How Analytic Network Process (ANP) Beats AHP for Complex Decisions

In comprehensive evaluation models, the Analytic Hierarchy Process (AHP) is commonly used, but it assumes independence among decision elements, which may not hold in complex real‑world problems. To overcome this limitation, Thomas L. Saaty proposed the Analytic Network Process (ANP), an extension of AHP that can handle inter‑dependencies and feedback among elements. This article introduces the basic idea of the ANP model, its mathematical formulation, and illustrates each step with a case study.

Introduction to the ANP Model

ANP replaces the hierarchical structure of AHP with a network structure, allowing elements in a decision problem to influence each other.

The basic steps of ANP include problem structuring, building the network model, constructing pairwise comparison matrices, calculating weight vectors, constructing and normalizing the supermatrix, and performing comprehensive evaluation and decision making.

Mathematical Model

Problem structuring

First, decompose a complex decision problem into different levels and elements, such as goals, criteria, and alternatives.

Suppose we want to select the best project‑management software, with criteria including cost, functionality, user‑friendliness, and technical support. The alternatives are Software A, Software B, and Software C.

Building the network model

In ANP, we must identify the mutual influence relationships among elements and construct a network model. Elements can affect each other within the same level and across levels.

Assume cost and functionality influence each other, user‑friendliness and technical support influence each other, functionality influences user‑friendliness, and user‑friendliness also influences technical support.

Constructing the pairwise comparison matrix

Using the pairwise comparison method, we build a judgment matrix to quantify the relative importance of each element, typically based on expert scoring.

Calculating the weight vectors

Weight vectors are obtained by the eigenvalue method: for each judgment matrix, compute the maximum eigenvalue and its corresponding eigenvector, then normalize the vector.

Constructing and normalizing the supermatrix

The supermatrix is assembled, with its columns composed of the weight vectors of all elements, reflecting the mutual influence relationships.

After normalizing the supermatrix, we obtain the limit supermatrix , representing the stable state of the system.

Comprehensive evaluation and decision

By applying the limit supermatrix, we calculate the overall weights of each alternative, perform a comprehensive assessment, and select the optimal option.

Overall, compared with the AHP model, the ANP model offers greater flexibility and accuracy because it can handle complex inter‑dependencies among decision elements. However, ANP is more complex to implement; this article only covers the basic process, with future work planned to discuss tool implementation such as Python.

Reference: Tang Xiaoli, Feng Junwen, Wang Xuerong. “Project Risk Management Based on Analytic Network Process.” Statistics and Decision, Issue 16, 2005.