Application of Poisson Regression Models to Activity Frequency Analysis and Prediction

Abstract

The Poisson regression model and its variants are used to estimate individuals’ daily activity frequencies by activity type using the first 4 years of the Puget Sound Transportation Panel. The estimated model coefficients are applied to the 5th-year observed data to predict daily activity frequencies by activity type. Forecasting accuracy is measured with indicators of deviation between observed and predicted values. Examples of model estimates and their forecasting performance are provided. Comparisons between the observed and predicted 5th-year data show the predictions to be fairly accurate. Subsistence activity is the most accurate among all types, followed by trip chains, out-of-home leisure, and maintenance activities. The analysis also indicates that different theoretical distributions should be used for different dependent variables.