Causal inference for qualitative outcomes
The causalQual package provides a suite of tools for estimating causal effects when the outcome of interest is qualitative - i.e., multinomial or ordered. Standard causal inference methods such as instrumental variables (IV), regression discontinuity (RD), and difference-in-differences (DiD) are typically designed for numeric outcomes. Their direct application to qualitative outcomes leads to ill-defined estimands, rendering results arbitrary and uninterpretable.
This package implements the framework introduced in Di Francesco and Mellace (2025), shifting the focus to well-defined and interpretable estimands that quantify how treatment affects the probability distribution over outcome categories. The methods remain compatible with conventional research designs, ensuring ease of implementation for applied researchers.
Why use causalQual?
| Feature | Benefit |
|---|---|
| Avoids misleading conclusions | Conventional estimands are often undefined or depend on arbitrary outcome coding. causalQual targets interpretable and meaningful estimands. |
| Provides well-defined estimands | Instead of relying on average effects, causalQual models how treatment shifts probabilities over outcome categories. |
| Wide applicability | Supports selection-on-observables, IV, RD, and DiD. |
| Extensible and open-source | Actively developed with planned support for staggered adoption, fuzzy regression discontinuity, and more. |