Answers are grounded in this paper and cite their sources.
The paper explores statistical information properties in estimation, focusing on likelihood functions and observations.
In the theory of statistical estimation put forward by Fisher, an essential feature is the concept of statistical information, defined for estimates normally distributed as the reciprocal of their sampling variance, and more generally by the variance of ∂L/∂θ, where L is the logarithm of the likelihood function corresponding to our estimate of a parameterθ. The expected value (∂L/∂θ) is zero, so that we may write I = E(∂L/∂θ)2. (1) If ∂L/∂θ corresponds instead to the original sample of n independent observations from which our estimate was derived, its variance will be n times the information per observation, since ∂L/∂θ is then the sum of n similar components, and (1) will give the information in our sample.
No takes yet. Share an insight, caveat, or question.
Maurice Stevenson Bartlett (1936) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: