Metric relationship

What "moves before" means. We look at how each measure changed, year by year, in each place compared with the city as a whole (so a citywide boom or slump doesn't count). If places where measure A rose faster than the city tend to see measure B rise faster a year or two later, A "moves before" B.

The grade adds up five checks, shown on every pair's page:

• Evidence (raises the grade): they move together within places; the first tends to move first, and that survives a correction for how many pairs we tested; the second responds after the first's sharp jumps.
• Conditions (cap it at "weak" when they fail): the link isn't explained away by a third measure that moves with both, and the two aren't tied together by how the data is built (for example both estimated from the same census boundaries).

Grades: We can't tell (they don't move together), Weak (they move together, nothing more), Suggestive (plus timing or a response to jumps), Moderate (all of it).

What it can't tell you. This is observational data from 50–300 places over 5–25 years. A strong grade means the timing and the checks line up with one measure affecting the other; it doesn't prove it. "We can't tell" is not "no effect".

Is it more than chance? Every check carries a significance test that respects shared shocks: each place's history is paired with other places' at random, so a citywide shock (COVID hit every neighborhood at once) can't pass for a link, and every count of place-years sits next to its effective N, the number of distinct years. Both directions are always shown. How every check works.

Worked example: 'L' ridership and restaurant density, graded suggestive: restaurants respond after sharp ridership changes, but ridership moving first doesn't survive the multiple-comparison correction.