Package 'assesslite'

Assume an invariance: the analysis relies on it and it should be attacked

Description

Assume an invariance: the analysis relies on it and it should be attacked

Usage

assume_invariance(audit, invariance, rationale, licenses)

Arguments

audit	a structural_audit object.
invariance	a name from 'invariance_vocabulary()'.
rationale	why the claim is scientifically defensible here.
licenses	what inferential step the claim buys (pooling, transport, ...).

Value

the audit with the ledger entry added.

---

Build the pooling assumption lattice

Description

Build the pooling assumption lattice

Usage

assumption_lattice(audit)

Arguments

audit

a structural_audit object (with a fitted estimate).

Value

the audit with a $lattice element.

---

Apply the decision rules to a tested audit

Description

Apply the decision rules to a tested audit

Usage

decide(
  audit,
  abstain_if = list(estimate_sign_changes = TRUE, effect_crosses_threshold = NULL)
)

Arguments

audit	a structural_audit after test_invariance().
abstain_if	list of user-declared abstention rules: estimate_sign_changes (logical) and effect_crosses_threshold (numeric on the natural scale, or NULL).

---

Declare a causal DAG for the graph_check and adjustment_check attacks

Description

Declare a causal DAG for the graph_check and adjustment_check attacks

Usage

declare_graph(audit, edges, latent = character())

Arguments

audit	a structural_audit object.
edges	character vector of directed edges, e.g. c("age -> adherence", "stage -> survival"). Nodes are the union of everything named.
latent	character vector of node names that are part of the causal structure but not measured (e.g. an unmeasured confounder). Latent nodes may not enter an adjustment set, and implications that touch them are not testable.

---

Canonical invariance identifiers (core spec v0.1)

Description

Canonical invariance identifiers (core spec v0.1)

Usage

invariance_vocabulary()

---

Reject an invariance: the analyst asserts it does not hold here

Description

Reject an invariance: the analyst asserts it does not hold here

Usage

reject_invariance(audit, invariance, rationale, licenses)

Arguments

audit	a structural_audit object.
invariance	a name from 'invariance_vocabulary()'.
rationale	why the claim is indefensible here.
licenses	what the rejection removes from scope.

Value

the audit with the ledger entry added.

---

Render the audit as a self-contained HTML report

Description

Render the audit as a self-contained HTML report

Usage

render_report(audit, path)

Arguments

audit	a structural_audit object after 'decide()'.
path	file path to write the HTML report to.

Value

the path, invisibly.

---

Declare the structure of a causal analysis and open its audit

Description

Declare the structure of a causal analysis and open its audit

Usage

structural_audit(
  data,
  outcome,
  exposure,
  covariates = character(),
  cluster = NULL,
  time = NULL,
  subgroups = character(),
  coords = NULL,
  unit_id = NULL,
  edges = NULL,
  unit = "unit",
  estimand = NULL
)

Arguments

data	data frame of the analysis sample.
outcome	a single column name for a GLM outcome, or c(time, status) column names for a Cox model (requires the survival package).
exposure	column name of the exposure of interest.
covariates	character vector of adjustment covariate names.
cluster	column name of the cluster variable (hospital, site), or NULL.
time	column name of the calendar-time variable, or NULL.
subgroups	character vector of subgroup variable names.
coords	length-2 character vector of coordinate columns c(x, y) for the spatial attack, or NULL.
unit_id	column naming each unit, required with 'edges', or NULL.
edges	a two-column data frame of undirected unit-id pairs defining a network for the interference attack, or NULL.
unit	what one row is (e.g. "patient").
estimand	plain-language statement of the target quantity.

Value

a structural_audit object with the full-sample estimate fitted.

---

Run attacks against the declared invariances

Description

Run attacks against the declared invariances

Usage

test_invariance(
  audit,
  tests = c("unit_permutation", "cluster_holdout", "temporal_split",
    "subgroup_stability"),
  seed = 1,
  confounding_benchmark = 1.25,
  outcome_node = NULL,
  spatial_k = 3,
  tip_ratio = NULL,
  confounder_prevalence = 0.2,
  spatial_knn = 8,
  exposure_map = "mean"
)

Arguments

audit	a structural_audit object with a populated ledger.
tests	character vector of attacks to run: unit_permutation, cluster_holdout, temporal_split, subgroup_stability, confounding_sensitivity, confounding_scenarios, graph_check, adjustment_check, positivity_check, spatial_holdout, interference_check.
seed	integer seed for the permutation test.
confounding_benchmark	plausible unmeasured-confounding strength on the E-value (risk-ratio) scale, used by confounding_sensitivity (default 1.25).
outcome_node	graph node to treat as the outcome for adjustment_check (default: the model's outcome column).
spatial_k	grid resolution (k x k blocks) for spatial_holdout (default 3).
tip_ratio	decision threshold on the ratio scale for confounding_scenarios, or NULL for the null (default NULL).
confounder_prevalence	assumed confounder prevalence for confounding_scenarios (default 0.2).
spatial_knn	neighbours for the spatial_autocorrelation weight matrix (default 8).
exposure_map	neighbour-exposure summary for interference_check: "mean" (default), "any", or "sum".

Value

the audit with the requested attacks recorded and ledger verdicts updated.

---

Write the audit to a JSON file conforming to the core audit schema

Description

Write the audit to a JSON file conforming to the core audit schema

Usage

write_audit(audit, path)

Arguments

audit	a structural_audit object after 'decide()'.
path	file path to write the JSON audit record to.

Value

the path, invisibly.

---