Seminar
Juli Atherton
Bayesian optimal design for multi-path changepoint problems
In many applied problems it is anticipated that n observations taken over a time interval will undergo
a change in their distribution at some unknown instant called a changepoint. For example, suppose that
a sequence of baseline blood pressure readings will be taken on a patient, and a treatment administered
which is to be followed by further blood pressure readings. If it were anticipated that the patient would
experience a drop in mean blood pressure at some unknown time (changepoint) after the treatment, our concern
might be to estimate the pre-and-post changepoint means. An important design issue is where to take the
n readings to "best" estimate these two means. I will discuss, in a more general setting, optimal designs
for such problems. Such issues may be useful in the design of clinical trials where the
time-to-treatment-effect is unknown.
