Department of Statistics Seminar
North Carolina State University
presents
Drs. Kimberly Weems, Jason Osborne & Kevin Gross
North Carolina State University
A Sampler of NCSU Faculty
- Dr. Kim Weems ksweems@stat.ncsu.edu
Title: Using Instrumental Variable Estimation in
Generalized Linear Measurement Error Models
Abstract: We consider instrumental variable estmation
in a class of generalized linear measurement error
models. The conditional-scores method of Stefanski
and Carroll (1987) is used. We obtain sufficient
statistics for the unobserved predictors and the
conditional distribution of the observed data given
these sufficient statistics. Unbiased score functions
that are free of the unknown predictors are then used
to derive unbiased estimating equations for the model
parameters. This work generalizes that of Buzas and
Stefanski (1996) to nonnormal instrumental variables.
This is joint work with Len Stefanski at the NC State University.
- Dr. Jason Osborne jaosborn@stat.ncsu.edu
Title: Linear Boolean Models for Assessing Particle Flow
Abstract: A probability model is developed for a problem involving measurement of
mass flow in the application of granular fertilizers or pesticides.
The model is developed for type II counter data like those collected
using a sensor system developed by Grift (2001). The sensor
turns on and off as particles flow past it during application
towards a targeted field. Some simplifying assumptions, including
Poisson arrivals of particles to the sensor, lead to linear Boolean
models for the observable passage times of clumps of particles. These
passage times, or clumplengths, contain information about flow rate
and (unobservable) total mass flow. Likelihood and M-estimation methods
are developed. Prediction for total particle flow is also investigated.
When applied to sensor data from experiments using BBs, the model
appears to provide a reasonable fit.
- Dr. Kevin Gross gross@stat.ncsu.edu
Title: Extending the scope of inference for biodiversity - ecosystem function experiments
Abstract: A central task facing ecologists is the need to articulate the ecological consequences of biodiversity loss. This need has spawned a recent collection of experiments that attempt to manipulate biodiversity by assembling communities at random from species drawn from an experimental species pool. Here, I show that customary analyses of data from these experiments only support inferences that extend to populations of other communities composed of species from the same experimental pool. Extending inference to communities that contain species outside the pool requires a new method for estimating the sampling variability attributable to selection of the pool. Preliminary work suggests that resampling techniques may be useful for this problem, but the implementation presents challenges.
This is joint work with Brad Cardinale at the University of Wisconsin - Madison.
Friday, January 23, 2004
3:35 - 4:35 pm
206 Cox Hall
Refreshments will be served on the second floor of Dabney Hall
(left of Room 222) at 3:00 pm.