Causalis

`multi_unconfoundedness`

Modules

dgp –
model –
refutation – Refutation utilities for multi-treatment unconfoundedness.

Classes

MultiTreatmentIRM – Interactive Regression Model for multi-treatment unconfoundedness.

`MultiTreatmentIRM`

Bases: BaseEstimator

Interactive Regression Model for multi-treatment unconfoundedness.

DoubleML-style cross-fitting estimator consuming MultiCausalData and producing pairwise ATE contrasts against baseline treatment (column 0). Model supports >= 2 treatments.

Parameters

Functions

`coef`

Return the estimated coefficient.

Returns

ndarray – The estimated coefficient.

`confint`

`data`

`diagnostics_`

Return diagnostic data.

Returns

dict – Dictionary containing 'm_hat', 'g_hat' and 'folds'.

`estimate`

`fit`

`ml_g`

`ml_m`

`n_folds`

`n_rep`

`normalize_ipw`

`orth_signal`

Return the cross-fitted orthogonal signal (psi_b).

Returns

ndarray – The orthogonal signal.

`pvalues`

Return the p-values for the estimate.

Returns

ndarray – The p-values.

`random_state`

`score`

`se`

Return the standard error of the estimate.

Returns

ndarray – The standard error.

`sensitivity_analysis`

`summary`

Return a summary DataFrame of the results.

Returns

DataFrame – The results summary.

`trimming_rule`

`trimming_threshold`

`dgp`

Functions

generate_multitreatment_binary_26 – Pre-configured multi-treatment dataset with Binary outcome.
generate_multitreatment_gamma_26 – Pre-configured multi-treatment dataset with Gamma-distributed outcome.
generate_multitreatment_irm_26 –

`generate_multitreatment_binary_26`

Pre-configured multi-treatment dataset with Binary outcome.

3 treatment classes: control + 2 treatments
8 confounders with realistic marginals
Binary outcome with logistic-link linear confounding
Heterogeneous treatment effects and correlated confounders via Gaussian copula

`generate_multitreatment_gamma_26`

Pre-configured multi-treatment dataset with Gamma-distributed outcome.

3 treatment classes: control + 2 treatments
8 confounders with realistic marginals
Gamma outcome with log-link linear confounding
Heterogeneous treatment effects and correlated confounders via Gaussian copula

`generate_multitreatment_irm_26`

`model`

Classes

MultiTreatmentIRM – Interactive Regression Model for multi-treatment unconfoundedness.

`HAS_CATBOOST`

`MultiTreatmentIRM`

Bases: BaseEstimator

Interactive Regression Model for multi-treatment unconfoundedness.

DoubleML-style cross-fitting estimator consuming MultiCausalData and producing pairwise ATE contrasts against baseline treatment (column 0). Model supports >= 2 treatments.

Parameters

Functions

`coef`

Return the estimated coefficient.

Returns

ndarray – The estimated coefficient.

`confint`

`data`

`diagnostics_`

Return diagnostic data.

Returns

dict – Dictionary containing 'm_hat', 'g_hat' and 'folds'.

`estimate`

`fit`

`ml_g`

`ml_m`

`n_folds`

`n_rep`

`normalize_ipw`

`orth_signal`

Return the cross-fitted orthogonal signal (psi_b).

Returns

ndarray – The orthogonal signal.

`pvalues`

Return the p-values for the estimate.

Returns

ndarray – The p-values.

`random_state`

`score`

`se`

Return the standard error of the estimate.

Returns

ndarray – The standard error.

`sensitivity_analysis`

`summary`

Return a summary DataFrame of the results.

Returns

DataFrame – The results summary.

`trimming_rule`

`trimming_threshold`

`refutation`

Refutation utilities for multi-treatment unconfoundedness.

Modules

Functions

compute_bias_aware_ci –
get_sensitivity_summary –
plot_m_overlap – Multi-treatment overlap plot for pairwise conditional propensity scores.
plot_residual_diagnostics – Plot residual diagnostics for multi-treatment nuisance models.
run_overlap_diagnostics – Run multi-treatment overlap diagnostics from data and estimate.
run_score_diagnostics – Run score diagnostics for multi-treatment baseline contrasts.
run_unconfoundedness_diagnostics – Run multi-treatment unconfoundedness diagnostics from data and estimate.
sensitivity_analysis –
sensitivity_benchmark –
validate_unconfoundedness_balance – Convenience wrapper returning the balance block only.

