Fisher information based progressive censoring plans
| Autor: | Erhard Cramer, Marco Burkschat, Glenn Hofmann, Narayanaswamy Balakrishnan |
|---|---|
| Rok vydání: | 2008 |
| Předmět: |
Statistics and Probability
Optimality criterion Applied Mathematics Statistical computation Simple random sample Censoring (statistics) Computational Mathematics symbols.namesake Computational Theory and Mathematics Law of large numbers Statistics symbols Econometrics Life test Fisher information Survival analysis Mathematics |
| Zdroj: | Computational Statistics & Data Analysis. 53:366-380 |
| ISSN: | 0167-9473 |
| Popis: | In life tests, the progressive Type-II censoring methodology allows for the possibility of censoring a number of units each time a failure is observed. This results in a large number of possible censoring plans, depending on the number of both censoring times and censoring numbers. Employing maximum Fisher Information as an optimality criterion, optimal plans for a variety of lifetime distributions are determined numerically. In particular, exact optimal plans are established for some important lifetime distributions. While for some distributions, Fisher information is invariant with respect to the censoring plan, results for other distributions lead us to hypothesize that the optimal scheme is in fact always a one-step method, restricting censoring to exactly one point in time. Depending on the distribution and its parameters, this optimal point of censoring can be located at the end (right censoring) or after a certain proportion of observations. A variety of distributions is categorized accordingly. If the optimal plan is a one-step censoring scheme, the optimal proportion is determined. Moreover, the Fisher information as well as the expected time till the completion of the experiment for the optimal one-step censoring plan are compared with the respective quantities of both right censoring and simple random sampling. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|
Full Text from ScienceDirect