Diversification Bias

For a group of twenty-five children who were trick-or-treating on the evening of October 31, 1993, Halloween was going exactly as planned.

Approach the house.

Ring the doorbell.

Select the candy.

Pretty typical, right?

What these kids didn’t know was they were the unwitting participants in an experiment.

On the same street, coming in the opposite direction, a smaller group of children was doing the same thing. This group, too, went from house to house, rang the doorbell, picked their candy from a bowl. At the end of the evening, each group of trick-or-treaters made an inventory of their candy.

But their inventories looked very different. While the first group got lots of their favorite candy, this the second group got less of their favorite candy—much less.

But why? After all, both groups had visited the same houses and been offered the same candy.¹ Why would one group choose candy they liked less?

Here’s why. One group of kids went one way down the street. When this group approached a house, the person on the other side of the door showed them a bowl and offered them a choice of one piece of candy. They then proceeded to the next house where they were given the same options—and the kids made the same choice. Two houses, two pieces of candy, the favorite chosen each time. Meanwhile, for the kids in the other group coming the opposite way, they were shown the same candy. However, instead of spreading their choice across two houses, they made their choice at only one house: they were offered a bowl of candy and asked to make two selections.

All the kids saw the same candy and had the chance to pick two; some just made two choices at two houses from two bowls, while others made two choices at one house from one bowl. The difference between these two groups is that some kids made their decisions all at once, at other people made their decisions over time. Some of these decisions were made simultaneously and others were made sequentially.

This is called diversification bias.

Here’s the full definition:

Diversification bias describes the tendency to choose more variety—to diversify—when making a simultaneous decision, and to choose less variety when making the same decisions sequentially.

Does diversification bias lead to worse choices?

At first glance, it seems that making simultaneous choices that result in diversification leads to worse outcomes. We have already seen that trick-or-treaters get less of their favorite candy, playlist creators listen to fewer of their favorite songs, and gamblers lose more money. A 1995 study by Daniel Read and George Loewenstein tried to quantify the various outcomes by comparing sets of simultaneous choices with sequential choices.² To do this, they asked people to pick three snacks from a set of six: Snickers, Oreos, milk chocolate with almonds, tortilla chips, peanuts, and cheese-peanut butter crackers. Everyone saw the same six snacks. The researchers split their subjects into seven cohorts. To calculate the level of variety seeking in each cohort, Read and Loewenstein took the median number of choices within that cohort. The higher the number, the more variety. Here are the results for each cohort:

As predicted, sequential choices produced less variety than simultaneous choices. The sequential choice group (cohort 1) chose a median of 1.61 snacks out of a possible three, while the simultaneous choice group (cohort 4) chose a median of 2.27 snacks. Or, to put this slightly differently: simultaneous choosers chose about 40.9% more variety.

On average, sequential choosers got more of their favorite snacks, and simultaneous choosers got less. It would appear that diversification bias leads to worse outcomes. 

Diversification bias comes with real-world consequences:

• When witnesses to a crime are asked to identify a suspect from a line-up of possible suspects, they are more likely to make a false identification when they see everyone at once than when they view suspects one at a time.³
• When hiring managers interview and make job offers one by one ( sequentially) they’ll get a less diverse workforce than if the hiring for multiple jobs at once (simultaneously). In the words of Rory Sutherland: “In assessing job applicants…. It is assumed that ten groups each choosing one candidate will make the same ‘optimal’ choice as one group choosing ten. They won’t.”⁴
• When grocery shoppers go to the store less often and do so with larger lists, they’ll buy more variety—and more food they don’t like.⁵
• When people plan for retirement, they tend to spread their investments equally across the array of funds presented to them—regardless of how risky each fund may be.⁶

In the rest of this post, we’ll explore why you are biased toward diversification when you make simultaneous choices. Here’s our roadmap:

1. Diversification bias is caused by judgment errors
   1. Satiation: you overestimate how long you’ll feel the pleasure generated by a good experience
   2. Time contraction: you misjudge how long an increment in the future will feel
   3. Affective forecasting errors: you misjudge how you will feel in the future
   4. Repeat experiences: you underestimate how much you’ll enjoy repeating an experience
2. Diversification bias is caused by the choice set
   1. Choice bracketing: you arbitrarily and equally divide your choices by the number of options.
   2. No strong favorite: you are more likely to diversify when there is not a favored option
3. Diversification bias is caused by environmental factors
   1. Sex ratios: you are more likely to diversify when you are surrounded by more people of your sex than people of the opposite sex.
   2. Public vs. private choices: you are more likely to diversify in public because you believe others like variety more than you.
   3. Sleepiness: you are more likely to diversify when you are tired.
   4. Time of day: you are more likely to diversify in the afternoons than in the mornings.
   5. Confined spaces: you are more likely to diversify if you’re in a confined space.
   6. Signalling: to signal expertise, novices choose variety

Let’s dive in.

Diversification bias is caused by judgment errors

Simultaneous decisions are riskier than sequential decisions. The riskier a decision, the more likely you are to rely on decision-making frameworks that are prone to bias.⁷ 

1. Satiation errors: you overestimate how long you’ll feel the pleasure generated by a good experience

Good feelings wear off. The second glass of water on a hot day doesn’t quench your thirst quite like the first. The second bite of cheesecake doesn’t taste quite as good as the first. The second time seeing a great movie isn’t quite as good. But how quickly do these feelings recede? They don’t go away in an instant. They also don’t go away at a steady rate.

How long a good feeling lasts–the time it takes to go from the peak back to the baseline–is called satiation. Part of the reason people choose to do something or buy something is because it makes them feel good. It makes them feel good, and that good feeling lasts. They feel satiated for a while.

The problem is that the good feelings you get from buying something or doing something get their maximum effect if they’re not interrupted by the lingering good experiences of the last time.

• We don’t want to have eggs for breakfast two mornings in a row.
• I like chicken more, but I’ll have a chicken taco and a beef taco because the second chicken taco won’t be as enjoyable
• I really liked Hamilton but I don’t want to watch it a second night in a row.

