How we read the data

Two things moving together is not a reason

When two trends line up on a chart, it's tempting to read one as causing the other. That temptation is almost always wrong. Correlation tells you that a relationship exists — it says nothing about direction, mechanism, or whether a third variable is driving both. Every story on this site that shows two datasets side by side is showing a pattern, not an explanation. We'll note when the research literature has something stronger to say; otherwise, we'll let the pattern be a pattern.

The number can be right and still mislead

Sometimes the data says exactly what it says — and that's still not enough to draw conclusions. A region with twice the rate of something might have twice the underlying condition, or twice the civic engagement, or simply a demographic skew that explains the difference entirely. Context doesn't undermine the data; it's what makes the data mean anything. When a finding is striking, that's when we slow down: what else would we need to know before this number tells a complete story?

What gets counted is rarely what we care about

Every dataset is a record of something that was easy to record — which is almost never the thing you actually want to study. Government data, commercial data, sensor data: all of it measures an instrument, and the instrument drifts from the underlying reality in ways that matter. We try to be explicit about what each dataset actually records, and careful about the distance between that instrument and the broader phenomenon we're exploring.

Absence of data is not absence of the thing

Data gets generated by people and systems with their own distributions, incentives, and blind spots. A dataset with no records from a region doesn't tell you that region is empty — it may tell you that the region wasn't surveyed, couldn't afford to participate, or had no reason to report. Gaps in data are themselves data, and the places with the fewest records are often the places that need the most attention.

What we won't claim

Some inferences are tempting and the data could technically support them — but honesty requires more than technical possibility.

• We won't draw causal conclusions from correlational patterns without external support from the research literature.
• We won't treat a high rate of recorded events as a proxy for need or severity without accounting for who does and doesn't generate records.
• We won't compare across categories or datasets when the underlying definitions make the comparison unfair.
• We won't present a pattern as an explanation. Patterns are starting points for questions, not answers.