Take the 2-minute tour ×
Cross Validated is a question and answer site for people interested in statistics, machine learning, data analysis, data mining, and data visualization. It's 100% free, no registration required.

Suppose I've a random variable $X$ and a sample of it with size $N$. I count how many element of the sample fall in a specific range $a<x<b$ and I found $n$ entries, so if I use Poisson statistics I can say that the error on $n$ is $\sqrt{n}$.

Now suppose I've a function $f$, quite close to the unit function, and I define a new random variable $X'=f(X)$. Now the number of entries in $a<x<b$ is $n'=n+\delta n$.

I don't have an analitical from of $f$. I only now $n$ and $n'$.

What is the error on $n'-n$? How to take into account the correlation between them?

More difficult: suppose $f$ depends on a parameter $\lambda$ with error $\delta\lambda$.

share|improve this question
add comment

1 Answer

Why would you use Poisson? If you have $n$ independent observations out of $N$ falling in $(a,b)$ with constant chance for each, wouldn't that be binomial rather than Poisson?

You can think of transforming not $X$ but (assuming invertibility, with $g = f^{-1}$), transforming $(a,b)$; the probability relates to the cdf of $X$ ($F(a') = F(g(a))$ and so on). The standard error of the binomial is $Np(1-p)$ where $p=F(b)-F(a)$ and $p' = F(b')-F(a')$ where the transformation of the endpoints works as above.

share|improve this answer
    
thanks. Poisson is a good approximation for my case. By the way I don't have an analitical form of $f$. –  wiso Feb 21 '13 at 11:23
add comment

Your Answer

 
discard

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.