The amount of pleasure you get from buying something or doing something can be measured by the difference between my baseline and my peak pleasure. But if you are still satiated from the last time you bought the thing or did the thing, then your baseline is higher–and the amount of pleasure you experience will be less. 

There are two ways to solve this problem. First, you could choose to repeat the experience and accept that the amount of pleasure will be a little less. Second, you could choose a different, second-best experience altogether. Which option you choose depends on which is greater: the smaller difference between the baseline and the peak from a repeated experience, or the difference between the baseline and the peak from a second-best experience.

People experience this when they listen to music. In an experiment, people listened to their favorite song nine times in a row. With each repeated listen, people liked the song a little less. That’s not surprising, because with each repeated listen, the difference between the baseline and the peak got a little smaller; they were still satiated from the previous time they listened. But when the experimenters added a one-minute interval of silence before repeating the song, people liked the song just as much as the first time they heard it. That extra minute allowed feelings of pleasure to recede just a little more. This created space for a bigger jump from the baseline to the peak the next time the song played. If you want to listen to your favorite song several times in a row, you’ll enjoy it more if you add one minute of silence after each time you listen. You can also add other songs. The experimenters had people listen to their three favorite songs on shuffle and got the same effect. People would rather hear a mix of their top three favorites than just their favorite on repeat. At a certain point, a second-favorite or a third-favorite brings more pleasure than the top favorite.⁸

But here’s the problem: people overestimate how long the pleasure from the first good experience will last. People think they will be satiated longer than they actually will be. This causes them to choose a different second experience–i.e. option two. People mistakenly think that the difference between the baseline and the peak of the different, second-best experience will be greater than the difference between the baseline and the peak of the repeated experience.

Predicting how long we will feel satiated is difficult. People get it wrong all the time. Incredibly, many people answer “no” to this question…

Should we [insert consumption experience] for two [insert time period] in a row?

…regardless of what the consumption experience is and regardless of the time period. For example, many of the people who would say no to this:

Should we have tacos two nights in a row? 

Would also say “no” to this:

Should we have tacos two weeks in a row? 

even though one of the future time periods is measured in days and the other is in weeks. This is because it’s difficult to predict how we will feel across varying future time periods. In an experiment, people choose three snacks to consume. One half would eat one snack per day for three days. The other half would eat one snack per week for three weeks—that’s six extra days for satiation to wane. Of the one-snack-per-day group, 47% chose three different snacks, while one-snack-per-week people chose three different snacks 33% of the time. On the one hand, it’s true that some people who thought they would still be satiated after a day realized they would be satiated after a week. But on the other hand, most of the people who assumed they would be satiated after a day also thought they would still be satiated after a week. Predicting the rate that satiation diminishes is difficult.⁹ 

One reason you are biased toward diversification when making simultaneous decisions is that you overestimate how long you will be satiated. This makes you more likely to choose variety, compared to sequential decisions–which involve no future prediction.

2. Time contraction errors: you misjudge how long an increment in the future will feel

We don’t just misjudge future satiation, we misjudge our perception of how we will experience time itself in the future. Namely, the farther a unit of time is in the future, the shorter we perceive it to be.

If this sounds a little abstract, then you’re in good company. Two researchers, B. Kyu Kim and Gal Zauberman of the University of Pennsylvania, tried to make it a little less abstract by having people indicate with a physical line how long a time period felt.¹⁰ The people in the experiment didn’t know they were participating in a study on time perception; they were just drawing lines on a screen—stretching or contracting a horizontal line. Kim and Zauberman gave participants a length of time and then asked participants to indicate how long a time period felt by adjusting the length of the line. They could shrink the line down to a pixel to represent a short period of time, or they could stretch the line off the edge of the screen, theoretically to infinity, to represent long periods of time. The lengths of time ranged from three months to thirty-six months. Sometimes, the periods of time occurred in the near future. Other times, the periods of time occurred far into the future. 

When the experiment was over, Kim and Zauberman compared line lengths for each time interval at various points in the future. They found that line lengths got shorter for time intervals farther into the future. Psychological future time is contracted compared to its equivalent calendar time unit.

Even more surprising, they discovered that the experiment subjects contracted future psychological time at a predictable rate:

Where T is the subjective perception of time and t is the objective length of time. This generated a graph that looks like this:

Kim and Zauberman’s study—drawing lines on a computer screen in response to experimenter prompts—doesn’t fully represent the complexities of lived experience. Thus, the results do not tell us that T = 1.05t^0.72 is the single universal rate humans compress future time for all possible experiences. Instead, these results tell us is that humans do not perceive future time at the same scale a clock or a calendar might dictate, and the rate psychological time and calendar time differ will vary depending on the stimuli and the context of the decision.

Time contraction causes a problem for decision-making. If future time contracts, then the inter-consumption time interval between two experiences will be perceived (from the vantage point of the present) as shorter than it really will be. You think the future will come more quickly than it really will. This, then, causes you to think you won’t be ready for repeat consumption so quickly. When you’re making a simultaneous decision about future consumption, you are more likely, then, to opt for a different experience rather than a repeat experience. 

You diversify.

Time contraction is related to satiation. Whereas satiation causes you to overestimate how long consumption will make you feel good, time contraction shrinks the experience of time itself. They work together to make you more biased toward diversification.

3. Affective forecasting errors: you misjudge how you will feel in the future

You can remember the past, but you cannot take the actual feelings from the past with you. You can take only the memories of those feelings. Likewise, you can predict the future, but you cannot accurately predict how you will feel in the future. You can only feel in the present–not in the past or in the future. 

The challenge is that we make decisions on how we think we’ll feel in the future. Those possible future feelings—or affects, as they’re called by psychologists—such as happiness, sadness, excitement, and joy, inform how we think and behave. They also inform how we decide.

