US Army Materiel Command AMCP 706-197 — AMCP 706-197 Reliability Prediction Handbook — US Army Engineering Design Guide PDF

Downloaded from https://www.everyspec.com AKIC 70-19 7 CHAPTER 4 REVIEW OF ELEMENTARY STATISTICAL THEORY 4-1 INTRODUCTION estimator is measured by the second moment of the estimator taken about the true value. If This chapter presents some of the statisti- the estimator is s-biased (per. 4-2.3), then this cal concepts which are useful in a reliability second moment is "variance + (bis) 2 ". If the context. The Bibliography at the end of Chap- estimator is s-unbiased (zero bias), s-efficiency ter 1 gives elementary, intermediate, and is measured by the variance of the estimator. advanced texts on probability and statistics.

It For a fixed sample size, the smaller the vari- is not the purpose of this chapter to write ance of the estimator, the more s-efficient it another textbook on statistics. is. The purpose of statistics is to help people There isalower bound tothevarianceof analyze real data and draw reasonable conclu- an estimator-the Cramer-Rao lower bound. sions from them.

In reliability engineering, s-Efficiencies often are measured relative to the function of- atistics most often will be in the Cramer-Rao lower bound; if this s-effi- showing an engineer what he does NOT know ciency is 100 percent, that's as a-efficient as from the data; i.e., statistics will provide an cn gt Mos estmatrs se in a one can get. Most estimators used in reliabili- engineer with a feeling for the uncertainty in ty work are quite s-efficient. the conclusions he wants to draw from the a-Efficiency is perhaps the most desirable data. sEfcec sprastems eial dathe fproperty of an estimator.

It tells you how The few concepts of statist~cs that are good or bad your estimate is likely to be. important in reliability ought to be carefully learned. It is better not to use them than to 4-2.2 s&CONSISTENT ESTIMATORS use them incorrectly. An s-consistent estimator is one which 42 ESTIMATION OF PARAMETERS "approaches" the true value as the sample size "goes to infinity". The reason for the quote It is usually convenient to summarize a marks is that the phrases are loose expressions mass of data by stating a distribution from of complicated mathematical concepts; for a which they might well have come. This usual- more exact definition, consult a textbook.

ly is done by choosing a distribution (on the s-Consistency is a very desirable attribute of basis of previous ideas, simplicity, massaging an estimator. Virtually all estimators in use in of the data, or something else) and then esti- reliability work are s-consistent. mating the parameters of the distribution. There are several popular methods of estimat- 42.3 s-BIAS ing parameters; they are not detailed here--but Part Six, Mathematical Appendix and Glossa- s-Bias is the difference between the ry, shows estimation methods for many of the s-expected (mean) value of an estimator (for a popular distributions, fixed sampling plan) and the true value.

It The important thing about an estimate is enters the measure of s-efficiency (par. 4-2.1); its properties, not how you got it. In these as long as the s-bias is less than about 50 per- days of readily available computers, the cost cent of the standard deviation, the contribu- of making estimates whose properties are tion of the s-bias can be neglected. Being s-un- good and well known is negligible compared biased is nice for theoretical work, but it is to the cost of getting the original data. vastly overrated as a criterion for goodness of reliability estimators.

The main reason for this 4-2.1 s-EFFICIENT ESTIMATOR is that if 0 Lq an s.unbiased estimator of 0, f(0) is an s-biased estimator of f(O) unless f() is a For engineers, s-efficiency is what estima- linear function. The most widespread misun- tion is all about. Any estimator uses a statis- derstanding of this principle is involved in the tic; that statistic has properties such as a mean estimate for the variance of an s-normal distri- value and a variance. The s-efficiency of an bution. The S 2 statistic, S 2 = SS(N-1) 4-1