Browsing by Author "Marc G. Genton, Committee Member"

Now showing 1 - 3 of 3

Controlling Variable Selection By the Addition of Pseudo-Variables
(2004-08-09) Wu, Yujun; Marc G. Genton, Committee Member; Leonard A. Stefanski, Committee Co-Chair; Dennis D. Boos, Committee Co-Chair; Hao Helen Zhang, Committee Member
Many variable selection procedures have been developed in the literature for linear regression models. We propose a new and general approach, the False Selection Rate (FSR) method, to control variable selection with the advantage of being applicable to a broader class of regression models; for example, binary regression, Poisson regression, etc. By adding a number of pseudo-variables to the real set of data and monitoring the proportion of pseudo-variables falsely selected in the model, we are able to control the model false selection rate, selecting as many important variables as possible while selecting a relatively low proportion of false important ones. We focus on forward selection because it is applicable in the case where there are more variables than observations. Due to the difficulty of obtaining analytical results, we study our approach by Monte Carlo and compare it with a variety of commonly used procedures. We first focus on linear regression models, and then extend the approach to logistic regression models. The new method is illustrated on four real data sets.
The Economic Effects of Federal Peanut Policy: The 1996 FAIR Act, the 2001 Farm Security Act, and the Federal Crop Insurance Program
(2002-08-01) Chvosta, Jan; Walter N. Thurman, Committee Chair; Blake A. Brown, Committee Member; Matthew T. Holt, Committee Member; Marc G. Genton, Committee Member
Government programs that restrict production and increase prices to particular groups of producers have a long history in the United States. The purpose of this research is to analyze the implications of such a program for peanuts in three independent essays. The first essay focuses on the development of a model of the effects of cross-county transfers on peanut quota after the 1996 farm bill. Using a spatial linear regression model, the hypothesis that the lifting of transfer restrictions tends to equilibrate lease rates across counties is tested. The results indicate that, after the 1996 bill, peanut quota moved out of counties that under produce their quota to overproducing counties, indirectly indicating a tendency for lease rates to equalize. The second essay studies the most recent changes to the peanut program, enacted by the U.S. Congress in 2002, and reviews important events that led to these changes. Several models are developed that analyze the costs and benefits of the revised program in domestic and foreign markets. It is concluded that farmers in most peanut producing states will incur losses due to the peanut program changes, with the exception of Texas and Florida. The impact of the transformation on the world price of edible peanuts is analyzed and shown to be theoretically ambiguous-- the world price could either increase or decrease depending on demand and supply elasticities. The essay explores numerically the influence of the relevant elasticities. The third essay reviews the U.S. federal crop insurance program and investigates its interaction with the peanut program. A model of a risk neutral profit maximizing farmer is developed and comparative static results are derived. The results show that in equilibrium peanut quota lease rates do not represent the full difference between the support price and world price and are affected by the cost of crop insurance.
Modeling and Prediction of Nonstationary Spatial Environmental Processes
(2002-08-19) Barber, Jarrett Jay; Montserrat Fuentes, Committee Chair; Peter Bloomfield, Committee Member; Jerry M. Davis, Committee Member; Marc G. Genton, Committee Member; Marcia L. Gumpertz, Committee Member
Spatial data are often collected for the purpose of producing spatial predictions (i.e., maps), the accuracy of which relies on a good estimate of the spatial covariance. Traditional geostatistical methods for spatial interpolation assume covariance stationarity. However, spatial data often exhibit nonstationary covariance, and traditional methods can produce maps that are misleading. Some existing approaches to nonstationarity feature process models which lead naturally to a globally defined covariance but do not retain a familiar interpretation in terms of local stationarity, while other approaches focus on local stationarity but rely on ad hoc methods for calculating covariance. We present a different approach with a relatively simple but useful model for space-time data. The model is simultaneously defined everywhere (globally) and leads immediately to a globally defined covariance, and, locally, the model behaves like a stationary process. A nonparametric approach to estimating the nonstationary spatial covariance is presented along with some asymptotic properties. The approach is particularly suited to time-rich, spatially-sparse networks. We illustrate this nonparametric approach for spatial prediction of atmospheric pollution data collected periodically from an EPA environmental monitoring network. We also propose an alternative, parametric approach to estimation and prediction using a Bayesian formulation of a nonstationary spatial model.