Unfortunately, people do not predict their future feelings very well. In one study, researchers asked college football fans how they would feel a few days after their team lost. Most people thought they would be upset for a few days. But when the researchers surveyed the fans two days after the game, they found that the loss had no discernible effect on the fans’ happiness.¹¹ People also overestimate how unhappy they will feel after their candidates lose elections. In 1994, surveyors in Texas stopped voters as they exited polling places and asked them to predict how they would feel a month later if their candidate lost. People who voted for the winner were about as happy as they predicted. But people who voted for the loser were happy, too—at least happier than they predicted they would be.¹² In another study, people were asked to predict how much they would enjoy eating plain yogurt seven days in a row. Most people thought they would enjoy the yogurt a little less each day. In fact, by the end of the seventh day, they didn’t like the yogurt any less than after the first day.¹³ The evidence from these and other studies points in one direction: people aren’t very good at predicting how they will feel in the future.

Worse, people don’t just mispredict future feelings. They do so in systematic ways. First, they overestimate the rate of acceleration: how quickly they will feel anything. Second, they overestimate the overall intensity: how strongly they will feel it. Finally, they underestimate the rate of deceleration: how quickly those feelings will go away. (This is related to the tendency to overestimate how long satiation will last, as we saw above.) These prediction errors make people think they’ll feel the effects of their choices longer than they actually will.

When people make a simultaneous decision, they do so on the basis of how they think they’ll feel in the future. Specifically, they will think they are still feeling the lingering effects of previous consumption. To compensate, they will be biased toward diversification.

4. Repeat experience errors: you underestimate how much you’ll enjoy repeating an experience

People underestimate how much they’ll enjoy a repeated experience compared to a new experience.

When you repeat an experience, you enjoy it less than the first time. This is called hedonic adaptation. Pleasant feelings become less pronounced with each successive exposure to whatever stimuli caused those feelings. This is akin to stepping from a dark room outside into the sun. At first, it’s blindingly bright and impossible to see. A few seconds later, your eyes adjust and you can make sense of your surroundings. A minute later, and your vision has returned to normal. The same is true of stimuli or events that make you feel good. They feel really good the first time because they are new. But they feel a little less good the second time, and still less good the third time. If there’s a maxim, it’s this: it’s not as great the second time.¹⁴

But it’s also not as bad as you think.

The musical Hamilton was released as a movie on Disney Plus on Friday, July 3, 2020. Let’s say you watched it the day it came out and loved it. The next day, Saturday, July 4, 2020, do you watch Hamilton again? After all, it’s your current favorite, so you know you’ll enjoy it. But on the other hand, part of the reason you enjoyed it so much last time was because it was new. Your decision whether to watch Hamilton again can be summed up like this: Will the amount of pleasure I get from a repeat viewing of Hamilton be greater or less than the amount of pleasure I get from watching a less-awesome movie the first time?

Based on research conducted by Ed O’Brien of the University of Chicago, you probably should watch Hamilton again.¹⁵ O’Brien discovered that repeat experiences don’t feel as repetitive as people think they will. In a study, 130 people watched a movie of their choosing. Some people then imagined how they would feel watching the same movie a second night in a row, while others actually watched the movie. People who watched the movie only the first night rated their enjoyment a 5.29. They imagined they would rate their enjoyment a 3.47 if they watched the movie again the next night. However, people who watched the movie two nights in a row rated their enjoyment a 5.40 the first night and 4.52 the second night. Repeating a good movie two nights in a row is more enjoyable than people predict. Researchers discovered the same effect among people who played the same video game multiple days in a row, and among people who made repeat visits to a museum exhibit.¹⁶

This research confirms that you enjoy repeat experiences more than you think you will. Thus, when you are faced with a series of choices that you must make simultaneously, you are likely to underestimate how much you’ll enjoy a repeated experience compared to a new experience. As a result, you’re likely to make your second choice different from your first.

Diversification bias is caused by the choice set

The way options are presented affects how people choose. We have already seen that people are more likely to diversify when making simultaneous decisions. Now, we’ll see that, among simultaneous decisions, people are more likely to diversify based on two factors related to how the choice options are presented.

1. Choice bracketing: you arbitrarily and equally divide your choices by the number of options.

People divide their choices equally among the options presented to them. This is called choice bracketing or partition dependence. Choice bracketing happens when choices have three features: 

1. A fixed number of resources
2. A proscribed way of distributing those resources
3. A defined list of grouping options.

In a series of studies, three psychologists, Craig R. Fox, Rebecca K. Ratner, and Daniel S. Lieb, asked people how they would allocate financial aid to students in the fairest way possible. This is a choice that involves a fixed number of resources (a pool of money); a proscribed way of distributing those resources (financial need, academic merit); and a defined list of grouping options (income level). The researchers asked people to determine the best way to distribute the money across six income brackets.¹⁷

1. People from families making less than $15,000 per year
2. People from families making between $15,000 and $30,000 per year
3. People from families making between $30,000 and $45,000 per year
4. People from families making between $45,000 and $60,000 per year
5. People from families making between $60,000 and $75,000 per year
6. People from families making more than $75,000 per year

When people saw these income brackets, they gave 95.9% of the financial aid to students from families making less than $75,000 per year.

Next, the researchers asked people to do the same task—allocate financial aid in the fairest way possible—but across different brackets:

1. People from families making less than $75,000 per year
2. People from families making between $75,000 and $85,000 per year
3. People from families making between $85,000 and $100,000 per year
4. People from families making between $100,000 and $120,000 per year
5. People from families making between $120,000 and $145,000 per year
6. People from families making more than $145,000 per year

This group gave only 47.7% of the financial aid to students from families making less than $75,000 per year.

Why did one group give so much more financial aid to students from families making less than $75,000 per year than the other group? Because of how the brackets were structured. The participants in the study thought they were allocating financial aid fairly based on student need. What they didn’t know was that the grouping options—the brackets—were influencing their choices in a specific way. As a default, people first tend to divide equally among the brackets presented to them—in other words, to diversify. Then, they modify their decision based on the content of the brackets.

