Predictive & Causal Relationships
With the rise of Big Data predictive modeling has grown explosively in the last decade, while causal models have taken a back stage, mostly confined to academia and small research teams. It’s plausible that in a long run a well researched causal models would be more stable and a better tool for prediction compared to ad hoc predictive models. But we can't wait that long and need to know the future right now. For example, marketing brand managers may be less interested in research that lets them understand what causes consumer purchase behaviour and more interested in research that will enable them to predict how much their product will be purchased next month.
So what is the difference between these two analysis:
Predictive relationships focuses merely on association, there need be no presumption or implication of causation, only that variations in A are related to variations in B. If we can identify and verify such a relationship, then we can use our knowledge of variation in A to predict variation in B, without any need to explain why the association occurs or what causes variability in B.
Distinct from predictive associational relationships are causal relationships. These relationships invoke the notion of causality, with the idea that one of the variables in the relationship, X, influences the other variable in the relationship, Y. We can generally think of causality in terms of change. By thinking in causal terms, we are able to identify systematic relationships between variables and manipulate those variables to produce desired changes.
Conceptualising causality and establishing causal relationships is much more complex compared to Predictive Modeling. And hence we fall back on the predictive modeling procedures to develop formula for making predictions about the dependent variable based on tens of thousands of independent variables.
Given the scope we should invoke Causal Analysis to known Predictive Models to understand in-depth about our environment, organise our thoughts, predict future events, and even change future events.
Product Owner at Citi | PGPM MDI Gurgaon | NIT Surat
6 年Nice thoughts! The reason is similar to why we work to meet our short-term goals rather than focus on long-term and greater fruits bearing goals. We have a job to do, quarterly results to publish, target pressures, etc. because of which we compromise with suboptimal results. If business acumen, statistical modelling and luxury of time are cooked with right ingredients like infrastructure, attitude, passion then I am sure we will see a magnificent causal relationship model.