`compute_bias_aware_ci`

`get_sensitivity_summary`

`overlap`

Modules

overlap_plot –
overlap_validation – Overlap diagnostics for multi-treatment unconfoundedness.

Functions

plot_m_overlap – Multi-treatment overlap plot for pairwise conditional propensity scores.
run_overlap_diagnostics – Run multi-treatment overlap diagnostics from data and estimate.

`overlap_plot`

Functions

overlap_plot – Convenience wrapper to match overlap_plot(data, estimate) API style.
plot_m_overlap – Multi-treatment overlap plot for pairwise conditional propensity scores.

`overlap_plot`

Convenience wrapper to match overlap_plot(data, estimate) API style.

`plot_m_overlap`

Multi-treatment overlap plot for pairwise conditional propensity scores.

For each comparison baseline (default 0) vs k, this plots P(D=k | X, D in {baseline, k}) = m_k(X) / (m_baseline(X) + m_k(X)) on the observed pair sample D in {baseline, k}, comparing:

units with D=k (treated for the pair),
units with D=baseline (control for the pair).

Parameters:

diag.d: (n, K) one-hot
diag.m_hat / diag.m_hat_raw: (n, K) propensity
treatment_idx:
- None -> plot all k != baseline_idx (multi-panel)
- int -> plot one comparison
- list[int] -> plot selected comparisons
ax: supported only for a single comparison (exactly one k)

Returns matplotlib.figure.Figure.

`overlap_validation`

Overlap diagnostics for multi-treatment unconfoundedness.

Functions

run_overlap_diagnostics – Run multi-treatment overlap diagnostics from data and estimate.

`run_overlap_diagnostics`

Run multi-treatment overlap diagnostics from data and estimate.

Diagnostics are computed pairwise between baseline treatment 0 and each active treatment k using pairwise conditional propensity P(D=k | X, D in {0, k}) as the comparison score.

`plot_m_overlap`

Multi-treatment overlap plot for pairwise conditional propensity scores.

For each comparison baseline (default 0) vs k, this plots P(D=k | X, D in {baseline, k}) = m_k(X) / (m_baseline(X) + m_k(X)) on the observed pair sample D in {baseline, k}, comparing:

units with D=k (treated for the pair),
units with D=baseline (control for the pair).

Parameters:

diag.d: (n, K) one-hot
diag.m_hat / diag.m_hat_raw: (n, K) propensity
treatment_idx:
- None -> plot all k != baseline_idx (multi-panel)
- int -> plot one comparison
- list[int] -> plot selected comparisons
ax: supported only for a single comparison (exactly one k)

Returns matplotlib.figure.Figure.

`run_overlap_diagnostics`

Run multi-treatment overlap diagnostics from data and estimate.

Diagnostics are computed pairwise between baseline treatment 0 and each active treatment k using pairwise conditional propensity P(D=k | X, D in {0, k}) as the comparison score.

`overlap_plot`

Functions

overlap_plot – Convenience wrapper to match overlap_plot(data, estimate) API style.
plot_m_overlap – Multi-treatment overlap plot for pairwise conditional propensity scores.

`overlap_plot`

Convenience wrapper to match overlap_plot(data, estimate) API style.

`plot_m_overlap`

Multi-treatment overlap plot for pairwise conditional propensity scores.

For each comparison baseline (default 0) vs k, this plots P(D=k | X, D in {baseline, k}) = m_k(X) / (m_baseline(X) + m_k(X)) on the observed pair sample D in {baseline, k}, comparing:

units with D=k (treated for the pair),
units with D=baseline (control for the pair).

Parameters:

diag.d: (n, K) one-hot
diag.m_hat / diag.m_hat_raw: (n, K) propensity
treatment_idx:
- None -> plot all k != baseline_idx (multi-panel)
- int -> plot one comparison
- list[int] -> plot selected comparisons
ax: supported only for a single comparison (exactly one k)

Returns matplotlib.figure.Figure.

`plot_m_overlap`

Multi-treatment overlap plot for pairwise conditional propensity scores.

For each comparison baseline (default 0) vs k, this plots P(D=k | X, D in {baseline, k}) = m_k(X) / (m_baseline(X) + m_k(X)) on the observed pair sample D in {baseline, k}, comparing:

units with D=k (treated for the pair),
units with D=baseline (control for the pair).

