Can We See John Stockton’s Heart Rate in the Box Score?

3-Post NBA Durability Series — Status 2026-10-11 (v3)

The Premise

John Stockton and Boris Diaw reportedly share a resting heart rate of ~35 BPM — well below the elite-athlete norm of ~50 BPM. The question: does that physiology leave a detectable trace in box-score data as statistical resilience — fewer missed games, steadier late-game performance, more durable decision-making under fatigue? Stockton’s durability is currently circulating on social media, providing a natural hook.

Post Structure

Post	Subject	Data Window
1	John Stockton + 14 PG comps	E1/E3 full career 1984–2003; E2/E4/E5 1996–2003 only
2	Boris Diaw + 5 PF comps (Nowitzki, Boozer, Jamison, Marion, Lewis)	2004–2016, full BRef
3	All qualifying BRef-era guards + forwards + centers	1980–2010 debuts

Stockton ships first to ride the current conversation; Diaw follows immediately as independent methodological confirmation — different position, different era, full BRef coverage, no age-regime caveat. Post 3 addresses the cherry-picking objection: placing both players in the full distribution of all BRef-era players who meet the career-length threshold.

Estimands

	Metric	Data Tier	Status
E1	Games missed per 82 — career durability	Game logs	Done — both cohorts
E3	Shot selection: 3PA/FGA, FTA/FGA on B2B vs rest	Game logs	Done — both cohorts
E2	4Q FT% degradation vs Q1–Q3	Quarter box scores	Done (PG) — Stockton +4.1 pp, only improver; PF pending
E4	4Q AST/TOV ratio vs full-game	Quarter box scores	Degenerate at quarter resolution (52% zero-TOV rows); dropped
E5	4Q TS% vs Q1–Q3	Quarter box scores	Degenerate (21% coverage per quarter); dropped
E6	Full Q1→Q2→Q3→Q4 scoring profile	Quarter box scores	Done — both cohorts

E1 and E3 are primary. E6 (profile plot) generated from existing quarter splits; committed to both posts.

Inference Approach

E1’s denominator corrects for lockout-shortened seasons: 1999→50 games, 2012→66, 2020→65, 2021→72.

n_treated = 1 makes bootstrap CIs invalid. For E1/E3: resample the treatment label across all N cohort players 5,000 times; report where the observed gap falls in the null distribution. For E2/E4/E5: mixed-effects model metric ~ quarter * is_treated + rest_days + (1|player_id) via statsmodels.MixedLM; permutation resamples is_treated at the player level (not game level) to preserve within-player correlation.

Current Status

Infrastructure: complete. 141 tests passing across four scripts:

Script	Purpose
fetch_bref_data.py	BRef game log scraper; HTML-cached per (player, season); 4 sec/request
analyze_durability.py	E1 + E3; OLS + permutation; emits {cohort}_comparators.csv
fetch_quarter_splits.py	Quarter box score scraper; cached by game ID; 3.5s/request + jitter
analyze_quarter_splits.py	E2/E4/E5; mixed-effects model; player-level permutation

Data fetch: complete for PG and PF cohorts; population fetch running locally.

Status	Players
Full PG game logs + quarter splits	Stockton, Payton, Jackson, Harper
Full PG game logs only	Thomas, Hardaway, K. Johnson, Porter, Bogues, Strickland, Kidd, Nash, Blaylock, Anderson, Douglas
PF game logs + quarter splits (all 6)	Diaw, Nowitzki, Boozer, Jamison, Marion, Lewis
Population (local fetch)	~109 players fetched, running — ETA ~Oct 14

Post drafts:

Post	E1/E3	E2/E4/E5	E6	Status
Stockton	Written	E2 written; E4/E5 dropped	Written	Published (draft removed; freeze committed)
Diaw	Written	E2 written (null); E4/E5 dropped	Written	Published (draft removed; freeze committed)
Population	—	—	—	Outline only — blocked on population fetch

Caveats (mandatory in every post)

