Simple question I believe, but I am not as sharp on my stats as i would like to be. I have 66 items with a calculated value associated with them, due to data restraints I was only able to find a "ground-truth" value for 10 of them and therefore calculate an error term for ten of the 66 items.
I found that the % error term can be strongly explained (R^2=0.85) by a power regression y=15.57x^(-0.627) where y is the error term and x is the calculated relevant value.
I would like to now apply this to the larger data set and am just tempted to plug all the x values into the found relationship, but since this is for a scientific paper, I would like to quantify this effect and see the strength of generalizing it to the whole population of 66. What are some ways I can adjust the R^2 to account for the fact that the relationship reflects data from only 10/66 samples?