Graduate a single age curve onto a complete age grid

Internal single-curve workhorse called by smooth_decrement_rates() once per (group_cols, status_col) combination. Given raw (age, rate) observations for one specific outcome curve (e.g. "male pensioner rate by age"), returns a smoothed rate for every age in full_ages – including ages that had zero raw observations at all.

The method used depends on how much data is actually available, since a local regression needs enough support to be stable:

• Exactly one distinct age observed: there is no trend to fit from a single point, so that one rate is repeated flat across the whole grid.

• Two or three distinct ages: too few for a stable loess fit, so falls back to stats::approx() (piecewise linear interpolation). Ages in full_ages outside the observed range get the nearest boundary value (rule = 2) rather than NA.

• Four or more distinct ages: fits stats::loess(rate ~ age, weights = weight, span = span, degree = 2) and predicts it onto full_ages. weight (exposure) means an age with many people at risk pulls the local curve toward its raw rate harder than a thin, noisy age.

Usage

.smooth_rate_curve(age, rate, weight, full_ages, span)

Arguments

age

Numeric vector. Ages with an observed rate.

rate

Numeric vector. The observed rate at each age (same length as age).

weight

Numeric vector. Exposure weight at each age (same length as age), used as loess() weights.

full_ages

Integer vector. The complete, gapless target age grid to return a smoothed rate for.

span

Numeric. The loess() smoothing span (only used when there are at least 4 distinct ages); larger values produce a smoother, more global fit, smaller values track local features more closely.

Value

A numeric vector of smoothed rates, one per element of full_ages, in the same order. Not clipped to [0, 1] – callers (e.g. smooth_decrement_rates()) are responsible for that.

See also

smooth_decrement_rates