SKIP TO CONTENT
ALBERTA REFERENDUM 2026

THE ANALYSIS LAYER2021 CENSUS · 2026 REFERENDUM · 87 DIVISIONS

The Portrait

Who Alberta is, division by division, and where the count rose and fell. A specimen plate, not a scoreboard.

GENERATED
October 20, 2026
DIVISIONS
87
POLLS AT CAPTURE
91.2%
ANALYSIS N
86
CAUSAL CLAIMS
NONE

PARTIAL COUNT5,089 / 5,583 POLLS (91.15%) · 43/87 DIVISIONS FULLY REPORTED · SNAPSHOT 2026-10-21 01:31Z

00KEY FINDINGS

Four things the divisions agree on

Descriptive associations from the analysis file. Each card carries its coefficient, sample size and the multiple-testing adjustment; none is a causal claim.

  1. F01ROBUST

    Agriculture and forestry industry has the strongest negative association with remain share among the non-sensitive portrait variables

    Pearson r = -0.8076 (p = 5.83e-21, BH q = 1.36e-18), Spearman rho = -0.7241, n = 86; fully-reported subset r = -0.8057 (n = 43), direction stable in the fully-reported subset: yes; magnitude stable: yes.

    Association across divisions only; 2021 Census portrait vs 2026 unofficial partial count; no causal interpretation.

    n = 86PEARSON + SPEARMANBH Q

  2. F02ROBUST

    Lone-parent families has the strongest negative association with turnout (population 18+) among the non-sensitive portrait variables

    Pearson r = -0.4985 (p = 1e-06, BH q = 8e-06), Spearman rho = -0.5464, n = 86; fully-reported subset r = -0.5461 (n = 43), direction stable in the fully-reported subset: yes; magnitude stable: yes.

    Association across divisions only; 2021 Census portrait vs 2026 unofficial partial count; no causal interpretation.

    n = 86PEARSON + SPEARMANBH Q

  3. F03FRAGILE

    Some associations change in the fully-reported subset of divisions

    Partial-results sensitivity across the fully-reported divisions: 1 association(s) flip sign with |r| >= 0.2 in the full set (for example Indigenous identity x Turnout (population 18+): r all = -0.2339, r complete = 0.1885); 15 association(s) keep their sign but shift by more than 0.20 (largest: Median individual income x Turnout (population 18+): r all = 0.1939, r complete = 0.7131).

    Treat these as reporting-sensitive until the feed completes and the analysis is rerun.

    n = 86PEARSON + SPEARMANBH Q

  4. F04SUGGESTIVE

    A 4-predictor demographic model co-varies with remain share (R2 = 0.911)

    OLS on z-scored variables, n = 86, R2 = 0.911, adj R2 = 0.9066, F p = 1.08e-41; predictors: pctAgricultureForestry, pctMiningOilGas, pctBachelorPlus, loneParentPct; all VIF <= 5.

    Exploratory selection with a stated rule; sensitive variables excluded; no causal interpretation.

    n = 86PEARSON + SPEARMANBH Q

The 15 strongest 2023 UCP areas

Q10 and Q5 voting methods, with Q5 No share and its provincial rank.

October 21, 2026 · 4:29 PM MT · Frozen unofficial, partial count · Areas, not individual voters

View comparison

01THE MAP

One plate, three readings

One plate, three colour modes: bivariate terciles, cluster membership, and the 2023 winning party. Hover, tap or keyboard-walk a division; the exact values are in the card, the tooltip and the table.

MAP MODE
OUTCOME (LIGHTNESS)

Map colour encodes demographics (hue) and outcome (lightness) — it does not use the live desk’s remain/referendum blue/gold sides.

PARTIAL COUNT5,089 / 5,583 POLLS (91.15%)

EdmontonCalgaryRed DeerLethbridgeGrande PrairieFort McMurrayMedicine Hat

HOVER / TAP A DIVISIONARROWS MOVE · ENTER PINS · ESC CLEARS

02THE SCATTER

Where the divisions sit

One point per division, labelled outliers only. The line is a least-squares fit and is descriptive only; the dashed rules are the median split.

Y · OUTCOME

REMAIN

↑ 25-44 →

CREAM DASH = LEAST-SQUARES LINE, DESCRIPTIVE ONLYGOLD DASH = MEDIAN SPLIT

TAB / HOVER A POINT

03THE EVIDENCE

What the coefficients say

Pearson and Spearman with two-sided p and a Benjamini–Hochberg q across the analysis family. The filter keeps only the analysis file’s reliable cards.

