What if you flipped a coin for every decision?

For close calls a coin costs you almost nothing and may even make you happier. For everything else, the probability compounds against you fast.

A coin spinning in the air above an outstretched hand
Photo: ICMA Photos, CC BY-SA 2.0, via Wikimedia Commons

If you flipped a coin for every decision, you would do fine on the close calls and badly on everything else. For choices where both options are nearly equal, a coin saves the time you would waste deliberating, and the largest real experiment on coin-tossed life decisions, Steven Levitt's study of 22,511 tosses, found that people the coin pushed into a big change were happier six months later. For any choice where you can tell the better option more than half the time, the coin throws that edge away, and over many decisions the losses compound.

Key takeaways
  • A hand-flipped coin lands on its starting face 50.8% of the time; heads versus tails is a clean 0.500.
  • Flip when the gap between options is below thinking cost ÷ your edge, \( t w/(q-0.5) \).
  • In Levitt's 22,511 tosses, "heads, make the change" made people about 25 points more likely to change.
  • For big decisions, those the coin pushed into change were happier six months later (imprecise estimate).
  • Over \( n \) flips the chance of all going your way is \( 0.5^n \): 1 in 1,048,576 for 20, India's ODI toss streak.

The question splits into three parts that can each be solved: is the coin itself fair, what does randomness cost you per decision, and what happens to real people who let a coin decide. The answers come from a 350,757-flip experiment, an expected-value model and a field study run through a website.

50.8%chance a fair coin lands on the side it started (Bartoš et al., 350,757 flips)
22,511usable coin tosses on life decisions in Levitt's experiment
+25 ptsmore likely to make a change after "heads" (2 months)
1 in 1,048,576odds of losing 20 fair tosses in a row, as India did in ODIs

Is a coin toss even a fair 50:50?

Almost, but not quite: a hand-flipped coin lands on the same face it started on about 50.8% of the time. That is the result of the largest coin experiment ever run, which tested a physical model of the toss. First, though, a reminder of how much can ride on one.

The story

On 5 June 1968 in Naples, Italy and the Soviet Union played the semi-final of the European Championship in front of an estimated 70,000 fans. After 120 minutes, extra time included, it was still 0-0. There were no penalty shoot-outs, so the rules called for a coin. Italy's captain Giacinto Facchetti and the Soviet captain Albert Shesternyov went down to the dressing rooms with the West German referee and two team officials.

"The referee pulled out an old coin and I called tails," Facchetti later told UEFA. Tails it was. He ran back upstairs into a stadium still full, and his celebration told the crowd. Italy went on to win the tournament, after a replayed final. UEFA notes it is the only competitive European international ever decided by a coin toss: two teams that 120 minutes could not separate, a perfect near-tie.

The Irish Times: Euro moments, tails the call as Italy reach '68 final via bizarre coin toss

The physics of a fair-looking flip

The theory. Persi Diaconis, Susan Holmes and Richard Montgomery argued in 2007 that a flipped coin precesses (wobbles) as it spins, so it spends slightly more time with its starting face up. Their model predicted a same-side probability of about 51%.

The bias comes from the physics of the spin, not from the coin's weight, so heads and tails overall remain equal.

Fair coins tend to land on the same side they started: Evidence from 350,757 flipsFrantišek Bartoš and 49 co-authors · arXiv preprint · 2023, revised 2025

A team of volunteers flipped coins by hand and recorded both the starting and the landing face, 350,757 flips in total. Same-side probability was 0.508 (95% credible interval 0.506 to 0.509, Bayes factor 2359), confirming the Diaconis-Holmes-Montgomery prediction. Heads versus tails was 0.500, so no heads bias. The same-side bias varied a lot between people and shrank as individuals flipped more, which the authors link to less wobbly technique. Limit: self-recorded flips by enthusiastic volunteers, not a machine.

Read the paper (free)

For our purposes that is a fair coin. If you do not look at which face is up before you flip, the 0.8-point edge is useless to anyone.

