Yep I know there's no actual limit for SD/RSD at all but I am really confused everytime I got stuck by a vast of assay values. I found this function in an SOP of process validation in my firm:
(see the attachment please)
Anyone knows about this? Anyone gives me explanation or function's origin?
Thanks
It's strongly looks like a transformation of the prediction interval of the next value of a Gaussian distribution, based on estimating its mean and standard deviation using a sample of size n of independant, identically distributed values from the same distribution.
The formula for the interval is
[m - t * s * sqrt( 1 + 1/n ) ; m + t * s * sqrt( 1 + 1/n ) ]
where m is the sample average and s the estimated standard deviation of the sample; t is the quantile of a Student distribution with n - 1 degrees of freedom.
No, let's assume you want your next value to lie between the target (which will be, let's say, considered to be the average, m) and the upper limit. That mean, you must have m + t * s * sqrt( 1 + 1/n ) < upper-limit, hence
s < (upper-limit - m) / (t * sqrt(1 + 1/n): you must have a maximal standard deviation given by the formula.
The fact that you use TINV(0.05) suggests that a 90% percent prediction interval is used, that is, there is a 90% probability that the next observed value will lie in this interval.
thank you Emmanuel Curis for helping me clarify this.
yet another problem is: should I define that maxSD before (in protocol) or after the assay (in report)? after having all assay results, the mean value will not be the same as ideal (100%) then maxSD may vary but much more adaptive to the current situation.
I think the answer will depend on which question you're trying to answer exactly...
- Badges
- Science topic