A Mesoscopic Framework for Modeling and Forecasting Population Age Distribution in Metropolitan Areas

Recent travel demand modeling practices focus on micro, disaggregate, and activity based travel behavior and patterns. The application of such practices requires detailed population data on socio-economic and demographic characteristics. For example, in a four-step travel demand model total household and employment within Traffic Analysis Zones (TAZ) are sufficient to model trip generation. However, in an activity based model more detailed information for each zone, such as population stratified by age and employment type, is desired to generate individual population details for the population synthesis step. In the past two approaches have been used for population estimation: (1) cohort-component method, and (2) population synthesizing approach. The cohort-component method requires detailed birth, migration, and survival information with the results at a macro geographic level. Such data at larger geography (macro level) may not be suitable for advanced travel demand modeling purposes. Population synthesizers are used to estimate individual characteristics required for Activity Based Models (micro-level), but suffer from a lack of evolution of control variables over time. In this paper, a new mesoscopic approach is presented where population by age cohort evolves over multiple time periods using a logistic regression technique. The proposed approach generates control variable which can be used as input to population synthesizers and provides simplistic alternative to a complex demographic evolution. The methodology is presented in three steps: coefficient estimation, forecast and validation. First, the age evolution trend from 1990 to 2000 is analyzed. The estimation result is applied to forecast 2010 and 2030 age composition. This evolutionary model is applied to the Baltimore Metropolitan Council (BMC) based on 1990 and 2000 Census data then validated with 2010 American Communities Survey data. The results show that the proposed model provides reasonable results and is feasible for generating future estimates. The important insights gained from this study are: (1) this model provides good estimation and prediction for the age cohorts 0-24 and 35-64; (2) the 25-34 and 65+ cohorts, prove less predictable as evolution trends among these groups are not consistent over time. The proposed tool can be used by small and large scale planning agencies to prepare detailed socio-economic and demographic profiles for input data into population synthesizer.