Module model 

DML models with linear scores: Plr, Irm, Pliv, and the LinearScoreModel trait for adding your own.

Structs

Apo

Average potential outcome E[Y(d)] of one treatment level d.

Apos

Average potential outcomes of several treatment levels, fitted on shared sample splits so that they (and their contrasts) have joint inference.

AposFit

Fitted Apos: the average potential outcome of every level, with joint inference, and contrasts between levels.

AutoDml

Auto-DML estimator of θ = E[m(W; γ)].

CateSignal

Ingredients for best-linear-predictor inference on effect heterogeneity (DmlFit::cate, DmlFit::gate): the BLP regresses signal − θ·theta_factor on the basis, with every row of the basis multiplied by basis_scale if given (partially linear models, where the signal is the outcome residual and the scale the treatment residual).

Cvar

Conditional value at risk of the potential outcome, E[Y(d) | Y(d) ≥ q_τ(Y(d))], as DoubleML’s DoubleMLCVAR. ml_g is a regressor, ml_m a classifier for P(D = 1 | X). The score is linear.

Did

Difference-in-differences with panel data (two periods): y must be the outcome change Y₁ − Y₀. ml_g learns E[ΔY | D = d, X] per group, ml_m the propensity P(D = 1 | X) (observational score only).

DidCs

Difference-in-differences with repeated cross-sections: the data need a 0/1 time indicator (DmlData::with_time_indicator). ml_g learns E[Y | D = d, T = t, X] separately in each of the four group × period cells, ml_m the propensity P(D = 1 | X) (observational score only).

ExternalPredictions

Out-of-fold nuisance predictions supplied by the user, as DoubleML’s external_predictions: per treatment and learner name (e.g. "ml_l", "ml_m"), one column per sample-splitting repetition. Learners with external predictions are not fitted, and have no out-of-sample loss.

Iivm

IIVM estimator of the LATE: ml_g learns E[Y | Z, X] (fitted separately on the Z = 0 and Z = 1 rows), ml_m the instrument propensity P(Z = 1 | X), and ml_r the treatment probability P(D = 1 | Z, X) (again per instrument arm).

Irm

IRM estimator: ml_g learns E[Y | D = d, X] (fitted separately on the treated and control rows of each training fold), ml_m learns the propensity score P(D = 1 | X).

Lplr

Logistic partially linear regression for a binary outcome, P(Y = 1 | D, X) = expit(θ·D + r(X)), as DoubleML’s DoubleMLLPLR; θ is a log odds ratio.

Lpq

Local potential quantile of compliers, with a binary instrument, as DoubleML’s DoubleMLLPQ. ml_g is a classifier for P(1{D = d}·1{Y ≤ θ̃} = 1 | Z = z, X), ml_m a classifier used for the instrument propensity P(Z = 1 | X) and for P(D = 1 | Z = z, X). Default sample splits are stratified by treatment and instrument.

NuisanceLearners

Learners for individual nuisance functions that replace a model’s default learner for them, e.g. different settings for ml_g0 and ml_g1 in Irm (DoubleML’s per-nuisance parameters, set_ml_nuisance_params). Set through the models’ nuisance_regressor / nuisance_classifier.

NuisanceTarget

The target a nuisance learner predicts; NaN marks rows where the nuisance does not apply (e.g. g₀ on treated rows).

Pliv

Partially linear IV regression. See PlivVariant and PlivScore.

Plpr

Partially linear regression on a transformed static panel, with folds and standard errors clustered by unit. See PanelApproach for the transformations and PlrScore for the scores.

Plr

PLR estimator: ml_l learns E[Y | X], ml_m learns E[D | X], and the optional ml_g learns g(X) for the IV-type score.

Pq

Potential quantile θ with P(Y(d) ≤ θ) = τ, as DoubleML’s DoubleMLPQ.

Qte

Treatment effects on quantiles (or CVaRs), as DoubleML’s DoubleMLQTE: for every τ, the potential quantiles of treatment 1 and 0 are fitted on the same sample splits and their difference is the effect, with joint inference across all quantiles.

QteFit

Fitted Qte.

RdFlex

RD with flexible covariate adjustment. A fuzzy design (RdFlex::fuzzy) also adjusts the treatment with a classifier.

RdFlexFit

Fitted RdFlex: the conventional, bias-corrected and robust RD estimates (rows), aggregated over repetitions as DoubleML.

ScoreEstimate

Score elements plus the cross-fitted nuisance predictions behind them.

ScoreInput

The data a model sees when estimating the effect of one treatment.

Ssm

Average treatment effect of a binary treatment when the outcome is observed only where the selection indicator S is 1 (attrition, non-response), as DoubleML’s DoubleMLSSM. Requires a selection indicator (DmlData::with_selection_indicator) and, for SsmScore::Nonignorable, an instrument for selection (DmlData::with_instrument).

StaticPanel

A static panel in long format: one row per unit id and period t.

Enums

DidScore

Assignment mechanism assumed by Did / DidCs.

FirstStage

Regressors of the first stage besides the covariates.

IrmScore

Target parameter of Irm.

LossKind

Loss used to evaluate a nuisance learner out of sample.

LplrScore

Score of Lplr.

PanelApproach

How Plpr handles the unit effects.

PlivScore

Score of Pliv.

PlivVariant

What Pliv partials out.

PlrScore

Orthogonal score used by Plr.

RdKernel

Kernel of the first-stage weights.

SsmScore

Selection mechanism of Ssm.

TreatmentLearner

A learner for E[D | X]: a regressor, or for a binary treatment a classifier whose class-1 probability is used.

Traits

LinearScoreModel

A DML model whose orthogonal score is linear in θ.

Type Aliases

Name

Name of a nuisance learner: a &'static str literal or an owned String (Cow holds either without copying literals).