Application of the delta method to functions of the sample mean when observations are dependent (Q1685282)

Let \(X_i\), \(i = 1,2,\dots\), be a sequence of real valued random variables with the same mean \(\mu\). A general technique applicable in nonparametric inference for estimating a known function \(f(\mu)\) is to substitute \(X_n'=(\sum_{i=1}^n X_i)/n\) for \(\mu\). The well-known `delta method' is used to examine the asymptotic behavior of the variance of \(f(X_n')\) when \(f\) is differentiable at \(\mu\). In this paper, the authors show that the asymptotic expansions for bias and variance are still possible even if the boundedness conditions are replaced by weaker conditions on \(f\) as well as when observations are not independent. Furthermore, the authors demonstrate the technique for the problems of estimation of \(|\mu|^r\) for some positive real exponent \(r\), reciprocal \(1/\mu\) as in the case of rate of replacement when \(X_i\) are lifetimes in service with mean \(\mu\) and exponential functions like \(e^\mu\) and hint at further generalizations.