Daniel Hernandez-Stumpfhauser, Preliminary Exam, Department of Statistics, Colorado State University
Thursday, February 3, 2011
9:00 a.m., room 006, Statistics Bldg
Typically, when dealing with survey data only summary information about the construction of survey weights is available. The weights can be inverses of inclusion probabilities, or they can also include calibration. Also, second order inclusion probabilities are typically unknown. In making inference about model parameters, variance estimators can be asymptotically biased if calibration variables are ignored. This effect is shown for several variance estimators for specific models.
An estimator was created for the distribution of time of day when anglers leave the fishing site for combinations of state, wave (time of the year), and mode (mode of fishing) based on the individual fishing trip information of coastal households collected during the MRFSS (Marine Recreational Fisheries Statistics Survey) telephone surveys during the years 1990 to 2008. The estimator is a convex combination of a direct estimator and an indirect estimator, this last one based on a Bayesian linear model for circular data using the projected normal distribution.
Dr. Jean Opsomer (Advisor)
Dr. Jay Breidt (Committee member)
Dr. Jennifer Hoeting (Committee member)
Dr. Sonia Kreidenweis (Outside committee member)