on 03-Jun-2022 (Fri)

Building on the structural approach to causality introduced by Haavelmo (1943) and the graph-theoretic framework proposed by Pearl (1995), the artificial intelligence (AI) literature has developed a wide array of techniques for causal learning that allow leveraging information from various imperfect, heterogeneous, and biased data sources (Bareinboim and Pearl, 2016). In this paper, we discuss recent advances made in this literature that have the potential to contribute to econometric methodology along three broad dimensions. First, they provide a unified and comprehensive framework for causal inference, in which the above-mentioned problems can be addressed in full generality. Second, due to their origin in AI, they come together with sound, efficient, and complete (to be formally defined) algorithmic criteria for automatization of the corresponding identification task. And third, because of the nonparametric description of structural models that graph-theoretic approaches build on, they combine the strengths of both structural econometrics as well as the potential outcomes framework, and thus offer an effective middle ground between these two literature streams

There are two categories of things that could go wrong if we condition on descendants of 𝑇:

1. We block the flow of causation from 𝑇 to 𝑌.

2. We induce non-causal association between 𝑇 and 𝑌.

Disadvantags of survey-based CX measurement

1. Limited: The typical CX survey samples only 7 percent of a company’s customers, providing an extremely limited view of what customers experience and value. In fact, only 13 percent of the CX leaders we surveyed expressed full confidence that their CX measurement system provides a representative view of their customer base

Predictive customer scores

The company develops analytics—often using several types of machine-learning algorithms—to understand and track what is influencing customer satisfaction and business performance, and to detect specific events in customer journeys. The algorithms generate predictive scores for each customer based on journey features. These scores allow the company to predict individual customer satisfaction and value outcomes such as revenue, loyalty, and cost to serve. More broadly, they allow CX leaders to assess the ROI for particular CX investments and directly tie CX initiatives to business outcomes

Rather, my outcome is only a function of my own treatment. We’ve been using this assumption implicitly throughout this chapter. We’ll now formalize it.

Assumption 2.4 (No Interference) 𝑌 𝑖 (𝑡 1 , . . . , 𝑡 𝑖−1 , 𝑡 𝑖 , 𝑡 𝑖+1 , . . . , 𝑡 𝑛 ) = 𝑌 𝑖 (𝑡 𝑖 )

Building on the structural approach to causality introduced by Haavelmo (1943) and the graph-theoretic framework proposed by Pearl (1995), the artificial intelligence (AI) literature has developed a wide array of techniques for causal learning that allow leveraging information from various imperfect, heterogeneous, and biased data sources (Bareinboim and Pearl, 2016)

For example, suppose that the causal DAG includes an unmeasured common cause of A and Y, U and also a measured variable L that is an effect of U. In those cases, it is generally better to adjust for L, because even though adjusting for L will not eliminate all confounding by U, it will typically eliminate some of the confounding by U. In those cases we say that L is a surrogate confounder or a proxy confounder. In summary, in any given causal DAG confounding is an absolute concept, but confounder is a relative concept

We present new research that brings clarity and a fact base to the shortcomings of survey-based measurement systems. We then examine how a few leaders have implemented data-driven CX systems and in turn reduced churn, boosted revenue, and lowered cost to serve.

Disadvantags of survey-based CX measurement

2. Reactive: Surveys are a backward-looking tool in a world where customers expect their concerns to be resolved increasingly quickly.

Disadvantags of survey-based CX measurement

1. Limited: The typical CX survey samples only 7 percent of a company’s customers

Inverse probability matching is in fact just one of the group of so called G-methods

So all these methods for confounding adjustment -- stratification, matching, inverse probability weighting, G-formula, G-estimation -- have two things in common. First, they require data on the confounders that block the backdoor path. If those data are available, then the choice of one of these methods over the others is often a matter of personal taste. Unless the treatment is time-varying -- then we have to go to G-methods

Inverse probability weighting

the most important take-home message: we need expert knowledge to determine if we should adjust for a variable. The statistical criteria are insufficient to characterize confounding and confounders. Of course, in many cases we don't have enough expert knowledge to draw the true causal DAG that represents a causal structure of treatment A, outcome Y, and potential confounder L. In those cases we may propose several possible causal DAGs without being able choose a particular one. And that's fine, because those causal DAGs that we propose allow us to identify inconsistencies between our beliefs and our actions. For example, suppose L is fetal death. We don't know the true causal DAG, we propose seven causal DAGs. Suppose that L does not help block a backdoor path in any of the seven DAGs, then we will not adjust for L, even if L were strongly associated with A and Y.

we have not been able to come up with a story in which L helps reduce confounding. In other words, we will not adjust for a variable unless we believe it may be a confounder. Our beliefs and our actions will be internally consistent when we use causal graph