This paper aims to estimate a Random Coefficient Logit model. This is a generalisation of the Standard Logit Model, and it is shown to be analogous to both the Mixed Logit Model and Error Component Logit Model. The development of the Random Coefficient Logit model was motivated by the effort to overcome the assumptions existing in the Standard Logit Model. The Random Coefficient Logit Model allows parameters to vary randomly across respondents and it does not exhibit the restrictive independence from irrelevant alternatives property as a result. Also, it allows efficient estimation when there are repeated choices by the same respondents by inducing the correlations in unobserved utility (error component) over repeated choices. We estimated both models with normal and log normal distributions for the coefficients of cost, time, reliability and alternative specific constants. Results show that the mean coefficients in the Random Coefficient Logit Model are consistently larger than those of the Standard Logit Model. This implies that the variances in the error term in the Standard Logit Model is greater than the variance in the extreme value component of the error term in the Random Coefficient Logit Mode. Also the estimated standard deviations of coefficients are highly significant, indicating that parameters do indeed vary among the population.
Abstract