The Stroop effect

Print the word BLUE in red ink and ask someone to name the ink colour. They will slow down, and some of the time they will say "blue". The effect has been studied for ninety years and reproduced in hundreds of experiments — and it is still not settled what it measures.

What is actually happening

For a fluent reader, recognising a word is not something you choose to do. The moment the letters land on your retina the meaning is retrieved, whether or not it is useful. Naming an ink colour, by contrast, is a task you have to hold in mind and execute deliberately. When the two disagree, the unwanted answer is already available before the wanted one is ready, and it has to be pushed aside — which costs time, and occasionally fails outright.

The detail that makes this theoretically interesting is the asymmetry. Conflicting words slow down colour naming a lot. Conflicting ink barely slows down word reading at all. Any account of the effect has to explain why the interference runs in one direction and not the other, and that constraint is what has killed most of the simple explanations.

The usual explanation, and why it is too simple

The textbook story is "automatic versus controlled": reading is automatic, colour naming is deliberate. That is a decent first pass, but it treats automatic as a category when it is really a matter of degree, built by practice. In one experiment, people were trained to name arbitrary shapes. After a few hours of practice those shapes began interfering with colour naming. After about twenty hours the interference had flipped direction — the shapes now disrupted colour naming more than the colours disrupted shape naming.

That result is why most researchers now think of it as a race rather than a hierarchy. Two processes compete to produce the answer, and whichever is faster tends to win. Reading usually wins because you have done far more of it, not because it belongs to a different category of process. Change the practice and you change the outcome — which is exactly what makes the cross-language version of this question interesting.

What is still argued about

This is the part usually left out of the introductory description, and it is the part that matters if you are planning to use the task.

If you are going to run it, get these right

• Decide between spoken and button responses on purpose. Saying the colour out loud produces the larger effect and is closest to the original. Pressing a key adds a colour-to-key mapping step in between, which changes what you are measuring. Both are defensible; they are not interchangeable, and you cannot compare across them.
• Include neutral trials. With only congruent and incongruent conditions, your single difference score quietly mixes two things: how much conflict slows you down, and how much agreement speeds you up. Adding ordinary non-colour words in coloured ink separates them, and they do not always move together.
• Fix the proportion of incongruent trials and report it. This has been known to change the effect since 1979, so a study running 25% incongruent is not comparable to one running 75%. An even split is the conventional default.
• Think about colour vision. A red/green response set does not work for everyone. The commonly quoted prevalence figures come from northern-European samples and do not transfer well elsewhere, so rather than rely on a number, either screen your participants or pick colours that differ in lightness as well as hue.
• Do not panic about browser timing. Response times collected in a browser run around 25 ms slower than lab hardware, but that offset is consistent enough that experimental effects come out the same size. What genuinely varies is the spread across your participants' assorted phones and laptops. A Stroop effect is a difference between two conditions measured in the same person on the same device, so it survives comfortably. Absolute response times are what you should not quote precisely.

Build a cross-language Stroop task: report the ink colour, ignore the word.
Colours red, green, blue, yellow, answerable by key or by on-screen button.

Three blocks, with block order counterbalanced across participants:
  1. English colour words (RED, GREEN, BLUE, YELLOW)
  2. Hindi colour words in Devanagari (लाल, हरा, नीला, पीला)
  3. The same Hindi words in Roman letters (LAAL, HARA, NEELA, PEELA)

Roman Hindi has no standard spelling — pick one spelling per word and keep it
fixed, or you are adding noise for nothing.

48 trials per block, split evenly congruent / incongruent / neutral, and keep that
split identical across the three blocks so the blocks stay comparable. A neutral
trial is an ordinary non-colour word in coloured ink: use a set of at least eight
per block (CHAIR, TABLE, WINDOW, BOTTLE, ...) rather than repeating one word,
otherwise "neutral" quietly becomes "familiar".

Tag every trial with its block and its script, so the data can be split by both.

Before the first block, ask which languages they read, the age they learned to read
each, and self-rated reading fluency from 1 to 7 for each. Then ask how often they
read Hindi written in Roman letters (daily / weekly / rarely / never) — keep that as
its own question, not folded into the fluency ratings.

Show each block's accuracy on the break screen after it.