Dice Question Streamline Icon: https://streamlinehq.com

Source of coverage shortfall for MCF IATE inference

Determine whether the observed coverage-probability shortfall for the Modified Causal Forest’s individualized average treatment effect (IATE) inference is primarily attributable to bias in the individual IATE estimates rather than to underestimation of standard errors, and characterize the implications for inference and potential debiasing corrections.

Information Square Streamline Icon: https://streamlinehq.com

Background

In simulations, all estimators showed coverage probabilities below nominal levels for IATEs. The authors hypothesize that, at least for the MCF, this deficiency is driven by estimator bias rather than by underestimated standard errors, since the same weights-based standard error methodology performs well for ATE and GATE settings.

Confirming the primary cause of coverage shortfall would guide the design of debiasing strategies and adjustments to inference procedures for MCF-based IATE estimation.

References

"At least for the mcf, we conjecture that this problem comes from the biases of the individual IATEs. It is less likely that it comes from a too small estimated standard error, because weights-based standard error estimation is performed in a very similar way as for the ATE and the GATEs, in which it turned out to be almost unbiased."

— Comprehensive Causal Machine Learning (2405.10198 - Lechner et al., 16 May 2024) in Section 5.3.3 (Conditional average treatment effects at the finest aggregation level)