A coin being tossed before a cricket match
Cricket settles who bats first with a coin. Across 44,224 matches, winning the toss lifted the chance of winning by about 2.8%.Photo: Sunsetsealove, CC BY-SA 4.0, via Wikimedia Commons

What does a coin cost you on each decision?

A coin costs you your judgement's edge over chance, multiplied by how much the options differ. That is an expected-value calculation, and it tells you exactly when flipping is the smart move.

A coin is not a bad decision-maker. It is a free one, and free beats expensive whenever the options are close.

Assumptions and variables. A decision has two options. The better one is worth \( \Delta \) more than the worse one (in rupees, hours or happiness, as long as you are consistent). If you think it through, you pick the better option with probability \( q \). Thinking takes \( t \) hours, and your time is worth \( w \) per hour. A fair coin picks the better option with probability 0.5.

  1. Expected value if you decide.
    \[ E_{\text{think}} = q\,\Delta - t\,w \]
    measured relative to the worse option.
  2. Expected value if you flip.
    \[ E_{\text{coin}} = 0.5\,\Delta \]
    (a flip takes no time).
  3. Flip when the coin wins.
    \[ E_{\text{coin}} > E_{\text{think}} \iff \Delta < \frac{t\,w}{q - 0.5} \]
  4. A worked example: which of two dosa places tonight. Say the better meal is worth ₹50 more to you, you guess right 60% of the time, deciding takes 5 minutes and you value your time at ₹300 an hour. Then \( t w = (5/60)\times 300 = ₹25 \) and the threshold is \( 25/(0.6-0.5) = ₹250 \). Since ₹50 is far below ₹250, flip.
  5. A second example: two job offers. The better job is worth ₹2 lakh a year more, you pick right 70% of the time, and a week of research is 20 hours at ₹300. Threshold: \( 6{,}000/0.2 = ₹30{,}000 \). The gap is far above it, so think.

Buridan's donkey

The model has a named ancestor. Buridan's ass, the paradox named after the 14th-century philosopher Jean Buridan, is a donkey placed exactly midway between food and water, equally hungry and thirsty, that dies because it finds no reason to prefer either. The equation says the donkey should flip: when \( \Delta \) is near zero, any thinking cost at all makes deliberation a loss.

When should you flip? The decision ruleValue gap between options (Δ)Your edge (q − 0.5)Δ = t·w / (q − 0.5)FLIPTHINKbig gap or clear edgenear-ties
Below the curve, deliberation costs more than it gains, so a coin is the rational choice. Above it, your judgement's edge is worth the time. Drawing by the author from the expected-value model in the text.

Use this to check what you just read. Put in a real decision and see which side of the line it falls.

Should you flip a coin for this?

What happened to people who actually let a coin decide?

The coin moved real decisions, and for big ones the people it nudged into change were happier. Economist Steven Levitt ran the experiment through a website where people stuck on a decision could flip a virtual coin and agree to be surveyed later.

Heads or Tails: The Impact of a Coin Toss on Major Life Decisions and Subsequent HappinessSteven D. Levitt · The Review of Economic Studies · 2021 (NBER working paper 22487, 2016)

About 165,000 unique visitors produced roughly 23,500 tosses, of which 22,511 were usable. Heads usually meant "make the change", tails "keep the status quo". Participants and a nominated third party were surveyed at two and six months. At two months about 63% of actions matched the coin; people who got heads were 24.9 percentage points more likely to have made a change (21.1 points at six months). For important decisions such as quitting a job or ending a relationship, the instrumental-variable estimate of making a change on six-month happiness was +2.15 on a scale averaging about 6.9, though imprecise (standard error 0.65). Limit: a self-selected online sample, self-reported happiness, and the coin only moved people who were already close to indifferent.

Read the paper (free working paper)
How often people did what the coin said (two-month survey)
See the data as a table
Followed the coin (%)
Told to keep the status quo75
Less important decisions67
All decisions63
Important decisions56.1
Told to make a change50

Source: Levitt, Review of Economic Studies (2021), Figure 4 and text · 50% would mean the coin had no effect. Less important decisions are reported as more than 67%.

