018 include more two-way wings, defensive specialists, and high-efficiency role players than rosters from the early 2010s. The fan-vote starters have shifted less because the inputs to fan voting have not changed structurally. The reserve picks are where the analytical influence has been most visible.
The takeaway, in one paragraph
NBA All-Star voting and analytical player evaluation are doing different jobs. The fan vote rewards visibility and accumulated identity. The analytics reward efficiency, role-adjusted impact, and on-court contribution. Most years they overlap on the obvious selections and diverge on the borderline ones. Reading the roster through both frames produces a sharper understanding than either does alone. For the broader vocabulary this conversation sits inside, our sports analytics field guide is the natural next read.