OUTCOME
234 TESTS · BH Q · n = 86

PARTIAL COUNT5,089 / 5,583 POLLS (91.15%) · 43/87 DIVISIONS FULLY REPORTED · SNAPSHOT 2026-10-21 01:31Z

  1. 01 · LABOUR

    Agriculture and forestry industry

    −0.81

    PEARSON R

    SPEARMAN
    −0.72
    Q
    < 0.0001
    N
    86
    DIRECTION
    NEGATIVE
    RELIABLEPEARSON + SPEARMAN · BH Q

  2. 02 · IMMIGRATION

    Immigrant population

    0.74

    PEARSON R

    SPEARMAN
    0.77
    Q
    < 0.0001
    N
    86
    DIRECTION
    POSITIVE
    RELIABLEDESCRIPTIVE ONLYPEARSON + SPEARMAN · BH Q

  3. 03 · EDUCATION

    Bachelor degree or higher

    0.72

    PEARSON R

    SPEARMAN
    0.76
    Q
    < 0.0001
    N
    86
    DIRECTION
    POSITIVE
    RELIABLEPEARSON + SPEARMAN · BH Q

  4. 04 · ETHNICITY

    Visible minority population

    0.71

    PEARSON R

    SPEARMAN
    0.79
    Q
    < 0.0001
    N
    86
    DIRECTION
    POSITIVE
    RELIABLEDESCRIPTIVE ONLYPEARSON + SPEARMAN · BH Q

  5. 05 · EDUCATION

    High-school diploma

    −0.64

    PEARSON R

    SPEARMAN
    −0.63
    Q
    < 0.0001
    N
    86
    DIRECTION
    NEGATIVE
    RELIABLEPEARSON + SPEARMAN · BH Q

  6. 06 · RELIGION

    Muslim

    0.62

    PEARSON R

    SPEARMAN
    0.79
    Q
    < 0.0001
    N
    86
    DIRECTION
    POSITIVE
    RELIABLEDESCRIPTIVE ONLYPEARSON + SPEARMAN · BH Q

  7. 07 · RELIGION

    Christian (total)

    −0.60

    PEARSON R

    SPEARMAN
    −0.66
    Q
    < 0.0001
    N
    86
    DIRECTION
    NEGATIVE
    RELIABLEDESCRIPTIVE ONLYPEARSON + SPEARMAN · BH Q

  8. 08 · EDUCATION

    Postsecondary certificate, diploma or degree

    0.60

    PEARSON R

    SPEARMAN
    0.56
    Q
    < 0.0001
    N
    86
    DIRECTION
    POSITIVE
    RELIABLEPEARSON + SPEARMAN · BH Q

  9. 09 · IMMIGRATION

    Canadian citizens aged 18+

    −0.54

    PEARSON R

    SPEARMAN
    −0.61
    Q
    < 0.0001
    N
    86
    DIRECTION
    NEGATIVE
    RELIABLEDESCRIPTIVE ONLYPEARSON + SPEARMAN · BH Q

  10. 10 · LABOUR

    Unemployment rate

    0.53

    PEARSON R

    SPEARMAN
    0.56
    Q
    < 0.0001
    N
    86
    DIRECTION
    POSITIVE
    RELIABLEPEARSON + SPEARMAN · BH Q

Method: Pearson and Spearman coefficients per division pair, two-sided p-values and Benjamini–Hochberg q across the 234 demographic × outcome tests in the analysis family. “Reliable” is the analysis file’s own flag (q-adjusted significance and sample size); the coefficient stays descriptive — a large r does not describe individuals and is not a causal claim. Reading r: sign is direction, magnitude is tightness — with n = 87, |r| ≥ 0.3 is a real pattern, ≥ 0.5 strong, ≥ 0.7+ very strong (guide in METHOD → READING THE STATISTICS). Sensitive variables are marked DESCRIPTIVE ONLY.

PARTIAL-RESULTS SENSITIVITY

ASSOCIATIONS RETESTED

78

STABLE ACROSS SUBSETS

54

REPORTING-SENSITIVE

24

Partial-results sensitivity: every outcome association is also computed on the 43 divisions fully reported at the frozen snapshot (reportingCompleteness == 1; 5089/5583 polls, 91.1517%). directionStable = same sign in both subsets; magnitudeStable = |delta r| <= 0.20; stable = both.

