Virginia Polytechnic Institute
Maximum likelihood estimation of parameters using grouped data samples
Abstract
dc:description.abstractThe method of maximum likelihood was employed to estimate the parameters of the univariate normal distribution when grouped data only are available. The term grouped data sample was used to describe a sample in which the only information available is that there are lilt observations less than or equal to r<sub>i</sub>, i = O, 1, …, k and = r<sub>0</sub> < r<sub>1</sub> < … < r<sub>k</sub> and m<sub>k</sub> = n. Three cases were considered: the variance of the underlying normal distribution known, the mean of the underlying normal distribution known, and the case where both parameters are unknown. The numerical method of successive approximations was studied and was compared to other iterative methods of solution of the likelihood equations. Theorems which establish the convergence of the sequence defined by the method of successive approximations to the unique root of the likelihood equations were developed for the first two cases. In the case where both parameters are unknown, numerical results indicate that a region can be ascertained in which there can be at most one joint solution of the likelihood equations and that, if there is a solution in this region, then it is a point at which the likelihood function has a relative maximum. A search was conducted to find a sample which satisfies the conditions insuring the existence of a joint solution of the likelihood equations and which might provide two solutions of the likelihood equations. No example was found with this property. It was concluded that, except possibly for extremely ungrouped data sample configurations, one should expect only one solution of the likelihood equations in the case of a normal distribution. Simply computed, consistent estimators were developed which provided good approximations to the solution of the likelihood equations and. which can be used tor starting values in any iterative method of solution. Modifications of the method of successive approximations were developed. These methods of solution define sequences which are rapidly convergent, and which enable the solution to be obtained. with the UH of tables of the standard normal probability distribution of motion and its derivative The maximum likelihood estimators obtained when both parameters are unknown were proved to be asymptotically efficient provided there are at least three groups, i.e. k = 3. The theorems were developed to establish this property were shown to have a much wider application. Other distributions such as the Poisson distribution and the bivariate normal distribution were considered. Sufficient conditions were established for a grouped data sample, from these as well as other distributions, to provide maximum likelihood estimators which are asymptomatically efficient.
Degree
thesis:*- Name thesis:degree_name
- Ph. D
- Level thesis:degree_level
- doctoral
- Discipline thesis:degree_discipline
- Statistics
- Department dc:contributor.department
- Statistics
- Grantor dc:publisher
- Virginia Polytechnic Institute
- Year dc:date.issued
- 1968
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Bowen, Jacob Van
- Chair dc:contributor.committeechair
-
- Jensen, Donald R.
- Committee members dc:contributor.committeemember
-
- Harshbarger, Boyd
- Krutchkoff, Richard G.
- Kramer, Clyde Y.
- Pace, W. Emory
Rights
dc:rights- Statement dc:rights
-
- In Copyright
- Licence dc:rights.uri
- Language dc:language.iso
- en
Identifiers
dc:identifier.*- Handle dc:identifier.uri
- https://hdl.handle.net/10919/138618