On-site Poisson Regression Modeling for Forecasting the Number of Visitors to City Center Retail Environment and Its Evaluation

Abstract

This paper proposes two types of on-site Poisson regression models to forecast the number of visitors to city center retail environment that correct the choice-based bias due to on-site sampling. One is the modified on-site weighted Poisson regression model and the other the truncated on-site weighted Poisson regression model. While the previous weighted Poisson regression model is formulated to require additional information not obtained from on-site survey, the novelty in these two models is in our proposed method that makes them estimable without any extra information other than on-site survey. We have applied these two models to the actual case of Fukuoka City and evaluated the accuracy of parameter estimates in comparison with those obtained from the usual Poisson regression model based on exogenous sampling data. The major findings are as follows: First, three explanatory variables are strongly significant to account for the number of visitors to the city center. They are the time distance from home to city center, the shop floor area both at city center (destination) and at residential area (origin). Second, the values of parameter estimates obtained from the modified on-site weighted Poisson regression model are accurate enough to be within the range of one standard deviation of those parameters obtained from the usual Poisson model based on exogenous sampling. With these results, the on-site Poisson regression modeling is shown to be an effective and promising method for forecasting the number of visitors to city center retail environment.