In any system we model we have to make assumptions that don’t quite match reality, hence the quote “all models are wrong, some are useful”. So you’re right in a sense, you can’t really have the value exist without the weather condition. At some point you have to diverge from reality and be pragmatic about your model. As long as the results are interpretable and “match” the reality that you are trying to predict/model/etc. then it’s ok to make some assumptions.

Additionally, my article on marginalisation explains how we deal with the dependence that you’ve explained without explicitly requiring “data from a store selling ice cream above the atmosphere with no weather conditions”.

    Jonny Brooks-Bartlett

    Written by

    Data scientist at Deliveroo, public speaker, science communicator, mathematician and sports enthusiast.