In full generality, the idea with jackknife resampling is that you take your full sample, remove one data point, compute your statistic of interest, x, remove the next point, recompute, etc. Once you've done this N times, you can use Eq. 2 and Eq. 3 here to compute the mean and variance of x for your sample.
In practice, the goal of this is to compute the effects of sample variance (sometimes incorrectly called cosmic variance) in your simulation. What you'd do is break the simulation into N boxes, compute the spearman coefficient on each set of N-1 boxes, and use those two equations. There's no rule that tells you what the right number of boxes is: 2^3 is probably fine, but I'd honestly test 2^3, 3^3, 4^3, ... etc. and see if the choice has any impact on your error estimates.
ismael-mendoza
commented
3 years ago
Phil: