The path to model calibration is getting easier as more calibration aids and techniques are developed. The automatic methods available in the full-featured distribution and assignment now handle externals calibration for you. The next step, in theory, is to get the screenline and cordon line crossings under control. These evaluate major directional movements. The screen line analysis module has long been available and works easily and quickly.

However, when screenlines are not acceptable (or even when they are) it is often necessary to proceed to a more detailed step to assess the situation. Volume comparisons on individual links of ground counts versus assigned volumes can point out where and what kind of problems exist, which can lead you to quicker and easier solutions. While we have been able to make these comparisons for years, TMODEL2 has been updated to make the comparison process and the evaluation of the comparisons much easier and more effective.

Menu choice 1.6.5 Allowable Link Calibration Errors, has new capabilities which evaluate volume comparisons. The choices are too varied to cover completely in this article. Therefore, we will paint a broad-brush scenario, and you can delve into the rest in your update documentation. Improvements such as these, by the way, are always coming out. This is why participation in the Maintenance Program is so valuable for those of you actively modeling.

In NCHRP 255 there is a graph on page 49 captioned "allowable link calibration errors." This graph shows, for example, that for a link with an ADT volume of 18,000, an allowable calibration error could be up to 22%. It suggests that this level of accuracy will produce a forecast which will not cause a facility to be misplanned. Because this is the only "federal guidelines" we have, it is incorporated as a standard into the analyses of link assignment errors.

From the Allowable Link Calibration Errors menu, choosing 1) will load a file containing ground counts and automatically insert in memory the maximum allowable percent deviations in the Select Zone 1 fields and maximum allowable absolute volume deviations in the Select Zone 2 fields. These values are taken from a look-up table derived from the NCHRP 255 graph.

Next, choosing 3) will load a loaded link file (.LLX) which resulted from a D&A run and will, again in memory, automatically compare the volume with the ground count and replace the assigned volumes with the differences between the assigned volume and the ground count. In the Select Zone 1 fields errors will be shown only for those links that exceed the maximum allowable error. Select Zone 3 will again receive the maximum allowable absolute volume deviations.

This file can be saved and viewed using the Network Graphic Editor. The volumes columns will show all the differences, and the SZ1 columns will point out the major sources of error.

Before that, though, choice 5) will display system measures of the comparison. Shown below is a scattergram of assignment volumes versus ground counts for each link where a count exists. Rather than assignment volume, this scattergram can also be plotted in terms of percent errors or of volume errors. These links can, as usual, be limited by Class, Area, and/or Type.

In addition to the points plotted, the desired "goal" line is plotted as well as curves derived from the allowable errors look-up table. Points which fall outside of the curves are problematic. The regression line for the assignment volume data (or percent error or volume error) is graphed as a dashed line. This regression line can show you that, while your overall percent error might be good, it may be that lower volume links are being over assigned and higher ones under assigned or vice versa. This can direct you to a better specification of variables or parameters which affect this distribution such as capacities or capacity constraint equations, or time and distance weights, or gravity model parameters.

Also calculated and displayed are some statistics such as the average value of the Y-axis variable (volume or percent error or volume error), the percent root mean squared error (%RMSE), the percent of the assignment points which fall outside of the allowable curves, and the R² of the regression line. The %RMSE is analogous to standard deviation from the "goal" line rather than from an average. The R² shows the goodness of fit for the regression line. If that regression line is not close to the "goal" line, it does not represent goodness of assignment.

Examining existing models, especially for the arterial and collector class roadways, may show some frightening assignment results. However, having this sophisticated tool should help us all do better models in the future.