LARGEST REPORTING-SENSITIVE ASSOCIATIONS · FULL COUNT VS 43 FULLY-REPORTED DIVISIONS
ASSOCIATIONr ALLr COMPLETEDELTA
Mining/oil-gas × Turnout 18+−0.160.48+0.63
Individual income × Turnout 18+0.190.71+0.52
Household income × Turnout 18+0.140.57+0.43
Indigenous × Turnout 18+−0.230.19+0.42
Family income × Turnout 18+0.220.62+0.40
Christian × Turnout 18+0.260.66+0.40

REGIONAL MEANS · DESCRIPTIVE CONTEXT

REGIONNREMAIN MEANTURNOUT 18+ MEAN
Calgary2881.8%61.8%
Central1259.8%65.7%
Edmonton26 of 2782.9%57.5%
North960.3%43.8%
South1167.0%66.5%

Unweighted means of the division values, for orientation only. Regions are the analysis file’s grouping, not an official census geography.

04THE SHAPE

Clusters, and what they are made of

Two clusterings: the outcome grouping is stable, the demographic one is below the analysis file’s stability floor. Profiles are mean z-scores; names appear only when supplied, and texture carries the cluster even without colour.

K = 3 · SILHOUETTE 0.26 · STABILITY 0.64

FRAGILE — STABILITY BELOW THE 0.70 FLOOR

HOVER A DIVISION TO READ ITS CLUSTER

  • Cluster 142
  • High Postsecondary / High Dwelling value / Low No certificate26
  • High Unemployment / High Lone-parent19

Cluster 1

n = 42
65+
+0.4
25-44
−0.5
Lone-parent
−0.1
Household income
−0.1
Postsecondary
−0.5
No certificate
+0.3
Unemployment
−0.5
Participation
−0.3
MEMBERS · 42
  • 11Calgary-Fish Creek
  • 13Calgary-Glenmore
  • 22Calgary-Peigan
  • 33Edmonton-Gold Bar
  • 40Edmonton-Riverview
  • 48Airdrie-East
  • 49Athabasca-Barrhead-Westlock
  • 51Bonnyville-Cold Lake-St. Paul
  • 52Brooks-Medicine Hat
  • 53Camrose
  • 54Cardston-Siksika
  • 55Central Peace-Notley
  • 56Chestermere-Strathmore
  • 57Cypress-Medicine Hat
  • 58Drayton Valley-Devon
  • 59Drumheller-Stettler
  • 60Fort McMurray-Lac La Biche
  • 62Fort Saskatchewan-Vegreville
  • 63Grande Prairie
  • 64Grande Prairie-Wapiti
  • 65Highwood
  • 66Innisfail-Sylvan Lake
  • 67Lac Ste. Anne-Parkland
  • 68Lacombe-Ponoka
  • 69Leduc-Beaumont
  • 70Lesser Slave Lake
  • 71Lethbridge-East
  • 72Lethbridge-West
  • 73Livingstone-Macleod
  • 74Maskwacis-Wetaskiwin
  • 75Morinville-St. Albert
  • 76Olds-Didsbury-Three Hills
  • 77Peace River
  • 79Red Deer-South
  • 80Rimbey-Rocky Mountain House-Sundre
  • 81Sherwood Park
  • 82Spruce Grove-Stony Plain
  • 83St. Albert
  • 84Strathcona-Sherwood Park
  • 85Taber-Warner
  • 86Vermilion-Lloydminster-Wainwright
  • 87West Yellowhead

High Postsecondary / High Dwelling value / Low No certificate

n = 26
65+
−0.7
25-44
+0.7
Lone-parent
−0.7
Household income
+0.6
Postsecondary
+1.1
No certificate
−0.9
Unemployment
−0.1
Participation
+0.9
MEMBERS · 26
  • 03Calgary-Bow
  • 04Calgary-Buffalo
  • 06Calgary-Currie
  • 08Calgary-Edgemont
  • 09Calgary-Elbow
  • 12Calgary-Foothills
  • 14Calgary-Hays
  • 16Calgary-Lougheed
  • 18Calgary-Mountain View
  • 19Calgary-North
  • 20Calgary-North East
  • 21Calgary-North West
  • 23Calgary-Shaw
  • 24Calgary-South East
  • 25Calgary-Varsity
  • 26Calgary-West
  • 31Edmonton-Ellerslie
  • 37Edmonton-Meadows
  • 42Edmonton-South
  • 43Edmonton-South West
  • 44Edmonton-Strathcona
  • 45Edmonton-West Henday
  • 46Edmonton-Whitemud
  • 47Airdrie-Cochrane
  • 50Banff-Kananaskis
  • 61Fort McMurray-Wood Buffalo

