Masterclass
Beschrijving
Causal decision making (CDM) at scale has become a routine part of business, and increasingly CDM is based on statistical models and machine learning (ML) algorithms. Businesses target offers, incentives, recommendations, and even content algorithmically with the goal of affecting consumer behavior. I highlight something important: deciding on an action for causal effect is not the same as causal effect estimation. In fact, accurate causal effect estimation is not necessary for accurate CDM. I will discuss three implications: (1) We should optimize ML for accurate CDM rather than for accurate effect estimation. (2) For CDM, it may be just as good or even better to learn with confounded data as with unconfounded data. Finally, (3) causal statistical modeling may not be necessary at all to support CDM, because there may be (and perhaps often is) a proxy target for statistical modeling that can do as well or better. This last observation helps to explain at least one broad common CDM practice that seems "wrong" at first blush--the widespread use of non-causal models for targeting interventions like advertisements and retention incentives. The last two implications are particularly important in practice, as acquiring (unconfounded) data on both "sides" of the counterfactual for modeling can be quite costly and often is impracticable. Understanding causal decision making is vital to modern data science practice, and is fertile ground for new data science research (there's been surprisingly little until just the past few years).
Programma
- 4 p.m. : Masterclass
- 5 p.m.: Sandwich lunch
