To estimate the variance of a parameter, I know that a simple bootstrap will not work for correlated data. I also read about block bootstrap. I am trying to get an intuition about why it is necessary to keep the correlation structure intact in resampling. Wouldn't you want to create an independent sample if possible? And how does keeping the correlation structure intact by block resampling, but then using the same basic inference methodology as in a normal bootstrap, help estimate the correct variance.