Statistics as a Distillation of Everyday Experience
Gerald van Belle
Professor, Department of Biostatistics
Professor and Chair Emeritus, Department of Environmental and Occupational Health Sciences
University of Washington
Seattle, WA, 98195-7232
The basic aim of statistics is to deal with variation and causation. In this presentation I argue that these are also basic considerations of human experience. Every day we face variation and issues of causation. Statistics is a distillation of everyday experience and, in return, provides a way for dealing with causation and variation.
Part I of this presentation gives everyday examples of variation and causation. Specific topics include description of variation in everyday life; the concept of collective or population behind variation; variation in time, space, and social structure; the challenge of selection in the face of variation; controlling variation; and inducing variation. Examples are given where improper handling of variation has led to misinterpretation of the real world.
With respect to causation I argue that humans are “hard-wired” to look for causes, to immediately interpret association as causation, and think of a chain of causation. Again examples from everyday experience will be given.
Part I concludes with the special challenge of dealing with causation in the face of variation. A brief contrast between observational and experimental studies concludes this section.
Part II deals with an application of the above considerations to consider the concept of “normal aging.” I begin by listing four questions that every scientific investigation needs to ask: (1) What is the question, (2) Is it testable, (3) Where will I get the data; (4) What are the data telling me? This immediately suggests that “normal aging” is too vague to be investigated. I will reduce the question to a consideration of cognitive change and aging. The data set is a cohort study (Adult Changes in Thought, ACT) of 2500 non-demented individuals, aged 65 and older, recruited in 1996 at the Group Health Cooperative in Seattle and continuously followed. Cognition is measured by the Cognitive Assessment Screening Instrument (CASI) obtained every two years. Incident dementia is determined by consensus conference.
The analysis assumes that cognition is a latent trait estimated by the CASI score. Some aspects of a latent trait analysis are discussed.Data are presented comparing the cognitive trajectory of ultimately demented subjects and those not demented at the last evaluation. The conclusion is that the cognitive trajectory in non-demented adults is relatively stable over a long period of time. In contrast subjects ultimately diagnosed with dementia show a steady decline in cognitive functioning several years before being diagnosed.
Part III reviews the strength of inference of Part II in view of the issues raised in Part I. In particular, where did the data come from? The recruitment into the cohort was a two-stage design leading to problems of regression to the mean in the face of measurement error. Other issues are discussed as well.
Conclusion. This presentation has argued that statistics is a distillation of everyday experience. It returns the compliment by providing ways of dealing with experience using everyday statistics.
Acknowledgements.
Eric Larson for permission to use the ACT data. Paul Crane for now mentoring me in the area of Item Response Theory. Doug Tommet for carrying out most of the analyses of Part II. My colleague, Lloyd Fisher, for 30 years of support and intellectual stimulation. Colleagues in the department of Biostatistics, and the department of Environmental and Occupational Health Sciences at the University of Washington.