GradedInterventionTimeSeries.effect#

GradedInterventionTimeSeries.effect(window, channels=None, scale=0.0)[source]#

Estimate the causal effect of scaling treatment channels in a time window.

This method computes a counterfactual scenario by scaling the specified treatment channels in the given window, reapplying all transforms with the same parameters, and comparing to the observed outcome.

For Bayesian models, returns posterior distributions of effects with credible intervals. For OLS models, returns point estimates.

Parameters:

window (Tuple[Union[pd.Timestamp, int], Union[pd.Timestamp, int]]) – Start and end of the effect window (inclusive).
channels (List[str], optional) – List of treatment channel names to scale. If None, scales all channels.
scale (float, default=0.0) – Scaling factor for the counterfactual (0.0 = remove treatment).

Returns:

result – Dictionary containing: - “effect_df”: DataFrame with observed, counterfactual, effect, cumulative effect - “total_effect”: Total effect in window (mean for Bayesian) - “mean_effect”: Mean effect per period in window For Bayesian models, also includes HDI bounds for counterfactual and effect.

Return type:

Dict

Examples

# Estimate effect of removing treatment completely
effect = result.effect(
    window=(df.index[0], df.index[-1]),
    channels=["comm_intensity"],
    scale=0.0,
)
print(f"Total effect: {effect['total_effect']:.2f}")