This package should be used as a backend by package developers. It allows developers to add a ::CovarianceEstimator argument in the fit method defined by their package. See FixedEffectModels for an example.

Each type defined in this package defines the following methods:

# return a vector indicating non-missing observations for standard errors
completecases(table, ::CovarianceEstimator) = trues(length(Tables.rows(table)))
# materialize a CovarianceEstimator by using the data needed to compute the standard errors
materialize(table, v::CovarianceEstimator) = v
# return the "meat" of the sandwich estimator
S_hat(x::RegressionModel, ::CovarianceEstimator)
# return variance-covariance matrix
vcov(x::RegressionModel, ::CovarianceEstimator)
# return degrees of freedom for t/F statistics
dof_residual(x::RegressionModel, ::CovarianceEstimator) = dof_residual(x)

Covariance Estimators

• Vcov.simple() — iid errors (classical OLS variance)
• Vcov.robust() — heteroskedasticity-robust (HC1, default)
• Vcov.hc(1), Vcov.hc(2), Vcov.hc(3) — explicit HC variant:
  • HC1: White's estimator with small-sample correction N/(N-K)
  • HC2: residuals scaled by 1/√(1 - hᵢᵢ), where hᵢᵢ is the leverage
  • HC3: residuals scaled by 1/(1 - hᵢᵢ)
• Vcov.cluster(:x) — one-way cluster-robust
• Vcov.cluster(:x, :y) — two-way cluster-robust

References

Kleibergen, F, and Paap, R. (2006) Generalized reduced rank tests using the singular value decomposition. Journal of econometrics

Kleibergen, F. and Schaffer, M. (2007) RANKTEST: Stata module to test the rank of a matrix using the Kleibergen-Paap rk statistic. Statistical Software Components, Boston College Department of Economics.