Markov chain models

maybe some (smoothed) logistic regression (bit more data on
non-succeeding players than the IFDA project, but some worries about
selection bias