How Groups Reach Consensus

The class before spring break is always hard. Students don’t want to be there; and, honestly, I don’t want to be there, either. Everyone has one eye on the clock counting down the minutes until vacation.

So rather than offering a half-hearted lecture that will be ignored by all, over the past six years I have used the time to do a simple experiment aimed at teaching the class how social networks (aka crowds) reach consensus - in the belief that the concepts are as applicable to economic decision making as they are to political choices and the financial markets.

The experiment is simple. I place a jar filled with 200 pennies on the table in front of the classroom and then ask the students to write down on a piece of paper their estimate of the number of coins as well as a simple justification – why they estimated the number they did. Once I have collected all the responses, I then ask the students to pair up in twos and, together, to provide a new estimate. After collecting those answers, I pair up the pairs to form groups of four and ask for another set of estimates. I repeat the group pairing process (groups of 4, 8, 16, 32 students etc.) until the entire class comes together as one to offer a final “collective” estimate.

Before revealing any of the estimates, I ask the class to talk about the process by which they individually and then collectively came up with their figures. I want the students to stop obsessing about whether their estimate is right and, instead, to analyze how they came up to consensus – to look at the means, rather than the end.

While over the years there have been slight variations, what is remarkable is just how consistent the process is from class to class.

At an individual level, students’ rationales for their initial estimate fall into six general categories. Some students use complex mathematical and geometric formulas to “calculate” their estimate. 

Some simply choose round numbers – 50, 100, 250 etc. – while others deliberately pick a number next to or near a round number in the belief that I would never put a round number of pennies in the jar. That would be too obvious, they say.

Some students pick a particular “lucky” number, while still others reference prior experience with coins – “It looks like three coin rolls when I worked as a bank teller,” or “It looks like what I had in the cash register each morning, when I worked at a supermarket, times roughly 2.” 

More than half of the students, though, admit to guessing, with most of them supporting their specific estimate with a rationalization that might be best be summed up as “It just felt like ___ pennies.” They went with their gut.

As the image just above shows, it is remarkable how specific we can be using self-admitted vague feelings to guide us. Our very precise answer to questions like "How many coins are in the jar," aren’t generated with precision. Our quickly-developed feelings routinely trump our slow-to-conclude cognitive processes.

Over the years, I have received individual estimates for the 200-penny jar ranging from about 35 to almost 700. Interestingly, the well-rationalized, mathematically-driven estimates have been no more accurate than the gut-driven figures. I have had students “calculate” a jar filled with 450 pennies and I have had other students “guess” 45. No single specific rationalization has offered greater accuracy than another over the years.

What has offered great accuracy, however, have been the averages of the individual estimates as well as the averages of the highest and lowest individual estimates. More than 75% of the time, both have come closer to the actual number in the jar, 200, than the group’s final estimate. Consistent with Sir Francis Galton’s 1906 bull-weighing statistical analysis, taken together independent agents are wiser than the collective crowd.

The class’ discussion on the process of group decision-making quickly explains why that is the case.

Even in just the first pairing, one can see how visceral the decision is to reject the extreme estimates. Unknown to the participants, the first conversation narrows the range of estimates dramatically – typically down to a range of roughly 125-275 coins in the jar. On more than one occasion, I have seen students with two similar extreme answers (coincidentally placed in a group of four during the second pairing) rapidly capitulate to a far more moderate “it just feels right” consensus. Unknowingly, "outliers" somehow sense that their view is extreme and quickly fall in line with the crowd, even when the crowd is just one or two other people.

But that isn’t the only dynamic at work. So, too is the dominance of extroversion and System 1 thinking in a crowd. In a group, loud is perceived to be more “accurate” than quiet; and simple more “accurate” than complex.

This may seem like a harsh statement – particularly for System 2-heavy introverts – but to a crowd, extroverted System 1 communication equates to confidence. Just turn on CNBC and you can see this in action every day. A crowd quickly loses interest when a System 2 thinking introvert tries to get his point across. Crowds communicate in easy-to -digest soundbites and you must comply to be heard. Crowds are a cocktail party, not a classroom.  In a group, emotional intelligence beats intellectual intelligence hands down.

