Abstract:In causal inference, the joint law of a set of counterfactual random variables is generally not identified. We show that a conservative version of the joint law - corresponding to the smallest treatment effect - is identified. Finding this law uses recent results from optimal transport theory. Under this conservative law we can bound causal effects and we may construct inferences for each individual's counterfactual dose-response curve. Intuitively, this is the flattest counterfactual curve for each subject that is consistent with the distribution of the observables. If the outcome is univariate then, under mild conditions, this curve is simply the quantile function of the counterfactual distribution that passes through the observed point. This curve corresponds to a nonparametric rank preserving structural model.
What problem does this paper attempt to address?
This paper attempts to solve a core problem in causal inference: how to make a conservative estimate of the causal effect when the joint distribution of counterfactual random variables cannot be fully identified. Specifically, the authors propose a method to infer a conservative version of the joint distribution, which corresponds to the minimum treatment effect. Through this method, given the observational data, the "flattest" form of the counterfactual dose - response curve for each individual can be inferred, that is, the curve that is consistent with the observational data and has the minimum treatment effect.
### Background and Problem of the Paper
In causal inference, the joint distribution of a set of counterfactual random variables usually cannot be fully identified. This means that although the marginal distribution of each counterfactual variable can be identified, their joint distribution is unknown. Therefore, it is very difficult to directly infer the complete counterfactual dose - response curve for each individual.
### Solution
The authors propose a conservative method to estimate these counterfactual curves. The specific steps are as follows:
1. **Conservative Joint Distribution**:
- Through the optimal transport theory, find a joint distribution \( J^* \) that is consistent with the known marginal distributions, and this joint distribution minimizes a certain treatment - effect parameter \( \psi \).
- This conservative joint distribution \( J^* \) corresponds to the case of the minimum treatment effect.
2. **Inference of Counterfactual Curves**:
- Under the conservative joint distribution \( J^* \), given the observed values \( (A, Y) \), the counterfactual dose - response curve \( Y^*(a) \) for each individual can be inferred.
- The specific formula is:
\[
Y^*(a) = F_a^{-1}(F_A(Y))
\]
where \( F_a(y) = P(Y(a) \leq y) \), and \( F_A(Y) \) represents the cumulative distribution function value at the observed values \( (A, Y) \).
3. **Optimal Transport Mapping**:
- This curve \( Y^*(a) \) is actually the optimal transport mapping from \( Y(A) \) to \( Y(a) \).
- It corresponds to the non - parametric rank - preserving structural causal model.
### Application and Extension
- **Conditional Effect**:
- The author also considers the conditional effect, that is, in the case of given covariates \( V \), inferring the conservative counterfactual curve \( E_{J_v}[Y(a) | Y(A) = Y, V = v] \).
- **Multivariate Treatment and Outcome**:
- When the treatment variable \( A \) and the outcome variable \( Y \) are multi - dimensional, a similar method is still applicable, but the specific inference process will be different.
### Related Work
- **Boundary Estimation**:
- This problem is related to the boundary estimation of the causal effect. For example, in the case of binary treatment, the upper and lower bounds of the causal effect can be estimated through the Fréchet - Hoeffding bounds.
- **Optimal Transport Theory**:
- This method utilizes the optimal transport theory, which has wide applications in economics, statistics, and machine learning, etc.
### Summary
This paper provides a new method for conservative estimation in causal inference. Through the optimal transport theory, a joint distribution that minimizes the treatment effect is found, so that given the observational data, the counterfactual dose - response curve for each individual can be inferred. This method is not only of great theoretical significance but also provides a new tool for practical applications.