AbilityBenchFree BART, 30 balloons, 5 cents a pump
Balloon analogue risk task scored on adjusted average pumps
Thirty balloons arrive one at a time. Every pump adds five cents to a pile that a burst would take with it, the burst point is drawn uniformly somewhere between the first pump and the 128th, and you bank the pile whenever you decide to. The result is your adjusted average pumps — the mean over the balloons that survived, which is the measure the original study used and the only one that does not credit an explosion as caution. Free, no account, no clock, and every burst point is shown to you afterwards.
- 100% free
- No signup
- 30 balloons
- Burst point 1-128
- Unadjusted mean shown too
Thirty balloons, one at a time. Every pump adds 5 cents to a temporary pile and might be the pump that bursts the balloon, and a burst balloon takes its whole pile with it. Bank what you have whenever you like — that decision, thirty times over, is the entire measurement.
One pump, twenty pumps and 64 pumps. The width goes as the square root of the count so that a hundred-pump balloon still fits on the screen; the picture is not trying to tell you how close to bursting it is, and it could not, because it does not know.
Everything this run does
- Balloons
- 30 that count, plus 2 optional ones that do not
- Burst point
- drawn uniformly over 1 to 128 for each balloon independently, before the run starts
- A pump
- 5 cents into a pile that a burst takes with it
- Your job
- pump or bank, as often as you like, with no clock running and no time limit on a balloon
- Score
- adjusted average pumps — the mean over balloons that survived. The unadjusted mean is printed beside it, never instead of it
- Optimum
- 64 pumps every time, worth $1.60 a balloon; 44-84 pumps is within a tenth of that
- Void
- any balloon this tab was hidden during, and a part-inflated one when the run is ended
The optimum row is arithmetic, not advice: pumping 64 times maximizes 100% of nothing if you cannot stand watching $3.20 vanish half the time, and that reluctance is the thing being measured rather than an error in your reasoning.
Space pumps and Enter banks, or use the two buttons. Nothing here is timed, so take as long over a balloon as you want; the run keeps no clock and reports none. The money is imaginary, which is a real difference from the published version and is dealt with under the result.
How to run the balloon task
Thirty stopping decisions, none of them on a clock, and one number at the end.
Pump, and watch the pile rather than the balloon
Space inflates and so does the Pump button. The counter under the balloon is the honest display: it shows how many pumps you have made and what the pile is worth at five cents each. The picture is drawn to fit on a screen rather than to signal danger, so a balloon that looks alarming at 60 pumps is telling you nothing the counter has not already said.
Bank it before it goes, or find out where it was
Enter banks the pile and moves to the next balloon; a burst takes the pile and moves on too. Either way the page then shows you the pump that would have burst that balloon, which is the piece of information every version of this task withholds during the run and most withhold forever. Two practice balloons are on offer first and appear in no figure.
Read the adjusted average, then the three blocks of ten
The headline is the mean pump count over balloons that survived, with a 95% interval and the unadjusted mean printed beside it so you can see what the adjustment removed. Underneath, the same figure split into balloons 1-10, 11-20 and 21-30, because a run that started cautious and ended bold is a different run from one that did the reverse, and a single average hides both.
