Comparing estimation results of land use development models using different databases available in Switzerland

Abstract

Land development models are often the weak part of land use transport interaction models. This is partly due to the in transparent real estate market, e.g. decisions on the realisation of real estate development projects are seldom recorded. We understand the term project in this context as prepared plan for building new built space which materialises in form of buildings. Since we are ultimately interested in explaining development of built space with discrete choice modelling, we would ideally observe the decision of realisation, which happens probably just before applying for construction permit. However, it is more likely that register information on existing buildings are available rather than data on project decisions. If the register contains information on the construction year we can take this as an observation of a development event. One thing we miss with this conception is the fact that sometimes multiple buildings are constructed in the same project. This paper discusses this issues in the context of an implementation of a parcel based UrbanSim land development model implemented for a land use transport model of Canton Zurich in Switzerland. We estimate four discrete choice models of developments choosing their location using UrbanSim. With in the whole modelling system these models have the purpose of updating the stock of real estate which is needed to provide alternatives for further models household or firm location choice models. Each development model is estimated on two differing data basis. One contains construction projects, also with multiple buildings, and the other contains only single buildings as observations. The estimation results are discussed focusing on the impact of disregarding project information. We find that using building registers is a viable option if no data on development projects is at hand, especially if there are few multi building projects in the considered area. However, register information has the disadvantage of not containing proper observations which leads to certain bias. Variables such as the fit to development constraints are especially critical when formulating development location choice models on the level of parcels.