This year's Montana chapter meeting of the American Statistical Association will be held on Friday, September 29th at the Procrastinator Theatre on Montana State University's campus. We would love for you to join us!
Abstract: In response to the spread of White-nose syndrome (WNS) and SARS-CoV-2, ecologists and wildlife managers have renewed interest in assessing the impact of disease on bat populations in North America. Historically, this assessment has been made by first modeling disease spread, then modeling species occupancy using the estimated disease spread as a covariate in the ecological process. This two-step approach does not acknowledge the joint information shared by the disease and ecological processes and fails to propagate the uncertainty in estimation of the disease process through to the ecological process, resulting in biased estimates of the effect of the disease on the ecological process. We present a framework capable of jointly modeling a spatial disease process and ecological process, allowing for appropriate propagation of uncertainty throughout the model. Using simulation, we show that the joint model results in more precise estimates of disease prevalence and unbiased estimates of the effect of the disease on the ecological process. Our method development is motivated by acoustic data collected in Montana, USA to assess potential impacts of WNS on susceptible bat species.
Abstract: Missing data is common in data sets in every field of science. There are three types of missingness: Missing Completely at Random (MCAR), Missing at Random (MAR) and Missing Not at Random (MNAR). I will discuss these three types and the methods that one can use to handle them. I will discuss in detail estimation under the MNAR condition which is less well studied. In particular, I will present a robust estimator for the parametric component in a semiparametric propensity model which leads to consistent estimation of outcome mean.
Abstract: After the biofilm: bacterial transfer, infections and hand hygiene
Abstract: Suppose a researcher wants to design an experiment that will either (i) allow the fitting many potential models (e.g., linear or nonlinear models found in the scientific literature) or (ii) be used to explore the design space for subregions that may contain an optimum predicted response. Suppose also the researcher only has enough time and money to collect N experimental trials. Due to the limitations of classical response surface designs in terms of design size and/or dependence on a specified model that can be fitted, a reasonable alternative is for the researcher to find a good N-point "space-filling" design. The N points should provide good coverage of the design space for combinations of the experimental variables and will allow for the fitting of many models. The design generating procedure begins by generating a set of space-filling points in a unit hypercube which are then transformed to the scale of the original design variables to generate a design the researcher will implement. Three methods are presented for generating good space-filling designs in hypercubes. These methods will be extended for applications in spherical design spaces and to designs of mixtures in the simplex and subspaces of the simplex.