Professor Paul Garthwaite
The Open University (UK)
Fri, 25/02/2011 - 4:00pm to 5:00pm
Statistics rarely tries to draw conclusions from a single individual. However, in neuropsychology an individual with brain injury may display unusual combinations of abilities that give insight into the architecture of the brain - perhaps showing that two tasks that appear similar are, in fact, performed by different parts of the brain as the brain injury has impaired performance on one task but not the other. Clinical neuropsychologists also need to evaluate whether a patient has a deficit on some specified task, asking the question "How extreme are the patient's scores compared with the scores that would be obtained in the general population?" This talk addresses the task of estimating the proportion of the general population that would obtain a lower score (or more extreme score combination) than an individual obtained. The task seems simple but it is important to do it well because it arises so often. Both frequentist and Bayesian methods are proposed along with methods that aim to have good frequentist properties while using Bayesian ideas. The work illustrates that agreement between frequentist and Bayesian methods is much easier to obtain in one-parameter problems than in multiparameter problems.