Mental Models for Decision-Making: The 7 Worth Knowing, Evidence-Graded

In April 1994, Charlie Munger stood in front of Guilford Babcock's investment class at USC and gave the talk that would quietly found a genre. His claim was not that the students needed more information — it was that information without structure is useless: if the facts don't hang together on a latticework of theory, you don't have them in a usable form. The fix: you've got to have models in your head — eighty or ninety of them, drawn from every discipline — and you've got to array your experience on that latticework.

Thirty years on, mental models is a content genre: viral threads listing seven models in seven paragraphs, none sourced, none graded. The models themselves deserve better — several have real intellectual pedigrees, and the one effect size the genre does quote turns out to measure something other than what it is quoted for. The honest version of the list has to say what the viral versions never do: where each model came from, what the evidence actually is, and where it breaks.

So here is the evidence-graded version: what a mental model actually is (a question cognitive science answered in 1943), the seven most worth having for decisions, and the uncomfortable finding about why knowing the list changes nothing — plus what does.

What a Mental Model Actually Is

The term predates the genre by half a century. In The Nature of Explanation (1943), the Cambridge psychologist Kenneth Craik proposed that the mind carries a small-scale model of external reality — and that this is the whole trick of intelligence: an organism with an internal model can try out alternatives, conclude which is best, and react to situations before they arise, in a much fuller, safer and more competent manner. Philip Johnson-Laird's Mental Models (1983) built that insight into a cornerstone of cognitive science: we reason not by formal logic but by constructing and inspecting internal models of situations.

Munger's contribution was normative: if thinking runs on models anyway, choose them deliberately — collect the big ideas from the big disciplines rather than inheriting whatever your trade happens to teach, and hang new facts on the lattice as they arrive. That reframe is what separates a mental model from a framework: a decision model prescribes a procedure; a mental model is a lens you carry into every situation. The seven below are the lenses with the strongest claims for decision work.

The Seven, Evidence-Graded

For each: what it says, where it actually comes from, what the evidence shows — and where it breaks. Three have full companion pages; use them for depth.

The Problem With Model Collecting

Now the part the viral lists omit. Knowing this list changes approximately nothing — the same finding that haunts bias awareness. A model you can recite is trivia; the latticework Munger described was not a reading list but a practice: models applied, case by case, to live decisions, until the lens becomes the default way of seeing. The mental-models genre's own critics have made this point for years — collectors accumulate vocabulary while their decisions run on autopilot, exactly as before.

Cognitive science gives the criticism a sharper edge. Resource-rational analysis (Lieder & Griffiths, 2020) models human thinking as the optimal use of limited computation — which means a mental model is a heuristic, and a heuristic's value depends on fit: the right lens on the right decision, at a cost the decision justifies. Running first-principles decomposition on lunch is as irrational as running analogy on an acquisition. The skill is not owning models; it is matching them — and matching is trained by application, not by reading. Even the machines agree on the direction: 2024 work on prompting LLMs with explicit mental models (arXiv:2402.18252) finds that giving a reasoner a structured model of the task beats leaving it to free-associate.

The Technique: One Decision, Three Lenses

The application habit fits in fifteen minutes. Take one live, significant decision and run exactly three lenses over it:

Where Argumentree Fits

The gap between knowing models and using them is a format problem, and a structured argument tree closes it by making the models' moves the default moves of the discussion. First principles: a proposal's premises are explicit nodes that must carry evidence and can be attacked. Inversion: the con side is a standing, legitimate home for the failure case. Second-order thinking: projected consequences enter as child arguments — claims with support, checkable later. Incentives: evidence fields make what backs this? a routine question rather than an accusation. Probabilistic thinking: ratings let the group express how much weight an argument deserves, instead of pretending arguments are binary.

That is the latticework as infrastructure: the lenses applied on every significant decision, by the format, whether or not anyone remembers the list that day.

The Latticework, Not the List

Munger's 1994 advice was never really collect models. It was: refuse to let experience pile up unstructured. The seven lenses above — graded honestly, three of them with full companion pages — are enough lattice for a decade of decisions. The eighth model, the one the genre never lists, is the meta-model: match the lens to the decision, at a cost the decision justifies.

So skip the next listicle and run the fifteen-minute drill on something real this week. The facts you already have are waiting for a structure to hang on.