This paper considers how measurement errors in the explanatory
variables affects the analysis of a Poisson regression model for frequencies
of recreational and shopping trips. Measurement errors can introduce
bias in the parameter estimates and the effects on this particular
data set and model is investigated. The structure of the data, with two
observations on each individual, makes it desirable to test for correlation
within individual. It is possible that tests of random effects are
sensitive to measurement error. The properties of tests of random individual
effects when there are measurement errors are therefore studied
in the paper. The results of a simulation study show that classical
measurement errors cause severe bias and Berkson measurement errors
produce little bias. The tests for random individual effects work
well both with measurement error and negatively correlated responses
according to the simulation study.
2006. Vol. 9, no 1, 39-48 p.