Block disaggregation

How Canvas turns a drawn shape into an election estimate.

When you draw a boundary in Canvas, we reaggregate 8 million census blocks to estimate the presidential result for that exact area — even if it follows no existing political geography.

Open Canvas Margin convention

Census blocks

8M+

Across 50 states (Alaska unavailable)

Precincts 2024

163,926

NYT geometry + results

Method

Population-weighted

Same as MEDSL and VEST

What it does

One tool, three steps

Canvas estimates presidential results for any user-drawn boundary by reaggregating census-block-level data. Draw a shape, get results.

1. Disaggregate

Precinct → block

Precinct-level vote totals are distributed to the census blocks that live inside each precinct, weighted by block population.

2. Reaggregate

Block → drawn shape

When you draw, Canvas finds every block whose geometry intersects your boundary and sums their estimated votes.

3. Compute

Margin + turnout

Margin and turnout are computed from the summed block values using our standard margin convention.

Sources

Where the precinct data comes from

2024 Presidential · New York Times

163,926 precincts with full boundary geometry covering 50 of 51 states. Alaska is unavailable because precinct geometry is not published at a usable level.

2020 Presidential · VEST (Harvard Dataverse)

Precinct results with geometry covering 46 of 51 states. The same source academic researchers rely on for precinct-level analysis.

Disaggregation

How precinct votes end up on blocks

Each census block's centroid is assigned to whichever precinct contains it via spatial join. Precinct votes are then distributed to blocks proportional to each block's share of its precinct's total population — from the 2020 PL 94-171 release.

Population weighting

A precinct with 1,000 voters and three blocks housing 600, 300, and 100 residents will allocate 60% of its votes to the first block, 30% to the second, and 10% to the third. Uninhabited blocks receive zero votes. Hare quota rounding reconciles totals back to certified precinct results.

Well-established method

The same approach MIT MEDSL and VEST use

This is not a bespoke Akashic invention. Population-weighted block disaggregation is the standard academic method for moving from precinct geography to block geography.

Reaggregation

From blocks to your drawn shape

When you finish drawing, Canvas finds every census block whose geometry intersects your boundary.

Small shapes (under 5,000 blocks)

Every intersecting block is counted at full weight. Fast, simple, and accurate for neighborhood-sized analysis.

Large shapes (over 5,000 blocks)

Blocks that are only partially covered by your boundary are weighted by the fraction of their area that falls inside. This keeps the estimate proportional without exploding cost.

Margins

How margins are computed

Canvas uses the standard Akashic margin convention.

margin = (Democratic votes - Republican votes) / total votes × 100

Positive values indicate a Democratic advantage. Third-party votes are included in the denominator — see the margin convention page for why.

Lean score

Multi-cycle aggregation

When three or more cycles of block-level data are available, Canvas computes a weighted Lean Score that blends recent elections.

Weights

60% on the most recent cycle, 25% on the second most recent, 15% on the third. Emphasizes current lean while smoothing single-election noise.

When it's used

On any shape for which we have three or more cycles of disaggregated block data. Single-cycle shapes report raw margin, not Lean Score.

Turnout

How turnout is estimated

Total votes cast divided by the estimated voting-age population (VAP) within the drawn area. VAP comes from the 2020 Census PL 94-171 data at the block level.

Turnout estimates carry additional uncertainty for boundaries that contain fast-changing populations. We label confidence accordingly.

Limitations

Where the method breaks down

Canvas is useful. It is not perfect. Here are the limits.

Block-level estimates are approximations, not exact counts. Accuracy is highest where precincts are small and demographically uniform, and lowest in large, diverse precincts.
Population weighting assumes voters within a precinct are distributed proportionally to total population. It does not account for demographic variation within precincts — a block with a retirement community may turn out differently than a neighboring block of young families in the same precinct.
Alaska (FIPS 02) has no block-level data because usable precinct geometry is not available. Shapes drawn over Alaska return no results.
When blocks are reaggregated to known boundaries (counties, states), totals match within rounding tolerance. State totals are exact.

Confidence

What the confidence label means

Medium confidence is the default

All block estimates carry medium confidence. This reflects the standard quality of population-weighted disaggregation. Confidence drops for very small areas (fewer than ten blocks), where individual block estimation errors have outsized impact.

Contact

Questions about the method

Researchers comparing Canvas output to other sources can reach out for reconciliation help.

team@akashicedge.com