Theta-Aware Full Variance-Covariance for NB Models

Returns the full sandwich variance-covariance matrix for NB models, including the theta (dispersion) parameter. For non-NB models, returns vcov(object) unchanged.

Usage

vcov_nb(object, vcov_type = "HC1", cluster = NULL)

Arguments

object

A fitted uncounted object.

vcov_type

HC type for the sandwich ("HC0" or "HC1"). Default "HC1". HC2+ are not available for the theta-aware path.

cluster

Optional cluster vector for cluster-robust variance.

Value

A square variance-covariance matrix. For NB models, dimensions are p + 1 (or p + 2 if gamma is estimated), where the extra row/column corresponds to log(theta). For non-NB models, returns vcov(object).

Details

This is the explicit interface for the theta-aware covariance that vcov() uses internally for NB fits. Use this when you need the full matrix including theta, or when you want to be explicit about which covariance you are requesting.

Examples

data(irregular_migration)
d <- irregular_migration[irregular_migration$year == "2019", ]
fit <- estimate_hidden_pop(d, ~m, ~n, ~N, method = "nb", gamma = 0.005)
vcov_nb(fit)  # includes theta row/column
#>              alpha        beta            
#> alpha 0.0008780469 0.001285950 0.001801256
#> beta  0.0012859499 0.002185531 0.002006601
#>       0.0018012560 0.002006601 0.044811875
vcov(fit)     # alpha/beta only (same as vcov_nb submatrix)
#>              alpha        beta
#> alpha 0.0008780469 0.001285950
#> beta  0.0012859499 0.002185531