Levitt's Coin-Toss Study: Those Who Made a Change Were Happier 6 Months On── Ending weighs more than beginning because it gives up sunk costs and the status quo, so set the standard for stopping on the day you start

Quit the job, or shut down the generative AI pilot? The economist Steven Levitt asked people stuck on major life decisions, such as whether to quit a job, to flip a coin, then followed what happened to them. Those who made a change reported being happier six months later. How does the courage to end something differ from the courage to begin? Ending differs from beginning because it means giving up what you have already paid and the state you are in. Both pull judgement toward carrying on, so the call to stop is best made against a standard set on the day you start.

01Ending weighs more because it gives up past costs and the present state

Ending is a heavier decision than beginning because there are two things to let go of. One is the money, time and effort already spent. The other is the present shape of one's life or work. A decision to begin carries neither.

Both weights have been measured, in psychology and in economics. A 1985 study using theatre season tickets showed that money already paid pulls people toward carrying on. A 1988 study showed that people keep choosing the state they are already in.

These weights are not confined to private life. A team that has brought generative AI into the workplace on trial will sooner or later face the day it decides whether to continue or stop. My view is that the standard for stopping should be written down on the day the trial begins. Once things are under way, both weights pull judgement toward carrying on.

02Theatre-goers who paid more for season tickets went to more plays over six months

Take the first weight, money already spent. In 1985 the psychologists Hal Arkes and Catherine Blumer gave it a name, the sunk cost effect: once money, effort or time has been invested in something, people show a greater tendency to keep going with it.

A sunk cost is one that has been paid and cannot be recovered. Since it is gone whether you stop or continue, it has no proper bearing on what to do next. People are pulled by it all the same.

Arkes and Blumer tested the effect with theatre season tickets. Customers who had paid a higher price for their tickets went to more plays over the following six months.

The second box in the figure is the one to look at. The moment the money was paid, it was settled that it would not come back. Even so, the customers who had paid more kept turning up at the theatre more often.

03AI pilots face the same call: Gartner expected 30% to be dropped after proof of concept

Decisions pulled by past spending do not happen only in theatre seats. In many workplaces, whether to continue a generative AI trial has become an everyday decision. The stage at which an organisation tries something on a small scale, before full rollout, to see whether it works is called a proof of concept. The research firm Gartner forecast how many projects would be dropped after that stage.

Both figures are forecasts, not counts of projects actually abandoned. Still, they show an expectation that many workplaces that start a trial will reach the day of deciding whether to stop.

A trial costs effort to prepare data, staff time and contract fees. By the day a decision to stop is on the table, all of that has been spent. The same force that pulled the theatre customers can pull this decision too.

04People keep the current state far more often than standard models predict

Past spending is not the only thing that makes a trial hard to end. The second weight lies in the present state itself. In 1988 the economists William Samuelson and Richard Zeckhauser described it as status quo bias.

In a series of experiments they found that people stick with the status quo disproportionately. They also looked at records from outside the laboratory: how university faculty chose their health plans and retirement plans, and how rarely they changed those choices.

A decision to end something removes the default option. A decision to begin leaves the present state in place and adds one new thing. If the default is chosen disproportionately, a decision that removes it has to work against that bias, and is heavier by that much.

The two weights were measured in separate studies. No single study has shown that both act at the same moment in the same decision. Still, both pull judgement the same way, toward carrying on.

05Those who changed course after the coin toss reported more happiness six months later

If people keep the state they are in, the natural next question concerns those who changed anyway. Levitt's study looked at exactly that, in a form that can be counted. Levitt, an economist at the University of Chicago, published the paper in 2016 through the National Bureau of Economic Research (NBER) in the United States.

The participants were people struggling with a major life decision, such as whether to quit a job or end a relationship. They flipped a coin on a website, and heads or tails gave them a nudge toward changing or staying.

Those whose coin came up "change" were more likely to actually make the change. And those who made a change reported being substantially happier two months later and again six months later. A coin toss is decided by chance. Because chance splits people into those nudged toward changing and those who were not, the effect of changing itself becomes easier to see.

The study was later published in the economics journal Review of Economic Studies. But the participants were people who volunteered for the study, and their happiness was what they reported themselves. That limit has to stay in view when reading the result.

06Write down the standard for stopping on the day you begin

From the research so far, three things follow for deciding when to stop.

The ground for the first card is the theatre study. Arkes and Blumer read the higher attendance of those who paid more as unrecoverable cost entering the decision. Taking the amount paid out of the decision weakens that pull.

The ground for the second is status quo bias. The present state gets chosen too often. Treating the option of continuing as "doing nothing" tilts the decision toward it before any comparison begins. The option of continuing should carry its own future costs, written in.

The ground for the third is that both weights grow after you start. On the first day, little has been spent and the present arrangement is still new. Since the theatre customers who had paid more were pulled harder toward carrying on, the weight of past spending is larger once costs have piled up. Among the reasons Gartner gave for abandoned projects were cost and unclear value. Start without a measure of value, and there will be nothing to decide by when the day to stop arrives.

Following the steps in the figure makes it harder for past spending to enter the decision on the deadline. For a generative AI trial, that means writing on the first day something like: "If staff time has not fallen by the agreed amount by the deadline, we stop." For a personal decision the content is the same: a date and a number.

07A 2021 replication found no escalation of commitment

The three conclusions about writing the standard on day one rest on research, and that research has limits. It helps to separate, study by study, what was found from what cannot be said.

In 1976 the organisational researcher Barry Staw showed in an experiment that people put more resources into a course of action after it had failed. The result came to be treated as an extreme form of the pull of past spending. Yet a 2021 replication found no evidence of escalation among its 365 participants. Another replication produced mixed results.

Levitt's result has limits too, as the right-hand column shows. The participants volunteered, and their happiness was self-reported. Whether the same would hold for people who never signed up is unknown. So the study does not show that quitting makes people happy.

What has been established is that, in the settings studied, such as theatre tickets and health and pension choices, money already paid and the present state pulled judgement toward carrying on, and no more. In my view that alone is reason enough to write the standard down first. Once the standard is on paper, the pull may still act, but the basis for the decision becomes harder to shift after the fact.

The courage to end differs from the courage to begin because it must let go of two weights that beginning never carries: what has been paid and the state you are in. At least the weight of past spending grows as more is paid after you start. Writing the standard for stopping on the first day means the decision on the deadline is made by that standard rather than by what has been paid.