Now having said that, I have never seen a single extroverted System 1 thinker successfully drive the class consensus to his/her individual estimate – even when the individual has estimated the number of pennies in the jar correctly. While, these “influencers” frequently succeed in persuading groups of 4, 8 and even 16 to settle on a given number, when the crowd gets larger than that, and the range of answers has narrowed – typically within no more than, say, a 50-penny range – the class always turns to averaging: A group of 16, for example, will average the two estimates from the groups of eight, with the group of 32 then averaging the two groups of 16’s figures.

For a crowd, the energy and cognitive attention needed to specifically conclude that the jar holds 182 rather than 187 pennies, for example, is way too great. Crowds quickly lose interest in System 2 thinking-required precision and instead resort to System 1-driven averaging. For a crowd, fast and done is far better is slow and accurate.

Now you might think that adding a financial incentive would change that dynamic, but it really doesn’t. In years when I have offered a cash prize, all adding an incentive has done is to increase the complexity of the rationalizations behind student’s individual estimates and to stretch the time (and added volume to) the discussion in the smaller groups. The incentivized introverts work harder, albeit no more accurately, to formulate their estimates, while the incentivized extroverted “influencers,” who were already going to dominate the discussion anyway, simply become more strident among the groups of 4 and 8. 

By the time we’ve reached groups of 16, though, the classes still resort to simple averaging, albeit now with more collective frustration. Even with a reward, done is far better than accurate for a crowd.

To be clear, what I have just shared are observations of a non-statistically significant sample. A dozen classes undertaking an uncontrolled experiment would quickly fail the rigor of academic research.

That said, the pattern of consensus forming has been consistent - from “individualized” to “influenced” to “averaged.” Like a piece of sea glass, the extreme sharp edges are quickly broken off and then the surface is shaped and smoothed - first by large objects and then by simple, repetitive abrasion in the sand.

For what it’s worth, as a financial market observer, I routinely see the same crowd consensus building process at work. Individual market themes are first brought forward by technical analysts, investors, traders and other subgroups, but they must "feel right" before they are then articulated and promoted in the financial media by “influencers” on CNBC, Bloomberg and Fox Business – where extroversion and confidence dominate. With time and repetition, these themes are then simplified and come together to form singular narratives, like “Buy the FANGs” or “Sell Retailers” that everyone seemingly understands. 

Just like my students, crowds of investors quickly lose interest in System 2 thinking-required precision and, instead, resort to System 1 thinking averaging. For Wall Street, fast and done is far better than slow and accurate.

And the same process appears in politics, too, as local grassroots candidates are propelled by "influencers" into national campaigns. Successful presidential candidates are "averaged." They "feel right" to voters. To get Main Street’s support politicians must use slogans and images that require the smallest use of voters’ System 1 thinking possible.

Finally, the same group decision making dynamic work will hold true for this blog post. For example, if this blog gets just 85 clicks, 3 likes and 1 reshare, I will know that I have sadly, (and once-again) offered the crazy System 2 ideas of an extreme introvert that went on far too long and that no one wanted to read.

I will have written 1,600 words when 500 better sculpted and applied to a different topic today would have felt far better.

On the other hand, if I get 10,000 clicks, 7,000 likes and 5,000 reshares, I will know that I have not just refined my crazy System 2 ideas into easily-understood System 1 concepts, but those thoughts have been eagerly embraced by “influencers” and then averaged into the broader conversation. Either way, I will know soon enough.

Whether you are growing a business, running a political campaign or investing in the stock market, understanding how your audience reaches consensus matters. Like it or not, their process - not just your idea - will determine your success or failure.

 Peter Atwater is the president of Financial Insyghts LLC, a research and consulting firm focused on the role of confidence in economic, financial, political and social decision-making. The author of "Moods and Markets" and an adjunct professor in the Economics Department at The College of William & Mary, he speaks and writes frequently on the impact of changing confidence on current events.