Distractor filtering

Flanker task run at two distractor spacings

Five arrows appear and only the middle one carries the answer: press F if it points left, J if it points right, and try to ignore the four beside it. Half the rows are drawn tight and half with the neighbors pushed 72 CSS pixels apart, so the result reports the cost of the distractors at each separation and the change between them — which is the comparison Eriksen and Eriksen built the task to make. 80 scored trials, about three minutes, free and with no account.

100% free
No signup
80 scored trials
Two spacings measured
Effect in ms, per spacing

Five arrows in a row. Only the middle one is your business: press F if it points left, J if it points right, and the other four are noise you are asked to ignore and will not manage to. Half the rows are drawn tight and half are drawn wide apart, because the size of the failure depends on how far away the noise is — that comparison is what this page exists for.

close · 34 px apart

wide · 72 px apart

Both example rows above are incongruent: the middle arrow points left and its four neighbors point right. Nothing on this screen is timed.

Everything this run does

Stimulus: five arrows in a row; only the middle one carries the answer
Your job: F for a middle arrow pointing left, J for one pointing right
Cells: four — agree or disagree, crossed with 34 px or 72 px between arrow centers, 20 rows in each
Timing: 400 ms cross, a gap drawn between 250 and 600 ms, then the row until you answer or 1500 ms passes
Rest: one, after row 40, for as long as you want it
Void: anything answered inside 150 ms, anything answered before the arrows painted, and any row the tab was hidden during

Rest your left index finger on F and your right on J before you begin. A hand that has to find its keys adds its search time to every trial equally, which does not change the difference this page reports but does make it noisier.

How to run the flanker task

Two fingers, one arrow that matters, and eighty rows that shuffle two things at once.

Put a finger on F and a finger on J
Left index on F, right index on J, and leave them there for the whole run. The keys are read by their position on the board rather than by the character they produce, so they are the same two keys on a French or a Dvorak layout even though the caps say something else. On-screen buttons do the same job for a touchscreen or a visitor who cannot use a keyboard.
Answer the middle arrow and nothing else
Every row holds five arrows. The four outer ones always point the same way as each other and are irrelevant to what you are asked; on half the rows they agree with the middle arrow and on half they disagree. You will be slower on the disagreeing rows and there is nothing you can do about it, which is the finding rather than a failure. Eight untimed practice rows are on offer first if you want the mapping in your fingers.
Read the two effects side by side
The result panel shows the cost of disagreement at close spacing and at wide spacing, each with a 95% interval, and then the difference between the two. A close row that costs more than a wide one is the spatial part of the original result — the distractors matter less once they are far enough from the thing you are looking at. Each cell's accuracy is printed beside its times, because a fast run with errors in it has bought its milliseconds.

Technical specifications

Trials	80 scored — 20 in each of four cells: agreeing or disagreeing arrows, crossed with close or wide spacing. Eight optional practice rows come first and appear in no figure
Stimulus	Five arrow glyphs, 28 CSS pixels each, in a horizontal row centered in the field. The middle one is the target and the flanking four always match one another
The two spacings	34 CSS pixels between arrow centers on close rows and 72 on wide ones — 9.0 mm and 19.1 mm at the CSS definition of 96 pixels to the inch, which is roughly 0.9 and 1.8 degrees of visual angle if you sit 60 cm from the screen. Neither viewing distance nor true physical pixel size is knowable from inside a browser, which is why pixels are the stated unit and degrees are stated as a conditional
Response	F for left, J for right, taken from physical key position. Both flanker types map onto a response, so the distractors compete for a hand rather than merely being different shapes — the arrangement that produces the larger effect
Trial structure	400 ms fixation cross, a blank interval drawn uniformly between 250 and 600 ms, then the row for up to 1,500 ms or until a key lands, then 350 ms blank. One rest after row 40
Discarded rows	Anything answered within 150 ms of the row painting, anything answered before it painted, and every row during which this tab lost the foreground. A voided row is asked again immediately, so the block delivers 80 answered rows however many attempts that takes
Timing floor	The display's frame interval is sampled while the start screen is up and printed with the result. Differences are given to the whole millisecond because a row cannot appear part-way through a frame, and one frame is about 16.7 ms on a 60 Hz panel
Reference figure	30-100 ms — Eriksen & Eriksen (1974), Effects of noise letters upon the identification of a target letter in a nonsearch task, Perception & Psychophysics. A span across published work rather than a distribution, so it sits beside your two effects and is never turned into a rank

Frequently asked questions

Why does the spacing change between rows?

Because the spacing is the experiment. The 1974 paper was not written to show that distractors slow people down — that much was already assumed — but to measure how the cost falls away as the distractors are moved off the target, and the answer, that the cost had largely gone by about a degree of visual angle, is what turned into the idea that attention has a width. A version of this task run at one spacing has reproduced the byproduct and skipped the finding.

Why arrows instead of letters like the original?

Because the original used letters that had been assigned to responses, and arrows do that job without a mapping to memorize. Eriksen and Eriksen gave one group of letters to one hand and another group to the other, so a flanking letter was not merely a different shape — it was actively calling for the wrong key. That is what makes the effect large. A flanker drawn from outside the response set produces a much smaller cost, and a page using neutral shapes as distractors is measuring something weaker while quoting figures from the stronger version.

