Correlation

: each council candidate against each measure

Bigger dot = stronger pattern. Green: the measure did better where the candidate did better. Purple: it did worse. Gray rings can't be told apart from zero at precincts. Candidates whose precincts looked alike are next to each other. Hover or tap a cell for details; click to pin it and see its precincts below.

and Measure , precinct by precinct

. Each dot is a precinct, sized by ballots cast. ; dropping any one precinct moves it between and .

Most measures line up with the council divide. The interesting ones don't.

Each row is one city measure. Right of zero: the measure did better in precincts that leaned toward the candidates at the top of that year's matrix; left of zero, toward those at the bottom. Bars are 95% ranges; hollow dots can't be told apart from zero.

Which statistic?

• Rank (Spearman), the default, asks whether precincts that rank higher for the candidate also rank higher on the measure. It isn't thrown off by one unusual precinct or by a curved relationship, which matters with only 15–54 precincts. It counts every precinct equally, so a 60-ballot precinct weighs as much as a 4,000-ballot one.
• Weighted by ballots (Pearson) measures a straight-line relationship and gives big precincts more say. A couple of large precincts at the extremes can pull it up, which is partly why it often looks stronger.
• Neither is "more correct". When the two agree, and the pattern survives dropping any single precinct (shown on hover), it's solid. When they disagree, treat the pattern as unsettled.
• With 15–18 precincts (2024 and 2026), a correlation has to be roughly ±0.5 or stronger before it can be told apart from chance.