Three Modeling Philosophies Explained: Positivism, Structuralism, Instrumentalism

Recently I discussed mathematical modeling with a friend and realized that, despite its broad scope, people often have very different assumptions and goals when they talk about modeling.

Assumptions are the foundation of any model; a wrong assumption can make a building collapse just as a faulty model can fail.

Different researchers hold different assumptions, which I loosely categorize into three "modeling schools": Positivism , Structuralism , and Instrumentalism .

Positivism: Data Science Believers

Positivists believe "data is truth" and that observing and statistically inferring the world can capture its regularities. With enough data they think facts can be reproduced. They favor regression, time‑series, and machine‑learning methods, placing data at the core of algorithmic models.

This school is intuitive and pragmatic, especially in the era of AI large‑language models driven by the three engines of "algorithm, compute, data".

Positivists simply let the data speak for itself.

However, this approach can over‑rely on historical data and ignore contextual information; sudden policy changes, for example, can render purely empirical forecasts ineffective.

Structuralism: Theory‑Driven Advocates

Unlike positivists, structuralists focus on the underlying "essence" of phenomena, seeking first‑principles. They believe the world follows internal rules that can be uncovered through logical reasoning or theoretical derivation, using differential equations, dynamical models, and other mechanism‑based methods.

This approach shines in physics, ecology, etc., such as using the Lorenz model for weather or Lotka‑Volterra equations for predator‑prey dynamics.

Their models offer high explanatory power and transparency, but can become overly complex and hard to apply directly to real‑world problems.

Instrumentalism: Pragmatic Problem‑Solvers

Instrumentalists have no fixed philosophical agenda; they care about solving the problem at hand. Their toolbox ranges from simple linear programming to sophisticated heuristics. Their core belief is "any model that solves the problem is a good model."

This flexibility makes them well‑suited for concrete, well‑defined tasks, though it may lack deep reflection on underlying causes. They act like engineers, quickly finding a "good enough" solution.

An Integrated Framework

Personally I lean toward structuralism in theory but adopt instrumentalism in practice. The classification is not meant to rank schools but to help modelers understand themselves and each other, fostering better communication and collaboration.

Top modelers are loyal not to a school but to the problem itself.

These three schools are not mutually exclusive; often a model blends elements from multiple schools.

We can map modeling approaches on a two‑axis diagram: data quantity (positivism) versus mechanism depth (structuralism). This yields four categories:

Data‑poor, mechanism‑poor : Simple, heuristic methods used early in a project when resources are limited.

Data‑rich, mechanism‑poor : Purely data‑driven models (e.g., recommendation systems, image recognition) that can be accurate but act as "black boxes".

Data‑poor, mechanism‑rich : Theory‑driven models that rely on strong assumptions and can explain phenomena even with limited data (e.g., classic physics or ecological models).

Data‑rich, mechanism‑rich : Ideal scenario combining abundant data with deep theory, such as climate modeling that merges global observations with atmospheric physics, though it demands high computational resources.

In practice, modelers often shift among these types based on problem nature, available resources, and time constraints, balancing accuracy, complexity, and interpretability.

Understanding these modeling schools clarifies the diversity of mathematical modeling and helps teams leverage complementary strengths to enhance overall modeling capability.