Leave-One-Out Influence on Coefficients

Computes the change in estimated coefficients when each observation (or country) is dropped. This is a convenience wrapper around loo.

Usage

# S3 method for class 'uncounted'
dfbeta(model, by = c("obs", "country"), ...)

Arguments

model

A fitted uncounted object.

by

Character: "obs" (default) drops one observation at a time, "country" drops all observations for one country at a time (requires countries in the original fit).

...

Additional arguments passed to loo.

Value

A matrix with one row per dropped unit and one column per coefficient. Each entry is the change in the coefficient estimate relative to the full model (\(\hat\beta_{(-i)} - \hat\beta\)).

See also

loo, dfpopsize

Examples

data(irregular_migration)
d <- irregular_migration[irregular_migration$year == "2019" & irregular_migration$sex == "Male", ]
fit <- estimate_hidden_pop(d, ~ m, ~ n, ~ N, method = "poisson",
                           gamma = 0.001, countries = ~ country_code)
db <- dfbeta(fit)
head(db)
#>           alpha          beta
#> 1  4.618242e-03  1.448704e-02
#> 2 -1.379703e-04 -4.409443e-04
#> 3  3.970809e-03  1.264425e-02
#> 4  3.043501e-04  9.516993e-04
#> 5  1.259357e-03  9.422888e-04
#> 6 -2.749349e-05 -8.519565e-05