The accessibility-based model’s estimates of trips by mode are generated by applying the mode choice models - estimate based on regional household survey data - to accessibility scores and household characteristics. The resulting estimates generally reflect the trends and tendencies revealed in the survey records. However, when compared to mode shares generated by the regional travel demand model, some discrepancies were observed. Notably, the accessibility-based model tended to estimate higher utilization of transit and non-motorized modes than the regional model.
To ensure consistency between the two modeling frameworks, the accessibility-based mode choice estimates were factored to more closely approximate the regional model’s outputs. The factors (called “K factor” in this analysis) were calibrated such that the LRTP scenario outputs from the accessibility model would resemble the corresponding estimates generated by the regional model. These factors were then applied in the same manner across all scenarios.
Below are several relevant notes for understanding, applying, and interpreting the K factors.
The K factors are provided as a scalar vector in the script that executes the distribution proces (Step 6).
The K factors were estimated based on daily trips. The distribution process also factors trip estimates down to period-specific (AM) trips using period factors that vary by purpose.
Period and K factors are applied when developing trip distribution seeds. In effect, this means the distribution process starts with production-end mode shares matching regional model production-end mode shares for the LRTP scenario (since the factors are calibrated to that scenario).
The distribution process balances the full multimodal trip table, constraining productions and attractions at the person trip level only. Trips by mode are allowed to float while total person trips must match the trip generation outputs.
The attraction-end mode choice models provide a guidepost for seeding the distribution process, which is then run across all modes together with the balancing factors focusing only on total person trips. Thus, the distribution is seeded with expectations about where trips are beginning and ending by mode and uses the trip generation outputs to ensure the right numbers of person trip productions and attractions are maintained through the iterative proportional fitting (IPF) process used for balancing the trip table.
As balancing unfolds, the factoring up or down of certain zones retains the mode splits to/from each zone, but it can result in aggregate shifts in mode shares relative to the mode choice model outputs.
The distribution process described above can dilute the effect of the K factors. For example, the non-motorized trips in the window area hover around 35-40 percent despite the initial factoring to lower those estimates by 5-10 percentage points.