faq 95911983

Typical Scoring Function Parameters and their Calibration

by Zahra Navidi on 2016-11-25 09:42:09

---

Dear all,

In page 32, the last step of calibrating the scoring function according to choice model is as follows:

"Set all other marginal utilities of travel time by mode relative to the car value.

For example, if your logit model says something like:

...-6/h.tt(car) - 7/h.tt(pt)...,

then

β(dur) = 6, β(tt,car) = 0, β(tt,pt) = -1,"

Now my question is what if β(tt,car) > β(tt,pt) (for example, the logit model is somehing like ...-7/h.tt(car) - 6/h.tt(pt)...,)? in that case,  β(tt,pt) = +1, which conceptually means traveling PT is desirable and increases the utility, which is wrong. So, I was wondering what would be the solution?

Does it make sense to assign the positive value of the smallest negative β to β(dur)? in this case β(tt,pt)? what if the smallest value was the coefficient of waiting time for PT?

Thank you

---

Comments: 2

---

Re: Typical Scoring Function Parameters and their Calibration

by Kai Nagel on 2016-11-25 21:44:52

Dear Zahra,

You are on the right track.  I would in your situation try

beta_perf = 6  (= from the mode with the least negative beta)

beta_pt = 0 (= the mode with the least negative beta)

beta_car=-1 (= offsets for all modes that are less agreable than the "best" mode)

etc.

Does this make sense?  If not, then there is either a typo or a knot in my head.

---

Recall that travel is also punished by the opportunity cost of time.  This is approximated by beta_perf (see the book).  So even if beta_pt were +1, together with beta_perf = 6 this would end up as an (approximate) penalty of -5/h.

Best wishes Kai

---

Re: Typical Scoring Function Parameters and their Calibration

by Zahra Navidi on 2016-12-01 11:10:15

Thank you vey much Kai.

To follow up with my previous question, my model is like:

-7/h.tt(car) -6/h.tt(pt) -5/h.tt(ptWait)  

So, running my scenario with the following numbers give me the correct mode share:

beta_perf = 7

beta_wait = -5

beta_pt = -6

beta_car = -7

However, running it with the following number result in a very different mode share:

beta_perf = 5

beta_wait = 0

beta_pt = -1

beta_car = -2

I understand that the first set of numbers are not the correct way of calibrating scoring function according to a choice model. However, I can't explain why it gives me such a perfect mode share. So, I'd like to ask your opinion about this, is it just a coincidence or is there a way to explain what's happening.

Thank you very much.