SpatialInference 0.1.0
Conley spatial HAC standard
errors
conley_SE() computes Conley (1999) spatial HAC
variance-covariance matrices for lfe::felm() models, with
support for cross-sectional spatial correlation, serial (temporal)
correlation, and the combined spatial HAC estimator.- Six kernel functions: Bartlett, Epanechnikov, Gaussian, Parzen,
Biweight, and Uniform.
- Haversine great-circle distances (default) and a 111 km/degree
approximation.
- Balanced-panel optimisation pre-computes the distance matrix
once.
compute_conley_lfe() convenience wrapper for quick
single-coefficient extraction.lm_sac() all-in-one workflow: regression, Moran’s I
tests, and Conley standard errors, with modelsummary
integration via custom tidy and glance
methods.
Bandwidth selection
covgm_range() estimates the spatial correlation range
from the empirical covariogram of regression residuals (Lehner
2026).extract_corr_range() extracts the zero-crossing
distance from a covariogram (gstat::variogram()) or
correlogram (ncf::correlog()).inverseu_plot_conleyrange() diagnostic plot showing how
the Conley SE varies with the bandwidth, revealing the inverse-U
relationship (Lehner 2026).
Spatial utilities
DistMat() kernel-weighted spatial distance matrix
(C++).coords_as_columns() extracts sf point
coordinates into tibble columns.gravity_centroid() computes the (optionally weighted)
geographic centroid of an sf object.grid_FE() assigns observations to spatial grid cells
for use as fixed effects.
- Distance matrix computation, kernel weighting, and variance
component accumulation (
XeeXhC, Bal_XeeXhC,
XeeXhC_Lg, TimeDist) are implemented in C++
via Rcpp and RcppArmadillo.
- Memory-efficient large-sample variant (
XeeXhC_Lg)
avoids constructing the full n x n distance matrix when n >
50,000.