High Unemployment / High Lone-parent

n = 19
65+
−0.0
25-44
+0.2
Lone-parent
+1.2
Household income
−0.7
Postsecondary
−0.5
No certificate
+0.5
Unemployment
+1.3
Participation
−0.5
MEMBERS · 19
  • 01Calgary-Acadia
  • 02Calgary-Beddington
  • 05Calgary-Cross
  • 07Calgary-East
  • 10Calgary-Falconridge
  • 15Calgary-Klein
  • 17Calgary-Bhullar-McCall
  • 27Edmonton-Beverly-Clareview
  • 28Edmonton-Castle Downs
  • 29Edmonton-City Centre
  • 30Edmonton-Decore
  • 32Edmonton-Glenora
  • 34Edmonton-Highlands-Norwood
  • 35Edmonton-Manning
  • 36Edmonton-McClung
  • 38Edmonton-Mill Woods
  • 39Edmonton-North West
  • 41Edmonton-Rutherford
  • 78Red Deer-North

Bars are mean z-scores within the cluster: zero is exactly the provincial average, +1 one typical spread above it, −1 one below — each card is a fingerprint of its divisions (reading guide in METHOD). kmeans (k-means++, fixed seed, 100 inits for the final fit)

Read this panel as a coarse descriptive split, not as types: the bootstrap ARI is below the 0.70 floor.

VOTER TYPOLOGY & COHEN’S D EFFECT-SIZE PROFILES · POLICY YES MINUS REMAIN GAP (±15PP)

RELUCTANT 9 (≥+15pp) · ALIGNED 26 (within ±15pp) · SKEPTIC 51 (≤−15pp) · DESCRIPTIVE ONLY

Ag/forestryd = +2.54
In: 12.8%Out: 2.3%
Bachelor+d = -1.31
In: 13.7%Out: 31.8%
Immigrantd = -1.31
In: 8.0%Out: 24.1%
DESCRIPTIVE ONLY
Visible minorityd = -1.26
In: 5.3%Out: 28.9%
DESCRIPTIVE ONLY
High schoold = +1.22
In: 31.5%Out: 25.0%
Postsecondaryd = -1.16
In: 53.4%Out: 65.1%
Dwelling valued = -0.96
In: $317,111Out: $416,740
Unemploymentd = -0.90
In: 9.9%Out: 11.7%

Standard pooled-SD Cohen’s d (df-weighted) comparing divisions in the selected quadrant against all other counted divisions (n = 86; ED 34 excluded). Ecological fallacy applies — division-level contrast is not an explanation of individual behaviour.

05THE QUESTIONS

How the ten answers move together

Row and column are the same ten questions. Hover or arrow through a cell for its coefficients. A high r means the two yes shares travel together across divisions — nothing more.

Pearson correlation between the ten question yes shares across the divisions. Q10 is the remain-versus-referendum question. Use the arrow keys to inspect a cell; the readout beside the matrix prints its coefficients and sample size.
Q × QQ1Q2Q3Q4Q5Q6Q7Q8Q9Q10
Q1
Q2
Q3
Q4
Q5
Q6
Q7
Q8
Q9
Q10

Read across a row: how one question’s yes share moves with the other nine. A positive cell means the two shares tend to rise together across divisions; a negative cell means they trade off.

The diagonal is fixed at 1 and is not evidence. 45 cells, n per cell printed in the readout. A cell describes co-movement only: no mechanism is claimed between questions.

06METHOD & CAVEATS

How to read this, and what it cannot say

Sources, formulas, the partial count and the limits of an ecological association.

READING THE STATISTICSr · d · z IN THREE SHAPES

Three shapes to hold in your head

The evidence cards, cluster bars and scatter are each one of these pictures. The written definitions sit in the accordion below.

WHAT r LOOKS LIKEWEAK PATTERN

A loose cloud: the two values barely travel together.

r = −0.21

n = 86 DIVISIONS

2.3%26%94%46%

x = population aged 65+ · y = remain share
20 dots shown, one per division; the line is least squares on all 86.

