I have, however, spent the last six hours working on my research. Yay for that. Even with the modifications the bootstrap algorithm is still off by quite a bit. But, that's a good thing. If it worked perfectly, there wouldn't be much point in deriving the exact distribution of the block variance. As you can see, in the uncorrelated case, it works great (to be expected - bootstraps like truly random data). In the other two cases, the variance is getting badly understated and there are too many misses.
So, that's pretty much best possible result there. It works, just not very well. Makes for a good baseline.
No comments:
Post a Comment