Nitaya Buntao. Statistical inference with imprecise random samples. Doctoral Degree(Applied Statistics). King Mongkut's University of Technology North Bangkok. Central Library. : King Mongkut's University of Technology North Bangkok, 2012.
Statistical inference with imprecise random samples
Abstract:
This thesis deals with statistical inference with imprecise random samples which actual values of samples are unknown. The lognormal, delta-lognormal, and heavy-tailed distributions are studied in some cases that measurements belong to interval [xi - ∆i, xi + ∆i]. According to statistical inference of imprecise random samples is quite complicated, firstly, we concern the traditional statistics with precise data for lognormal and delta-lognormal distributions and then new confidence intervals for statistical characteristics are proposed by several methods. Next, we formulate the general problem in the case of imprecise random samples and show how to estimate different characteristics and produce estimates based on these input intervals. The ideas are similar to the traditional statistical approach, we look for point estimates with imprecise random samples to find the appropriate point estimates for different parameters using the Maximum Likelihood approach. We first provide general formulas for using this approach to produce point estimates under interval uncertainty. It is therefore desirable to find the range of possible values for these estimates. The formulas and algorithms for computing these ranges are provided. In section of heavy-tailed distributions, we use decision theory to describe which characteristics are the best for different objective functions. For the resulting characteristics, we provide efficient methods for computing their estimates with imprecise data. Finally, we extend statistical inference to the fuzzy uncertainty in the way to establish well-formalized models for elements combining probabilistic information expressed by a probability measure P (S) and expert-related possibilistic information expressed by a possibility measure M(S).