Multinomial Logit Processes and Preference Discovery: Inside and Outside the Black Box

Abstract We provide two characterizations, one axiomatic and the other neuro-computational, of dependence choice probabilities on deadlines, within widely used softmax representation $$\begin{align*} p_{t}\left( a,A\right) =\dfrac{e^{\frac{u\left( a\right) }{\lambda\left( t\right) }+\alpha\left( }}{\sum_{b\in A}e^{\frac{u\left( b\right) }}, \end{align*}$$ where $p_{t}\left( $ is probability that alternative $a$ selected from set $A$ feasible alternatives if $t$ time available to decide, $\lambda$ a time-dependent noise parameter measuring unit cost information, $u$ time-independent utility function, $\alpha$ an alternative-specific bias determines initial (reflecting prior information memory anchoring). Our analysis provides behavioural foundation (also known as Multinomial Logit Model when constant). neuro-computational derivation biologically inspired algorithm may explain emergence in behaviour. Jointly, approaches thorough understanding softmaximization terms internal causes (neuro-physiological mechanisms) external effects (testable implications).