Fig. 4
From: Path-level interpretation of Gaussian graphical models using the pair-path subscore
Dependence of PPS estimation accuracy on sample size n and K. Results are stratified by length of the max PPS path and sample size to aid visualization. Note that in this simulation, \(K = 6\) corresponds exactly to \({\hat{s}}_p\) from (13). Error decreases with n, but actually increases for a given n if K is large enough. However, if n is not too small (at least 1000, say) the increase in error if K is too large is quite small compared to the increase in error if K is too small. It is therefore better to err on the side of choosing K too large than too small