Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

I have independent random values $x_1 < ... < x_n$ with the same distribution function. I want to find expected value of this distribution by my . Of course it will be $X = \frac{1}{n} \sum_i x_i$

Probability $P(|X - E(x)| < \epsilon) > 1- \delta$

The question is how to find the number $n$ to be accurate about $\epsilon$ and $\delta$ (they equal 0.01)? I want to find FULL proof for that accuracy. And I don't want to calculate dispersion, because it will cause same difficulties.

I had 2 variants now - Chebyshev Inequality and Hoeffding's Inequality.

First is to complicated - it needs $n> 1000000$

Second needs a precise interval for $x$ . I can estimate it theoretically, but It becomes too big, much bigger than on practice. Maybe there is some variant of HI where bounds for $x$ are changed to $x_1$ and $x_n$ ?

I also looked at Central limit theorem, but I didn't find the proof for accuracy for my $X$.

Thanks, Vasily.

share|improve this question

1 Answer 1

up vote 0 down vote accepted

Stated that way, there is no answer to your question. Consider a distribution that gives some finite, but small, probability $\eta$ to a huge value $v$, while with probability $1-\eta$ the value is near $0$. If $v$ is large enough, the mean will be more than $1$.
But with probability $(1-\eta)^n$ the value $v$ won't occur in a sample of size $n$. So an estimate that is based only on the sample values will probably be inaccurate by more than $1$ for such a distribution.

share|improve this answer
    
Yea, maybe you're right –  146 percent Russian May 25 '12 at 18:28

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.