Reference

This page documents the public spur-stata command surface.

Overview

Command	Description
spur	Run the full SPUR workflow and return diagnostics plus levels and transformed branches
spurtest	Run one of the four SPUR diagnostic tests
spurtransform	Transform variables to remove spatial unit roots
spurhalflife	Estimate a spatial half-life confidence interval

Conventions

Coordinates

All commands in the package expect the spatial coordinates to be stored in variables named s_*.

• with latlong, s_1 is latitude and s_2 is longitude
• without latlong, the available s_* variables are treated as Euclidean coordinates

Dependencies

• moremata is required by the SPUR commands
• scpc is a separate Stata package and handles the inference stage after the branch has been chosen

Simulation Controls

The SPUR tests and half-life interval are simulation-based.

• q(#): number of low-frequency weighted averages; default q(15)
• nrep(#): number of Monte Carlo draws; default nrep(100000)
• seed(#): random seed for the simulation draws; default seed(42)

Full Workflow

spur

Run the full SPUR workflow for a linear regression.

Syntax

spur depvar indepvars [if] [in] [, q(#) nrep(#) seed(#) latlong avc(#) uncond cvs replace]

Behavior

• runs spurtest i0 and spurtest i1 on the dependent variable
• runs spurtest i0resid and spurtest i1resid for the regression
• estimates the levels regression with regress ..., robust
• runs the external scpc postestimation command
• transforms the dependent and independent variables with spurtransform, using prefix(h_) and transformation(lbmgls)
• estimates the transformed regression with regress ..., robust
• runs scpc again

The wrapper always computes both the levels and transformed branches. It does not encode the practitioner-guide branch decision as a returned field.

Options

• q(#), nrep(#), seed(#), latlong: passed to the SPUR diagnostic tests
• avc(#), uncond, cvs: passed to scpc
• replace: allow overwriting existing transformed h_* variables

Stored results

spur stores the following main results in r():

• r(diagnostics): rows i0, i1, i0resid, i1resid; columns teststat, p, ha_param
• r(levels_scpcstats): SCPC inference table from the levels regression
• r(transformed_scpcstats): SCPC inference table from the transformed regression
• r(levels_scpccvs) and r(transformed_scpccvs): critical-value matrices when cvs is specified
• diagnostic scalar aliases such as r(i0_p) and r(i1resid_teststat)
• model-statistic scalars such as r(levels_N) and r(transformed_r2)

The wrapper is OLS-only. IV and fixed-effect workflows should be estimated directly and followed by scpc.

Diagnostics

spurtest

Run one of the four SPUR tests.

Syntax

spurtest i1 varname [if] [in] [, q(#) nrep(#) seed(#) latlong]
spurtest i0 varname [if] [in] [, q(#) nrep(#) seed(#) latlong]
spurtest i1resid depvar [indepvars] [if] [in] [, q(#) nrep(#) seed(#) latlong]
spurtest i0resid depvar [indepvars] [if] [in] [, q(#) nrep(#) seed(#) latlong]

Subcommands

• i0: variable-level I(0) test
• i1: variable-level I(1) test
• i0resid: residual-based I(0) test for a linear regression
• i1resid: residual-based I(1) test for a linear regression

Options

• q(#): number of weighted averages used in the test statistic
• nrep(#): number of Monte Carlo draws used to simulate the reference distribution
• seed(#): random seed for the simulation draws; default seed(42)
• latlong: interpret s_1 and s_2 as latitude and longitude

Stored results

spurtest stores the following in r():

• r(teststat): LFUR or LFST test statistic
• r(p): p-value
• r(ha_param): alternative-hypothesis parameter (c_a or g_a)
• r(cv): matrix of critical values at 1%, 5%, and 10%

The residual tests automatically include a constant in the underlying regression.

Transformation

spurtransform

Transform one or more variables and write the transformed versions back into the current dataset.

Syntax

spurtransform varlist [if] [in], prefix(string) ///
    [transformation(string) radius(#) clustvar(varname) latlong replace separately]

Required option

• prefix(string): prefix applied to the transformed variables, for example prefix(h_) to create h_am

Options

• transformation(string): transformation mode
• lbmgls: GLS-style transformation; default branch
• nn: nearest-neighbor differencing
• iso: isotropic differencing within a radius
• cluster: within-cluster demeaning
• radius(#): radius for transformation(iso); required there and not allowed otherwise
• clustvar(varname): cluster variable for transformation(cluster); required there and not allowed otherwise
• latlong: interpret s_1 and s_2 as latitude and longitude
• replace: allow overwriting previously created transformed variables
• separately: construct the transformation separately for each variable rather than on the common non-missing sample

Behavior

• the command transforms every variable in varlist
• transformed variables are written back into the active dataset
• there are no stored results; the output is the newly created variables

Persistence

spurhalflife

Estimate a confidence interval for the spatial half-life of a variable.

Syntax

spurhalflife varname [if] [in] [, q(#) nrep(#) seed(#) level(#) latlong normdist]

Options

• q(#): number of weighted averages; default q(15)
• nrep(#): number of Monte Carlo draws; default nrep(100000)
• seed(#): random seed for the simulation draws; default seed(42)
• level(#): confidence level in percent; default level(95)
• latlong: interpret s_1 and s_2 as latitude and longitude
• normdist: report the interval as a fraction of the maximum pairwise distance rather than in distance units

Stored results

spurhalflife stores the following in r():

• r(ci_l): lower interval bound
• r(ci_u): upper interval bound
• r(max_dist): maximum pairwise distance in the sample

If normdist is not specified, the interval is reported in metres under latlong, or in the units of the Euclidean coordinates otherwise.

Workflow Note

Use spur for the full OLS pipeline in one command. Use spurtest, spurtransform, spurhalflife, and direct scpc calls when you want only selected steps or when you need IV / fixed-effect estimation.