Estimation of parameters of the Gompertz distribution using the least ...
Estimation of parameters of the Gompertz distribution using the least ...
Estimation of parameters of the Gompertz distribution using the least ...
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
146 J.-W. Wu et al. / Appl. Math. Comput. 158 (2004) 133–147<br />
than ML estimates in c ¼ 2:0, k ¼ 0:01; (e) for m ¼ 10, n ¼ 30, <strong>the</strong> estimates<br />
obtained by Method 1, Methods 5–7 and Methods 9–11 are better than ML<br />
estimates in c ¼ 2:0, k ¼ 0:02; (f) for m ¼ 30, n ¼ 10, <strong>the</strong> performance <strong>of</strong><br />
UWLS and WLS estimates is better than ML estimates in c ¼ 0:1, k ¼ 0:01 and<br />
c ¼ 0:01, k ¼ 0:02, respectively; (g) for m ¼ 30, n ¼ 10, <strong>the</strong> UWLS estimates<br />
obtained by Method 3 and <strong>the</strong> WLS estimates obtained by Method 7 and<br />
Method 9 are better than ML estimates in c ¼ 0:01, 0.1, 2.0, k ¼ 0:01; (h) for<br />
m ¼ 30, n ¼ 30, <strong>the</strong> WLS estimates obtained by Method 8 are better than ML<br />
estimates in c ¼ 0:01, 0.1, 2.0, k ¼ 0:01; (i) for m ¼ 30, n ¼ 30, generally <strong>the</strong><br />
UWLS and WLS estimates are better than ML estimates in c ¼ 0:01, k ¼ 0:01;<br />
(j) for m ¼ 30, n ¼ 30, <strong>the</strong> WLS estimates obtained by Methods 6–10 are better<br />
than ML estimates in c ¼ 2:0, k ¼ 0:01 and (k) for m ¼ 30, n ¼ 30, <strong>the</strong> WLS<br />
estimates obtained by Methods 6–9 are better than ML estimates in c ¼ 0:1,<br />
2.0, k ¼ 0:02. In addition, from <strong>the</strong> above results, it is suggested that Method 9<br />
is useful for estimating c and k under <strong>the</strong> first failured-censored data.<br />
4. Concluding remarks<br />
In summary, <strong>least</strong> squares methods <strong>of</strong>ten provide simple and fairly effective<br />
ways <strong>of</strong> obtaining estimates with complete data and <strong>the</strong> first failured-censored<br />
data. The procedures described were based on transform <strong>of</strong> F ðxÞ, which is <strong>the</strong><br />
<strong>Gompertz</strong> cumulative <strong>distribution</strong> function. Results from simulation studies<br />
illustrate <strong>the</strong> performance <strong>of</strong> <strong>the</strong> WLS estimates is acceptable.<br />
Acknowledgement<br />
This research was partially supported by <strong>the</strong> National Science Council,<br />
ROC (Plan No. NSC 89-2118-M-032-013).<br />
References<br />
[1] U. Balasooriya, Failure-censored reliability sampling plans for <strong>the</strong> exponential <strong>distribution</strong>,<br />
Journal <strong>of</strong> Statistical Computation and Simulation 52 (1995) 337–349.<br />
[2] V. Barnett, Probability plotting methods and order statistics, Applied Statistics 24 (1975) 95–<br />
108.<br />
[3] Z. Chen, Parameter estimation <strong>of</strong> <strong>the</strong> <strong>Gompertz</strong> population, Biometrical Journal 39 (1997)<br />
117–124.<br />
[4] R.B. DÕAgostino, M.A. Stephens, Goodness-<strong>of</strong>-fit Techniques, Marcel Dekker, New York,<br />
1986.<br />
[5] B. Faucher, W.R. Tyson, On <strong>the</strong> determination <strong>of</strong> Weibull <strong>parameters</strong>, Journal <strong>of</strong> Materials<br />
Science Letters 7 (1988) 1199–1203.