The price elasticity of demand doesn’t magically become causal by using DoubleML instead of regression. Similarly, we can’t estimate the causal effect of a treatment if a condition is always treated or never treated. We need to treat sometimes and not treat other times.
Causal modeling starts with bespoke data and continues with assumptions. The methods follow the data and assumptions and are useful only if the right data and assumptions are available. This is different from predictive modeling, where brute force bias reduction using the most complex method can be successful.
We offer a reminder in this solo piece at Data Duets. You can read or listen (just scroll to the end).