Parameters:

diag.d: (n, K) one-hot
diag.m_hat / diag.m_hat_raw: (n, K) propensity
treatment_idx:
- None -> plot all k != baseline_idx (multi-panel)
- int -> plot one comparison
- list[int] -> plot selected comparisons
ax: supported only for a single comparison (exactly one k)

Returns matplotlib.figure.Figure.

`plot_residual_diagnostics`

Plot residual diagnostics for multi-treatment nuisance models.

Panels

1..K. Arm-specific residual-vs-fitted: u_k = y - g_k vs g_k within arm D_k=1. K+1. Binned calibration error for each arm: E[D_k - m_k | m_k in bin] vs binned m_k.

`run_overlap_diagnostics`

Run multi-treatment overlap diagnostics from data and estimate.

Diagnostics are computed pairwise between baseline treatment 0 and each active treatment k using pairwise conditional propensity P(D=k | X, D in {0, k}) as the comparison score.

`run_score_diagnostics`

Run score diagnostics for multi-treatment baseline contrasts.

`run_unconfoundedness_diagnostics`

Run multi-treatment unconfoundedness diagnostics from data and estimate.

This implementation currently supports ATE diagnostics only and computes pairwise balance between baseline treatment 0 and each active treatment k.

`score`

Modules

residual_plots – Residual diagnostic plots for multi-treatment nuisance models g_k and m_k.
score_validation – Score diagnostics for multi-treatment unconfoundedness.

Functions

plot_residual_diagnostics – Plot residual diagnostics for multi-treatment nuisance models.
run_score_diagnostics – Run score diagnostics for multi-treatment baseline contrasts.

`plot_residual_diagnostics`

Plot residual diagnostics for multi-treatment nuisance models.

Panels

1..K. Arm-specific residual-vs-fitted: u_k = y - g_k vs g_k within arm D_k=1. K+1. Binned calibration error for each arm: E[D_k - m_k | m_k in bin] vs binned m_k.

`residual_plots`

Residual diagnostic plots for multi-treatment nuisance models g_k and m_k.

Functions

plot_residual_diagnostics – Plot residual diagnostics for multi-treatment nuisance models.

`plot_residual_diagnostics`

Plot residual diagnostics for multi-treatment nuisance models.

Panels

1..K. Arm-specific residual-vs-fitted: u_k = y - g_k vs g_k within arm D_k=1. K+1. Binned calibration error for each arm: E[D_k - m_k | m_k in bin] vs binned m_k.

`run_score_diagnostics`

Run score diagnostics for multi-treatment baseline contrasts.

`score_validation`

Score diagnostics for multi-treatment unconfoundedness.

Functions

run_score_diagnostics – Run score diagnostics for multi-treatment baseline contrasts.

`run_score_diagnostics`

Run score diagnostics for multi-treatment baseline contrasts.

`sensitivity_analysis`

`sensitivity_benchmark`

`unconfoundedness`

Modules

sensitivity –
unconfoundedness_validation – Unconfoundedness diagnostics for multi-treatment settings.

Functions

compute_bias_aware_ci –
get_sensitivity_summary –
run_unconfoundedness_diagnostics – Run multi-treatment unconfoundedness diagnostics from data and estimate.
sensitivity_analysis –
sensitivity_benchmark –
validate_unconfoundedness_balance – Convenience wrapper returning the balance block only.

`compute_bias_aware_ci`

`get_sensitivity_summary`

`run_unconfoundedness_diagnostics`

Run multi-treatment unconfoundedness diagnostics from data and estimate.

This implementation currently supports ATE diagnostics only and computes pairwise balance between baseline treatment 0 and each active treatment k.

`sensitivity`

Functions

`compute_bias_aware_ci`

`get_sensitivity_summary`

`sensitivity_analysis`

`sensitivity_benchmark`

`sensitivity_analysis`

`sensitivity_benchmark`

`unconfoundedness_validation`

Unconfoundedness diagnostics for multi-treatment settings.

Functions

run_unconfoundedness_diagnostics – Run multi-treatment unconfoundedness diagnostics from data and estimate.
validate_unconfoundedness_balance – Convenience wrapper returning the balance block only.

`run_unconfoundedness_diagnostics`

Run multi-treatment unconfoundedness diagnostics from data and estimate.

This implementation currently supports ATE diagnostics only and computes pairwise balance between baseline treatment 0 and each active treatment k.

`validate_unconfoundedness_balance`

Convenience wrapper returning the balance block only.

`validate_unconfoundedness_balance`

Convenience wrapper returning the balance block only.

`validate_unconfoundedness_balance`

Convenience wrapper returning the balance block only.