The Stroop Effect: When a Word and Its Colour Disagree

The Stroop Effect: When a Word and Its Colour Disagree

by Julio Song

Picture the word RED printed in blue ink. Your job is to say the colour of the ink, out loud, as fast as you can.

You will get to "blue". You will also get there late, and you may feel "red" forming in your mouth first.

That stumble has a date and a number attached. In December 1935, the Journal of Experimental Psychology ran a paper by John Ridley Stroop, drawn from the dissertation he wrote at George Peabody College in Nashville. He had timed 100 students doing exactly this. Naming the ink took them 74% longer when the ink was wrapped around the wrong word.

If you put text labels on coloured things for a living, in buttons, badges, chart legends or theme pickers, Stroop's cards are about your work. So is the pile of research that followed, which Colin MacLeod counted at more than 700 articles by 1991.

Fifty years of people being slow at colours

Stroop did not start the argument. He walked into one that was already half a century old.

In 1886 James McKeen Cattell, working on his doctorate under Wilhelm Wundt, reported something odd. People could read the word "red" aloud faster than they could say "red" to a patch of red. Cattell's explanation still sounds current. With words, he wrote, "the association between the idea and name has taken place so often that the process has become automatic," whereas with colours "we must by a voluntary effort choose the name."

For the next few decades psychologists argued about why. Was it practice, since adults have read far more words than they have named colours? Or was it something about colours themselves, each of which can call up several possible answers while a word calls up one?

Stroop and his supervisor, Joseph Peterson, had been running reading-versus-naming studies of this kind. Then they put the two things on the same card. Stroop was less interested in how fast people name things than in what happens when two habits collide. He may not have been first. According to an earlier review cited by MacLeod, the German psychologist Erich Jaensch combined words and colours in a conflicting way in 1929. MacLeod's verdict on sharing the credit is that "the Jaensch-Stroop effect" would be an extremely unwieldy label.

What Stroop did with five inks

The materials were simple. Stroop chose five colours: red, blue, green, brown and purple. He printed their names on cards in a ten by ten grid, in 14-point Franklin lowercase type, with every word in an ink that did not match it. Each colour appeared twice in every row and column.

He then ran two separate experiments, and the first one usually gets left out of the story.

In Experiment 1, 70 students simply read the words aloud. Half the cards were in black ink and half were in the clashing colours. The clashing ink cost them 2.3 seconds across 100 words, about 5.6%, and the difference was not statistically reliable. Wrong colours barely slow reading down.

Experiment 2 flipped the task. Now 100 students named colours. On the control cards the colours were solid squares. On the test cards they were those same mismatched words. The squares took an average of 63.3 seconds per 100. The words took 110.3 seconds. That is 47 seconds lost, the 74% figure, and 99% of the times on the word cards were slower than the times on the squares.

A few things Stroop is often credited with never happened. MacLeod, who replicated both experiments decades later and got nearly the same numbers, lists them. Stroop never tested a matching condition, with RED printed in red. He never timed a single word either; every result came from whole cards read top to bottom.

Why the traffic only runs one way

Put the two experiments side by side and you get the finding that matters most. A wrong word wrecks colour naming. A wrong colour does almost nothing to reading.

The usual reading of this is Cattell's. You have read so many words that you cannot switch reading off. The word arrives whether you asked for it or not, and then you have to push past it to say the colour.

Stroop tested that idea in a third, less famous experiment. He had 32 students practise naming the inks of clashing words for eight days. They got faster, from 49.6 seconds to 32.8 seconds on a card of 50 words. In his own summary, practice decreased the interference "but did not eliminate it." Something stranger happened too. After all that colour practice, the students were suddenly slower at plain reading when the ink clashed, 34.8 seconds against 19.4 before. The effect had started to run backwards. It faded again by a second test.

Risograph print of a dark river sweeping round a wide bend past rocks, while a thin coral stream curls in from the side and is pulled into the main current

In 1988 MacLeod and Kevin Dunbar rebuilt this without words at all. They taught people to call four meaningless shapes by colour names, then showed the shapes in clashing colours. With a little training, the ink colour interfered with naming the shape. With moderate training, each interfered with the other equally. With a lot of training, the shape names interfered with the ink and the ink no longer bothered the shapes. Whichever skill has had more practice gets to do the interfering.

Children show the same thing in slow motion. MacLeod's summary of the developmental studies is that interference is minimal in first grade, peaks around second and third grade as reading takes hold, declines through adulthood and rises again after about age 60.

A lemon is also a colour word

The obvious fix is to keep actual colour names off your coloured things. In 1964 the psychologist G. S. Klein showed that this is not enough.

