Learning Individual Causal Effects from Networked Observational Data

Abstract

The convenient access to observational data enables us to learn causal effects without randomized experiments. This research direction draws increasing attention in research areas such as economics, healthcare, and education. For example, we can study how a medicine (the treatment) causally affects the health condition (the outcome) of a patient using existing electronic health records. To validate causal effects learned from observational data, we have to control confounding bias -the influence of variables which causally influence both the treatment and the outcome. Existing work along this line overwhelmingly relies on the unconfoundedness assumption that there do not exist unobserved confounders. However, this assumption is untestable and can even be untenable. In fact, an important fact ignored by the majority of previous work is that observational data can come with network information that can be utilized to infer hidden confounders. For example, in an observational study of the individual-level treatment effect of a medicine, instead of randomized experiments, the medicine is often assigned to each individual based on a series of factors. Some of the factors (e.g., socioeconomic status) can be challenging to measure and therefore become hidden confounders. Fortunately, the socioeconomic status of an individual can be reflected by whom she is connected in social networks. With this fact in mind, we aim to exploit the network information to recognize patterns of hidden confounders which would further allow us to learn valid individual causal effects from observational data. In this work, we propose a novel causal inference framework, the network deconfounder, which learns representations to unravel patterns of hidden confounders from the network information. Empirically, we perform extensive experiments to validate the effectiveness of the network deconfounder on various datasets. CCS CONCEPTS• Mathematics of computing → Causal networks; • Networks → Online social networks; • Human-centered computing → Social network analysis.

Counterfactual Evaluation of Treatment Assignment Functions with Networked Observational Data

“…As a result, it is extremely challenging to collect data with ground truth of counterfactual outcomes. Therefore, we follow [27,11] to synthesize the treatments and outcomes based on the observed features and network information which are extracted from two real-world datasets. Specifically, we introduce two networked observational datasets for evaluating the utility of treatment assignment functions.…”

“…The advanced big data technologies have granted us the convenient access to massive observational data in a plethora of highly influential applications such as social networks [11], online advertising, and recommender systems [23]. For example, an online blogger community generates massive log data containing bloggers' keywords (features), users' browsing devices (the treatments), and users' opinions (the outcomes).…”

“…The importance of network structures in catching hidden confounders' patterns have been shown in other tasks of causal inference [27,11]. For example, while the writing style of a blogger is hard to quantified, her representation can be implicitly captured by her social network patterns such as which bloggers are her friends.…”

“…We can see the three types of counterfactual evaluation methods cannot utilize the network information. The success of using network information to handle hidden confounders has been demonstrated in other tasks of causal inference [27,11] (e.g., causal effect estimation). Motivated by the success, we investigate incorporating network information in counterfactual evaluation.…”

Matching in Selective and Balanced Representation Space for Treatment Effects Estimation

“…Let (Φ( )| = 1) and (Φ( )| = 0) denote the empirical distributions of the representation vectors for the treatment and control groups, respectively. In FSRM, we adopt the IPM defined in the family of 1-Lipschitz functions, which leads to IPM being the Wasserstein distance [6,34,36]. In particular, the IPM term with Wasserstein distance is defined as…”