When we estimate the multivariate mean of a random vector of where we believe the individual random variables are meaningfully related, shrinkage estimators are known to improve on the sample mean when the size of the random vector is at least 3. This talk shows that we can improve on the best-known shrinkage estimator by shrinking towards a weighted mean of the random variables instead of their mean. It shows the new estimator has the properties of scale and translation invariance (measuring in Celsius instead of Fahrenheit won’t change the result) and can be used even if the covariance matrix of the random variables is singular.
The seminar will start at 4
pm in Room 115,
Health Sciences Building room 115, Foresterhill Campus, University of Aberdeen with tea and biscuits from 3.30
pm. If you wish to join online and do not already have a Teams invitation, please email
Highlandslg@rss.org.uk and we will add you to the seminar invitation.
John D. Lamb, University of Aberdeen