Picture it now: You’re at a casino watching someone play roulette. The ball lands on red four times in a row. A gambler at the table quickly points out, “Oh boy! Reds are hot!” and places their chips accordingly. The next spin occurs, and the ball settles on red once again. The gambler is convinced they’re on to something and continue this strategy until they’re certain reds have officially “cooled off.” Assuming a fair roulette wheel and croupier, the non-random pattern identified by the gambler was simply non-existent. They were under the dreaded influence of the clustering illusion — a tendency to identify as non-random the inevitable (and most certainly random) streaks or clusters that arise in small samples.[i To put it another (shorter) way: The clustering illusion is the tendency to see patterns in purely random events.HeadsTailsAt its core, this cognitive bias is simply a sample-size issue. Our brain attempts to make the most of whatever information it has. When the amount of info is small, patterns will present themselves, and the mind tends to view these patterns as tendencies or even causal relationships. However, if you’re able to see the result of a large number of coin flips, the mind will quickly determine the coin has no bias associated with it, no memory and no preference. Again, to put it another (shorter) way: When you can see the entire picture, random streaks of consecutive heads or tails are viewed as just that — random streaks.
Flip outDue to the nature of truly random events, there’s bound to be some clustering of specific outcomes. For example, if you flipped a coin 100 times, you almost certainly won’t get a perfect pattern of heads, tails, heads, tails and so on. You’re far more likely to see streaks of consecutive heads or tails within the overall sample. In fact, clustering to some extent is evidence of an event being random. In a sample as small as 20 coin flips, there’s a 50% chance of getting four heads in a row.[ii] Don’t believe us? Grab a coin and try it. Don’t have a coin? No problem. We created a random coin-flip generator that will toss 20 fair coins all at once. Repeat it as many times as you like, and notice all the various clustering that occurs in a sample even as small as 20 flips.
A small problemUnderstanding the probability of an event’s various outcomes can help you better prepare for what’s likely to come. But this can just as easily lead to being fooled by small samples. For example, assume an individual flips a coin a bunch of times as others guess what it will land on. The first three are tails, and one of the participants — knowing the overall distribution of heads and tails in a series of flips is 50/50 — says, “The next flip has got to be heads.” That person has identified a pattern in pure randomness. The next flip is no more likely to land on heads than any of the previous flips or any flips to follow, so the likelihood of heads remains 50%. This concept was dubbed “Belief in the Law of Small Numbers.”[iii] It points out that humans have a tendency to expect small samples to still produce outcomes that match the probabilities identified in very large samples. In large samples — let’s say 1,000 flips or more — you’d expect the coin to land on tails in about 50% of the coin flips and heads for the other 50%. But in small samples, a result different from 50/50 is quite likely to occur. Identifying this cognitive bias in a coin-flipping exercise is relatively simple. Some circumstances are far less clear, which is what makes being aware of this potential bias so valuable. By keeping it in mind, you’re far less likely to be fooled by small groupings of event outcomes that are actually a result of chance.
Not-so-sound investmentInvesting presents a great real-world example of this. Everyone is looking for that get-rich-quick investment option, and “beating the market” is a constant goal for investors big and small. But while the stock market and other investment options are surely not as random as flipping a coin, the clustering illusion can still be present. In any given year, several mutual funds, individual stocks, commodities and other investment vehicles will outperform the S&P 500. Some will even pull this off multiple years in a row. Similarly, an investor might view a series of market-betting returns as a trend or evidence of a well-managed mutual fund, when really they’re just observing a small sample of positive results within the natural ebb and flow of varying returns. This doesn’t mean certain mutual funds aren’t being managed in a way that could result in consistently strong returns; it just means a savvy investor would be wise to not be fooled by short-term returns and would benefit from considering larger samples. After all, even a bunch of blindfolded monkeys with darts have managed to beat the market over the years.[iv]
But what about you?The same concept can apply when you evaluate your next business decision. Whether it’s estimating the impact of a new initiative or promotion, segmenting customers by common characteristics, or evaluating RFP submissions for a needed service, understanding this potential bias can save you time and money. To make things a little easier, here are a few things to keep in mind so you don’t fall prey to the clustering illusion when it really matters:
- The larger the sample size, the better.
- For analysis that involves a temporal aspect, make sure to evaluate over a long period of time. Depending on the circumstances, you might benefit from not even peeking at results in early stages of a new initiative.
- Consider the natural variance in the metrics you’re evaluating.
- Is the recent uptick/downtick within the range of past variation?
- Control for as much as possible so you can limit the number of potentially “random” influencers and hone in on the impact of your efforts or the relationships among key factors.
Looking for more?
- Esther Ingliss-Arkell, “How a cognitive bias made people think their neighbors were spies”
- George Johnson, “Cancer Cluster or Chance?”