Skip to contents

Compute ENR points (Excess Normalized Residuals over a LOESS trend)

Source: R/enr.R
computeENR.Rd

Trait-agnostic residual standardization for any y vs x relationship. Fit a LOESS mean trend ŷ(x), compute residuals r = y - ŷ, and standardize them to unit-free, z-like scores ENRz. Positive ENRz = above-trend; negative = below-trend.

Usage

computeENR(
  tb,
  span = 0.7,
  family = "symmetric",
  degree = 1,
  use_weights = TRUE,
  std_mode = c("robust_global", "global_sd", "local_loess"),
  span_scale = 0.7
)

Arguments

tb

A data.frame/tibble with at least columns x and y.

span

LOESS span for the mean trend (default 0.7).

family

LOESS family; "symmetric" is robust to outliers (default).

degree

Local polynomial degree for LOESS (1 = local lines).

use_weights

If TRUE, use column n as LOESS weights when present.

std_mode

One of "robust_global", "global_sd", or "local_loess".

  • robust_global: global MAD*1.4826 (robust σ)

  • global_sd: global standard deviation

  • local_loess: LOESS on abs(resid) to model local σ

span_scale

LOESS span used only for the local_loess sigma fit.

Value

tb with added columns:

  • y_hat: LOESS fitted value at each x

  • resid: raw residual y - y_hat

  • ENRz: standardized residual (z-like score)

Examples

# out <- MRCAs(df, focal) |> computeENR()
# head(out)