Skip to content

The global baseline

Description

The seventh component of effector's report: section 3 of the HTML page. The split features' global curves, kept at the end as the honest before picture.

Reading time

Approx. 2' to read.

What you see

For every feature whose split was accepted, section 3 repeats the global mean effect: the curve you would have read if you had never split anything.

hr, the global baseline

hr's global effect. The ICE cloud around the evening peak is the spread the accepted split just explained; in this counterfactual it still hides inside the band.

The chips here say global importance and global heterogeneity: the before numbers (0.48 for hr), not the snapshot means of the ranked features (0.29). The section only exists when at least one split was accepted; a purely additive model has nothing to compare against.

Why keep the before picture

So the report can be checked, not just trusted. Put hr here against its subregions in the regional analysis: the global curve is not wrong, it is the correct average; the baseline shows what that average was hiding, and the ledger prices the difference (+15.5%). It is also the explanation you would have shipped with a purely global method; keeping it visible is what makes the regional claim falsifiable.

Where to next