Stroop Test

What the three tasks ask

In each task you answer one feature of what you see and ignore another. On congruent trials the ignored feature points to the same answer, on incongruent trials to a different one. How much slower and less accurate are you when the two disagree, and how precisely can one short run tell?

• Stroop. A word in coloured ink: press the key of the ink, not the word. Congruent trials show a colour name in its own ink (RED in red), neutral trials a word that names no colour, incongruent trials the name of another colour. The neutral words are LOT, SHIP, CROSS and ADVICE, as in the lab study below; in Slovak STANICA, KNIHA, POLICA and OKNO, each as long as one colour name. 32 trials of each kind.
• Flanker. An arrow in the middle with one symbol above it and one below: arrows pointing the same way (congruent), straight lines (neutral) or arrows pointing the other way (incongruent). Answer the direction of the middle arrow. 32 trials of each kind.
• Simon. A blue or orange square left or right of the centre. Blue is always F or the left button, orange always J or the right button. On congruent trials the square is on the side of its key, on incongruent trials on the other side. 48 trials of each kind.

Each stimulus stays until you answer, at most 2.5 s, followed by a small cross for 750 ms, the interval of the lab study. The 2.5 s limit is ours, so a missed trial cannot stall the run. The order is random, but never more than three trials of one kind come in a row and never the same stimulus twice in a row. The first run of a task starts with 12 practice trials with feedback, which are not counted.

Things to try

• Run Stroop twice in a row. In the lab study the effect was smaller in the second session.
• If you read English and Slovak, run Stroop in both. Does the language you read less often get in the way as much?
• Do one run as fast as you can and one as accurately as you can, then compare the effect with the error cost.
• Compare a keyboard run with a touch run: all times change, but does the effect?
• After a Stroop run, open Show the numbers: against the neutral words, is the effect mostly slowing on incongruent trials or speeding on congruent ones?

How the result is calculated

Only correct answers give a time. Answers under 100 ms (too early to be a reaction) and trials without an answer within 2.5 s are left out of the times. If the screen skipped a frame just as a stimulus appeared, the result says on how many trials. For each condition the page shows the median time and the error rate, where a wrong key and no answer both count as errors. The effect is incongruent minus congruent, in two forms:

• Trimmed mean. In each condition, times further than three median absolute deviations from the median are dropped and the rest are averaged; the deviation is scaled by 1.4826 to match the standard deviation of a normal distribution. Hedge and colleagues used the same 100 ms floor and three median absolute deviations, but per participant and with no scale factor stated: close to their trimming, not identical, and the number to set beside their means.
• Difference of medians. No trimming; one slow answer hardly moves a median.

Here RT is the time from the frame that showed the stimulus to your key press or touch, and M is the trimmed mean or the median of the correct answers of one condition.

The 95% interval is a percentile bootstrap: the page draws 2000 new sets of trials from yours, with replacement and separately for each condition, computes each set's effect and keeps the middle 95%. If the interval does not include 0 ms, this run shows a clear effect. The error cost is the error rate on incongruent trials minus the rate on congruent ones. A small time effect with a large error cost means that part of the conflict went into errors instead of time.

Compared with a laboratory study

Hedge, Powell and Sumner (2018) tested undergraduate students in a lab with 240 trials per condition, answers on a keyboard (Z X N M for Stroop), each stimulus shown until the answer and 750 ms between trials. Their mean effects (Supplementary Table B1; the standard deviation between people in brackets):

Both studies found smaller effects in the second session, 3 weeks after the first. A run here has 32 trials per condition, so its interval is much wider. For the Simon task no openly available study with comparable values was found for this page, so a Simon effect is compared only with your own runs.

Why there is no percentile

The same study tested every person twice, and people kept their order from one session to the next only moderately: the intraclass correlation (ICC; 1 means the same order and values both times, 0 means no relation) was 0.60 and 0.66 for the Stroop effect and 0.40 and 0.57 for the flanker effect, and it ranged from 0 to 0.82 across seven well-known tasks. The authors call this the reliability paradox: an effect that almost everyone shows, which is what makes it easy to replicate, leaves little difference between people to rank. With only a few dozen trials per condition, a rank among other people would mostly reflect chance, so History compares you with your own runs instead.

Where the tasks come from

J. Ridley Stroop (1935) printed colour names in inks that did not match them. In his Experiment 2, 100 students named the colours of 100 items: 63.3 s on average when they were solid squares, 110.3 s when they were mismatched colour names, 74% longer. Reading the words aloud, by contrast, was hardly slowed by the ink (his Experiment 1).

Barbara and Charles Eriksen (1974) showed a target letter for 1 s just above a fixation point, with other letters beside it; H and K meant one lever, S and C the other. Letters that called for the opposite lever slowed the answer more than letters of the same response. The arrows here follow the version of Hedge and colleagues.

In the Simon task the side of the square does not matter, yet it still pulls towards the key on that side.

Limits

• Colour vision. Stroop uses red, green, blue and yellow, Simon blue and orange. With red-green colour blindness the Stroop task measures something else; the flanker task uses no colour.
• Few trials. 32 or 48 trials per condition give a wide interval, so two runs whose intervals overlap may not differ at all; read the history rather than one run.
• Your device. The screen, the keyboard or touch screen and the browser add a delay to every time. Added to both conditions, it largely cancels in the effect; its variation from trial to trial does not, and widens the interval. Compare runs on the same device.
• The words. The Slovak neutral words match the colour names only in length; the words of the lab study were also matched for how common they are.
• A hidden tab. The trial in progress is discarded and repeated later, and the run waits for Resume.
• Not a medical test. This is not a test for ADHD or any other condition, and the size of your effect says nothing about your health.

Controls

• Keyboard: Stroop D red, F blue, J green, K yellow; flanker F or ← left, J or → right; Simon F blue, J orange. Enter or Space on the focused button goes on; Esc stops a run.
• Touch and mouse: the answer buttons under the display.
• Pauses: for 0.7 s after a result, the end of a part or Stop the buttons are dimmed and ignore keys still pressed from the task. A second click within 0.7 s of a start does not stop the run, so a double click is safe.
• Language: the link at the end of History switches the app to Slovak; the Stroop words change with the next run.

Sources

• Stroop, J. R. (1935). Studies of interference in serial verbal reactions. Journal of Experimental Psychology, 18, 643–662. Full text at Classics in the History of Psychology
• Eriksen, B. A., & Eriksen, C. W. (1974). Effects of noise letters upon the identification of a target letter in a nonsearch task. Perception & Psychophysics, 16, 143–149. doi:10.3758/BF03203267
• Hedge, C., Powell, G., & Sumner, P. (2018). The reliability paradox: why robust cognitive tasks do not produce reliable individual differences. Behavior Research Methods, 50, 1166–1186. Full text and supplementary tables (PMC5990556)