Enables the registration of custom fi statistics that can be easily summoned with the function fitstat.

fitstat_register(type, fun, alias = NULL, subtypes = NULL)

## Arguments

type A character scalar giving the type-name. A function to be applied to a fixest estimation. It must return either a scalar, or a list of unitary elements. If the number of elements returned is greater than 1, then each element must be named! If the fit statistic is not valid for a given estimation, a plain NA value should be returned. A (named) character vector. An alias to be used in lieu of the type name in the display methods (ie when used in print.fixest_fitstat or etable). If the function returns several values, i.e. sub-types, you can give an alias to these sub-types. The syntax is c("type" = "alias", "subtype_i" = "alias_i"), with "type" (resp. "subtype") the value of the argument type resp. (subtypes). You can also give an alias encompassing the type and sub-type with the syntax c("type.subtype_i" = "alias"). A character vector giving the name of each element returned by the function fun. This is only used when the function returns more than one value. Note that you can use the shortcut "test" when the sub-types are "stat", "p" and "df"; and "test2" when these are "stat", "p", "df1" and "df2".

## Details

If there are several components to the computed statistics (i.e. the function returns several elements), then using the argument subtypes, giving the names of each of these components, is mandatory. This is to ensure that the statistic can be used as any other built-in statistic (and there are too many edge cases impeding automatic deduction).

Laurent Berge

## Examples


# An estimation
base = iris
names(base) = c("y", "x1", "x2", "x3", "species")
est = feols(y ~ x1 + x2 | species, base)

#
# single valued tests
#

# say you want to add the coefficient of variation of the dependent variable
cv = function(est){
y = model.matrix(est, type = "lhs")
sd(y)/mean(y)
}

# Now we register the routine
fitstat_register("cvy", cv, "Coef. of Variation (dep. var.)")

# now we can summon the registered routine with its type ("cvy")
fitstat(est, "cvy")
#> Coef. of Variation (dep. var.): 0.141711

#
# Multi valued tests
#

# Let's say you want a Wald test with an heteroskedasticiy robust variance

# First we create the function
hc_wald = function(est){
w = wald(est, keep = "!Intercept", print = FALSE, se = "hetero")
}
# This test returns a vector of 4 elements: stat, p, df1 and df2

# Now we register the routine
fitstat_register("hc_wald", hc_wald, "Wald (HC1)", "test2")

# You can access the statistic, as before
fitstat(est, "hc_wald")
#> Wald (HC1): stat = 122.7, p < 2.2e-16, on 2 and 145 DoF.

# But you can also access the sub elements
fitstat(est, "hc_wald.p")
#> Wald (HC1), p-value: 6.573e-32