Skip to content

Non-parametric tests in R

2 messages · Subhrangshu Nandi, Stefan Grosse

Original
Subhrangshu Nandi
# 
Lets say I have a population of 500 datapoints, which is clearly not normal.
(Please see attached text file for an example population). These datapoints
can be assumed to be IID. Now, I get 30 more datapoints and want to test if
these have been drawn (with replacement) from the population. I'm doing the
followin:

1. Since the population is non-normal, I'm unable to use regular t-tests. I
am trying to use Wilcoxon's rank sum test in R. I'm not sure if it only
compares the new sample with the median of the population or does it also
compare the central tendency of the dataset around the median.

2. I'm bootstrapping 5000 samples (resampling), of size 30 from the
population and recording their means and standard deviations and trying to
infer if my sample mean is within acceptable range. I'm using the concept of
Law of Large Numbers, however, I'm not sure if this is an acceptable
methodology.

Any thoughts on this will be great.
-Nandi
2 days later
Stefan Grosse
# 
On Fri, 15 May 2009 14:22:01 -0500 Subhrangshu Nandi
<nands31 at gmail.com> wrote:
SN> Lets say I have a population of 500 datapoints, which is clearly
SN> not normal. (Please see attached text file for an example
SN> population). These datapoints can be assumed to be IID. Now, I get
SN> 30 more datapoints and want to test if these have been drawn (with
SN> replacement) from the population. I'm doing the followin:

How about the Kolmogorov-Smirnov Test?
http://en.wikipedia.org/wiki/Kolmogorov%E2%80%93Smirnov_test
?ks.test


Stefan