Choice brackets and experiences

People spread their choices across brackets not just with resources, but with experiences, too.¹⁸ At the beginning of an academic year, students were offered three free lunches at a fancy restaurant. But there was a catch: students had to choose in advance when they wanted them. 

College students love free food, so it wasn’t a surprise that all students chose their free lunches earlier in the academic year instead of later. All students generally chose two free lunches during the fall semester and one free lunch during the spring semester. 

But the researchers also discovered some students moved their free lunches into the spring semester if they saw their options in a slightly different way. Here’s how. One group of students was shown an academic calendar split into three sections: the first term of the fall semester, the second term of the fall semester, and both terms of the spring semester. This group of students picked an average of 2.58 lunches in the fall and 0.42 lunches in the spring. A second group of students saw a slightly different academic calendar split into two sections: a fall semester and a spring semester. This group of students picked an average of 2.15 lunches in the fall and 0.85 in the spring. The third group of students saw yet a different version of the calendar split into three sections: the fall semester, the first term of the spring semester, and the second term of the spring semester. This group opted to have lunch 2.10 times on average in the fall and 0.90 times on average in the spring.

In other words, students were twice as likely to eat lunch in the spring (0.90 times versus 0.42 times) when the spring semester was partitioned. The desire to split up the experiences between partitions did not completely overcome the desire to eat a great lunch sooner instead of later, but it did have a noticeable effect.

Choice brackets as a substitute for expertise

People are more likely to divide their choices by partitions when they lack subject matter expertise. In a study, people were given a list of six wines and asked to choose three. Some people saw lists that divided the wines by grape type: two Chardonnays, two Pinot Grigios, and two Sauvignon Blancs. Other people saw lists that divided the wines by region: two from Italy, two from Australia, and two from California. As expected, people were more likely to choose three different kinds of grape when they saw lists that divided the wines by grape, and they were more likely to choose wines from three different regions when the lists were divided by region. However, people who considered themselves wine novices were more likely to divide by grape type or by region compared to wine experts. Non-experts look to the partitions for clues to experiences that provide the most variety.¹⁹

If there’s an area where people look to choice brackets to compensate for lack of expertise, it’s retirement planning. In a 2001 study, Shlomo Benartzi and Richard Thaler discovered that people divide up their retirement savings among the funds presented to them—regardless of what those funds are. They noticed that a retirement plan offered by TWA had five stock funds and one bond fund. People who invested in this plan put 75% of their money in stocks. A different retirement plan, offered by the University of California, offered one stock fund and four bond funds. People invested only 34% of their money in stocks and the rest in bonds. Both numbers deviate sharply from the national average of 57%. Stocks are riskier than bonds; people are willing to take more (or less) risk with their retirement savings not by weighing risks and returns, but by splitting up their savings across the choices presented to them.²⁰

In a follow-up experiment, Benartzi and Thaler showed people three retirement plan options. Each option contained two funds: Fund A and Fund B. People could split their money however they wanted across Fund A and Fund B.

1. In the first option, Fund A was comprised of all stocks and Fund B was comprised of all bonds.
2. In the second option, Fund A comprised of was bonds, and Fund B was comprised of half stocks and half bonds.
3. In the third option, Fund A was comprised of half stocks and half bonds, and Fund B was comprised of all bonds.

As expected, people divided their savings across the options presented to them. People who saw the first group allocated 54% of their savings to stocks. People who saw the second plan allocated 73% of their savings to stocks. People who saw the third plan allocated 35% of their savings to stocks.²¹

Part of the reason we fall for the diversification bias is that simultaneous decisions are subject to the effects of choice bracketing. When evaluating items in a choice set, we evaluate the attributes of the items in that set and group them together by subset.²² For example, when looking at snack options, we group salty snacks together in one subset and sweet snacks together in another subset. Then, when we distribute consumption over time, we select from each bracket—and the less familiar we are with the items, the more likely we are to choose from each bracket.

Sequential choices, on the other hand, are not subject to bracketing effects because they are made one at a time and evaluated in isolation from one another.

2. No strong favorite: you are more likely to diversify when there is not a favored option

People are more likely to diversify when all the options in a choice set are similar, and they are less likely to diversify when there is a single, strong, stand-out, favorite option.

For example, imagine you must choose today what you will drink with your breakfast over the next seven days.

For most people, the differences between Italian, French, and Spanish roast coffee are small. It’s difficult to know which is your favorite now; it’s almost impossible to predict what your future self will prefer. The similarity of the options in the choice set produces uncertainty, and you compensate by diversifying.

In the second choice set, the options are very different from one another. Most people would find beer with breakfast disgusting. And drinking bleach, apart from culinary considerations, is a death sentence. This decision is easy. There’s no uncertainty about what your present—or future—self will want, and you’ll choose Italian roast coffee every time. No diversification.

To what extent does diversification increase between these two extremes—a choice set with the subtlest differences possible compared to a choice set with only one favorable option among a set of extremely bad options? In a 2012 study, Linda Salisbury and Fred Feinberg tried to find out. First, they had people rank a set of 12 snacks in the order they like them on a scale of 1 to 12. This produced a list of 12 snacks, ranked from liked-most to liked-least. Next, a week later, Salisbury and Feinberg split the participants into three groups and asked them to pick snacks:

1. The first group saw mostly favorite snacks with one stand-out least-favorite snack: snacks ranked 1, 8, 9, 10, 11, and 12.
2. The second group saw snacks spread evenly across their rankings: snacks ranked 1, 8, 9, 10, 11, and 12.
3. The third group saw mostly least favorite snacks with one stand-out most-favorite snack: snacks ranked 1, 2, 3, 4, 5, and 12.

