I have > 1,500 bootstrapped data points, each a pair of values (x, y). y is calculated by some function of x (the function being a variance component estimation of genetic data). The data points are generated as such:
1) x is sampled uniformly from 0 - 20.000,
2) y is calculated as some function of x, the ultimate result begin a float between 0 and 1.
I am interested in a 95% percentile threshold of these data, i.e. for a value x', what is the lowest value y' can take and still be significant?
So far, I have managed to use a smooth quantile regression model in R ("quantreg" package), but only applying it directly to the mass of bootstrap replicates.
I am reading the bootstrap-book by Efron and Tibshirani, and I think that I somehow should be doing a bootstrapping on the 95% percentile regression, but I am not sure.
Is there anything I should be doing to improve my results?