Status quo bias and the torn decider

Two findings matter for our what-if. First, a status quo bias: only about half of those told to change did so, while 75% of those told to stay put stayed. Status quo bias is the tendency to prefer the current state simply because it is current, and Levitt reads his results as evidence that people are too cautious about big changes. Second, the coin only works on people who are genuinely torn. Levitt's estimates describe the marginal decider, the person for whom \( \Delta \) is already close to zero. That is exactly the "flip" zone of the model above.

How fast does chance compound over many decisions?

Very fast. The chance that \( n \) independent fair flips all go your way is \( 0.5^n \), which halves with every extra decision.

\[ P(\text{all } n \text{ go your way}) = 0.5^{n}, \qquad 0.5^{10} = \frac{1}{1{,}024}, \qquad 0.5^{20} = \frac{1}{1{,}048{,}576} \]

Indian cricket gave everyone a lesson in this. India lost the toss in 20 consecutive ODIs, a run that began at the 2023 World Cup final on 19 November 2023. Wisden reported the 18th loss at Sydney in October 2025 as a 1-in-262,144 event (that is \( 2^{18} \)), and ESPNcricinfo reported the streak finally ending in Visakhapatnam in December 2025 when KL Rahul called right.

Run of lost tossesProbability for one named teamOne in
5\( 0.5^{5} \)32
11 (Netherlands, previous record, 2011 to 2013)\( 0.5^{11} \)2,048
15\( 0.5^{15} \)32,768
18 (India, Oct 2025)\( 0.5^{18} \)262,144
20 (India, final length)\( 0.5^{20} \)1,048,576

Why streaks are expected

Before you suspect a rigged coin, the Erdős-Rényi law of long runs is the theory to know: in \( N \) fair tosses the longest run of one outcome is typically about \( \log_2 N \). A sequence of 1,000 tosses usually contains a run of about 10, and cricket has played many thousands of tosses across all teams. Someone, somewhere, was due a long streak; it just happened to be India in a highly watched stretch. The table gives the odds for one named team looking forward, which is the wrong way to judge a streak spotted after the fact.

Two county cricket captains and an umpire at the toss before a match
A county-match toss. Each one is a 50:50 event, but string enough of them together and long streaks become almost certain somewhere.Photo: John Sutton, CC BY-SA 2.0, via Wikimedia Commons

What does winning a toss actually buy you?

A little. Winning the toss in cricket raises the chance of winning the match by about 2.8%, which is a fair estimate of what a coin's choice is worth when the options are close.

Fairly Random: The Impact of Winning the Toss on the Probability of WinningGaurav Sood, Derek Willis · arXiv preprint · 2016

An analysis of 44,224 cricket matches, treating the toss as a natural random experiment. Winning the toss raised the chance of winning by about 2.8%: small but statistically significant. The advantage varied with how evenly matched the teams were and with conditions, larger in day-night one-day games and smaller in T20s. Limit: the toss decides only who chooses to bat or bowl, so this measures the value of one small, informed choice, not a whole match.

Read the paper (free)

That 2.8% is the coin version of \( q - 0.5 \) in our model: the toss winner's decision about batting order is a small edge, and it rarely decides the game. Most of life's daily choices look like this. A few, like whom to marry or whether to take a loan, do not.

Chance that n fair flips all go your way50%25%12.5%6.3%3.1%1.6%0.8%1234567n = 20:1 in 1,048,576number of flips (n)
Each extra flip halves the chance that every one goes your way. Drawing by the author; values are \( 0.5^n \).

What would a full day run by a coin look like?

A day of coin decisions would be cheap on the trivial choices and dangerous on the few that matter, because the damage is concentrated where \( \Delta \) is large.

My estimate, with working. Suppose a day has 50 genuinely two-way choices you notice (a number I picked for illustration; popular claims of 35,000 daily decisions have no source I could find, so I do not use them). Say 45 are near-ties worth about ₹20 each in difference, and 5 are real choices worth ₹2,000 each, where your judgement is right 80% of the time.

