ABSTRACT
Coarsening At Random is not quite everything
Eric Cator
For a while now, CAR, i.e. Coarsening At Random, has been a popular model
assumption. It prescribes a certain class of allowed random coarsening
mechanisms of the data, in such a way that the likelihood of the data
factorises and one can apply maximum likelihood on the parameter of
interest. It has been conjectured that the CAR-assumption cannot be tested,
but we will show that this conjecture is false even in the well known case
of current status data. We will also present a theorem stating exactly when
the CAR-assumption cannot be tested, i.e. the model allows a dense set of
distributions on the dataspace.
Back to the history of the seminar
or the Colloquium Stochastiek homepage.
Martijn Pistorius
(pistorius@math.uu.nl)
Last Updated: November 19, 2001