Usually in Stated Preference (SP) surveys a number of choice observations is taken from each individual to reduce the cost of data collection. This raises the so-called "repeated measurements" problem. Observations taken from the same individual are not independent, which means that simple analysis methods may be biased. In this paper the application of the Jackknife and the Bootstrap techniques to attack this problem are described in the case of a simple Logit model, and their relative performance compared. The Jackknife and Bootstrap are used to investigate bias due to serial correlation, the influence of the size of the Jackknife sample, the influence of the error distribution for each respondent respondent individually and across respondents, and the variance estimate of the model coefficients obtained in the course of Maximum-likelihood estimation. The estimation of variance in Logit models is important in the assessment of the accuracy of estimated parameters based on SP data. These procedures also have applications in joint SP-RP modelling.
Abstract