1. Selection-on-outcome bias — >=600-game threshold conditions on the dependent variable; E1 estimates durability among players who completed long careers, not all players.
2. B2B state selection — playing a B2B selects healthier player-states; within-player pairing removes stable effects but not selection.
3. Blowout/garbage time — starters sit in non-competitive games; 4Q minutes are not random. Compresses treatment effect toward zero.
4. Intentional fouling — close-game FTA inflation in Q4; FT% estimates partly reflect pressure context, not just physical state.
5. Timeout-adjusted rest — unmeasured confounder; coaches use Q4 timeouts strategically.
6. Age-regime — Stockton’s quarter data covers ages 34–40 (career decline). A null within-game result does not contradict a prime-years resilience claim.
7. Pre-1996 quarter gap — BRef play-by-play starts 1996-97; only Stockton ages 34–40 are covered. Justin Jacobs (Squared Statistics) has video-reconstructed pre-1996 quarter logs that would extend the window to his prime (ages 22–33). Outreach drafted, not sent.
8. No causality — frame consistently as “consistent with” and “does not contradict.”

What’s Left

Step	Blocker	Est.
Population fetch (local)	Running — ETA ~Oct 14	—
Population analysis + post	Population fetch complete	1 day
Jacobs outreach (#22) — manual browser form	Human action required	20 min

---

NOTE: Causal language forbidden. Null results are valid findings — state them plainly. Sourcing asymmetry: Diaw’s 35 BPM comes from documented post-retirement interviews; the Stockton figure is less sourced — acknowledge this explicitly in the Stockton post.

---

Changelog

v3 — 2026-10-11

• E6 plots added — plot_e6_quarter_profile() added to analyze_quarter_splits.py; generates Q1→Q4 FT% line chart highlighting treated player. PG: Stockton is the only player with a monotonic Q1→Q4 rise (76.9% → 80.6% → 83.2% → 84.7%). PF: Diaw shows slight Q4 decline (67.7%), Marion is the Q4 outlier (82.2%).
• E6 figures generated for both cohorts (data/figures/pg/e6_quarter_profile.png, data/figures/pf/e6_quarter_profile.png); figures committed.
• E6 section added to both posts — Stockton post: updated from “pending” placeholder to real figure with caption; conclusion updated with monotonic-increase note. Diaw post: new E6 section inserted after E5.
• Diaw E2/E4/E5 complete — PF quarter splits were already fetched (Oct 10); analysis complete; post draft finished.
• Posts 1 + 2 published — draft: true removed from Stockton and Diaw posts; jupyter: blog added to frontmatter; _freeze/ outputs generated locally and committed; CI will serve frozen outputs without re-executing Python.
• CI fixed — kernel registration (python -m ipykernel install --user --name blog) added to publish.yml; last green run 2026-10-09.
• Population fetch running locally — PID 39001; ~109 population players fetched so far; ETA ~Oct 14.
• 141 tests passing (was 131 after Oct 10 session).

v2 — 2026-10-10 (updated same day)

• BRef ID fixes: Bogues (boguemu01), Harper (harpede01), Strickland (stricro02) — PG cohort grew from 12 → 15 players
• E1 + E3 analysis complete for both PG (15 players) and PF (6 players)
• Stockton post drafted with E1/E3 results; E2/E4/E5 sections stubbed pending figures
• Diaw post drafted with E1/E3 results; E2/E4/E5 sections stubbed
• Quarter splits fetched for Payton, Jackson, Harper, Stockton (5,585 rows)
• E2/E4/E5 analysis run for PG cohort; E2 interpretable (Stockton Q4 FT% +4.1 pp, only improver, p=0.494 at floor); E4/E5 dropped — degenerate at per-quarter resolution (52% zero-TOV rows; 21% TS% coverage)
• Stockton post now complete draft (E1/E3/E2 written; E4/E5 explained as methodologically degenerate)
• Population fetch scope expanded to include centers; running on Apollo (232/1,995 players, 99,883 rows)
• Population cohort size: 1,995 players (guards + forwards + centers, debuts 1980–2010)
• Fixed 5 bugs in fetch_quarter_splits.py: Age column KeyError, rate limit (0.35s→3.5s+jitter), 429 backoff, game_id Tm→Team column, season range (1996→1997)
• Fixed parallel --player race condition in fetch_bref_data.py with fcntl.flock()

v1 — 2026-10-09

• Initial status document; infrastructure complete (131 tests); data fetch incomplete