How wide is the spacing in degrees of visual angle?

About 0.9 and 1.8 degrees if you are 60 cm from the screen, and unknown otherwise. A browser has no access to your viewing distance and no reliable access to the physical size of a pixel, so degrees cannot be computed here in the way a laboratory computes them from a chinrest and a measured screen. What the page can state exactly is the separation in CSS pixels and the millimeter figure that follows from the CSS definition of 96 pixels to the inch, so anyone who wants the angle for their own setup can work it out from their own distance.

Does this run have neutral rows?

No, and that is a deliberate trade. A neutral condition — distractors that call for no key at all — would split the effect into help and harm, but the trials have to come from somewhere, and spending them on a third congruency level would have cost the second spacing. This page keeps the spacing manipulation because the spacing is what makes it a flanker task rather than a generic conflict task. If the help-versus-harm split is what you are after, it is on the Stroop page, which has the neutral cell instead.

My effect got bigger at the wide spacing. Is the run broken?

Almost certainly not — 20 trials per cell is a small sample for a difference between two differences. The quantity at the bottom of the result panel is built from four separate cell means, so the noise in all four lands on it at once, and a single stretch of inattention in one cell can flip its sign. The two individual effects and their intervals are the reliable part of the page; the change between them is worth reading as a direction over several runs rather than as a measurement from one.

Should I answer as fast as possible or as accurately as possible?

As fast as you can while still being right, and the result checks whether you managed it. Speed and accuracy trade against each other continuously, so a run can produce a small congruency effect simply by answering the hard cell before it has been read — which shows up as a hole in the accuracy column rather than in the times. The per-cell accuracy figures are printed next to the per-cell medians for exactly this reason, and a disagreeing cell below about 90% correct is a run where the milliseconds were bought with mistakes.

Is this the same test as the attention network task?

It is the core of it, without the cueing. The attention network task wraps this exact flanker arrangement in a set of warning and location cues so that three separate quantities can be pulled out of one run, and the flanker part is the piece that carries the executive-control measure. This page runs the flanker manipulation on its own and adds spacing instead of cues, which keeps a single question in view rather than dividing 80 trials three ways.

What Eriksen and Eriksen were measuring, and what a browser cannot copy

The 1974 experiment is usually summarized as “distractors interfere,” which gets the result and loses the point. Participants were told exactly where the target letter would appear — there was no searching involved — and the manipulation was the distance between that known position and the noise letters beside it. The cost of the competing letters was large when they sat close and had shrunk substantially by the time they were a degree of visual angle away, and the conclusion drawn from that gradient is the one everybody has since inherited: attention is not aimed at a point but spread over an area, and things inside the area get processed whether or not they were asked for. The published span for the congruency cost itself is 30-100 ms, per Eriksen & Eriksen (1974), Effects of noise letters upon the identification of a target letter in a nonsearch task, Perception & Psychophysics, and the width of that span is mostly a matter of which separation and which distractor set a given study used.

Two design details decide whether a flanker page is measuring the strong version of the effect or a weak imitation of it, and both are worth checking on any version you meet. The first is whether the distractors belong to the response set. A flanking arrow points somewhere, and where it points is a key; a flanking square or a flanking hash mark is merely visual noise. The competing-response arrangement produces a much larger cost than the merely-different one, which means a page using inert distractors and quoting a published range has mismatched its own task to its own reference. The second is the separation, and it is the one that cannot be reproduced faithfully in a browser at all. Degrees of visual angle need a known distance and a known physical pixel; a web page has neither, and a device pixel ratio does not supply them. So the honest move is to state the geometry in the unit the page actually controls — 34 and 72 CSS pixels between arrow centers here — give the millimeter equivalent that follows from the CSS reference pixel, and offer the angle only as an if.

What the result is for, then, is the comparison inside one visitor on one machine. Absolute reaction times from a browser carry the monitor, the keyboard, the operating system and how awake the person is; the difference between two cells measured minutes apart on the same equipment carries almost none of that, which is why this page leads with differences and prints the raw cell medians underneath as evidence rather than as the answer. It is also why no rank appears: essentially every adult produces a congruency effect, so having one distinguishes nobody, and the size of one person’s effect moves enough between sittings that a single run cannot place them against anyone. For the same interference staged inside a single object rather than between neighboring ones, run the Stroop test; for a distractor that competes only with the hand and not with the eye, the Simon task is the cleaner case. The filtering question without a speeded key at all is the selective attention test, the same control mechanism sampled across several paradigms is the executive function assessment, and what happens to filtering over a long unbroken stretch is the continuous performance test.

A reaction time here is the interval between the frame that painted the stimulus and the timestamp the browser attached to your key, both read from the same monotonic clock. What neither can see is the display pipeline behind it, so on a 60 Hz screen roughly 16 ms of every figure below is the machine rather than you. That is the timing floor: two numbers closer together than that are the same number, and this page reports no precision it cannot support.

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 ADHD, an autism spectrum condition or any disorder of attention. Only a qualified professional, working with more than a browser, can make that judgment.

What happens to 80 keypress timestamps

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 eighty times exist as an array in this tab, are read once to produce four cell means, and are discarded when the tab closes. They are never written to storage and never posted anywhere, so a finished run cannot be recovered after a reload — the copy button is the only way to keep the numbers, and it carries the summary rather than the trials.