From Rogeberg on rational addiction:
Theories of rational addiction make assumptions concerning the choice rule, preferences and beliefs of people, and derive the resulting consumption plans. … Some economists claim support from the famous as-if methodology of Friedman 1953 … which explicitly identifies prediction as the only aim of “positive economics.”…[Such] explanations may prove excellent predictive devices as long as the empirical regularities they describe remain stable… [but] this defense comes at a cost: Since someone acting as-if he was rationally solving some decision problem would not behave optimally unless this was the actual decision problem he faced, assumptions matter when we turn to welfare analysis. Nor can as-if theories claim to explain in the sense of describing the mechanism or causal process underlying a phenomena, their aim is just to relate observable quantities in the simplest, most empirically successful way. To use a metaphor… an as-if theory of a car would be of no help to a mechanic if the car broke down.
The reference point for optimality depends upon the model you’re using. If it’s not a good causal explanation of the problem then it cannot tell you about the optimality of the agent’s decision.
It’s just another reminder that we need to be really careful when moving into the realm of welfare economics because it implies numerous additional assumptions that we don’t need elsewhere.