The attribute X's included those that required carriers' immediate attention as identified in the second stage survey. After formulating the choice model, the next step of experiment design was to specify attribute levels to be combined to describe alternative service bundles inchoice situations. To enhance the realism of the survey, selected attribute levels should cover a range similar to that actually experienced by shippers. Both the highest possible level and the lowest possible level should be included.Typically two and five levels should be assigned to each attribute but the same number
of levels should be included for each attribute to avoid bias in the estimated importance (Lilien and Rangaswamy, 2003). In our study,three levels were assigned to each attribute. Given the data provided by the shippers participating in our survey, the plausible values for attributes to be used in constructing service bundles were shown in Tables 3 and 4.
Having established the relevant factors and their levels, the next step was to develop hypothetical service bundles by combining attribute levels presented in the above table. All possible combination of these attribute levels gave rise to a total of 81 (34) possible service bundles for the delivery of automotive parts and 243 (35) bundles for the delivery of
consumer goods. Obviously, these were too many for us to handle effectively. We therefore applied a fractional factorial
design to reduce the number of service bundles to a manageable level while maintaining an orthogonal main-effects design. Using this design, the numbers of service bundles were significantly reduced to 9 and 16 for the delivery of automotive parts and consumer goods respectively. These descriptions of trucking service are shown in Tables
5 and 6