To find out if stand-out options produced more diversification for simultaneous decisions, Salisbury and Feinberg had half of the people in each group choose snacks to consume immediately, while the others consumed snacks over the subsequent three weeks. As expected, the researchers discovered that diversification disappeared when people saw a choice set that included a strong favorite. When people choose in advance from options that are similar, they are more likely to diversify. When people choose in advance from options that are different, or from options where there’s a strong, stand-out favorite, then they are less likely to diversify.²³

Diversification bias is caused by environmental factors

We have already seen that diversification bias is more likely when making certain judgments under uncertainty; it is also more likely based on factors related to how choice sets are constructed. Now, we’ll see that the context of the decision itself can lead to more diversification.

1. Sex ratios: you are more likely to diversify when you are surrounded by more people of your sex than people of the opposite sex.

When people are surrounded by members of the opposite sex, they are less likely to take risks and more likely to diversify.

People who are at a disadvantage take riskier bets. 

If you are surrounded by fewer members of the opposite sex and more members of your own sex, then from the standpoint of reproductive success—one of the chief aims of your genes—you are at a competitive disadvantage. The ratio is not in your favor: you are one of few men competing for many women, or you are one of a few women competing for many men. This is something you are not conscious of. Few people walk into a conference room, a restaurant, or a Christmas party and deliberately calculate sex ratios with optimal mate choice in mind. Instead, it happens in the background—a subtle unconscious cue that reminds us we are operating in an environment that is unfavorable in this specific way. This causes us to take more risks: the lone guy at a bar filled with beautiful women (a sex ratio that’s favorable for him) is less likely to take a risk than at a bar where he’s in a big group of guys and there are only a few women (a sex ratio that’s unfavorable for him).

The problem is that the response to an environment with these kinds of unfavorable sex ratios affects human behavior in other ways. In a series of studies, Joshua Ackerman and his colleagues asked people to make risky choices like bets and investment decisions. In their first experiment, they asked people to choose either one $10 lottery ticket to win $10,000 (more risky) or ten $1 lottery tickets to win $1,000 each (less risky). The expected value is identical, but the ten $1 lottery tickets spreads out the risk. Some people made the bets in the context of an unfavorable sex ratio—men in groups of mostly men and few women, and women in groups of mostly women and few men. People made bets that were far riskier when they were surrounded by people of the same sex. Incredibly, the odds of people buying a single $10 ticket were 4.46 times greater when people made the choice in the context of an unfavorable sex ratio.²⁴

Something similar happened when people were asked to make investment decisions. People were given a pool of money and told to create a portfolio. When surrounded by unfavorable sex ratios, their portfolios contained more shares in fewer companies on average—less diversification. The same thing happened when people created a retirement plan. When surrounded by more members of their sex compared to members of the opposite sex, they selected fewer funds for their retirement savings—they went all in on fewer funds instead of spreading their assets around.²⁵

There’s something deep in our biology that prompts us to make risky decisions with respect to mate choice in such situations, and this same urge seems to drive decisions completely unrelated to mate choice.

2. Public vs. private choices: you are more likely to diversify in public because you believe others like variety more than you.

People are more likely to diversify in the presence of other people. In an experiment, a professor offered sixty-five of her students five kinds of candy. Some people chose in private, while others chose in front of others. When people were alone, they chose 2.67 kinds of candy on average. But when they were in the presence of others, they chose 3.31 kinds on average. This tells us that people are more likely to diversify when they make a choice in the presence of others.²⁶

Here’s why: everyone else thinks they like more variety than they do. In a follow-up study, researchers found that people do this because they think others like variety more than they do. Researchers asked people to choose ten appetizers to eat at a reception at a fancy hotel. The group who imagined making the decision in private chose 6.81 appetizers on average. But the group who imagined making the decision in the presence of others chose 7.38 appetizers on average. Then the researchers asked participants to guess how many appetizers of the ten they would choose compared to how many a “typical person” would choose. The results were surprising: people thought they would choose 6.74 appetizers on average, while the “typical person” would choose 7.63 kinds of appetizers.²⁷

Everyone thinks everyone else likes variety, so everyone behaves as if everyone actually does. You’re more likely to diversify when you make a choice in the presence of others.

3. Sleepiness: you are more likely to diversify when you are tired.

People are more likely to choose variety when they are tired. That’s because variety—It’s new! It’s shiny!—acts as a form of stimulation. 

At a deep level, we subconsciously rely on variety to raise our alertness when we are tired. People who drive late at night flip through radio stations to stay alert. Students who pull all-nighters to finish a paper might go for a walk once an hour. Parents who stay up late with newborns pace back and forth. In each case, variety acts as a stimulus.

It’s difficult to measure the effects of variety-as-stimulus when we are tired because it’s difficult to measure tiredness itself: it’s not possible to control for all the variables that make you sleepy. Fortunately, there’s a naturally occurring experiment once every year when everyone loses an hour of sleep: the beginning of Daylight Savings Time each spring. Zhongqiang (Tak) Huang and his colleagues compared transactions from the Sunday morning after Daylight Savings Time. Using data from the Chicago Nielsen consumer panel data set, which tracks detailed consumption patterns across 60,000 households, along with transaction data for 1.4 million grocery-related UPC codes, they isolated a specific kind of transaction: purchases that included more than one candy bar. You might be wondering how many people buy more than one candy bar on the Sunday morning following Daylight Savings Time, and it turns out: quite a lot. They identified 3,903 transactions. Of those, transactions with more than one candy bar included 0.41 more UPCs the morning after Daylight Savings Time began compared to a normal Sunday morning. People who are tired were less likely to get two of the same kind of candy bar and more likely to get two different kinds of candy bars.²⁸

In a different study, Huang and his colleagues asked people a series of questions to determine if they were morning people or night owls. Then they offered four pieces of candy and asked people to pick three. People chose more variety when they were tired: morning people chose variety in the evenings, and night owls chose more variety in the morning. In a different version of the study, people were asked to relax before they chose their candy—and this caused them to choose even more variety.²⁹

When you are tired, you might make a choice as a result of deliberate calculation. Or—more likely—you choose diversity because, at a subconscious level, it serves as a substitute for the kind of stimulus that might wake you up.