Technical specifications
| Balloons | 30 that count, plus 2 optional practice balloons that appear in no figure. Each balloon's burst point is independent of every other, so a run of three explosions carries no information about the fourth |
|---|---|
| Burst point | Drawn uniformly over 1 to 128 for every balloon, from the run seed, before the first pump is made. The balloon that burst on your third pump was always going to burst there, and every burst point is revealed in the results |
| A pump | Five cents into a temporary pile. A burst takes the whole pile; banking moves it to the total. Nothing is real money and the page says so beside the result rather than in a footnote |
| Score | Adjusted average pumps: the mean over balloons that did not burst, with a 95% Student-t interval. The unadjusted mean over all 30 is printed next to it, never instead of it, so the size of the correction is visible |
| Optimum | 64 pumps every time is the best fixed strategy, worth $1.60 a balloon in expectation, because the expected value of stopping at k pumps is 5k(128-k)/128 cents and that is a parabola with its peak at 64. Anything from 44 to 84 pumps earns within a tenth of the maximum |
| Input | Space pumps and Enter banks, or the two on-screen buttons for a touchscreen. The run holds the keyboard while it is active so that Space inflates instead of scrolling the page; Tab is deliberately left through, so focus still moves and the stop control is still reachable |
| Void | A balloon this tab was hidden during, and a part-inflated balloon when a run is ended early. Voided balloons lose their pile and are replaced at the end of the run, so 30 still count. Nothing here is discarded for being fast: there is no clock to be fast against |
| Reference figure | 1-128 pumps — Lejuez et al. (2002), Evaluation of a behavioral measure of risk taking: the Balloon Analogue Risk Task (BART), Journal of Experimental Psychology: Applied. It is a task parameter rather than a measurement of anybody, which is exactly why it can be quoted: it fixes the arithmetic of the task without ranking the person doing it |
Frequently asked questions
Why does the score throw away the balloons that burst?
Because a burst balloon records where the balloon stopped rather than where you would have. If you were prepared to go to 70 pumps and it burst at 22, the number 22 is the balloon's, not yours — and the more willing you are to keep pumping, the more often that censoring happens. Averaging bursts in therefore drags the score of a bold player down toward the score of a cautious one, so the measure would read most conservative exactly where behavior was most extreme. Dropping the burst balloons removes that, at the cost of a smaller sample, which is why the interval on the headline figure is as wide as it is.
Is there a right number of pumps?
There is a best fixed strategy, and it is 64. Stopping at k pumps pays 5k cents if the balloon survives, and it survives with probability (128-k)/128, so the expected return is 5k(128-k)/128 cents — a parabola whose peak sits at exactly half of 128. The peak is also broad: anything between 44 and 84 pumps stays within ten percent of the maximum, which is worth knowing because almost nobody plays anywhere near that band. That is not a criticism of the player. Watching $3.20 disappear on half your balloons is unpleasant in a way the arithmetic does not model, and the gap between the optimum and what people actually do is the thing the task was built to measure.
Does the balloon get more dangerous as it gets bigger?
Yes, and this is the part almost every description gets backwards. The burst point is uniform, so before a balloon starts, each of the 128 pumps is equally likely to be the fatal one. But once you have survived 100 pumps you know the burst point is one of the remaining 28, so the chance that the next pump is the one has risen from 1 in 128 to 1 in 28. The hazard climbs steeply near the end even though the distribution is flat. This is also why the drawing on this page does not try to look strained: a picture that swells alarmingly is making a claim about a specific balloon that the page has no way to support.
Why five cents a pump if the money is imaginary?
Because keeping the published parameter is the only thing that makes the arithmetic on this page comparable to the arithmetic in the paper, and changing it would move the optimum without making the money any more real. Five cents a pump over a 1-to-128 burst range is what Lejuez and colleagues ran, and it fixes both the $1.60 expected value at the optimum and the $3.20 that sits at risk on a 64-pump balloon. What the imaginary money does change is your behavior, not the sums: an unpaid session usually produces bolder pumping than a paid one, which is one of two reasons a figure from this page is not comparable to a published mean.
Can I just bank every balloon without pumping?
Yes, and the run will report an adjusted average of zero, which is an accurate description of what happened. The task permits banking at any point including the first, so there is no minimum and no coercion, and the page does not treat a zero-pump run as invalid — it is a real strategy that earns nothing and never loses anything. The three blocks of ten are the place this shows up most clearly, because a run that was genuinely cautious throughout looks quite different from one where somebody lost interest halfway and started banking immediately.
Does knowing that 128 is the ceiling change how I play?
Almost certainly, and that is a real difference from the standard administration. Participants in the original are told nothing about the distribution: they are told each balloon can burst at any point and left to build their own model of it from experience, which is what makes the early balloons different from the late ones. This page states the range in the method because the alternative — withholding it while inviting you to trust the numbers — is the shape this site refuses to take. If you would rather meet the task cold, run it before reading the specification below; the run itself never displays the range while a balloon is on screen.
