My research interests include multiresponse and model-robust experimental design, statistical disclosure limitation for tabular data, and statistical estimation in angler surveys. See also my CV (updated October 29, 2009).
Multiresponse and model-robust optimal experimental design
For my dissertation, I’m studying multiresponse optimal experimental design and how this may possibly help us to find model-robust experimental designs. Many standard designs exist (i.e. fractional factorial or central composite) and are adequate for most experimental situations. However, such designs are occasionally unacceptable and even impossible to implement. For example, if there is some constraint imposed upon the experimental region (such as in mixture designs), classical designs with cuboidal or spherical design regions are not appropriate. In cases such as these, univariate optimal designs are a well-developed alternative.
But most often a system under study has more than one response of interest, which leads to the concept of multiresponse optimal designs in which the factors have a different functional relationship with each of the responses. The chief criticism of univariate optimal designs is that the functional form must be known in advance, which is rarely the case in practice. This criticism is even more relevant to the multiresponse optimal design situation, and so for such designs to be useful we must find multiresponse model-robust designs.
My advisor is Professors Enrique del Castillo and the chair of my dissertation committee is James Rosenberger.
- Smucker, B. (presenter), del Castillo, E., and Rosenberger, J. “Multiresponse Exchange Algorithms for Model-Robust Experimental Design.” Fall Technical Conference, Indianapolis. October 2009.
Statistical disclosure limitation for tabular data
Statistical disclosure limitation (SDL) attempts to honor pledges of privacy given to individuals by data-gathering agencies while allowing valid statistical inferences to be drawn based on the collected data. For my Master’s thesis, I examined data released in the form of rates (or conditional probabilities) in contingency tables and calculated bounds on cells based upon the released rates and the sample size. Further questions remain, particularly with respect to algorithms which may more efficiently calculate these bounds, as well as to the effect of rounding on the bounds.
This is joint work with my Master’s advisor, Aleksandra Slavković.
- Smucker, B. and Slavković, A. B. (2008). Cell Bounds in Two-Way Contingency Tables Based on Conditional Frequencies. In Domingo-Ferrer, J. and Saygin, Y., editors, PSD 2008. LNCS, Volume 5262, pp. 64-76. Springer-Verlag Berlin Heidelberg. [PDF]
- Smucker, B. (2007). Calculating Cell Bounds in Contingency Tables Based on Conditional Frequencies. Master’s Thesis, The Pennsylvania State University. [PDF]
Statistical estimation in angler surveys
I have worked in the Statistical Consulting Center at Penn State from 2007-2009, and my primary responsibility has been to work on a project for the Pennsylvania Fish and Boat Commission–an extensive survey of the Juniata and Susquehanna rivers which utilized both aerial and interview data collection mechanisms–and to produce desired statistics such as catch rate, daytime angler effort, total catch, etc., along with associated standard errors. Through this process, the shortcomings of standard estimation methods have become apparent and I wish to explore improved estimation methods for such surveys.
