Imagine you’re cruising down the highway and notice that you are running low on fuel. Your GPS shows 10 gas stations ahead on your route. Naturally, you want the cheapest option. You pass the first few and observe their prices before approaching one with a seemingly good deal. Do you stop, not knowing how sweet the bargains could be up the road? Or do you continue exploring and risk regret for rejecting the bird in hand? You won’t double back, so you face a now-or-never choice. What strategy maximizes your chances of picking the cheapest station?
Researchers have studied this “best-choice problem” and its many variants extensively, attracted by its real-world appeal and surprisingly elegant solution. Empirical studies suggest that humans tend to fall short of the optimal strategy, so learning the secret might just make you a better decision-maker.
The scenario goes by several names: two examples are “the secretary problem,” which involves ranking job applicants by their qualifications (instead of gas stations or the like by prices), and “the marriage problem,” in which you rank suitors by eligibility. All incarnations share the same underlying mathematical structure, in which a known number of rankable opportunities present themselves one at a time. You must commit yourself to accepting or rejecting each of them on the spot with no take-backs (if you decline all, you’ll be stuck with the last choice). And the opportunities can arrive in any order.
On supporting science journalism
If you're enjoying this article, consider supporting our award-winning journalism by subscribing. By purchasing a subscription you are helping to ensure the future of impactful stories about the discoveries and ideas shaping our world today.
Let’s test your intuition. If the highway were lined with 1,000 gas stations and you had to evaluate them sequentially and choose when to stop, what are the chances that you would pick the absolute best option? If you chose at random, you would find the best only 0.1 percent of the time. Even if you tried a strategy cleverer than random guessing, you could get unlucky if the best option happened to show up quite early, when you lacked the comparative information to detect it, or quite late, by when you might have already settled for fear of dwindling opportunities.
Amazingly, the optimal strategy results in the number-one pick being selected almost 37 percent of the time, and its success rate doesn’t depend on the number of candidates. Even with a billion options and a refusal to settle for second best, you could find your needle in a haystack more than a third of the time. The winning strategy is simple: Reject the first approximately 37 percent of the choices no matter what. Then pick the first option that is better than all the others you’ve encountered so far. (If you never find such an option, take the final one.)
Adding to the fun, mathematicians’ favorite little constant, e = 2.7182..., rears its head in the solution. Also known as Euler’s number, e holds fame for cropping up all across the mathematical landscape in seemingly unrelated settings—including in the best-choice problem. Under the hood, those references to 37 percent in the optimal strategy and corresponding probability of success are actually 1/e, or about 0.368. The magic number comes from the tension between wanting to see enough samples to feel informed about the distribution of options and not wanting to wait too long lest the best pass you by. The proof argues that 1/e balances these forces.
The first known reference to the best-choice problem in writing appeared in Martin Gardner’s beloved Mathematical Games column in Scientific American. The problem had spread by word of mouth in the mathematical community in the 1950s, and Gardner outlined it in the February 1960 issue by describing a puzzle game called Googol, following up with a solution the next month.
Today the problem generates thousands of hits on Google Scholar as mathematicians continue to study its many variants: What if you’re allowed to pick more than one option, and you win if any of your choices are the best? What if an adversary chose the ordering of the options to trick you? What if you don’t require the absolute best choice and would feel satisfied with second or third? Researchers study such when-to-stop scenarios in a branch of math called optimal stopping theory.
Looking for a house—or a spouse? David Wees, who designs math curricula, applied the best-choice strategy to his personal life. While apartment hunting, Wees recognized that to compete in a seller’s market, he would have to commit to an apartment on the spot at the viewing before another buyer snatched it. With his pace of viewings and six-month deadline, he extrapolated that he had time to visit 26 units. And 37 percent of 26 rounds up to 10, so Wees rejected the first 10 places and signed the first subsequent apartment that he preferred to all the previous ones. Without inspecting the remaining batch, he couldn’t know whether he had in fact secured the best, but he could at least rest easy knowing he had maximized his chances.
When in his 20s, Michael Trick, now dean of Carnegie Mellon University in Qatar, applied similar reasoning to his love life. He figured that people begin dating at 18, and he assumed that he would no longer date after 40 and would meet potential partners at a consistent rate. Taking 37 percent of this time span would put him at age 26, at which point he vowed to propose to the first woman he met whom he liked more than all his previous dates. He met Ms. Right, got down on one knee and was promptly rejected. The best-choice problem doesn’t cover cases where opportunities may turn you down. Perhaps it’s best we leave math out of romance.
Empirical research finds that people tend to stop their search too early. So applying the 37 percent rule could improve your decision-making, but be sure to double-check that your situation meets all the conditions: a known number of rankable options is being presented one at a time in any order, you want the best, and you can’t double back. Nearly every conceivable variant of the problem has been analyzed, and tweaking the conditions can change the optimal strategy in ways large and small. For example, Wees and Trick didn’t really know their total number of potential candidates, so they substituted in reasonable estimates instead.
If decisions don’t need to be made on the spot, that nullifies the need for a strategy entirely: simply evaluate every candidate and pick your favorite. If you can settle for a broadly good outcome instead of the absolute best option, then a similar strategy still works, but a different threshold, typically lower than 37 percent, becomes optimal. Whatever dilemma you face, there’s probably a best-choice strategy that will help you quit while you’re ahead.