Accuracy of Zonal Socioeconomic Forecasts for Travel Demand Modeling: Retrospective Case Study

Abstract

Modeling for urban travel demand begins with 20-year forecasts of population, households, vehicle ownership, and employment for a region's individual transportation analysis zones. Yet even though these models rely on socioeconomic forecasts, the long-term accuracy of such models has not received attention, especially for smaller regions with limited planning staff. This paper reports on a case study of the socioeconomic predictions made in 1980 for a horizon year of 2000, by comparing predicted and actual results in Lynchburg, Virginia. The region percentage error reflects the difference between forecast and observed values for the entire region. Although regional forecasts for the number of vehicles and employment showed errors of less than 10%, those forecasts for population and households showed errors of 48% and 14%, respectively. The failure of planned development in two of the region's 68 zones accounted for much of this error, such that removal of these two zones lowered population and household region percentage errors to 10% and 1%, respectively. The zone percentage error is the average of all individual zone percentage errors. Even after removal of the two aforementioned zones, population, households, vehicles, and employment had strikingly large zone percent errors of 39%, 48%, 45%, and 136%, respectively. These results make a compelling case for executing the regional travel demand model twice: once with the given socioeconomic forecasts and once with forecasts modified on the basis of expected errors. For regions that have not conducted an assessment such as that presented here, the expected errors from this paper may be used.

Large-scale spatial population synthesis for Denmark

“…This may reflect the development of new urban settlements not reflected in the first-stage fitting. Not surprisingly, the prediction error is significantly smaller per year than has been observed in Krishnamurthy S, Kockelman KM [21] and in McCray et al [24]. However, these studies considered the total observed error and also included uncertainty in the population targets.…”

“…This gives an indication of a "best-case" benchmark for the synthesizer. Others have looked at various sensitivity investigations when targets and the starting solutions have been varied [24]. However, this in itself involves a rather comprehensive Monte-Carlo scheme which may not be of any particular interest in general and is certainly outside the scope of the present paper.…”

“…Step 2: Set k = k + 1 and iterate eq. 17- (24) below. P kþ1 i; l; f ; c; zjz 0 ; a; g ð Þ ¼ T k a; g; i; l; f ; c; z ð Þ P i;l; f ;c;z∈z 0 T k a; g; i; l; f ; c; z ð Þ ð17Þ…”

“…That is, the second-stage IPF is carried out by iterating (17)- (24) and 29 Another option which is also supported by the model is to superimpose the Q(z) target such that it is not adjusted in the harmonisation stage. In this case we simply impose the level provided by the target to the conditional distribution for the population for the specific cell.…”

Forecast accuracy of demand for registered nurses and its determinants in South Korea

“…Studies have reported findings on nurse demand and supply forecasts, but they have been insufficient for evaluating the forecasting of nurses. Post hoc tests of forecasting accuracy effectively help individuals understand forecasts and results, measure the availability of data, and make policy decisions [ 16 , 17 ]. We aimed to assess forecasting accuracy and investigate its determinants through post-evaluation of demand forecasts reported in past studies in terms of demand–supply forecasting.…”