Significance of UL/LL in sample size calculation [Power / Sample Size]
Hi BE-proff,
The use of confidence limits on the CV not in particularly wide practical use, but it is advocated by some. If you go with an upper limit (as compared to the CV itself), that is going to be conservative, sample size will increase, power will go up. If you go with the lower limit, then you may be risking a too low sample size, the risk of failure increases. If you use limit then you need a level like 80% or 90% or whatever. It is really not set in stone and guidelines do not give an answer.
Error is a tricky word, and it has a negative feel doesn't. You will always have some uncertainty on the CV before you plan the trial. Whether you want to translate this uncertainty into a confidence limit is of course up to you.
If I were you and having some uncertainty I would do the following:
a. Try and generate a curve of power versus N for some value of CV and GMR.
b. Now adjust the GMR so that it is a little farther from 1.0 than the GRM under a. and generate the curve.
c. Now try and increase CV a generate the curve.
You will quickly be able to derive a few important, but mostly non-quantitative relationships, from such experiements. It will give you some ideas about the "what if"-scenario that you are alluding to, one way or another.
At the end of the day all this stuff about power, confidence limits on GMR, confidence limits on CV, type I errors etc only works if the residual in the linear model is normal. While it is mandatory to use that model (normal residual) for evaluation and assessment, usually/often/sometimes the residual is really not normal. We never test for it, and that in itself translates directly into additioanal uncertainty on whatever estimate you are working on, regardless of how fancy your formulae otherwise are.
❝ How significant is a use of upper/low limits for calculation of sample size?
The use of confidence limits on the CV not in particularly wide practical use, but it is advocated by some. If you go with an upper limit (as compared to the CV itself), that is going to be conservative, sample size will increase, power will go up. If you go with the lower limit, then you may be risking a too low sample size, the risk of failure increases. If you use limit then you need a level like 80% or 90% or whatever. It is really not set in stone and guidelines do not give an answer.
❝ Will I have a serious error using only CV without limits?
Error is a tricky word, and it has a negative feel doesn't. You will always have some uncertainty on the CV before you plan the trial. Whether you want to translate this uncertainty into a confidence limit is of course up to you.
If I were you and having some uncertainty I would do the following:
a. Try and generate a curve of power versus N for some value of CV and GMR.
b. Now adjust the GMR so that it is a little farther from 1.0 than the GRM under a. and generate the curve.
c. Now try and increase CV a generate the curve.
You will quickly be able to derive a few important, but mostly non-quantitative relationships, from such experiements. It will give you some ideas about the "what if"-scenario that you are alluding to, one way or another.
At the end of the day all this stuff about power, confidence limits on GMR, confidence limits on CV, type I errors etc only works if the residual in the linear model is normal. While it is mandatory to use that model (normal residual) for evaluation and assessment, usually/often/sometimes the residual is really not normal. We never test for it, and that in itself translates directly into additioanal uncertainty on whatever estimate you are working on, regardless of how fancy your formulae otherwise are.
—
Pass or fail!
ElMaestro
Pass or fail!
ElMaestro
Complete thread:
- Significance of UL/LL in sample size calculation BE-proff 2016-08-30 08:35 [Power / Sample Size]
- Significance of UL/LL in sample size calculationElMaestro 2016-08-30 09:04
- Significance of UL/LL in sample size calculation BE-proff 2016-08-30 13:53
- Significance of UL/LL in sample size calculation ElMaestro 2016-08-30 14:07
- a very nice function of powerTOST mittyri 2016-08-30 14:25
- nice functions of powerTOST d_labes 2016-08-30 16:39
- a very nice function of powerTOST mittyri 2016-08-30 14:25
- Significance of UL/LL in sample size calculation ElMaestro 2016-08-30 14:07
- Significance of UL/LL in sample size calculation BE-proff 2016-08-30 13:53
- Significance of UL/LL in sample size calculationElMaestro 2016-08-30 09:04