The current way to estimate value of travel time is to use a mode-specific sample and hence to estimate mode-specific value of travel times. This approach raises certain questions concerning how to generalise the values to a population. A problem would be if there is an uncontrolled sample selection mechanism. This is the case if there is correlation between mode choice and the value of travel time that is not controlled for by explanatory variables.

What could confuse the estimated values is the difficulty to separate mode effects from user effect. An example would be the effect of income which is a user effect. Suppose all high income individuals use car while low income individuals use bus. Then the effect of income would be reflected in the values for car and bus, and not as a user effect.

            Here we present a model that explicitly distinguishes mode effects from user effects. This makes the model useful for representing a general population with no specific mode. The cost of this approach is that we introduce an endogenous mode dummy. We therefore have to deal with this in our estimation. We use two different ways to deal with this endogeneity inspired by Lewbel (2004).

            To model the value of travel time we use a stated choice dataset. These data include binary choice within mode for car and bus. The first approach is to use a probit model to model mode choice using instruments and then use this in the estimation of the value of travel time. The second approach is based on the use of a very exogenous regressor.