# Base Rates ## The Idea in Brief Before considering the specifics of any situation, ask: how often does this type of thing happen in general? The base rate is the overall frequency of an event in a reference class. Starting with the base rate and then adjusting for specific factors is more accurate than reasoning from specifics alone. Most people skip this step entirely—and their predictions suffer. --- ## Key Concepts ### The Outside View Kahneman and Tversky call this "the outside view"—putting a problem into a comparative perspective that downplays its uniqueness. What's the success rate for startups in this sector? What's the typical timeline for projects of this type? What percentage of acquisitions create value? The inside view asks: what's special about this case? The outside view asks: what usually happens? ### Reference Classes A reference class is the set of similar cases you use for comparison. Choosing the right reference class is crucial. Is your software project like "all software projects" or "software projects in regulated industries with legacy systems"? The more specific and appropriate the reference class, the more useful the base rate. ### Updating from the Base Rate The base rate is your starting point, not your final answer. Once you have it, adjust based on specific factors that make this case different. The Bayesian approach: start with the prior (base rate), update with new evidence, arrive at a posterior probability. But you need the base rate first—without it, you're just telling stories. ### Why We Ignore Base Rates We're drawn to the vivid specifics of a situation. The compelling narrative, the confident expert, the unique circumstances. Base rates are abstract and boring. But abstract and boring often beats vivid and wrong. --- ## Implications **In forecasting:** Superforecasters start with the outside view before considering inside factors. "What usually happens in situations like this?" before "What's special about this situation?" **In planning:** The planning fallacy exists because we focus on the specific plan rather than asking how long similar projects actually took. Check the base rate for projects of this type before committing to a timeline. **In investing:** Before analysing a specific company, ask: what's the typical return for this asset class? What's the success rate for this type of investment? Your specific analysis should update from that baseline, not replace it. **In hiring:** Before getting excited about a candidate's specific qualities, ask: what's the success rate for hires from this background into this role? What do the numbers say about interview predictions vs actual performance? --- ## Sources - [[Everything Is Predictable]] — Bayes' theorem as the mathematical framework for updating from priors (base rates) to posteriors - [[Superforecasting]] — Tetlock's superforecasters adopt the outside view first, then adjust for inside factors - [[Antifragile]] — Taleb on the importance of empirical base rates over theoretical models - [[The Most Important Thing]] — Howard Marks on using base rates to avoid overconfidence in specific predictions --- ## See in Field Notes - [Decision Architecture](https://www.anishpatel.co/decision-architecture/) — Statistical discipline: weight evidence proportionally, don't lurch in response to single data points