FORECASTING EMPLOYMENT & POPULATION IN TEXAS: An Investigation on TELUM Requirements, Assumptions, and Results, including a Study of Zone Size Effects, for the Austin and Waco Regions

The FHWA’s publicly available Transportation, Economic, and Land Use Model (TELUM) predicts the future spatial distributions of employment and households using Putman’s Integrated Transportation Land Use Package (ITLUP) equations. TELUM was used here to generate the employment and household location forecasts for the Austin and Waco regions by district. While reasonably well documented, TELUM remains something of a black box, in that the research team could not duplicate its parameter predictions or its forecasts. It also is restrictive in application, requiring application at the district level, rather than any zone size (such as the more common, and smaller, traffic analysis zones [TAZs]). The research team developed its own “open source” Matlab code, which will be referred to as Gravity Land Use Model (G-LUM). G-LUM is based on Putman’s documentation of ITLUP equations as well as Caliper’s own (not yet publicly available) ITLUP user manual. This was done in order to provide transparency, try to corroborate TELUM’s results, and overcome TELUM’s zone size restrictions. Parameters for all ITLUP equations (five for each household type, five for each employment type, and 19 for the land consumption model) are estimated by solving a non-convex, non-linear optimization problem, which maximizes the entropy and thus the likelihood of the data. In such cases the solution algorithm can get trapped at a local optimum, so our code solves the calibration problem using different starting assumptions on parameter values, thus reducing the chance of non-globally optimal solutions. TELUM does not address this issue. The estimation results of our code differ from those of TELUM and yield higher entropy values. An analysis of the two models’ forecasts, from 2005 through 2030 (in 5year increments), for the Austin and Waco regions is presented here. TELUM’s employment and household predictions differ significantly from those of the G-LUM code for the Austin region. In the case of Waco, many similarities exist in the employment forecasts, across the models, but household forecasts differ considerably. Such distinctions engender analyst hesitation in pursuing the use of already compiled code, such as TELUM’s. More transparency is needed, to deduce the source of such distinctions. The influence of zone size on the forecast produced by the ITLUP equations also was investigated in this study. Future forecasts of the spatial distribution of employment and households by TAZ were obtained by using our own code for the Austin and Waco regions. These forecasts were aggregated by districts and then compared with the district-based forecasts. The comparison showed some stark differences. For example, in the case of Waco, forecasts by district showed more total employment in the eastern and western parts, while forecasts by TAZ showed more total employment in southern parts of the region. Also the spatial distribution of low income households in Austin was completely different for districtand TAZ-based forecasts. This report contains detailed descriptions and illustrations of applying both the TELUM and GLUM to the three-county Austin region and the Waco region. The report begins with background on the ITLUP model, and then results from calibration are discussed as well as results from model runs. A summary of findings, including advantages and drawbacks of the model, concludes the report. Table of