WHAT r LOOKS LIKESTRONG PATTERN

A clear tilt: when one is higher, the other tends to be higher too.

r = 0.60

n = 86 DIVISIONS

39%88%94%46%

x = postsecondary certificate, diploma or degree · y = remain share
20 dots shown, one per division; the line is least squares on all 86.

WHAT r LOOKS LIKESTRONG NEGATIVE

A steep downward cloud: high on one value, low on the other.

r = −0.81

n = 86 DIVISIONS

−1.1%23%94%46%

x = agriculture and forestry industry · y = remain share
20 dots shown, one per division; the line is least squares on all 86.

WHAT d LOOKS LIKESCHEMATIC

Two groups, each about one typical spread wide. d is how far apart their centres sit, measured in those spreads.

d = 0.2 / 0.5 / 0.8

SMALL / MEDIUM / LARGE

0.2SMALL0.5MEDIUM0.8LARGE

ILLUSTRATIVE CURVES — NOT DATA · d = how many typical spreads apart the two groups sit.

WHAT z LOOKS LIKEREAL DIVISIONS

Zero is the provincial average; each step is one typical spread. The whole distribution is the tick strip; three divisions are called out.

z = −3.03 → +2.02

TURNOUT PROXY (2021 18+ DENOMINATOR) · N = 86

+2+10-1-2PROVINCIAL MEAN35 · z −3.0341 · z 0.0047 · z +2.02
  • 35 Edmonton-Manning16.00% · z −3.03
  • 41 Edmonton-Rutherford59.69% · z 0.00
  • 47 Airdrie-Cochrane88.92% · z +2.02

ONE DOT = ONE DIVISIONTHE d CURVES ARE ILLUSTRATIVENO CAUSAL CLAIM ANYWHERE

SOURCES & FETCH DATES
VINTAGE — THREE DIFFERENT DATES

Demographics: 2021 Census (published 2022-2023). Results: 2026 referendum unofficial count. Boundaries: 87 Alberta provincial electoral divisions (2019 outline; 2026 Elections Alberta names).

A 2021 census value and a 2026 vote count describe the same division five years apart. The page labels every axis with its source; it does not model population change between the two dates.

HOW THE NUMBERS WERE MADE
CORRELATION
Pearson and Spearman (tie-corrected ranks), n per cell, two-sided exact t-based p-values; Benjamini-Hochberg FDR within the exported demographic x outcome family; no causal interpretation.
REGRESSION
OLS on z-scored outcome and predictors; complete cases only; block-wise forward selection (entry p < 0.05, max 2 per block, VIF <= 5, max 8 predictors); 95% CI from the t distribution; sensitive variables excluded; exploratory, not confirmatory.
CLUSTERING
k-means with k-means++ seeding, fixed seed 20261021, 100 inits; k chosen by maximum silhouette among k=2..8 with bootstrap mean ARI >= 0.70; stability from 200 bootstrap resamples (ARI vs full-sample partition).
SPATIAL
Queen contiguity from data/geo/alberta-eds.geojson (boundary distance tolerance 1e-3 degrees), row-standardised weights, Moran's I with 9999 permutations; expected value -1/(n-1); upper-tail permutation p.
TURNOUT
Primary = ballots / population 18+ (2021 total population minus ages 0-4 and 5-17); secondary = ballots / Canadian citizens 18+ (2021, 25% sample count). Both are labelled; neither is a registered-voter turnout rate.
SENSITIVITY
Every outcome association is recomputed on the 43 divisions fully reported at the snapshot (reportingCompleteness == 1); directionStable requires the same sign and |delta r| <= 0.20.
TURNOUT PROXIES & DENOMINATORS

Turnout proxy (2021 18+ denominator) = valid ballots ÷ 2021 Census population aged 18+ × 100.

Turnout proxy (citizens 18+) = valid ballots ÷ 2021 Census citizens aged 18+ × 100.

turnout18 and turnoutCitizen18 divide 2026 ballots by the 2021 Census 18+ population. Where the population grew between 2021 and 2026 (fast-growing urban divisions) the 2021 denominator is smaller than the 2026 electorate, so the proxy is biased upward; it is not a registered-voter turnout rate. The 18+ counts are 2021 vintage and the citizens 18+ numerator is a 25% sample count.

Partial counts lower turnout mechanically; the map and scatter label the completeness of the count.

OPPORTUNITY MIX — PROVISIONAL

advance/electionDay/vbm/mobile shares use the all-ten-question valid-vote base (not Q11 ballots) and are provisional until the count completes, as poll types report at different speeds. opportunity.other (1,213 votes in ED 40) is excluded from the four-way mix.

