Bootstrapping, part 2

I still do not really know how to decide on the width of the kernel for the smooth bootstrap,
and have to find some accesible literature. The the paper I browsed through did give a suggestion, but I am not sure if I am happy with it.

Anyway, the multi-dimensional non parametric resampling scheme has been implemented as well, including a line fitting example. The non parametric bootstrap is used to estimate standard deviations of the refined parameters:

-------------------------------------------
True and fitted coeffcients
-------------------------------------------
a 1.0 1.9029
b 2.0 1.9270
c 3.0 3.005
-------------------------------------------------------------------
Bootstrapped mean and standard deviations
-------------------------------------------------------------------
a 1.9173 0.22509
b 1.9259 0.05107
c 3.005 0.00292

Compare this with the output of GNUPLOT fitting:

a = 1.90288 +/- 0.2356
b = 1.92697 +/- 0.05748
c = 3.00469 +/- 0.002921

The estimated standard deviation via the bootstrap and full matrix inversion matches nicely.


The computational statistics book seems to be a nice, basic and readable introduction and set of references for areas such MCMC, EM, Simulation and Bootstrapping. They don't mention the smooth bootstrap however.