4. Time of day: you are more likely to diversify in the afternoons than in the mornings.

For similar reasons, people are more likely to choose variety in the afternoons—when they are tired. Your body follows a circadian rhythm. The most obvious rhythm is your sleep–wake cycle: you sleep at night and you’re awake during the day. But there are other forms of physiological variance as well:

• Blood pressure is higher in the morning and lower in the afternoon (and much lower while sleeping).³⁰ This is partly why heart attacks in the morning tend to be more severe.³¹ 
• Your body temperature also fluctuates throughout the day, starting at about 36.5°C three hours before you wake up, rising to 37.2°C by 9:00am and peaking at 37.4°C in the early evening, before falling back to its middle-of-the-night bottom.³²
• Even skin conductance levels—“electrodermal activity”—predictably rises and falls throughout each day.³³

Just as you experience a physiological circadian rhythm, you also experience a behavioral rhythm. You seek less variety in the mornings when you’re alert, and you seek more variety in the afternoons when you’re tired. Specifically, variety increases sharply from 5:00am to 11:00am, and then continues increasing from 11:00am to 11:00pm, although much more slowly. This discovery was made by Kelley Gullo and her colleagues, who looked at scanner data from a major grocery chain. First, they tracked the transactions of 1,115,133 households over a 25-month period.³⁴ Then, they calculated variety by looking at the unique items within a given category compared to the numbers of items purchased within that category. For example, if someone bought five yogurts, were they five separate flavors, or five of the same flavor? The results showed a clear pattern to variety-seeking. Variety increases sharply from 5:00am to 11:00am, and then continues increasing from 11:00am to 11:00pm, although much more slowly. For example, of transactions that included five yogurts, people shopping at 7:00pm were 6.5% more likely to buy four kinds of yogurt than people shopping at 7:00am. It didn’t matter how big the transaction was, how many items it included. It didn’t matter if the transactions happened on weekends or weekdays. This daily pattern of variety seeking held steady: starting slow, rising sharply, and then rising slowly throughout the day.

What causes this circadian behavioral pattern? The researchers noticed a correlation between sunrise and variety seeking. When the sun comes up earlier, variety-seeking starts earlier, too. The pattern holds for the beginning and end of Daylight Savings Times, too. When the clock abruptly jumps ahead or behind by one hour, so does variety-seeking. In the same way that exposure to light alters our physiology, light seems to alter our behavior, too. This makes you more likely to diversify in the afternoons.

5. Confined spaces: you are more likely to diversify if you’re in a confined space.

Humans don’t like tight spaces. Most people have had the experience of crowding into a subway or airplane or have felt the claustrophobia of a crawl space or an attic. Confined spaces produce a physiological response. Your palms sweat, your heart rate jumps, you feel stressed, and so on. These contexts of confinement produce a strong desire for freedom. When you’re stuck, you want out.

But they also generate a behavioral response in the form of seeking freedom through choices. Imagine you’re stuck in a confined space, and someone offers you six kinds of candy bars and asks you to choose three. Does your confinement cause you to make a different choice from what you would make elsewhere? In other words: would the choice on a crowded subway be the same as the choice on the street above? Would the choice in an airplane be the same as the choice in the airport or on the drive home?

Through a fascinating series of three experiments, Jonathan Levav and Rui Zhu decided to find out how feelings of confinement affected choices. They discovered that people make more varied choices when they make those choices in confined spaces.³⁵ They asked eighty people to walk down an aisle and choose three candy bars from a selection of six. Some people walked down an aisle that was seven feet wide, while others walked down an aisle that was three and a half feet wide. The group that walked down the wide aisle chose 2.66 kinds of candy bars on average, while the group that walked down the narrow aisle chose 2.90 candy bars on average.

In their next study, Levav and Zhu used the same setup: a wide aisle and a narrow aisle. This time, they asked people to select from six charities to donate to. People who walked down the wide aisle donated to 3.74 charities on average, while people who walked down the narrow aisle donated to 4.37 charities on average. Tellingly, both groups donated roughly the same amount; those who made the donation within a more confined space spread their amount across more charities.

Would the results obtained in a lab transfer to the real world? To find out, Levav and Zhu examined a dataset of 94,110,967 transactions of two or more items across 455 stores of a major grocery chain. They cross-referenced these transactions with the floor space and customer shopping patterns throughout the day to determine how crowded a store felt at the time of each transaction. In other words, they could roughly calculate how crowded the store felt at the time the customers were shopping. In most categories, customers bought a greater variety of products in a single transaction when stores were busier and more crowded.

Why would confined spaces cause us to seek more choice options? Levav and Zhu hypothesized that we “might attempt to regain freedom by making more varied and unique choices.” The same physiological desire to physically get out of a confined space exerts a subconscious influence on our behavior, it seems. When we’re confined or feeling stuck, we choose variety.

6. Signalling: to signal expertise, novices choose variety

Everyone wants to look smart. But what if you don’t know what you’re talking about—but you want to look like you do? It turns out that non-experts signal expertise by seeking more variety, while real experts signal expertise by seeking less variety. In one study, people assembled bags of gourmet chocolate truffles that would be auctioned off for charity. They were told that bags that appeared to be created by people who understood gourmet chocolate received a higher price. Novices assembled bags with lots of variety, while actual chocolate connoisseurs assembled bags with just a few kinds of chocolate. In a follow-up study, people were asked to assemble gift baskets containing an assortment of craft beers that would be judged by a panel of beer experts. People who were unfamiliar with beer chose a greater variety of craft beers than beer connoisseurs.³⁶

Why would novices signal expertise by choosing more variety, while actual experts signal expertise by choosing less variety? To answer this, we need to take a step back and explore two modes of understanding a domain:

1. The first is the breadth of categories within that domain. 
2. The second is depth within a category. 

Novices and experts interact with these domains in different ways:

1. Novices are capable of knowing many of the categories within a domain, and perhaps a little about a few of the categories. They signal expertise by choosing variety, thereby communicating that they know all the categories. 
2. Experts, however, not only know all the categories within a domain, but also know a great deal about each of the categories themselves. They signal their expertise by choosing from fewer—but better—categories. 

