6 thoughts on “All models are wrong …

  1. That is only one half of the issue; the benefits (savings, etc.) that the model provides, despite its incompleteness is also important. The judgment on the model needs to weigh the frequency and severity of both costs and benefits.

    Of course, for that we need a model…

    (Quis modoliet ipsos modeles 😎 )

  2. You can roll the benefits into the cost and just speak of cost. I assume that’s what Taleb is thinking. A hypothetical ideal model has some benefit, and you pay a cost for deviations from that ideal.

  3. “All models are wrong…” Agreed, and all models are more or less valid and reliable that others. The predictive power of a model can be judged normatively and pragmatically by whether or not the model works better than the next best model (or not). However, all models will fail the “perfection” criterion test. The goal cannot be to create models that “perfectly” model reality (if there is such a thing), but rather to get just close enough to reality in order to facilitate decisions (the closer a model gets to reality, the more costly the model usually is to implement and maintain). Thus, the “close enough” imperative remains operative from an efficiency standpoint. Next, low-level models are descriptive, useful models are predictive, but valuable models are prescriptive. Finally, simplicity reduces the threats to a model’s validity and reliability, which makes simplicity the modeler’s friend. Lots more can be said. Thanks for touching on one of my favorite topics…

  4. The way I learned it was “All models are wrong, some are useful.” If a model is deadly or costly (to us), I would argue that we don’t find it useful.

  5. That was the original quote, most often attributed to the late George Box. Taleb was going one step beyond.

Leave a Reply

Your email address will not be published. Required fields are marked *