Maximum-Likelihood Non-Decreasing Response Estimates

TL;DR

This paper introduces a method for estimating a non-decreasing sequence of parameters from independent data, maximizing likelihood under monotonicity constraints, with an efficient algorithm for computation.

Contribution

It provides a theoretical foundation and an efficient algorithm for maximum-likelihood estimation of monotonic response sequences.

Findings

Unique non-decreasing response estimate exists under mild conditions.

Algorithm efficiently computes the maximum-likelihood estimate.

Method applies to both non-decreasing and non-increasing cases.

Abstract

Let $x_{i,j}$, $1 \le i \le m$, $1 \le j \le n_i$, be observations from a doubly-indexed sequence $\{X_{i,j}\}$ of independent random variables (all of them discrete, or all of them absolutely continuous). Suppose that each $X_{i,j}$ has the PDF $f(x\mid\theta_i)$ from a one-parameter family of PDFs $f(x\mid \theta)$. Mild assumptions are described under which there is a unique, non-decreasing compound response estimate of $\mathbf \theta=<\theta_1, \hdots \theta_m>$ that maximizes the compound likelihood function among all non-decreasing response estimates. An efficient algorithm is described to compute this unique estimate. The same theory and algorithm also give the unique non-increasing compound response estimate that maximizes likelihood among all non-increasing response estimates. One simply reverses the order represented by the index $i$.