The Stroop effect

John Ridley Stroop published the experiments in 1935 as Studies of interference in serial verbal reactions, in volume 18 of the Journal of Experimental Psychology, on pages 643 to 662. It was the published version of his dissertation.

In the basic task, a person names the colour a word is printed in and ignores the word itself. When the word is a colour name printed in a different ink, such as the word GREEN in red ink, saying "red" is slow and error-prone.

Stroop's tests used five colours: red, blue, green, brown and purple. Black and yellow were tried and replaced by brown and purple.

Each test sheet was timed with a stop watch to a fifth of a second.

In Stroop's first experiment, 70 college undergraduates read 100 colour names aloud. Reading them took a mean of 2.3 seconds (5.6 percent) longer when the words were printed in clashing colours than when they were printed in black, a difference Stroop found not reliable.

In the second experiment, 100 students named the ink colours of 100 colour words printed in clashing ink, and named 100 plain colour squares. Of the 100 students, 88 were undergraduates and 12 were graduate students.

In Stroop's 1935 experiment, the mean time rose from 63.3 seconds for the colour squares to 110.3 seconds for the clashing words. That is 47 seconds longer, an increase of about 74 percent.

The medians were close to the means, at 61.9 and 110.4 seconds. The standard deviations were 10.8 and 18.8 seconds.

Words got in the way of naming colours, but colours barely got in the way of reading words. The interference ran mainly in one direction.

Stroop's third experiment used undergraduates at the University of Arizona, who practised on the clashing-word sheets over several days.

With practice, their time on the clashing-word sheet fell by 16.8 seconds, or 33.9 percent of the first day's mean. Time for naming plain colours improved by 4.0 seconds, or 13.9 percent.

Stroop put the gap down to training. He wrote that the difference in speed between reading colour names and naming colours could be accounted for by the difference in training in the two activities.

James McKeen Cattell had already argued in 1886 that reading is faster than naming because of practice. Stroop cited Cattell, and the idea that word reading is automatic traces back to Cattell's dissertation.

On the automaticity account, reading cannot be turned off, so an incompatible word is bound to interfere when someone is trying to name its colour.

The other long-standing account is relative speed of processing: faster processes can affect slower ones but not the reverse, and words are read faster than colours can be named.

The speed account does not hold up alone. When experimenters rotated words upside down and backwards, word reading became slower than colour naming, yet the words still caused as much interference as upright ones.

In another set of studies, people trained over several days to give colour names to unfamiliar white shapes. As shape naming became more practised, the shapes interfered more with naming print colours.

Almost no work followed Stroop's paper for about 30 years. Research resumed in the 1960s, helped by computer-controlled experiments that could time each trial, and a 1964 study by Klein is credited with relaunching it.

Klein found that swapping in colour words that were not among the print colours cut interference in half. Interference was smallest, but still reliable, for unpronounceable nonsense syllables.

Researchers have built Stroop-like tasks from the same template, including naming a picture with a conflicting word on it, stating the direction of an arrow with an incompatible word, and counting digits whose identity clashes with their number.

Many hundreds of studies have tried to explain the effect, and many more have used the method to study attention, learning, memory, reading and language.

Timed on a computer, the gap is measured in milliseconds and depends on the procedure. In a 2014 study of 35 English speakers, mean response time was 790 milliseconds for clashing words and 737 milliseconds for matching words when trials were 3,500 milliseconds apart.

In that study the interference measured against neutral words was 45 milliseconds with the long interval between trials, and 15 milliseconds, not statistically significant, with a short 200 millisecond interval.