WORK PLANS - Laserfiche WebLink

Home Browse Search

estimate the EPC term. The basis and theoretical justification for those recommendations are summarized in Singh and Singh(2003)for full data sets without ND observations. UCLs for Full Uncensored Data Sets without ND Observations 1. Student's-t UCL: to be used for normally (or at least approximately normally) distributed data sets. Student's-t UCL is available for all confidence coefficients, (1-a) in the interval [0.5, 1.0). 2. Approximate Gamma UCL: to be used for gamma distributed data and is typically used when k hat(ML estimate of the shape parameter,k) is greater than or equal to 0.5. Approximate gamma UCL is available for all confidence coefficients (1-a) in the interval [0.5, 1.0). 3. Adjusted Gamma UCL: to be used for gamma distributed data sets and should be used when k hat is greater than 0.1 and less than 0.5. Adjusted gamma UCL is available only for three confidence coefficients: 0.90, 0.95, and 0.99. 4. H-UCL based upon Land's H-statistic: to be used for lognormally distributed data sets. In ProUCL, H-UCL is available only for two confidence coefficients: 0.90 and 0.95. ProUCL can compute H-UCL for samples of size up to 1001. Caution: For highly skewed data sets,the use of H-UCL should be avoided as the H- statistic often results in unrealistically large, impractical and unusable H-UCL values. ProUCL provides warning messages and recommends the use of alternative UCLs for such highly skewed lognormally distributed data sets. 5. Chebyshev (MVUE) UCL: to be used for lognormally distributed data sets. This UCL computation method uses the MVU estimates of the standard deviation of the mean and of other parameters of a lognormal distribution. Chebyshev (MVUE) UCL is available for all confidence coefficients, (1-a) in the interval [0.5, 1.0). 6. Central Limit Theorem(CLT)based UCL: to be used when the sample size is large. 7. Adjusted-CLT (adjusted for skewness)UCL: may be used for mildly skewed data sets of large sizes. 8. Modified-t statistic (Adjusted for skewness)based UCL: may be used for mildly skewed data. Caution: UCLs listed in 6, 7, and 8 do not provide adequate (e.g., 95%) coverage when the data are moderately to heavily skewed, even when the sample size is large such as greater than 50. 9. Chebyshev(Mean, Sd)UCL: based upon the sample mean and standard deviation, Sd. 10. Jackknife UCL for mean(same as Student's-t UCL). 11. Standard Bootstrap UCL. 12. Bootstrap-t UCL. 13. Hall's Bootstrap UCL. 14. Percentile Bootstrap UCL. 15. Bias-corrected accelerated(BCA) Bootstrap UCL. UCLs Based Upon Left Censored Data Sets with ND Observations In order to compute UCLs, one has to first obtain estimates of population mean, standard deviation, and standard error of the mean based upon data sets with single or multiple detection limits. ProUCL 4.0 has a couple of estimation methods such as the ROS methods and Kaplan- Meier(KM)method that can handle multiple detection limits. The following methods for estimation of population mean and the standard deviation have been incorporated in ProUCL 4.0