On corrected Poisson approximations for sums of independent indicators

Abstract

Let $S_n=I_1+\cdots+I_n$ be a sum of independent indicators $I_i$, with $p_i=\Pr(I_i=1)=1-\Pr(I_i=0)$, $i=1,\ldots,n$. It is well-known that the total variation distance between $S_n$ and $Z_\lambda$, where $Z_\lambda$ has a Poisson distribution with mean $\lambda=\sum_{i=1}^n p_i$, is typically of order $\sum_{i=1}^n p_i^2$. In the present work we propose a class of corrected Poisson approximations, which enable the second order factorial moment distance (and hence, the total variation distance) to be bounded above by a constant multiple of $\sum_{i=1}^n p_i^3$ and $\sum_{i=1}^n p_i^4$, hence improving the order of approximation.