combining surrogates and block bootstrapping

[arfriedman]

Greetings,

This question is a follow-up to my question on using surrogates for composites:

Does it make sense to apply block bootstrapping -- used to compare segments within autocorrelated time series -- along with phase-shifted resampling?

In my case, I am comparing the means of subsets within a longer series, e.g., decadal means of a century-long record. For this type of comparison, would it make sense to also apply block bootstrapping within the surrogates (randomly sampling decades within each surrogate) to generate my null distribution of decadal means? Or is it recommended to simply compare the same fixed periods from a large number of surrogates?

[CommonClimate]

Hi @arfriedman , since you are comparing means on decade-long blocks, a block-bootstrap procedure with 10y block size would make sense. I am not sure if it would give you different results from using phase-randomized composites, depending on the nature of the series you use. You should definitely try that on synthetic data first. However, I do not believe that mixing and matching both approaches is advisable. Good luck!

[arfriedman]

Thank you @CommonClimate!