>So naturally, if all the examples you have of successful startups involve people with characteristic X, and you have no data about people without characteristi

>So naturally, if all the examples you have of successful startups involve people with characteristic X, and you have no data about people without characteristic X, the rational thing to do as a VC or potential early employee looking for startups, or a founder or hiring manager looking for employees, is to look for people with characteristic X. Let someone else do the research; it's not up to you to invest your money or time on figuring out if people without characteristic X can also be successful. I tend to take what you are saying a little differently. It is 'rational' in the sense of the heuristics (e.g. the recognizability heuristic) that people commonly use to deal with probability. However, the underlying philosophy (i.e. constraints/assumptions) rely on the certain characteristics of the problem for them to actually be rational instead of just seem rational. We use probability heuristics a lot, poorly, and uncorrelated to the statistical expertise of those using the heuristic. We use them exactly because they seem rational, even when they are not. To go way back in economics history (Frank Knight), they rely on analytical uncertainty...which the human brain has a strong bias towards. Analytical uncertainty relies upon a potential outcome space being knowable and independent of the actions of those making the predictions. Knight argues that in entrepreneurship the future performance is fundamentally not predictable in this way. The result is that the classical conceptions of uncertainty, probability, and (which is grounded on their being a tangibly solvable or characterizable uncertainty) risk do not apply/are not appropriate. Generally these more 'Knightian' conceptions are where we see truly new ideas (i.e. disruption) come from in ways that look obvious and form new markets in retrospect but are not foreseeable using a priori prediction tools. Saying that X is stable is accurate...it is choosing to make a prediction, which is fine. I would argue that it is better approached in a Bayesian sense to try and figure out whether X is really the thing you are looking for or merely a symptom that matches your algorithm for the underlying nugget. However what is often problematic to success is a reliance on these attempts to predict what is fundamentally unpredictable, especially for startups[1]. The successes of prediction (what successes there are) are more traceable to the existence of Nash/Nash-like equilibrium in these newly created, previously unpredictable markets where there is space for competition simply because the new markets crystallize very quickly once they are initiated. The speed of crystallization (and the ability to make money via prediction) has always been a product of the rate of information dissemination being greater than the ability of companies to grow. Your Harvard drop out may incrementally improve NoSQL, she might be convincing, and she might be worth a large long term investment with a measurable/calculable risk for a target ROI. However, your Juliard student may have a higher potential to fundamentally shift NoSQL thinking and be worth a small (hypothesis driven)investment to test the potential that we cannot calculate. What you are interested in as an investor is largely up to you but they are not interchangeable. This stuff feeds into so much of the parts of the startup ecosystem that work and don't. [please forgive the state of this response, it was written quickly at my bedtime...mostly a random statistical aside to your larger point, I kind of tangented at 'market like the startup world'] [1]Wiltbank, R., Read, S., Dew, N., & Sarasvathy, S. D. (2009). Prediction and control under uncertainty: Outcomes in angel investing. Journal of Business Venturing, 24(2), 116–133. http://doi.org/10.1016/j.jbusvent.2007.11.004 http://doi.org/10.1016/j.jbusvent.2007.11.004