The page marks these outcomes PROVISIONAL wherever they can be selected and keeps them out of the hero findings. Their shares are not a validated decomposition of the final count.

RELIGION — 25% SAMPLE, CONTEXT ONLY

Religion shares are 25% sample, population in private households; they are descriptive context only and are never used as regression predictors.

Religion is never used as a headline finding on this page and never as a regression predictor in the analysis file. Where it appears in a picker it carries the inline descriptive-context note.

PARTIAL RESULTS AT CAPTURE

PARTIAL COUNT — 5,089 of 5,583 polls reported (91.2%) and 43 of 87 divisions fully reported at the frozen snapshot (2026-10-21 01:31Z). Division results are not final and small margins can still move.

No unreported poll is imputed. Every coefficient in the analysis file is computed on the divisions with a counted value, and the sample size is printed with it.

ZERO-VOTE DIVISIONS
  • 34 · Edmonton-Highlands-Norwood
  • These divisions had no valid ballots counted at capture. Their outcome values are null, they are excluded from every n and every coefficient, and nothing is imputed.
ROUNDING, 25% SAMPLE & TOTAL NON-RESPONSE

Census counts are random-rounded and the sample files carry sampling error plus total non-response (TNR). Percentages in this page are as published or computed from published counts; they are not re-derived from rounded counts.

Variables from the 25% sample file are flagged in the analysis metadata; on the plate they are still descriptive context, not precise measurements.

READING THE STATISTICS — r, d AND z

r (correlation) — on the evidence cards. “Across the 87 divisions, when a division has more of X, does it also have more of Y?” Sign is direction; magnitude is tightness. With n = 87, |r| beyond about 0.21 is statistically significant; 0.3 a real pattern, 0.5 strong, 0.7+ very strong. One point per division — never per person.

d (Cohen’s d) — in the analysis record’s typology. Group mean minus everyone-else mean, divided by the pooled standard deviation: how many typical spreads apart the two groups sit. 0.2 small, 0.5 medium, 0.8 large, above 1.2 very large. Positive d: the group sits higher on that variable.

z (standard score) — on the cluster bars. Value minus provincial mean, divided by the provincial standard deviation. Zero is exactly average; +1 one spread above; −1 one spread below.

None of these numbers says why anyone voted. They describe how divisions co-vary — see the ecological-fallacy note.

ECOLOGICAL FALLACY — READ THIS BEFORE ANY COEFFICIENT

A division-level association does not describe individuals. That high-income divisions lean toward remain does not mean high-income voters lean that way; the pattern could reverse at the individual scale. Every scatter, correlation and profile on this page is ecological.

SENSITIVE VARIABLES — DESCRIPTIVE CONTEXT

Indigenous identity, visible-minority identity, religion, language and immigration are shown as descriptions of the electorate. They are never framed as explanations of vote choice, and the plate marks them with an inline note when selected.

Descriptive context — division-level association, not an explanation of how or why people voted.

WHAT THIS PAGE DOES NOT DO
  • No causal inference — associations only, and only at the division scale.
  • No individual-level inference or prediction.
  • No forecast; the count is the count at capture.
  • No imputation of missing values and no silent dropping of divisions.
  • No religion-by-imputation; only published census categories are shown.
LICENCE & ATTRIBUTION

Census data is published by the Government of Alberta; the publisher’s current terms apply and are stated on the dataset page linked above. Election results are Elections Alberta’s unofficial count. This page is not affiliated with, or an official publication of, either body.

GUARDRAILS FROM THE ANALYSIS FILE
  • Aggregate data only - no individual inference.
  • Ecological fallacy applies: division-level association does not describe individuals.
  • Demographics are the 2021 Census; results are the 2026 referendum; no causal claims.
  • Sensitive variables (Indigenous identity, ethnicity, language, immigration, religion) are descriptive context, never explanations of vote choice.
  • Random rounding and 25% sample/TNR caveats apply to census variables.
  • Results completeness caveat: 5089/5583 polls reported (91.15%) and 43/87 divisions fully reported at the frozen snapshot; numbers will change.
  • n = 87 divisions (86 with counted ballots); multiple-testing correction (BH) reported for the correlation family.
  • No imputation: missing values are null and excluded from n.

THE PORTRAIT · 87 DIVISIONS · UNOFFICIAL RESULTS · DESCRIPTIVE ONLY