the shape of the vote

SCOTUSstats

Each case is scored per justice: +1 in the majority or a concurrence, −1 in dissent, 0 for no part or a concur-in-part/dissent-in-part. With justices as variables and cases as observations, a plain covariance PCA(no scaling, no regularization) recovers the Court's voting axes.

← the full board

All time (current court) · 250 cases · 137 dividedprovisional — includes Oyez data for the in-progress term (incomplete until the Supreme Court Database releases); numbers will shift

loadings — PC1 (44%) × PC2 (24%)

RobertsThomasAlitoSotomayorKaganGorsuchKavanaughBarrettJacksonPC1 (ideology) →↑ PC2

loadings — PC1 · PC2 · PC3 (3-D)

PC1 44%PC2 24%PC3 9%KaganSotomayorJacksonKavanaughRobertsBarrettAlitoThomasGorsuchdrag to rotate

ideological ranking (PC1 loading, liberal → conservative)

Sotomayor -0.53Jackson -0.52Kagan -0.49Roberts 0.03Kavanaugh 0.06Barrett 0.07Gorsuch 0.16Thomas 0.30Alito 0.30

frequency in the majority — share of participated cases in the winning coalition

Roberts
95.6% · 249
Kavanaugh
94.4% · 249
Barrett
92.0% · 249
Gorsuch
81.3% · 246
Alito
81.2% · 245
Thomas
79.8% · 248
Kagan
77.0% · 248
Sotomayor
75.5% · 249
Jackson
73.6% · 246

How often each justice landed with the Court, among the cases they took part in. Toggle the denominator: all cases (unanimous included), only non-unanimous decisions, or just the closely dividedones (6–3 / 5–4). The count after each bar is that justice's denominator. The Chief and the median conservatives sit at the top; the further left a justice, the more the number falls as the denominator narrows to contested cases.

agreement matrix — % of shared cases on the same side (ordered by ideology)

So
Ja
Ka
Ro
Ba
Kv
Go
Al
Th
So
93
94
72
71
71
63
57
57
Ja
93
89
70
67
70
62
57
55
Ka
94
89
76
74
73
65
59
59
Ro
72
70
76
89
94
79
82
79
Ba
71
67
74
89
90
80
83
84
Kv
71
70
73
94
90
77
84
80
Go
63
62
65
79
80
77
83
82
Al
57
57
59
82
83
84
83
91
Th
57
55
59
79
84
80
82
91

Share of cases (in which both participated) where the two justices landed on the same side — both in the majority or both in dissent. Brighter = more agreement; the ideological blocs show up as bright squares.

pivotal power — empirical Banzhaf index (28 closely decided cases)

Roberts
15.9% · 22 cases
Kavanaugh
13.8% · 19 cases
Sotomayor
13.0% · 18 cases
Jackson
13.0% · 18 cases
Kagan
10.9% · 15 cases
Gorsuch
9.4% · 13 cases
Barrett
8.7% · 12 cases
Alito
8.0% · 11 cases
Thomas
7.2% · 10 cases

Among decisions won by a margin of two or fewer votes, how often a justice was a swing vote— one whose flip would have reversed the result. Bars show each justice's share of all such pivotal instances (an observed-coalition Banzhaf index); the count is their raw number of pivotal cases. Only majority-side votes can be pivotal, so a justice who dissented through all the term's knife-edge cases scores zero. This is the same vote-flip signal behind “most / least decisive,” spread across the whole bench.

most maverick (farthest from center, PC1 removed)
Gorsuch — stands apart from the pack on the secondary axes (PC-importance-weighted distance 0.279, ignoring left/right)
variance explained
PC1 44.2% · PC1+PC2 68.4%
effective # of components (participation ratio)
3.7 of 9 — the term votes like ~3.7 independent factors
splits on party lines
29% of divided cases
splits on any ideological line
55% of divided cases
unanimous (no dissent)
45% of decisions
most redundant (vote-flip changes fewest)
Thomas — flipping their vote changes 10 outcomes
least redundant (most decisive)
Roberts — flipping their vote changes 22 outcomes
twins (closest by maverick distance)
Roberts & Kavanaugh (d 0.224)
most unexpected lineup (largest Mahalanobis distance in PC space)
T. M. v. University of Maryland Medical System Corporation · 5–4 · decided 2026-06-18
Sotomayor +Jackson +Kagan −Roberts −Kavanaugh +Barrett −Gorsuch −Thomas +Alito +
minimum set to infer the winner — 7 justices · a simple majority vote

Take Roberts as the reference; each other justice is coded by whether they voted the same side or the opposite, and the winner is Roberts's side or the opposite. Knowing just these 7 determines who won in every case — and it reduces to a plain majority vote of the set (excludes 1 case where Roberts recused).

