Using Measurement error model to access the effect of metal pollution to Species in Arkansas River
| Yao Huang , M.S. Candidate
Department of Statistics, Colorado State University
3:00 p.m. January 29, 2008
Ecotoxicology aims to offer data, tools, and methods to assess the impact of pollutants on natural populations. In the research field of ecotoxicology, it is important to propose proper models based on the collected data to access the effect of pollution on natural populations.
Many useful models in ecotoxicology field have been able to contribute to pollutant management for environment preservation and conservation biology. However, to date, there are no models considering measurement errors in which should be a common problem in the ecological data collection. Measurement errors will cause biased coefficient estimations and distract the true effects of pollution to the natural species.
In this paper, we propose a two-step procedure to fit different models with measurement errors. In the first step, we make use of the auxiliary variable information to denoise the associated variables with measurement errors by kernel-type or spline smoother. In the second step, we fit the models based on these denoised variables. Compared to the traditional directly modeling method without considering the measurement errors in these variables, the simulation results show better performance of the two-step procedure. Moreover, the two-step procedure is applied to assess the effect of metal pollution to the metal-sensitive species HEPTAG in Arkansas River from 1989 to 2004. In this real case, three different models such as parametric, nonparametric and semi-parametric models are investigated. For each model, we fit two different versions with and without considering measurement errors. Our two-step procedure performs uniformly better among the version comparisons. There are also some problems needed for further research such as the asymptotic properties of our two-step procedure estimators.