1
$\begingroup$

I have an experimental sample, size of about 1000 values​​. I need to generate a much larger sample for simulation. I can create a samples like this:

library(ks) x<-rlnorm(1000) y<-rkde(fhat=kde(x=x, h=hpi(x)), n=10000, positive = TRUE)# z<-sample(x, 10000, replace = TRUE) par(mfrow=c(3,1)) hist(x, freq=F, breaks=100, col="red") hist(y, freq=F, breaks=100, col="green") hist(z, freq=F, breaks=100, col="blue")

What fundamental limitations when using KDE or bootstrap? How else can I create such a sample?

Improve this question
$\endgroup$
1
  • 2
    $\begingroup$ Please, don't cross-post on multiple sites. $\endgroup$ Commented Apr 21, 2013 at 8:30

1 Answer 1

Reset to default
2
$\begingroup$

With the KDE you are smoothing your data to get guess what the distribution is. Then you are sampling from that distribution.

With bootstrapping, you are resampling from the data that you already have.

One difference between the two methods is that KDE can result in values that you did not observe in your original sample. In bootstrap this cannot happen. Whether this is good or bad depends on what you know about this distribution. If you think your sample already represents well the distribution, bootstrapping might be better.

A third approach might be to look at the histogram or use some prior knowledge about the distribution to fit some parametric distribution.

You have to understand though, that there are general limitations to what you are trying to do. In general, unless you have some prior beliefs about your distribution, you will not be able to "discover" a phenomenon that you did not observe in your sample by generating a larger sample artificially in this manner.

Improve this answer
$\endgroup$
1
  • 1
    $\begingroup$ Thank you! You may suggest other ways to create big sample with properties experimental sample? $\endgroup$ Commented Apr 20, 2013 at 22:17

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.