Use this URL to cite or link to this record in EThOS: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.637241
Title: Problems of parameter estimation arising from sequential hypothesis testing
Author: Hassan, H. N.
Awarding Body: University College of Swansea
Current Institution: Swansea University
Date of Award: 1981
Availability of Full Text:
Access through EThOS:
Abstract:
Sequential hypothesis testing lays so much emphasis on the problems of choosing a good and economical hypothesis test that the equally important problem of estimating the parameter becomes more complicated. The thesis is devoted to the investigation of the problems of (point and interval) estimation arising from binomial sequential sampling. Various methods of point estimation have been considered including the maximum likelihood and a method generating unbiased estimators. For the latter two, the probability distributions have been graphed for different values of the parameter and the characteristics of these distributions have been calculated and used in the assessment of the corresponding estimators. A new method of point estimation based on the credibility of the median of any distribution as a representative value is proposed. This method is found to behave rather well in comparison with the former methods. Likewise, the case of interval estimation is discussed using four methods. These methods are evaluated individually; their exact and normal confidence coefficients are compared. Another comparison among methods is made using the length and mean-length of the intervals as criteria of goodness of interval estimation. These methods of (point and interval) estimation are applied to different sequential designs. It turns out that the problem of estimation tend to be more severe in the case of skew and open plans than in the restricted and repeated significance tests designs. It is generally found that the more economy in the number of observations the hypothesis test achieves, the more serious the problems become. The new method of point estimation, in our view, offers a more realistic alternative to the maximum likelihood and unbiased estimators, because its performance is intermediate between these two, bearing in mind that none of them is superior to the other in the whole range of the parameter.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.637241  DOI: Not available
Share: