Probability Bounds on the Sum of Independent Nonidentically Distributed Binomial Random Variables (Q3704669)
From MaRDI portal
!
This is the item page for this Wikibase entity, intended for internal use and editing purposes. Please use the normal view instead:
scientific article; zbMATH DE number 3932098
| Language | Label | Description | Also known as |
|---|---|---|---|
| default for all languages | No label defined |
||
| English | Probability Bounds on the Sum of Independent Nonidentically Distributed Binomial Random Variables |
scientific article; zbMATH DE number 3932098 |
Statements
Probability Bounds on the Sum of Independent Nonidentically Distributed Binomial Random Variables (English)
0 references
1985
0 references
symmetric function
0 references
successive sums of powers
0 references
0.7628756
0 references
0.75477165
0 references
0 references
0.74907994
0 references
0.73977345
0 references
0.73589975
0 references
0.7324027
0 references
Let \(X_ i\), \(i=1,...,n\) be independent binomial random variables with \(P(X_ i=1)=p_ i\) and \(P(X_ i=0)=q_ i\equiv 1-p_ i,\) \(p_ i\neq 1\), and set \(F(\alpha)=P(\sum^{n}_{i=1}X_ i<\alpha).\) \textit{W. Hoeffding} [Ann. Math. Stat. 27, 713-721 (1956; Zbl 0073.139)] found that \(0\leq F(\alpha)\leq \sum^{\alpha}_{k=0}\left( \begin{matrix} n\\ k\end{matrix} \right)\quad p^ kq^{n-k},\) \(0\leq \alpha \leq np-1\) and \(\sum^{\alpha}_{k=0}\left( \begin{matrix} n\\ k\end{matrix} \right)p^ kq^{n-k}\leq F(\alpha)\leq 1,\) np\(\leq \alpha \leq n\), where \(p=(p_ 1+...+p_ n)/n.\) The present authors first note that \(P(\sum^{n}_{i=1}X_ i=\alpha)=C_{\alpha}(p_ 1/q_ 1,...,p_ n/q_ n)/P,\) where \(P=1/\prod^{n}_{i=1}q_ i\) and \(C_{\alpha}\) is the \(\alpha\) th elementary symmetric function of the arguments, and calculates exact results for P(\(\sum^{n}_{i=1}X_ i=\alpha)\), \(\alpha =1,...,4\), in terms of \(\prod^{n}_{i=1}q_ i\) and the successive sums of powers \(\sigma_{\alpha}=\sum^{n}_{i=1}(p_ i/q_ i)^{\alpha}\). The authors then proceed to derive tight bounds for F(\(\alpha)\), (a) given \(C_ 1=\sum^{n}_{i=1}(p_ i/q_ i)\); (b) given \(C_ 1\) and \(C_ 2=\sum_{i<j}(p_ i/q_ i)(p_ j/q_ j)\); (c) given (F(0) and F(1). Their first theorem is the following. Set \(Q(\alpha)=F(\alpha)/\prod^{n}_{i=1}q_ i\). Then \[ 1+C_ 1(1- \delta_{\alpha,0})\leq Q(\alpha)<\sum^{\alpha}_{r=0}\left( \begin{matrix} n\\ r\end{matrix} \right)\quad (C_ 1/\alpha)^ r, \] and neither bound can be improved. Here \(\delta_{\alpha,0}=1\) if \(\alpha =0\) and 0 if \(\alpha\) \(\neq 0.\) The new bounds compare very well with the above-mentioned ones, as well as with bounds of \textit{J. L. Hodges} jun. and \textit{L. Le Cam} [ibid. 31, 737-740 (1960; Zbl 0100.143)].
0 references