12.07.2015 Views

Fundamentals of Probability and Statistics for Engineers

Fundamentals of Probability and Statistics for Engineers

Fundamentals of Probability and Statistics for Engineers

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

272 <strong>Fundamentals</strong> <strong>of</strong> <strong>Probability</strong> <strong>and</strong> <strong>Statistics</strong> <strong>for</strong> <strong>Engineers</strong>Example 9.4. Problem: determine the CRLB <strong>for</strong> the variance <strong>of</strong> any unbiasedestimator <strong>for</strong> in the lognormal distribution8 >0;>:0; elsewhere:Answer: we haveq ln f …X; †qq 2 ln f …X; †q 2 ˆ 12 2E q2 ln f …X; †q 2ˆ 12 ‡ ln2 X2 2 ;ˆ 12 2ln 2 X 3 ; 3 ˆ12 2 :It thus follows from Equation (9.36) that the CRLB is 2 2 /n.Be<strong>for</strong>e going to the next criterion, it is worth mentioning again that, althoughunbiasedness as well as small variance is desirable it does not mean that we shoulddiscard all biased estimators as inferior. Consider two estimators <strong>for</strong> a parameter ,^ 1 <strong>and</strong> ^ 2, the pdfs <strong>of</strong> which are depicted in Figure 9.2(a). Although ^ 2 is biased,because <strong>of</strong> its smaller variance, the probability <strong>of</strong> an observed value <strong>of</strong> ^ 2 beingcloser to the true value can well be higher than that associated with an observedvalue <strong>of</strong> ^ 1. Hence, one can argue convincingly that ^ 2 is the better estimator <strong>of</strong>the two. A more dramatic situation is shown in Figure 9.2(b). Clearly, based on aparticular sample <strong>of</strong> size n, an observed value <strong>of</strong> ^ 2 will likely be closer to the truevalue than that <strong>of</strong> ^ 1 even though ^ 1 is again unbiased. It is worthwhile <strong>for</strong> us toreiterate our remark advanced in Section 9.2.1 – that the quality <strong>of</strong> an estimatordoes not rest on any single criterion but on a combination <strong>of</strong> criteria.Example 9.5. To illustrate the point that unbiasedness can be outweighed byother considerations, consider the problem <strong>of</strong> estimating parameter in thebinomial distributionp X …k† ˆ k …1 † 1 k ; k ˆ 0; 1: …9:43†Let us propose two estimators, ^ 1 <strong>and</strong> ^ 2, <strong>for</strong> given by^ 1 ˆ X;^ 2 ˆ nX ‡ 1n ‡ 2 ; 9>=> ;…9:44†TLFeBOOK

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!