Present and Current Project Descriptions:
In the following we describe the types of problems in which we have expertise. Brief descriptions of some past consulting problems are provided.
- The first area of problems includes sample surveys, sampling methods and design of experiments. We have carried out sample surveys for a national organization over the past three years as well as for Dalhousie. We help with the design of the questionnaire, the sampling design and the interpretation of the results. In most cases we have provided a formal report on the survey results. We have worked with researchers on several large studies in the health domain which involved use of a number of instruments and repeated measures over time. The sampling plan had to accommodate the difficulties caused by individuals moving in and out of the sampling frame. In many studies the issue of sample size has to be addressed. We have carried out the calculations required for a number of complex studies including many in the area of health.
- In many problems the data collected is complex and requires considerable organization before meaningful results can be obtained. We have experience with such issues from handling data collected from school classrooms over a three year period. The students in a class did not remain the same from year to year and we had to develop strategies to handle missing data caused by students leaving or entering the study or simply by being absent when certain tests are run. In a study on epilepsy in children, we had children moving on and off various medications with highly variable time periods. Such complex short time series arise often in clinical settings and require careful thought before being subject to statistical tests. Another problem involves new techniques to measure the thickness of sea ice from measurements obtained along the flight path of a helicopter. We constructed a data base of Cytogenetic data for 17,000 patients with leukemia and lymphoma.
- A number of studies lead to observations repeated in time, whether on individuals as in clinical trials or on biological or physical systems as in ecology or oceanography. We have worked successfully with a number of clinicians over the years analysing data on surgical techniques, kidney transplants, pediatric epilepsy and others in which we have measurements over time on the same individual. In biology we have analysed data arising from a study evaluating treatments for cod sealworm and a study on social behavior in fish. In each case we had to develop non-standard techniques to analyse the data. We have the expertise to handle the more standard time series arising from the observation of physical systems over time where the analysis of spectra is the usual approach.
- In many settings, it is of interest to determine how a particular response is affected by some explanatory variables. For this purpose, data is often modelled using multiple regression and analysis of variance (ANOVA). We have used these techniques for a large number of studies including one on multiple sclerosis in which the response was a measure of mental health and several nursing studies in which the interest was on coping skills. At times the relationship is non-linear and we have to develop specialized methods as in a recent chemistry problem. In most studies of this kind, the issue of model selection is critical. Motivated by several consulting problems, we have developed some robust selection techniques which work well in the presence of outliers.
- In some problems, we have a multivariate response which requires the techniques of multivariate analysis to handle the data effectively. >From work on a study involving bottom grazers in the ocean, we were led to develop techniques which worked well for multivariate analysis of variance when standard assumptions are not met. The Heart Health study was highly multivariate and required that we use dimension reducing techniques to make the analysis manageable. In a more complex setting involving opthalmology, we needed to compare images of the eye for deterioration due to glaucoma. The data consisted of a 256x256 image and we developed some non-standard analysis of variance methods to handle the comparisons.
- In laboratory or production work, the issue of quality assurance is very important. We have worked out quality control procedures in several settings. The first involved a sampling method of the standard alcohol solution used in the breathalyser for the RCMP lab. This method is used in all the labs across Canada and has never been successfully challenged in court. Work on resolving issues on gas meters led to a new area of research for one of our statisticians. More recently we are implementing ISO9000 procedures for a biological company.