Let’s imagine a novice and an expert at a brewery together. The novice communicates hey, I know something about beer by sampling from a variety of styles and commenting on their desire to try as much as they can. The expert, however, takes a different approach. They know what they want. They already know what style they like, whether the brewery makes it, and whether it’s any good. (And if not, they have a second and third choice just in case.)

The bottom line: novices are likely to choose more variety than experts. When making a simultaneous choice, they are more likely to diversify if they are unfamiliar with the choice options compared to experts.

If you fall for diversification bias, are you irrational?

We have seen that diversification bias makes people make bad choices. We have also seen that people are more likely to make these bad choices as a result of judgment errors, choice set construction, and environmental factors.

If diversification bias leads to bad outcomes, then why does the bias persist?

Diversification is the rule, not the exception

The diversification bias causes people to make bad choices. People who fall for it choose candy, snacks, and music they don’t like. They make bad bets and even worse investment choices. They allocate financial aid unfairly. They do so because of judgment errors—like incorrectly predicting how they’ll feel in the future, or how long they’ll enjoy a good experience—or because components of the choice set trick them into making bad decisions. They also make bad choices because of environmental or contextual cues, such as unfavorable sex ratios, how tired they are, whether they’re making the choice in a confined space, or one of countless other known and unknown factors that influence decisions at a subconscious level.

Diversification bias is irrational.

…Or is it?

There is an assumption that the default setting of the human brain is a state of conscious, calculated, deliberate consideration. According to this line of thinking, biases compromise the default—optimal—setting. They do so in this way: external variables make these biases more pronounced. For example, we have seen that people are more likely to be biased toward diversification as a result of judgment errors, choice set construction, and environmental factors. Thus, the solution to overcoming bias is to mitigate these factors—do what we can to return our brain to its default state of conscious, calculated consideration.

This assumption is wrong.

Here’s why: it is premised on an incorrect assumption about defaults.

Our default state is not one of conscious, calculated consideration. We do not start from a state of non-diversification that becomes biased toward diversification in the presence of judgment errors, choice set factors, and environmental factors.

Instead, our default state is one of bias. We start from a state of diversification and adjust toward non-diversification in the absence of those same factors.

Ensemble averages and time averages

Why, then, does it seem like bad decisions lead to such bad outcomes?

Because, in a way, it does. Diversification bias does show that people make bad decisions on average. Let me quote myself from earlier in this post:

Diversification causes you to make worse choices, on average.

But what do we mean by average? 

• Is it the ensemble average: the average of what a group does once? 
• Or is it the time average: the average of what a single person does through time?

Compare these two averages:

This isn’t a semantic trick. One is the average of a group in a specific instance—such as an experimental outcome. The other is the average of a single person—you—over time. It’s the latter that you care about more, and it’s the latter that is the real driver of your behavior.

The core issue is that what may seem like a silly decision as an ensemble average is a perfectly rational decision as a time average. The time average matters because, unlike the ensemble average, it represents real human experience. It is the average of the decisions made by a single person in time. And because it moves through time, it cares a great deal about getting to the next decision. In fact, getting to the next decision is the most important thing for the time average. The diversification bias (and other biases) exist—indeed, are our default setting—because they remove risk in exchange for slightly worse choice outcomes, and in doing so, make it slightly more likely you’ll make it to the next decision. Thus, it’s not the ensemble average, but the time average and the biases that come with it that more closely represent real human experience: a single line on a graph—a single person living day-to-day—making a series of decisions, moving through time, staying alive, optimizing for outcomes that make it just a little more likely they’ll live another day to make another decision.

In an upcoming post, I’m going to take a deep dive into why and how diversification bias leads to better outcomes across a time average even as it simultaneously leads to worse outcomes across an ensemble average, and then explore some ways that understanding this important difference can translate into better (and sometimes more irrational) decisions.