Why 30 balloons rather than 10?
Because the adjusted average is computed over the survivors only, and ten balloons can easily leave four or five of those. The interval printed beside the headline figure is built from that count, and on five surviving balloons it is wide enough to include almost any conclusion. Thirty is the number the 2002 paper used, and it is also what makes the three blocks of ten legible as a sequence rather than as noise. It costs about five minutes, and the run has no time limit on any single balloon, so it can be left mid-flight for as long as this tab stays in front of you.
Why the BART is scored on survivors, and what a browser version cannot reproduce
The task was published by Lejuez et al. (2002), Evaluation of a behavioral measure of risk taking: the Balloon Analogue Risk Task (BART), Journal of Experimental Psychology: Applied as a behavioral alternative to asking people how much risk they take. The mechanism is a balloon, a pump worth a small fixed amount, and a burst point the participant cannot see; the interesting quantity is not how much money comes out but how far somebody is willing to push a sequence of decisions whose downside they cannot estimate. The scoring rule matters more than anything else in the design. A balloon that bursts truncates the pump count at whatever the balloon happened to choose, so the recorded number is the balloon’s decision rather than the participant’s, and the effect of that truncation is worst for precisely the people the task exists to identify. The standard index is therefore adjusted average pumps, computed over unexploded balloons only. Any version reporting the plain mean has built a measure that improves when you fail, and this page prints both figures side by side so the difference is a visible quantity rather than a claim.
The uniform burst range of 1-128 pumps is the second load-bearing parameter, and it is what lets a page compute rather than assert. Stopping at k pumps returns five cents a pump if the balloon survives, and survival has probability (128 minus k) over 128, so the expected return is proportional to k times (128 minus k) — a parabola peaking at 64 pumps and worth $1.60 a balloon. Two facts follow that are worth carrying into the run. The peak is broad, so anywhere between 44 and 84 pumps is within ten percent of the best available, and the optimum is far above where most sessions land. And although the distribution is flat, the hazard is not: surviving to pump 100 means the burst point is one of 28 remaining possibilities rather than one of 128, so the next pump really is about four times more dangerous than the first. A description that says every pump carries the same risk has confused the unconditional probability with the conditional one, and it is the conditional one a player is actually facing.
Two departures are worth stating plainly. The first is the money: the published sessions paid real cents and lost real ones, and an incentive is not a decoration on this task but part of what it measures, so a figure from this page is not comparable to a published mean and no percentile appears beside it. The second is that nothing here is timed at all. There is no reaction time, no response window and no timing floor on this page, because a decision to stop pumping is not a latency and reporting one would be inventing a measurement to look rigorous. What the page does report is thirty stopping decisions, the burst point each balloon was holding, and the three blocks of ten in the order they happened. For a task where the rule changes underneath you and the evidence has to be re-read, run the Wisconsin card sorting test; for stopping measured as a latency instead of a count, the stop signal task estimates how long your internal brake takes. Switching between two rules under a stopwatch is the trail making test, and the other half of this pair — attention moved by a cue rather than money staked on a guess — is the Posner cueing task.
This is a measurement exercise, not a clinical assessment. It reports what you did on this page against a stated reference and nothing more — it cannot establish an impulse-control problem, ADHD or a substance use disorder. Only a qualified professional, working with more than a browser, can make that judgment.
Where 30 pump counts go
Every number on this page is worked out by JavaScript running in the tab you are reading it in. Your answers, your reaction times and your score are never uploaded, logged or kept — which is also why the test carries on working after you disconnect from the network, and why nothing here can be held back behind an email address.
The thirty pump counts, the thirty burst points and the running total live as arrays in this tab and are read once to produce the figures above. None of it is written to storage, so a reload discards a finished run outright: if the thirty balloons are to outlive the tab, they have to leave through the copy button. The burst points are drawn from a seed inside the page before the first balloon, which is also why re-running gives a genuinely new set rather than the same thirty in a different order.