The Less-Is-More Decision Algorithm
Make better high-stakes decisions by using fewer variables, not more
The Less-Is-More Decision Algorithm is the principle that in many complex decision domains, focusing on a small number of critical variables produces more accurate outcomes than gathering and weighing all available information. This counterintuitive finding is demonstrated most powerfully through the Goldman algorithm for diagnosing heart attacks in emergency rooms, which used only three factors to outperform experienced physicians who gathered extensive patient histories.
The framework draws on the research of Lee Goldman, who developed a decision tree for chest pain triage that relied on just three pieces of information: the ECG pattern, whether the patient had fluid in the lungs (indicating unstable angina), and whether systolic blood pressure was below 100. When Cook County Hospital in Chicago implemented this algorithm, they found it was seventy percent better than the old method at recognizing patients who were not having heart attacks, while being equally effective at identifying those who were.
The key insight is that extra information beyond the critical variables does not just fail to help; it actively hurts. When doctors knew a patient was elderly, smoked, and had high cholesterol, they overrode normal ECG readings and admitted patients unnecessarily. The additional context created noise that obscured the signal from the truly predictive variables. This principle applies far beyond medicine to any domain where decision-makers are tempted to gather more data than they need.
- In many complex domains, a small number of critical variables carries nearly all the predictive power.
- Extra information beyond the critical variables often introduces noise that degrades decision accuracy.
- Simple algorithms consistently outperform expert judgment in domains with identifiable predictive patterns.
- Human experts tend to overweight vivid, contextual information (age, lifestyle, demeanor) at the expense of objective diagnostic indicators.
- The hardest part of implementing less-is-more is overcoming the deeply held belief that more information always helps.
- Identify the Critical Variables in Your DomainResearch or analyze which small set of factors carries the most predictive weight for the decisions you face. Goldman found that ECG patterns, lung fluid, and blood pressure predicted heart attacks better than dozens of other factors combined. Every domain has its equivalent vital signs.Pro tipLook for variables that are objective and measurable rather than subjective and interpretive. The Goldman algorithm deliberately excluded physician impression and patient history.WarningIdentifying the right critical variables requires rigorous analysis; getting them wrong creates overconfident but inaccurate decisions.
- Build a Simple Decision TreeStructure the critical variables into a clear decision flow. Goldman's algorithm was a branching tree: first check the ECG, then check for unstable angina, then check blood pressure. Each branch leads to a clear action. The tree must be simple enough to use under time pressure.Pro tipThe best decision trees fit on one page and can be followed by someone under significant stress.
- Deliberately Exclude Tempting But Non-Predictive InformationThe most difficult step is actively ignoring information that feels relevant but is not predictive. At Cook County, doctors had to stop considering patient age, lifestyle, medical history, and personal impression when following the algorithm. This requires discipline and institutional support.Pro tipMake a specific list of information you will not consider, and review it regularly to reinforce the discipline of exclusion.WarningResistance will be fierce. Experienced professionals feel that ignoring available information is irresponsible. The data must speak louder than the feeling.
- Test the Algorithm Against Expert JudgmentBefore full deployment, run the algorithm alongside existing decision processes on historical cases. Goldman tested his algorithm against years of actual emergency room data. The data must demonstrate that the simpler approach matches or exceeds the complex one before people will adopt it.Pro tipPresent the comparison data to the experts who will use the algorithm. Seeing their own judgment outperformed by a simple chart is the most powerful persuasion tool.
- Implement with Ongoing CalibrationDeploy the algorithm with continuous monitoring of outcomes. Track both the cases where it performs well and the cases where it fails. Use failures to refine the critical variables and branching logic over time.Pro tipBuild in a formal exception process for cases that feel truly anomalous, but track how often exceptions are invoked and whether they improve or degrade outcomes.WarningOver time, there is a tendency to add variables back into the algorithm, which gradually erodes its effectiveness. Guard simplicity aggressively.
Cook County Hospital implemented Lee Goldman's simple decision tree for triaging chest pain patients, replacing the traditional approach of extensive history-taking and physician judgment. The algorithm used only ECG readings, presence of lung fluid, and blood pressure to categorize patients into risk levels.
Brendan Reilly gave twenty identical chest pain case histories to a group of experienced cardiologists, internists, and emergency physicians and asked them to estimate the probability of heart attack for each case. The estimates ranged from zero to one hundred for the same patient across different doctors.
Coca-Cola conducted hundreds of thousands of sip tests that consistently showed consumers preferred New Coke. But the sip test measured only one variable (initial taste preference) in an artificial context. The full experience of drinking Coke involved brand associations, packaging, and sustained consumption that the simplified test excluded.
Gladwell tells the story through Cook County Hospital's Emergency Department, where chairman Brendan Reilly discovered that doctors showed almost no agreement on chest pain diagnosis when given identical case histories. Estimates of heart attack probability ranged from zero to one hundred for the same patient. Reilly then discovered the work of Lee Goldman, a mathematician and physician who had spent years developing and testing a simple algorithmic decision tree. When Reilly introduced Goldman's algorithm to Cook County, it initially met fierce resistance from physicians who believed their clinical judgment, informed by extensive patient data, must be superior to a simple chart. The data proved them wrong.