Learn it → Apply it → Save as Skill
We search where the light is good, not where the answer actually is. Convenience quietly substitutes for relevance: the easy-to-measure number replaces the meaningful one, the dataset already on hand replaces the one the question needs, and the thing that can be checked in an afternoon replaces the thing that was actually asked.
The Dr. Wallace’s Favorite Mental Models is designed to turn useful ways of thinking into reusable tools. Each model gives you a fast visual overview, deeper interactive learning, practical application exercises and an AI-ready Markdown skill you can save for future use.
Instead of merely reading about mental models, you can gradually build your own AI skills library: a collection of portable reasoning skills that can help ChatGPT, Claude, Grok and other AI assistants apply the mental models you value.
When evidence is hard to reach, there is a strong pull toward examining whatever is nearby, already logged, or simple to count, even when it has little to do with the real question. The search feels productive because something is being measured. It is just the wrong something.
Important: this is not a claim that convenient evidence is useless. It is a warning that convenient evidence quietly gets treated as sufficient, when it was only ever available.
The model is most useful whenever an investigation, a metric, or a decision has quietly narrowed to whatever is easiest to observe, rather than what was actually asked.
The method separates where you are looking from where the answer is likely to be, then asks whether that gap can be closed.
State the actual question before opening any data. Write it down before you open a dashboard, a log file, or a survey.
List where you are currently searching, and name what makes that location convenient: already collected, already visible, easy to measure, already at hand.
Ask whether the convenient location is also the relevant one. If they diverge, name what sits on the other side of the street.
If the relevant evidence is reachable at a reasonable cost, go get it. If it is not, say so plainly instead of letting convenient evidence stand in for it.
Open a use case to see what the model changes.
Administrative and platform data are seductive because they already exist. Before treating a convenient dataset as an answer, check whether it can actually observe the outcome in question, or only a nearby proxy for it.
Engineers tend to investigate the system with the most verbose logging, not necessarily the one that failed. Before closing an incident, check whether the explanation fits the evidence, or the evidence is just what happened to be captured.
Detection systems are tuned to known signatures, which are well lit by design. Novel fraud or intrusion routes tend to happen in the dark, in traffic nobody is logging closely. Ask what an attacker who knew your monitoring would deliberately avoid.
Interviews measure interview performance well and job performance only indirectly. Before trusting the signal, ask whether it measures the thing you are hiring for, or the thing that happens to be easy to observe in an hour.
A reported statistic, such as arrests, hospital visits, or click-throughs, often measures what an existing system happens to record, not the underlying problem. Ask what would be captured if nobody were already counting it.
Asking five people nearby feels like doing research, but proximity is not expertise or relevance. Before acting on the feedback you gathered, ask who would need to be consulted deliberately rather than conveniently.
This is an illustrative example designed to teach the method.
Expand for where the parable comes from and who turned it into a documented research idea.
Two origins are documented. A Western version, a person searching under a streetlight for something lost elsewhere because the light is better there, appears in print by the 1920s, including a July 1926 Alabama State Bar Association report. A related teaching story is older: Idries Shah’s 1964 collection “The Sufis” records a Nasreddin tale in which a man searches his yard for a ring he dropped inside a dark room, because the yard has much more light.
Philosopher of science Abraham Kaplan gave the joke its academic name, the principle of the drunkard’s search, in his 1964 book “The Conduct of Inquiry: Methodology for Behavioral Science,” using it to warn that researchers often study what current methods make easy to study, rather than what the question actually requires. Linguist Noam Chomsky later used the same joke to describe how entire fields of science can gravitate toward tractable problems over important ones.
The specific phrase gained wide circulation through science journalist David H. Freedman’s 2010 Discover magazine article, “Why Scientific Studies Are So Often Wrong: The Streetlight Effect,” which applied the parable to the tendency of published research to cluster around convenient, well-instrumented questions. The underlying idea is older than the phrase. The phrase is what made it citable.
Having a number is not the same as having the right number. A metric that exists is not automatically the metric that matters.
Hours spent analyzing the convenient dataset can feel like progress on the real question, even when it is not.
Data that would show a problem, but was never collected, looks identical to data showing no problem at all.
Chasing the theoretically right evidence at an unreasonable cost can be worse than honestly qualifying the convenient evidence you already have.
The convenient dataset is usually accurate for what it records. The mistake is asking it a question it was never built to answer.
Convenient data is frequently also the right data. The model calls for checking the fit, not discarding anything that happens to be easy to obtain.
it genuinely covers the population and outcome in question, not just a related one that happens to be easier to reach.
the issue is specifically that failures or dropouts have gone missing from view, a close cousin of this model with its own worked cases.
the problem is which tool or method gets reached for out of habit, rather than where the evidence itself is being sought.
This browser tool structures your thinking. It does not send your information anywhere or perform AI analysis.
Save as Skill converts the model into portable Markdown instructions that teach an AI when and how to apply the Streetlight Effect, what questions to ask, what evidence to seek, what mistakes to avoid and how to structure its analysis.
Save the skills you find useful and organize them into your own reusable library. The Markdown is intentionally model-neutral, so it can be used wherever an AI system accepts uploaded files, project knowledge, custom instructions or reusable text context. Platform capabilities differ, so this is a portable skill file rather than a claim of one-click installation.
No. The Law of the Instrument is about which tool or method gets used out of habit, a hammer sees nails everywhere. The Streetlight Effect is about which location or evidence gets searched out of convenience. A team can reach for exactly the right tool and still point it at the wrong data.
No. Convenient data is frequently also relevant data, which is exactly why it became convenient in the first place. The model is a check on fit, not a blanket suspicion of anything that happens to be easy to obtain.
Say so plainly, and treat the convenient evidence as a partial answer or a lower bound rather than a complete one. An honest gap, clearly labeled, is more useful than an answer that quietly substituted a different, easier question.