What sample a reading belongs to. Notes and expressions are taken
apart by default, since in the present material they scatter
differently.
That split is a stopgap and no distinction in the model. What it
stands in for is a skew: the offset between two copies runs across
the width of the paper, and the expression punches sit at the two
margins, where the gradient is largest and of opposite sign, while
the notes sit between them. Grouped by where the punch sits rather
than by what it says, the expression punches are no noisier than the
notes. Once the alignment carries a skew term, one sample will do,
which is why this is a parameter of the estimator and not a field
anywhere.
What sample a reading belongs to. Notes and expressions are taken apart by default, since in the present material they scatter differently.
That split is a stopgap and no distinction in the model. What it stands in for is a skew: the offset between two copies runs across the width of the paper, and the expression punches sit at the two margins, where the gradient is largest and of opposite sign, while the notes sit between them. Grouped by where the punch sits rather than by what it says, the expression punches are no noisier than the notes. Once the alignment carries a skew term, one sample will do, which is why this is a parameter of the estimator and not a field anywhere.