Roberts (ref)ThomasSotomayorGorsuchKavanaughBarrettJackson→ winner
sameoppoppsamesameoppsameRoberts’s side
sameoppoppsamesamesameoppRoberts’s side
sameoppsameoppoppsamesameRoberts’s side
sameoppsameoppsameoppsameRoberts’s side
sameoppsameoppsamesamesameRoberts’s side
sameoppsame—samesamesameRoberts’s side
sameoppsamesameoppoppsameRoberts’s side
sameoppsamesameoppsamesameRoberts’s side
sameoppsamesamesameoppsameRoberts’s side
sameoppsamesamesamesameoppRoberts’s side
sameoppsamesamesamesamesameRoberts’s side
same—opp—samesame—Roberts’s side
samesameoppoppsamesameoppRoberts’s side
samesameoppoppsamesamesameRoberts’s side
samesameoppsamesameoppoppRoberts’s side
samesameoppsamesameoppsameRoberts’s side
samesameoppsamesamesameoppRoberts’s side
samesameoppsamesamesame—Roberts’s side
samesamesameoppoppsamesameRoberts’s side
samesamesameoppsamesameoppRoberts’s side
samesamesameoppsamesamesameRoberts’s side
samesamesame—samesamesameRoberts’s side
samesamesamesameoppsameoppRoberts’s side
samesamesamesameoppsamesameRoberts’s side
samesamesamesamesamesameoppRoberts’s side
samesamesamesamesamesamesameRoberts’s side
sameoppoppoppoppoppoppopposite
sameoppoppoppsameoppoppopposite
sameoppoppoppsameoppsameopposite
sameoppoppoppsamesameoppopposite
sameoppoppsameoppoppoppopposite
sameoppoppsameoppsameoppopposite
sameoppsameoppsameoppoppopposite
samesameoppoppoppoppoppopposite
samesameoppoppsameoppoppopposite
samesameoppsameoppoppoppopposite

scree — variance per PC (all time)

44%
PC1
24%
PC2
9%
PC3
6%
PC4
6%
PC5
4%
PC6
3%
PC7
2%
PC8
2%
PC9

variance explained — PC1 (solid) · PC1+PC2 (light)

All
44% · 68%
OT2022
35% · 61%
OT2023
47% · 70%
OT2024
41% · 67%
OT2025
53% · 76%

effective # of components over time (participation ratio)

all-time 3.74.5OT20223.4OT20233.8OT20242.9OT2025

Effective components = 1 ÷ Σ(variance shareᵢ²) — the participation ratio (reciprocal of the Herfindahl index of variance concentration). It reads voting dimensionality directly: near 1 is a one-factor Court (everything rides the ideological axis), up to 9when variance is spread evenly across independent blocs. So a term here “behaves like ~N independent factors.”

It weights every PC by its size, so it isn't hostage to any single noisy eigenvalue — a steady read on how many dimensions the Court's divisions really span. Rising = more independent voting blocs; falling = collapsing toward a single axis.

coalition sizes by term — how the votes split

All
112
20
28
25
37
27
OT2022
29
5
7
6
4
7
OT2023
26
4
11
10
5
OT2024
28
4
13
4
6
8
OT2025
29
8
17
7
9–0 8–1 7–2 6–3 cross 6–3 party-line 5–4 tie

Every decided case placed by the size of its dissent, so a term reads as one stacked bar. The two 6–3 shades are the story of this Court: a party-line 6–3 (the six Republican appointees over the three Democratic ones, or vice-versa) versus a cross-cutting 6–3where at least one justice breaks from their bloc. The party-line band swelling over time is the empirical face of the “ideological split” the StatPack flags.

Bucketed by number of dissents (recusals fold into the nearest split, so an 8-vote 6–2 counts as 7–2); ties and equally-divided cases fall in tie.

ideological drift — each justice's PC1 loading across terms

OT2022OT2023OT2024OT2025RobertsThomasAlitoSotomayorKaganGorsuchKavanaughBarrettJackson↑ conservative · liberal below

PC1 is the Court's main left–right axis. Tracing each justice's loading on it term by term shows who is drifting and who holds station. Each justice keeps their own colour — running cool (liberal) to warm(conservative) — so a line stays traceable through the crowded conservative cluster. Each term's PC1 is its own unit eigenvector, oriented so a conservative vote loads positive, so the vertical scale is comparable across terms.

Lines near the center are the swing justices; a justice crossing another signals a term where their voting patterns converged. Watch the spread between the blocs widen or narrow — a compact read on polarization that complements the participation-ratio line above.