using [gpt](https://chatgpt.com/share/67f7ab4e-af0c-8002-adee-8d5cb4c728e8) given [[Discrete_Choice_Case_Study.pdf]] on multinomial logit and forecasting
nested logit
- know apriori or diagnose from data (coefficient scale up when comparing MNL vs flat only)
- infinite scale = deterministic choice = high similarity among subchoices (corr() = 1-mu^2/mu_d^2)
- simulation (within the range of known parameter) vs forecasting (with new price)
Swissmetro: SM's appeal persists, especially for low-income users; high-income users react more to price changes—counterintuitive insight.
Telephone Plans: Flat plans suffer after price hikes; BM (Budget Measured) gains across all segments. Policy impact is more pronounced among low-income users.
![[cs3 2025-04-10-5]]