The Model Thinker: What You Need to Know to Make Data Work for You
One book, placed in its field — what Scott E. Page's book gets right, where it goes further than the rest, and what the rest of the field adds.
Mike West, Editor · Leading With Data · peopleanalyst.com/guides/profile-the_model_thinker
Arrow keys or swipe to move · N for notes · F for full screen
The Model Thinker: What You Need to Know to Make Data Work for You
One book, placed in its field — what Scott E. Page's book gets right, where it goes further than the rest, and what the rest of the field adds.
Mike West, Editor · Leading With Data · peopleanalyst.com/guides/profile-the_model_thinker
The problem
Quality of Reasoning and Explanation erodes when it is left to instinct instead of method.
You were taught the moves piecemeal, never the whole model.
The Model Thinker: What You Need to Know to Make Data Work for You · Mike West, Editor · Leading With Data
The model
- 01Breadth of Model Repertoire
- 02One-to-Many Application Skill
- 03Grounding Models in Data
- 04Context-to-Model Matching
- 05Diversity of Models Engaged
- 06Logical Coherence of Reasoning
- 07Recognition of Conditionality
- 08Epistemic Humility
The Model Thinker: What You Need to Know to Make Data Work for You · Mike West, Editor · Leading With Data
Step 1 of 5 · Breadth of Model Repertoire
Breadth is measured by the number of distinct assumption-structures you command, not the number of model names you recognize.
Misconception
Practitioners believe mastering one powerful, general-purpose model (usually regression or expected-value) is enough because a good model flexes to fit any problem.
Reality
Every model encodes assumptions that make some phenomena visible and others invisible; a linear model literally cannot see a tipping point, and no amount of fitting will make it. Breadth is coverage of blind spots, not redundancy.
The Model Thinker: What You Need to Know to Make Data Work for You · Mike West, Editor · Leading With Data
Step 2 of 5 · One-to-Many Application Skill
One deeply mastered model applied across ten domains beats ten models each used once, because transfer skill compounds while shallow coverage does not.
Misconception
Practitioners believe a model belongs to the field that taught it — the SIR contagion model is for epidemiology, network centrality is for sociology — and applying it elsewhere is loose metaphor rather than rigorous analysis.
Reality
Models are abstract structures whose variables are labels, not commitments; the same diffusion equations that track a virus track a rumor, a product adoption, or a bank run, and the transfer is legitimate precisely when the underlying mechanism matches, not when the surface story does.
The Model Thinker: What You Need to Know to Make Data Work for You · Mike West, Editor · Leading With Data
Step 3 of 5 · Grounding Models in Data
In-sample fit is table stakes; a model is only grounded once it survives data it did not see during construction.
Misconception
Practitioners believe that if a model fits the historical data well, it has been validated and can be trusted to explain and predict.
Reality
Fit measures how well a model reproduces data you already have, not whether its mechanism is right or its predictions hold out of sample; a flexible enough model fits anything, which is why calibration on effect magnitudes and testing on withheld data matter more than in-sample fit.
The Model Thinker: What You Need to Know to Make Data Work for You · Mike West, Editor · Leading With Data
Step 4 of 5 · Context-to-Model Matching
Match the model to the class of outcome you need (predict, explain, or explore), not to the tool you find most impressive.
Misconception
Practitioners believe the more sophisticated or higher-fidelity model is always the safer choice.
Reality
Granularity is a tradeoff, not a virtue: a coarse model can outperform a detailed one when data is noisy or stakes reward robustness over precision, and behavioral assumptions (rational vs. rule-based vs. adaptive) must fit the actors, not your preference.
The Model Thinker: What You Need to Know to Make Data Work for You · Mike West, Editor · Leading With Data
Step 5 of 5 · Diversity of Models Engaged
Error cancellation requires models with independent, not correlated, mistakes — structural difference matters more than model count.
Misconception
People think diversity means running variations of the same model — three regressions or three versions of supply-and-demand.
Reality
Diversity that reduces error comes from models that emphasize different causal forces (incentives, networks, feedback, information) so their errors point in different directions; near-duplicate models add computation but not insight.
The Model Thinker: What You Need to Know to Make Data Work for You · Mike West, Editor · Leading With Data
The Model Thinker: What You Need to Know to Make Data Work for You · Mike West, Editor · Leading With Data
The cheat sheet
- Breadth of Model RepertoireBreadth is measured by the number of distinct assumption-structures you command, not the number of model names you recognize.
- One-to-Many Application SkillOne deeply mastered model applied across ten domains beats ten models each used once, because transfer skill compounds while shallow coverage does not.
- Grounding Models in DataIn-sample fit is table stakes; a model is only grounded once it survives data it did not see during construction.
- Context-to-Model MatchingMatch the model to the class of outcome you need (predict, explain, or explore), not to the tool you find most impressive.
- Diversity of Models EngagedError cancellation requires models with independent, not correlated, mistakes — structural difference matters more than model count.
- Logical Coherence of ReasoningMake assumptions explicit first — hidden premises are where most reasoning errors hide.
- Recognition of ConditionalityA claim without stated conditions is not more general—it is just less honest about where it fails.
- Epistemic HumilityThe point of a model is not to be right but to be wrong in an informative, correctable way.
The Model Thinker: What You Need to Know to Make Data Work for You · Mike West, Editor · Leading With Data
Sources
- The Model Thinker: What You Need to Know to Make Data Work for You — Scott E. Page
The Model Thinker: What You Need to Know to Make Data Work for You · Mike West, Editor · Leading With Data