Changelog • medoutcon

medoutcon 0.2.4

Updated required versions of some dependencies to their most recent stable versions, including hal9001 (v0.4.6) and sl3 (v1.4.5).

Added a new named argument cv_strat to est_onestep() and est_tml() and to the estimator_args list-argument in medoutcon(), which allows for stratified folds to be generated for cross-fitting (by passing these to the strata_ids argument of make_folds() from the origami package). This is also triggered by an override in est_onestep() and est_tml() when the proportion of detected cases is less than 0.1, a heuristic for rare outcomes.
Increased the default number of folds for cross-fitting from 5 to 10, setting cv_folds = 10L in named arguments to est_onestep() and est_tml() and to the estimator_args list-argument in medoutcon().
Changed default propensity score truncation bounds specified in g_bounds to c(0.005, 0.995) from c(0.01, 0.99) (in v0.22), based on sanity checks and manual experimentation.
Wrapped instances of sl3_Task() in which outcome_type = "continuous" is specified in suppressWarnings() to sink warnings when the outcome variable for a given nuisance estimation task fails sl3’s check for continuous-ness.

Change iterative targeting procedures in est_tml() to use glm2::glm2 instead of stats::glm to avoid issues with erratic IRLS in the latter; see https://journal.r-project.org/archive/2011/RJ-2011-012/ for details.
Changed default propensity score truncation bounds specified in g_bounds by an order of magnitude, from c(0.001, 0.999) to c(0.01, 0.99), to mitigate potential stability issues.

Added support for a semiparametric correction for outcome-dependent two-phase sampling designs with known or estimated sampling weights.
Tightened sanity checks for estimation of natural direct and indirect effects by requiring that EIF scores related to intermediate confounders uniformly be zero when Z = NULL is specified.
The above required minor changes to est_tml() so as to avoid fluctuation models for the TMLE step from updating the intermediate confounder nuisance components (i.e., q_prime_Z_one, q_prime_Z_natural) and causing numerical issues that violate the above internal checks.

For user clarity, the name of the argument for providing externally computed observation-level weights has changed (from ext_weights) to svy_weights.
Support for the natural direct and indirect effects has been added, requiring the addition of the new internal argument effect_type across functions for estimation, including cv_eif(), est_onestep(), and est_tml(). When Z = NULL is set in medoutcon(), a natural effect estimate corresponding to the argument effect is returned instead of an interventional effect.
The summary() and print() methods have been updated to allow handling of natural effects and counterfactual means under arbitrary contrasts.

An initial public release of this package, version 0.1.0, which includes support for external observation-level weights.