Tell me more ×
Cross Validated is a question and answer site for statisticians, data analysts, data miners and data visualization experts. It's 100% free, no registration required.
up vote 0 down vote favorite

I need a sort of review of all (or most of) available methods to create null distributions (as reference) to use to compare a result. For ex: if I want to validate a network I can create a set of 1000 random networks starting from the initial dataset, by shuffling 1000 times my data. Then I obtain a null distribution of the 1000 nets toward which I can compare my own net (bootstrap) to score the significance (if my net is in the tails of the null distributions). Since this is not the best solution according to the data I start with, I need to know all the available methods used to create null distributions to use as a reference. Any suggestion about papers, links etc. useful to me to choose the best method?

PS: I have to create the null distribution starting from a synthetic dataset (a dataset totally created in vitro) so I need to know how to build it first.

Thanks in advance

E.

share|improve this question
This question is too broad to be answerable: it covers a great deal of all of the theory and practice of simulation. To narrow it, could you describe in more detail precisely what distribution you are trying to simulate? – whuber♦ Sep 21 '12 at 19:24
Okkk, soon! I would like to start with a totally independent dataset (sample on columns and genes on rows). I suppose to create ex novo this dataset in silico starting from random numbers. After this, I would like to create a network applying an algorithm for net rewiring (for ex. mutual information), on this synthetic dataset. At this point I need to validate the net I build. My question is: how apart from bootstrap? – elb83 Sep 21 '12 at 19:30
I am not clear on much of this but the bootstrap always interests me. How does this bootstrap enter into this. I know that you say the bootstrap is related to your network. But how is the bootstrap applied and for what purpose? – Michael Chernick Sep 21 '12 at 20:24
Ok, in other words Michael, I need first of all to build a synthetic dataset (I don't know how apart using the sample function in R). Then I infer a network of gene regulation. Then I need to know if that network is due by chance or not. How? I suppose doing bootstrap (a shuffle data 1000 times to create a null distribution) . Is there another way to create this null distribution instead of bootstrap? – elb83 Sep 21 '12 at 21:18
As I heard more than once from people in editorial positions, "If you cannot find a review of a literature on the Best Local Analytical Hyperparameters (BLAH, for short), and you really expected that there is one given that everybody talks about BLAH, it means you can write one of your own, and chances are it will be valuable to the community." – StasK Sep 22 '12 at 0:34
show 2 more comments

Know someone who can answer? Share a link to this question via email, Google+, Twitter, or Facebook.

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Browse other questions tagged or ask your own question.