The multinomial logit model with last choice feedback

This paper proposes a method to develop the multinomial logit(MNL) model for a sequence of choices behavior that incorporates the probability of last choice in the current representative utility.The function of last probability could be additionally used as the coefficients of some part of observed component, while other alternative-specific coefficients are assumed to be variable only with their initial coefficients which are derived from the MNL model.Without loss of simplicity and readily interpretation, the model reflects the impact of last choice on observed component via last probability.The experiment based on synthetic data shows the fact that a person is aware of a alternative is willing to choose the alternative in following periods.The probability approaches a constant, and the bigger the proportion of observed component that influenced by last probability is, the higher the final probability become.