Skip to contents

Local Inference for LHWR Coefficient Surfaces

Source: R/local_inference.R
local_inference.Rd

Compute exploratory local t-tests and p-values for the coefficient surfaces estimated by an LHWR model. The function uses the local coefficient estimates, local standard errors, and local t-values stored in the fitted object; it does not refit the model.

Usage

local_inference(
  object,
  alpha = 0.05,
  df = c("edf", "residual", "normal"),
  adjust = c("BH", "none", "holm", "bonferroni", "BY", "fdr"),
  adjust_scope = c("coefficient", "global"),
  ...
)

local_inference.hgwr(
  object,
  alpha = 0.05,
  df = c("edf", "residual", "normal"),
  adjust = c("BH", "none", "holm", "bonferroni", "BY", "fdr"),
  adjust_scope = c("coefficient", "global"),
  ...
)

Arguments

object

An LHWR model object returned by lhwr().

alpha

Significance level used to create the significant flag.

df

Degrees-of-freedom rule. "edf" uses the model effective degrees of freedom stored in object$LHWR.diagnostic$edf; "residual" uses n - tr(S); "normal" uses a standard normal reference distribution.

adjust

Multiplicity adjustment passed to stats::p.adjust(). Use "none" for raw p-values.

adjust_scope

Adjustment scope. "coefficient" adjusts p-values separately within each coefficient surface; "global" adjusts over all local coefficient tests at once.

...

Unused.

Value

An object of class "lhwr_local_inference" with components: local, a long data frame of local estimates, standard errors, t-values, raw p-values, adjusted p-values, and significance flags; summary, a coefficient-level summary table; and settings.

Details

These p-values should be interpreted as local exploratory diagnostics. Because LHWR produces many local tests over dependent neighborhoods, multiplicity and local dependence matter. Use adjust = "BH", "holm", or "bonferroni" when the goal is to identify robust local patterns.

Inference is conditional on the Lorentz-model coordinates and neighborhood weights stored in object. If the representation was estimated, the reported standard errors and p-values do not include first-stage representation uncertainty or uncertainty from bandwidth selection.

Examples

set.seed(5)
H <- simulate_H2(80, r_max = 3)
Z <- H$Z
x1 <- rnorm(80)
y <- 1 + 2 * x1 + rnorm(80, sd = 0.3)
dat <- data.frame(y, x1)
fit <- lhwr(y ~ x1, dat, Z = Z, bw = 35, adaptive = TRUE, diagnostics = "full")
inf <- local_inference(fit, adjust = "BH")
inf
head(inf$local)