Klein used four inks, red, green, yellow and blue, and swapped the words underneath them. His subjects took about 44 seconds to name a card of plain colours. Then he measured how much each kind of word added:

  • The four ink names themselves, mismatched: about 37.5 seconds
  • Other colour names (tan, purple, gray, black): 18 seconds
  • Words tied to a colour (fire, grass, lemon, sky): 15.5 seconds
  • Common unrelated words (put, heart, take, friend): 12 seconds
  • Rare words (sol, helot, eft, abjure): 7.5 seconds
  • Unpronounceable letter strings: 5 seconds

The more a word means, and the closer that meaning sits to colour, the more it gets in the way. "Lemon" in blue ink is a small Stroop test. E. C. Dalrymple-Alford put numbers on single words in 1972: against an unrelated word, "red" in the wrong ink cost 131 milliseconds and "blood" cost 43.

Risograph print of a lemon hanging from a leafy teal branch, the lemon printed in dark navy blue instead of yellow

Language does not offer a way out. In bilingual speakers, MacLeod's review reports, a word in one language still interferes with naming the colour in the other, typically at about 75% of the within-language strength. If you want more on how colour names shape what people do with colours, we cover it in how language carves up colour.

Matching helps less than clashing hurts

If a wrong word slows people down, a right word should speed them up. It does, a little. When Glaser and Glaser measured both in 1982, a matching word bought 16 milliseconds and a clashing word cost 72. MacLeod's general conclusion is that the help from agreement "is much less than" the harm from conflict.

For anyone designing a labelled colour, that lopsidedness is the part to remember. Agreement between colour and word earns you a small bonus. Disagreement costs several times as much.

Two more findings from the same review set the limits. Moving the word away from the colour, so that it sits beside the patch instead of being printed in it, reduces the interference but does not remove it. And when people answer by pressing a key instead of speaking, the effect shrinks but stays statistically reliable. Most interface decisions are made with a finger, so expect the lab numbers to overstate what happens on a screen. Expect the direction to hold.

The one time someone tried it on charts

Most Stroop research uses cards of words, or single words shown one at a time. In 2013 a Stanford and Tableau team led by Sharon Lin asked the applied question. Does a bar chart read faster when each category gets the colour people already associate with it?

They recruited 144 people through Amazon Mechanical Turk and showed them bar charts of fruits, vegetables, drinks and company brands. Some charts used colours assigned in the default order of the Tableau 20 palette. Others used colours that fit the category, chosen either by the palette's own designer or by an algorithm that analysed image search results. Each person answered questions such as which of two bars was larger.

With the default order, a comparison took 9.1 seconds on average. With fitting colours it took 8.2 seconds for the algorithm's picks and 8.1 for the expert's. The authors call the effect small and note that it is still more than 10% of the task time. They also report where it came from. Nearly all of the saving was in the concrete categories, where a cherry really is red. For brands and soft drinks, whose colours are a marketing decision, the authors found "little or no effect", about a quarter of a second.

That is a fair picture of how much this matters in practice. A legend whose colours fight its labels will not make a chart unreadable. In Lin's study it added roughly a second to each comparison, and people reading a dashboard make a lot of comparisons.

What to do about it on a screen

None of the studies above tested a settings page or a status badge, so treat this list as reasoning from the lab findings, not as results.

Never let a colour name sit on a different colour. Theme pickers, tag colours and plan tiers are where this happens. A swatch labelled "Blue" that renders teal after a rebrand, or a "Green" plan with an orange badge, is Stroop's test card rebuilt in a product. If the colour changes, change the name, or stop naming things after colours. Our colour name finder will tell you what a hex value is commonly called before you commit to a label.

Count colour-associated words as colour words. Klein's list included fire, grass, lemon and sky. Your category names may include Ocean, Forest, Sunset or Mint. If a label already suggests a colour, either use that colour or expect a small cost.

Give chart categories their obvious colour when they have one. Lin's result was strongest for things with a physical colour. Bananas should not be purple in your legend. For categories with no natural colour, there is nothing to match, so spend your effort on telling the series apart, as we describe in our guide to honest colour systems for data visualisation.

Make status colours agree with status words. A red badge reading "Success" asks people to push past one signal to read the other. Which colour means success is a convention that changes across cultures, so check the pairing in each market you ship to.

Do not count on distance or habit to save you. Putting the label next to the swatch softens the clash and leaves some of it in place. Practice softens it too, but Stroop's students still showed interference after eight days of drills.

When the two disagree, the word wins. This is what Stroop's first experiment found. People read the label almost as fast in the wrong colour as in black. So the text is what users will act on, and the colour is what will slow them down. Fix the colour, or make it neutral.

The cheapest check is Stroop's own. Take your legend or your badge set and name the colours aloud, at speed, with the labels in view. If you trip, your users are spending that time too, on every screen that carries it.

Sources