Hi all,I'm currently calibrating the scoring function parameters set out in Section 3.4 for use in a road pricing study, and as such am looking to implement a marginal utility of money (Bm) that is dependent on income.I am wondering if anyone would have advice on whether this is better achieved by (1) adopting a single Bm that is then divided by an individual's income (i.e. single regression using income as a parameter), or (2) using different values for Bm for income groupings (i.e. multiple regressions for data disaggregated by income ranges).The (dis)utility functions would look something like the following:1) Bm = 100 where S = c + (Bm / Income_i) . Cost + etcfor Individual daily income 'i' (100, 125, 160)or2) Bm,i = 1, 0.8, 0.6 for income groups i (low, medium, high)where S = c + Bm,i . Cost + etcAt the moment I am leaning towards approach 1 as it ensures Bm is the same coefficient as that used in the mode choice model that was used to estimate the mode parameters (i.e. marginal utility of travelling). Any advice from those that may have tackled this issue before would be greatly appreciated.Thanks & regards,James
I would prefer (1) for the same reasons you say (one model for everything) but it depends how well your data is approximated by 1/Income. I am not an expert in that area but it seems that the "clean" solution is to estimate both models and see which one has the better fit (i.e. which one explains the data better).
Hi Kai, thank you very much for your response!! I have progressed with the first approach as the cost coefficient using this method was statistically significant, and more broadly the coefficients all made intuitive sense. Thank you again, I appreciate you taking the time to assist.Regards,James Brooks