【求助】 - 医博园

» 医博园 » 英语咖啡屋 » 翻译

 打印话题 寄给朋友

【求助】

angelpig

发贴: 13
积分: 6

得票: 0

状态: 离线

2008-07-14 22:29

此是一份文件中的一段，求翻译
13 Full Data Analysis Procedures
The design of the experiment allows estimation of the effect of the slope (proportional bias), intercept(constant bias), sample carry-over, nonlinearity, and linear drift. Section 11 and Appendix C can be used to estimate the components of variance. Appendix C also includes a multiple regression procedure that may be used if the initial data analysis indicates possible unacceptable performance of the method. Themultiple regression procedure consists of the formulae in Appendix C and Data Summary Sheets #4, #5,and #6. It is, however, necessary to go through this procedure only if use of the details concerning the individual components contributing to measurement error is desired.

13.1 A Comment on the Model
The EP10 multiple regression model might not always seem to be appropriate. For example, sample carry-over may be virtually impossible due to the design of an instrument. One solution to this of course is to contact a statistician to develop a different design. Yet using the NCCLS EP10 design in the above case where sample carry-over is unlikely could still yield valuable information for the following reason.From a model calculation standpoint, it is irrelevant whether effects are likely or unlikely, since all parameters in the model are automatically (i.e., mathematically) estimated. This means that variability in the data is explained as either nonzero parameter effects or as imprecision (residual error). So it is possible that some effect other than sample carry-over, which one has not thought of, is estimated as sample carry-over by the model. This would be a signal to investigate the measurement procedure to determine the origin of this error source.

13.2 Summarizing the Five Runs
The t-test procedure that had been used to summarize the five runs has been changed. With only five runs,the problem is that variation in a parameter estimate from run to run could be due to sampling variation(implies that the parameter is a constant that cannot be perfectly estimated with only ten samples) or that the parameter is truly changing from run to run. Of the possible ways to summarize the runs, the subcommittee chose a one-sample sign test.This is likely to be significant in many cases as it requires all estimates values to have the same sign for the five runs (1 is subtracted from the slope estimates).
If a result has been detected as statistically significant, it may still be of no practical importance as an
error source. Users must decide on how much error to allow to each error source.

14 An Alternative Procedure
A modification of the protocol can be carried out that allows the estimation of three interfering substances in addition to all of the other parameters estimated.

• 【原创】前两天照的小乌龟，呵呵

已读帖子

新的帖子

被删除的帖子

Welcome to YBY123.COM Forum