We present a framework to identify factors affecting loss aversion in a stated choice experiment where respondents trade travel time for money. Other studies have shown how to use stated choice data to estimate loss aversion separately from the marginal rate of substitution between travel time and money, but have not investigated how loss aversion varies with individual characteristics and features of the experimental design. The main contribution of the paper is thus to fill this gap.

A major problem when modelling reference-dependent preferences is to establish what the reference point is. In our case, the design of the stated choice experiment provides a natural candidate: In one experiment, all choice alternatives are variations of a base trip the respondent has recently made and is instructed to describe in detail in the beginning of the experiment. The explicit focus on the base trip makes it a likely reference point in the following choice situations. In another experiment, the respondent is asked to describe the base trip as it would have been if another transport mode was used instead - this trip is a likely reference point in the following choice situations, where choice alternatives are trips using the alternative transport mode.

Using a discrete choice model, we model observed behaviour as a function of the marginal rate of substitution between travel time and money (which is the value of travel time, VTT), and the perceived values of time/money savings and losses. We allow for asymmetry between gains and losses by means of piecewise linear value functions that map absolute attribute values to perceived values. The degree of loss aversion is determined by the degree of asymmetry between gain and loss valuations, which is parameterised as a function of socioeconomic variables and design characteristics. For estimation, we use a fixed effects logit estimator, reducing the need to make assumptions concerning the distribution of unobserved heterogeneity in the VTT.

Overall, we find significant loss aversion in both the time and cost dimensions. Loss aversion in the time dimension is larger than in the cost dimension, and depends on age, education, income, and gender, while we find no significant effect of occupation. We also find evidence suggesting that loss aversion depends on how well-established the reference point is: Loss aversion is smaller in the experiment where the reference point is the hypothetical version of the base trip, and smaller yet if this hypothetical trip is one the respondent rarely makes.