A Bayesian Uncertainty Analysis for Nonignorable Nonresponse 425
‘burn-in’ and took each iter ate thereafter. We found negligible a utocorrela-
tions among the iterates, and so it is good that ‘thinning’ is not needed. We
have also run the Geweke test of stationarity on µ
o
, µ
1
, µ
2
, µ
3
and τ and found
no evidence of nonstationarity. For numerical summaries, we use the poste-
rior mean (PM), posterior standard deviation (PSD) and CI with coverage
probability 0.95.
Let P denote the finite population prop ortion of Slovenians who would
attend the plebiscite and vote for independence. We show how to make infer-
ence about P under the nonignorable nonresponse model; the idea is the same
under similar models. [10] and [19] made inference about the superpopulation ...