Long-Term Effect Estimation with Surrogate Representation

Abstract

There are many scenarios where short-and long-term causal effects of an intervention are different. For example, low-quality ads may increase short-term ad clicks but decrease the long-term revenue via reduced clicks; search engines measured by inappropriate performance metrics may increase search query shares in a short-term but not long-term. This work therefore studies the long-term effect where the outcome of primary interest, or primary outcome, takes months or even years to accumulate. The observational study of long-term effect presents unique challenges. First, the confounding bias causes large estimation error and variance, which can further accumulate towards the prediction of primary outcomes. Second, short-term outcomes are often directly used as the proxy of the primary outcome, i.e., the surrogate. Notwithstanding its simplicity, this method entails the strong surrogacy assumption that is often impractical. To tackle these challenges, we propose to build connections between long-term causal inference and sequential models in machine learning. This enables us to learn surrogate representations that account for the temporal unconfoundedness and circumvent the stringent surrogacy assumption by conditioning on time-varying confounders in the latent space. Experimental results show that the proposed framework outperforms the state-of-the-art.

Evaluation Methods and Measures for Causal Learning Algorithms

“…This is because counterfactual outcomes cannot be observed at the individual level, otherwise, ITEs would not be needed. In time series causal effect estimation, data collection is even more challenging because outcomes and time-varying covariates may take months or even years to observe [68,24].…”

Deep Learning of Potential Outcomes

“…While models developed by Robins et al for estimating effects with time-varying treatments and confounding have existed for decades, they require strong parametric assumptions for estimation (Robins 1994;Robins et al 2000Robins et al , 2009. Within the machine learning community, researchers have begun to extend the above-described representation learning approaches to temporal settings using recurrent neural networks (RNN) (Box 7) (Lim et al 2018;Bica et al 2020a;Cheng et al 2021). For example, Cheng et al (2021) propose to replace each outcome modeling head of CFRNet with an RNN, one modeling the treated outcomes over time and one modeling the control outcomes.…”

“…Within the machine learning community, researchers have begun to extend the above-described representation learning approaches to temporal settings using recurrent neural networks (RNN) (Box 7) (Lim et al 2018;Bica et al 2020a;Cheng et al 2021). For example, Cheng et al (2021) propose to replace each outcome modeling head of CFRNet with an RNN, one modeling the treated outcomes over time and one modeling the control outcomes. Bica et al (2020a) more explicitly deal with time-varying confounding by having each unit of the RNN take time-varying X, non-varying features V , T , and representations from the temporally-previous unit as inputs.…”

Deep Learning for Causal Inference

“…While models developed by Robins et al for estimating effects with time-varying treatments and confounding have existed for decades, they require strong parametric assumptions for estimation (Robins 1994;Robins et al 2000Robins et al , 2009. Within the machine learning community, researchers have begun to extend the above-described representation learning approaches to temporal settings using recurrent neural networks (RNN) (Box 7) (Lim et al 2018;Bica et al 2020a;Cheng et al 2021). For example, Cheng et al (2021) propose to replace each outcome modeling head of CFRNet with an RNN, one modeling the treated outcomes over time and one modeling the control outcomes.…”

“…Within the machine learning community, researchers have begun to extend the above-described representation learning approaches to temporal settings using recurrent neural networks (RNN) (Box 7) (Lim et al 2018;Bica et al 2020a;Cheng et al 2021). For example, Cheng et al (2021) propose to replace each outcome modeling head of CFRNet with an RNN, one modeling the treated outcomes over time and one modeling the control outcomes. Bica et al (2020a) more explicitly deal with time-varying confounding by having each unit of the RNN take time-varying X, non-varying features V , T , and representations from the temporally-previous unit as inputs.…”