  • Near-ties: thinking would earn you \( (0.6-0.5)\times 20 = ₹2 \) each, roughly 45 × ₹2 = ₹90 a day, at a cost of maybe 2 minutes each, 90 minutes. A coin wins easily.
  • Real choices: the coin costs \( (0.8-0.5)\times 2{,}000 = ₹600 \) each, ₹3,000 a day, or about ₹11 lakh a year.

The sensitivity check is simple. If the real choices are worth ₹200 instead of ₹2,000, the coin's annual cost falls to about ₹1.1 lakh. If your judgement is only 55% right, it falls by another five-sixths. The coin looks better the worse your judgement is, which is a useful thing to know about yourself. Our biases, covered in cognitive biases ranked by evidence, are one reason \( q \) may be lower than we think. The older Indian question of whether outcomes are ours to choose at all runs through karma and free will.

Verdict: possible with changes. Flip the near-ties, think about the rest

A coin is effectively fair (50.8% same-side at worst) and costs you \( (q - 0.5)\Delta \) per decision. Flip whenever the gap between options is smaller than your thinking cost divided by your edge: for a ₹50 dinner choice with a 60% hit rate and five minutes of dithering, that threshold is ₹250, so flip. Levitt's 22,511 tosses suggest coins also break status quo bias on big, genuinely balanced decisions. Used for every decision, though, a coin would throw away about ₹600 of value on each consequential choice in my illustration, and \( 0.5^n \) guarantees long losing streaks.

What if next?

Questions people ask

Is a coin toss really 50/50?

For heads versus tails, yes: Bartoš and colleagues measured 0.500 across 350,757 flips. But a hand-flipped coin lands on its starting face about 50.8% of the time, so if someone sees the coin before tossing, they have a tiny edge.

Does flipping a coin help you decide?

It can. Levitt found people who got "heads, make the change" were about 25 percentage points more likely to change, and for important decisions those who changed reported higher happiness after six months. It works best when you are truly torn.

Is it irrational to let chance decide?

Not for near-ties. If the options differ by less than your thinking cost divided by your edge, random choice has a higher expected value than deliberation. Buridan's ass is the classic warning about refusing to choose.

Why do long losing streaks happen with a fair coin?

Because the longest run in \( N \) fair flips is typically about \( \log_2 N \). Across thousands of cricket tosses, a run of 15 to 20 somewhere is expected, even though it looks shocking for one team.

Sources

  1. Levitt S.D.: Heads or Tails, NBER working paper 22487 / Review of Economic Studies (2016/2021) (source for 22,511 tosses, 63% adherence, 24.9 and 21.1 points, +2.15 happiness estimate, status quo bias figures)
  2. Bartoš F. et al.: Fair coins tend to land on the same side they started, arXiv (2023, revised 2025) (source for 0.508 same-side, 0.500 heads)
  3. Sood G., Willis D.: Fairly Random, arXiv (2016) (source for 44,224 matches, 2.8% toss advantage)
  4. Wisden: India stretch world record ODI toss losing streak to 18 (2025) (source for start date 19 Nov 2023, 1 in 262,144, Netherlands' 11)
  5. ESPNcricinfo: India end their rotten luck at the toss (2025) (source for the 20-toss streak ending in Visakhapatnam)
  6. The Irish Times: Euro moments, tails the call as Italy reach '68 final via bizarre coin toss (2016) (source for the 1968 story, Facchetti's account and the 70,000 crowd)
  7. UEFA: 5 June, the coin-toss semi-final (only competitive European international decided by a coin)
  8. Wikipedia: Buridan's ass (background on the paradox, Jean Buridan, 14th century)
  9. Arratia R., Gordon L., Waterman M.S.: The Erdős-Rényi law in distribution, for coin tossing and sequence matching (source for the longest run growing like log2 N)

More on the science of the mind on the Manthan world page.

coin tossdecision makingSteven Levittstatus quo biasprobabilityexpected valuecricket tossBuridan's ass
Sapta

Sapta (Saptarshi Sarkar) is a mechanical engineer turned marketer in Bengaluru and a solo touring rider. He writes the seven Saptaverse blogs. Work with him or read more of Mindblog.

Keep reading in Mindblog

From the other blogs