1. Read, D. Loewenstein, G. (1995). “Diversification bias: Explaining the discrepancy in variety seeking between combined and separated choices.” Journal of Experimental Psychology: Applied 1:1, 34–49.
2. Read, D. Loewenstein, G. (1995). “Diversification bias: Explaining the discrepancy in variety seeking between combined and separated choices.” Journal of Experimental Psychology: Applied 1:1, 34–49.
3. Kneller, W., Memon, A., & Stevenage, S. (2001). Simultaneous and sequential lineups: Decision processes of accurate and inaccurate eyewitnesses. Applied Cognitive Psychology: The Official Journal of the Society for Applied Research in Memory and Cognition, 15(6), 659-671.
4. Sutherland, R. (2015). “Want more diversity? Hire groups, not individuals.” The Spectator 28 February 2015 https://www.spectator.co.uk/article/want-more-diversity-hire-groups-not-individuals
5. Simonson, I. (1990). “The effect of purchase quantity and timing on variety seeking behavior.” Journal of Marketing Research, 32, 150–162.
6. Benartzi, S., and Thaler, T. (2001). “Naive Diversification Strategies in Defined Contribution Saving Plans.” American Economic Review, 91 (1): 79-98.
7. Tversky, A., & Kahneman, D. (1974). Judgment under uncertainty: Heuristics and biases. Science, 185(4157), 1124-1131.
8. Galak, J., Kruger, J., & Loewenstein, G. (2011). Is variety the spice of life? It all depends on the rate of consumption. Judgment and Decision Making.
9. Read, D. Loewenstein, G. (1995). “Diversification bias: Explaining the discrepancy in variety seeking between combined and separated choices.” Journal of Experimental Psychology: Applied 1:1, 34–49
10. Kim, B. K., & Zauberman, G. (2009). Perception of anticipatory time in temporal discounting. Journal of Neuroscience, Psychology, and Economics, 2(2), 91–101.
11. Wilson, T. D., Wheatley, T. P., Meyers, J. M., Gilbert, D. T., & Axsom, D. (2000). Focalism: A Source of durability bias in affective forecasting. Journal of Personality and Social Psychology, 78, 821-836.
12. Gilbert, D.T., Wilson, T.D., Pinel, E.C., Blumber, S. J., and Wheatley, T. (1998). Immune Neglect: A Source of Durability Bias in Affective Forecasting. Journal of Personality and Social Psychology, 75, 617-638.
13. Kahneman, D., & Snell, J. (1992). Predicting a changing taste: Do people know what they will like? Journal of Behavioral Decision Making, 5(3), 187-200.
14. Frederick, S. and Loewenstein, G. (1999), “Hedonic Adaptation,” Well-Being: Foundations of Hedonic Psychology, eds. Daniel Kahneman, Edward Diener, Norbert Schwarz, 302–329.
15. O’Brien, E. (2019). Enjoy it again: Repeat experiences are less repetitive than people think. Journal of Personality and Social Psychology, 116(4), 519–540.
16. O’Brien, E. (2019). Enjoy it again: Repeat experiences are less repetitive than people think. Journal of Personality and Social Psychology, 116(4), 519–540.
17. Fox, C. R., Ratner, R. K., & Lieb, D. S. (2005). “How subjective grouping of options influences choice and allocation: Diversification bias and the phenomenon of partition dependence.” Journal of Experimental Psychology: General, 134, 538-551.
18. Fox, C. R., Ratner, R. K., & Lieb, D. S. (2005). “How subjective grouping of options influences choice and allocation: Diversification bias and the phenomenon of partition dependence.” Journal of Experimental Psychology: General, 134, 538-551.
19. Fox, C. R., Ratner, R. K., & Lieb, D. S. (2005). “How subjective grouping of options influences choice and allocation: Diversification bias and the phenomenon of partition dependence.” Journal of Experimental Psychology: General, 134, 538-551.
20. Benartzi, S., and Thaler, T. (2001). “Naive Diversification Strategies in Defined Contribution Saving Plans.” American Economic Review, 91 (1): 79-98.
21. Benartzi, S., and Thaler, T. (2001), “Naive Diversification Strategies in Defined Contribution Saving Plans,” American Economic Review, 91 (1): 79-98. It’s worth noting that in the short run, investors who diversify outperform investors who calculate the risk, variance, and potential return. Benartzi and Thaler’s experiments are excellent descriptions of choice bracketing, but their conclusions that naive diversification is suboptimal are incorrect. See DeMiguel, V., Garlappi, L., & Uppal, R. (2009), Optimal versus naive diversification: How inefficient is the 1/N portfolio strategy?, The review of Financial studies, 22(5), 1915-1953. See also this flock of sheep that outperformed professional stock-pickers by defecating in specific quadrants of a field: https://twitter.com/RobinWigg/status/1218146789947334667.
22. Read D., Loewenstein G., Rabin M., Keren G., Laibson D. (1999) “Choice Bracketing.” In: Fischhoff B., Manski C.F. (eds) Elicitation of Preferences. Springer, Dordrecht.
23. Salisbury, L. C., & Feinberg, F. M. (2012). “All Things Considered? the Role of Choice Set Formation in Diversification.” Journal of Marketing Research, 49(3), 320–335.
24. Ackerman, J. M., Maner, J. K., & Carpenter, S. M. (2016). “Going All In: Unfavorable Sex Ratios Attenuate Choice Diversification.” Psychological Science, 27(6), 799–809.
25. Ackerman, J. M., Maner, J. K., & Carpenter, S. M. (2016). “Going All In: Unfavorable Sex Ratios Attenuate Choice Diversification.” Psychological Science, 27(6), 799–809.
26. Ratner, R. K., & Kahn, B. E. (2002). The impact of private versus public consumption on variety-seeking behavior. Journal of Consumer Research, 29, 246–257
27. Ratner, R. K., & Kahn, B. E. (2002). The impact of private versus public consumption on variety-seeking behavior. Journal of Consumer Research, 29, 246–257
28. Huang, Z. (Tak), Liang, Y. (Sky), Weinberg, C. B., & Gorn, G. J. (2019). The Sleepy Consumer and Variety Seeking. Journal of Marketing Research, 56(2), 179–196.
29. Huang, Z. (Tak), Liang, Y. (Sky), Weinberg, C. B., & Gorn, G. J. (2019). The Sleepy Consumer and Variety Seeking. Journal of Marketing Research, 56(2), 179–196.
30. Uzu, T., & Kimura, G. (1999). Diuretics shift circadian rhythm of blood pressure from nondipper to dipper in essential hypertension. Circulation, 100(15), 1635-1638.
31. Suárez-Barrientos A, López-Romero P, Vivas D, et al. (2011). Circadian variations of infarct size in acute myocardial infarction. Heart 97, 970-976.
32. Refinetti, R., & Menaker, M. (1992). The circadian rhythm of body temperature. Physiology & behavior, 51(3), 613-637.
33. Hot P, Naveteur J, Leconte P, Sequeira H. Diurnal variations of tonic electrodermal activity. International Journal of  Psychophysiology. 1999;33(3):223-230.
34. Kelley Gullo, Jonah Berger, Jordan Etkin, Bryan Bollinger, Does Time of Day Affect Variety-Seeking?, Journal of Consumer Research, Volume 46, Issue 1, June 2019, Pages 20–35,
35. Levav, Jonathan and Rui Zhu (2009), “Seeking Freedom through Variety,” Journal of Consumer Research, 36 (4), 600–10.
36. Sela, Aner & Hadar, Liat & Morgan, Siân & Maimaran, Michal. (2019). Variety-Seeking and Perceived Expertise. Journal of Consumer Psychology.