(33d) Distributed Statistics – Some Challenges

When one estimates the values of the parameters in a knowledge-driven or a data-driven model using a set of experimental data, the steps that need to be followed are well established in the literature. They comprise the minimization of the sum of squares of the prediction errors, the test of significance of each parameter, and the Lack-of-Fit test. Unfortunately, the latter, also called the Goodness-of-Fit test, is not performed that frequently. With the current availability of a plethora of time-resolved data, it is often necessary to estimate parametric functions, say of time, instead of just parameters. The issue then arises about the appropriate test of significance for these estimated parametric functions. This talk will define this open research problem and will discuss some preliminary solutions. The author will also express his regrets that Professor Tunde Ogunnaike is not with us to contribute to the solution with his insightful and accomplished expertise in Statistics.