Getting Started

Installation

using Pkg
Pkg.add("SpatialEconometrics")

Basic Workflow

A typical spatial econometrics workflow follows these steps:

1. Construct a spatial weight matrix

using SpatialEconometrics

# From a regular grid
W = make_rook_weights(20)   # 20×20 grid → 400 observations
W = make_queen_weights(20)  # includes diagonal neighbors

# From coordinates
coords = hcat(lon, lat)
W = knn_weights(coords, 5)                          # 5 nearest neighbors
W = distance_weights(coords, 100.0; decay=:inverse)  # inverse distance, 100km cutoff
W = band_weights(coords, 10.0, 50.0)                # binary, 10-50km band

# Inspect
sc = spatial_connectivity(W)

2. Test for spatial dependence

# Moran's I on raw data or OLS residuals
mr = moran_global(y, W)

# LM tests to choose between lag and error specifications
lm = lm_tests(y, hcat(ones(length(y)), X), W)

3. Estimate a spatial model

# Cross-sectional SAR
m_sar = fit_spatial(SAR(), W, X, y)

# Cross-sectional SDM (Spatial Durbin)
m_sdm = fit_spatial(SDM(), W, X, y)

# Spatial Error Model
m_sem = fit_spatial_error(W, X, y)

4. Compare and select models

# Information criteria
spatial_aic(m_sar)
spatial_bic(m_sdm)

# Likelihood ratio test (SAR nested in SDM)
lr = lr_test(m_sar, m_sdm)

# Formatted comparison table
println(model_comparison_table(m_sar, m_sdm, m_sem; names=["SAR","SDM","SEM"]))

5. Compute spatial effects

effects = spatial_effects(m_sdm)
# effects.direct, effects.indirect, effects.total

Specification Tags

Model variants, fixed-effects specs and log-determinant methods are single shared types, exported directly. You never qualify them with a module name:

fit_spatial(SAR(), W, X, y)                                  # SAR, SDM
fit_spatial_panel(SDM(), W, X, y, N, T; fe = TwoWayFE())     # NoFE, IndivFE, TimeFE, TwoWayFE
fit_spatial(SAR(), W, X, y; method = ChebyshevApprox(order = 8))  # ExactLU, ChebyshevApprox

Submodules

The submodules organise the source and are not exported, so they cannot collide with registered packages that share a name. Reach one explicitly if you need an internal helper:

τ = SpatialEconometrics.CrossSection.precompute_chebyshev_traces(W,
        ChebyshevApprox(order = 12))