Nonparametric regression for problems involving lognormal distributions / Haipeng Shen.
xii, 170 p. ; 29 cm.
- Local subjects:
- Penn dissertations -- Statistics. (search)
Statistics -- Penn dissertations. (search)
- The lognormal distribution has a very long history and is almost as important as its sister distributions, the normal and the binomial. It is commonly used to model continuous positive quantities in several fields. Our particular interest in the lognormal distribution comes from a cross-disciplinary call center project. One specific characteristic of the call center data is that the service times of the calls are lognormally distributed. The mean of the service times is one of the key inputs for understanding and modelling call center operations. It can be used to derive system workload and delay, and ultimately to set staffing levels for call centers. Motivated from the project, we look at statistical inference for problems involving lognormal distributions. Techniques for providing point estimates and confidence intervals for a lognormal mean are studied. New methodologies are developed and compared with existing procedures. A new methodology based on local polynomial regression is then proposed to carry out nonparametric regression and provide corresponding confidence bands when the errors are lognormally distributed. A cross-validation type data-driven bandwidth selection procedure is also proposed to select the bandwidths needed. The developed methodologies then are applied to the call center data.
- Supervisor: Lawrence D. Brown.
Thesis (Ph.D. in Statistics) -- University of Pennsylvania, 2003.
Includes bibliographical references.
- Local notes:
- University Microfilms order no.: 3095940.
- Brown, Lawrence D., advisor.
University of Pennsylvania.
|Location||Notes||Your Loan Policy|
|Description||Status||Barcode||Your Loan Policy|