Linear Regression slopes- can you predict an acceptable slope range?

kimd

New Member
#1
I am planning to compare a test method to a reference method for viral loads. I've been told by my supervisor that the correlation between the 2 assays will be deemed acceptable if R2 >0.9 and if the slope of the regression line is between 0.9375 and 1.0625.
I can't figure out how he got to that range (slope)??
 

Dason

Ambassador to the humans
#2
Probably it isn't a statistical thing. That amounts to 1 +/- (1/16). That's too clean to be something derived from data.
 

ondansetron

TS Contributor
#3
Like many things in medicine (I'm guessing, unless you're more PhD in virology type), the "standard" is arbitrary and often misses the intended mark.

Can you ask your advisor why an R-squared of >.9 (or did he actually mean correlation of .9 and r-squared of >.81?) and how .9375 and 1.0625 for slope are chosen and what that means in clinical terms?
 

hlsmith

Not a robit
#4
Do you know which method is more accurate? I would also plot the values so you specifically know where they differ, in case it is systematic.