Single or combined location measurements of the same parameter without prior probability. (Altern. title, Parametric inference as useful illusion; Part 1)
mag(2002)
摘要
Motivation. This version is based solely on the calculus of probability,
excluding any statistical principle. "Location measurement" means the pdf of
the error is known. When the datum is obtained, intuition suggests something
like a pdf for the parameter; here we attempt a critical examination of its
meaning.
Summary. In default of prior probability the parameter is not defined as a
random variable, hence there can be no genuine prior-free parametric inference.
Nevertheless prior-free predictive inference regarding any future datum is
generated directly from the datum of a location measurement. Such inference
turns out as if obtained from a certain pdf ("fiducial") indirectly associated
with the parameter. This false pdf can expedite predictive inference, but is
inappropriate in the analysis of combined measurements (unless they all are
location measurements of the same parameter). Also it has the same distribution
as the ostensible Bayesian posterior from a uniform "prior". However, if any of
these spurious entities is admitted in the analysis, inconsistent results
follow. When we combine measurements, we find that the quantisation errors,
inevitable in data recording, must be taken into consideration. These errors
cannot be folded into predictive inference in an exact sense; that is, we
cannot render a predictive distribution of a future datum except as an
approximation.
Keywords: location measurement; combination of observations; parametric
inference; predictive inference; prior-free inference; quantisation error;
digitisation; frequentist interpretation; the fiducial argument; fiducial
probability; pivotal inference; intuitive assessment; prior-free assessment
更多查看译文
关键词
random variable,predictive inference,data analysis,predictive distribution
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络
Chat Paper
正在生成论文摘要