An analysis of the first two waves of the German mobility panel (which is still in an experimental stage) has produced three basic results. In spite of the motivation of the interviewees, some mobility data-sets showed incomplete or wrong information. Methods and algorithms are shown which can improve the quality and completeness of the data. A software tool is presented, which makes it possible to visualise the mobility and activity behaviour in the course of one week in order to identify typical errors (detection of faults or missing trips) or to correct and complete the data by comparing with other days or other household members. Other methods are discussed which can be used for the same purpose (neural networks, same purpose (neural networks, additional information) and probability measures for any corrections/completions. Resulting from the typical mistakes, some hints are given for improving the design of the panel survey. The results of the corrected and completed data showed a considerable difference in the mobility level compared with the results of the cross-sectional surveys of the KONTIV design. The reasons for the differences are analyzed (randomly chosen interviewees versus motivated people, completion of missing trips, correction of missing values, compensation of attrition biases in the panel data). The results help to understand the reasons why the classical cross-section surveys tend to underestimate the amount of mobility and show in which way cross-sectional and longitudinal mobility surveys can be used to complement each other. Panel data give an insight into internal behavioural changes within a population which remain hidden by the comparison of the net differences between two cross-sectional surveys performed at different times.
Abstract