R-Squared Values for a Standard Curve - (May/11/2010 )
Sorry I have no idea where to post this one, as I guess it will apply to people across the board. I have a quick question -
When running standards which are then fitted with a linear regression line, what is an acceptable R2 value, allowing me to accept the standard curve? It is basically a 6 concentration standard line, each run in duplicate. I can get this as high as 0.9996, but I would like to know a lower cut iff point as occasionally the GC machine goes a bit crazy! I get uneasy about anything under 0.985, what about anyone else?
> 0.98 is enough
Thank you very much!
So would you reject samples where the standard curve only read 0.9700 or 0.95 for example? I am using this report 'exact' quantities in unknown samples so I guess it's quite important.
Me? In fact no . 0.97 is still good correlation for my purposes.
It all depends how precise you want to be. If it is enough to know the order of magnitude then it's OK, but if you need to be as exact as possible then reject it.
The topic you may want to look up is called ‘inverse prediction’. It converts the uncertainty in a standard curve to error in your unknowns.