Skip to main content

adjustment.AdjustmentModel

Facade for a backdoor-adjusted single-equation outcome model.

Usage

adjustment.AdjustmentModel(
    *,
    treatment,
    outcome,
    adjustment_set,
    formula,
    graph_info,
    outcome_model,
    parent_spec,
    parent_families
)

Wraps a reduced PathModel built from Y ~ treatment + Z while retaining the original causal DAG for identification context. Construct via pathmc.modeling.PathModel.adjustment_model(), not by calling this class directly.

After fit(), query the designated treatment-outcome pair with ate(), att(), atu(), and cate(), or use the shared interpret methods predictions(), comparisons(), and slopes().

Parameters

treatment: str

Treatment variable in the designated query.

outcome: str

Outcome variable in the designated query.

adjustment_set: frozenset[str]

Validated backdoor adjustment set used to build the formula.

formula: str

Reduced regression formula string.

graph_info: GraphInfo

Original causal DAG (not the reduced estimation graph).

outcome_model: PathModel

Inner data-bound model for the reduced equation.

parent_spec: Spec

Original structural specification for DAG visualization.

parent_families: dict[str, str]
Outcome family inherited from the structural parent.

Examples

Build from a fitted structural model::

adj = m.adjustment_model("X -> Y")
adj.adjustment_set
adj.formula
adj.fit(draws=1000, chains=4)
ate = adj.ate(values=(0, 1))
ate.mean()

See Also

adjustment_model()

Constructor on the structural model.

InterpretResult

Unit-level interpret draws.

Attributes

Name Description
adjustment_set Validated backdoor adjustment set.
formula Reduced regression formula string.
graph_info Original causal DAG (not the reduced estimation graph).
idata Posterior samples from the inner model, if fitted.
outcome Outcome variable in the designated query.
outcome_model Inner PathModel for the reduced equation.
pymc_model Compiled PyMC model of the inner outcome regression.
treatment Treatment variable in the designated query.

adjustment_set

Validated backdoor adjustment set.

adjustment_set: frozenset[str]

formula

Reduced regression formula string.

formula: str

graph_info

Original causal DAG (not the reduced estimation graph).

graph_info: GraphInfo

idata

Posterior samples from the inner model, if fitted.

idata: xr.DataTree | None

outcome

Outcome variable in the designated query.

outcome: str

outcome_model

Inner PathModel for the reduced equation.

outcome_model: Any

pymc_model

Compiled PyMC model of the inner outcome regression.

pymc_model: pm.Model

treatment

Treatment variable in the designated query.

treatment: str

Methods

Name Description
ate() Average treatment effect via the inner model’s do() path.
att() Average treatment effect on the treated.
atu() Average treatment effect on the untreated.
cate() Conditional average treatment effect.
comparisons() Interventional contrasts on the reduced outcome model.
datagrid() Build a covariate grid from the inner outcome model’s data.
fit() Fit the inner outcome model via MCMC.
from_path_model() Build an adjustment model from a structural PathModel.
graph() DAG of the original causal system (not the reduced equation).
predictions() Response-mean predictions on the reduced outcome model.
slopes() Finite-difference slopes on the reduced outcome model.

ate()

Average treatment effect via the inner model’s do() path.

Usage

ate(outcome=None, treatment=None, values=(0.0, 1.0), **do_kwargs)

att()

Average treatment effect on the treated.

Usage

att(
    outcome=None,
    treatment=None,
    values=(0.0, 1.0),
    treated_value=1.0,
    kind="mean"
)

atu()

Average treatment effect on the untreated.

Usage

atu(
    outcome=None,
    treatment=None,
    values=(0.0, 1.0),
    untreated_value=0.0,
    kind="mean"
)

cate()

Conditional average treatment effect.

Usage

cate(
    outcome=None,
    treatment=None,
    values=(0.0, 1.0),
    condition=None,
    **do_kwargs
)

comparisons()

Interventional contrasts on the reduced outcome model.

Usage

comparisons(
    outcome=None,
    variable=None,
    *,
    contrast=(0.0, 1.0),
    comparison="diff",
    conditional=None,
    average_by="all"
)

Defaults outcome and variable to the designated query. causal is True only when variable is the designated treatment.

datagrid()

Build a covariate grid from the inner outcome model’s data.

Usage

datagrid(**cols)

fit()

Fit the inner outcome model via MCMC.

Usage

fit(**kwargs)

from_path_model()

Build an adjustment model from a structural PathModel.

Usage

from_path_model(
    parent,
    *,
    query=None,
    treatment=None,
    outcome=None,
    adjustment_set=None,
    formula=None,
    data=None,
    families=None,
    priors=None
)

This is the implementation behind pathmc.modeling.PathModel.adjustment_model(). Prefer that method from user code.

graph()

DAG of the original causal system (not the reduced equation).

Usage

graph()

predictions()

Response-mean predictions on the reduced outcome model.

Usage

predictions(outcome=None, *, set=None, newdata=None)

Defaults outcome to the designated query outcome. Without set, predictions are associational. causal is True only when set intervenes on the designated treatment alone.

slopes()

Finite-difference slopes on the reduced outcome model.

Usage

slopes(
    outcome=None,
    wrt=None,
    *,
    slope="dydx",
    eps=0.0001,
    conditional=None,
    average_by="all"
)

Defaults outcome and wrt to the designated query. causal is True only when wrt is the designated treatment.