Cognitive Psychology

Visual Search Laboratory

Find the tilted solid bar. When it is the only tilted thing on screen, the number of items barely matters. When half the tilted things are hollow, it matters a great deal.

Original geometric displays — illustrative browser timing, not laboratory measurement

Learning objective

By the end you should be able to read a search slope in milliseconds per item, and say what a flat slope and a steep one each imply about how the display was searched.

About 25 minutes. Nothing you do here is saved or sent anywhere.

Two warnings about the words

"Parallel" and "serial" are theories, not observations. A flat slope is consistent with checking everything at once, and also with a very fast limited-capacity process. A steep slope is consistent with checking one item at a time, and also with a graded process that gets less efficient as competition grows. Slopes constrain accounts; they do not choose between them.

Browser timing is approximate. Display refresh, keyboard latency and background load add tens of milliseconds. Treat every number here as illustrative.

  1. Answer the prediction below — it unlocks the rest of the page.
  2. Look at some displays first, untimed, with the target revealable on request.
  3. Run the feature block, then the conjunction block. Neither trial is ever timed out.
  4. Read the slopes: milliseconds added per extra item, separately for target-present and target-absent trials.
  5. If you would rather not do a speeded task, load the worked example: it is a documented simulated dataset covering both conditions.

First, a prediction

You will search for one target — a tilted solid bar — in two kinds of display. In the feature displays every other item is an upright solid bar, so the target is the only tilted thing there. In the conjunction displays half the other items are upright solid bars and half are tilted hollow bars. The target is the only item that is tilted and solid, and neither property alone picks it out.

As the number of items on screen goes from 8 to 32, what happens to search time in each condition?

Look before you time anything

Nothing here is timed. Build a display, look for the target, and reveal it when you are ready. Do this a few times at 32 items in each condition — the difference is obvious long before any reaction time is recorded.

Key terms
Feature search
Looking for an item that differs from everything else on a single property — tilt on its own, or fill on its own.
Conjunction search
Looking for an item defined by a combination that other items each share one half of: here, a bar that is both tilted and solid.
Set size
How many items are on screen on a given trial.
Search slope
Milliseconds added to your response for each extra item in the display. It is reported separately for target-present and target-absent trials, because those are different tasks.

Build a display

The target is always a bar that is both tilted and solid. Every display is drawn fresh from a documented generator; positions are jittered on a grid so nothing sits in a predictable place.

What to show
Actions
Answer the question above to unlock this panel.

A description of the display will appear here.

Run the experiment

A display appears and stays until you answer. Press J if the tilted solid bar is there and F if it is not, or use the two buttons. Set size varies from trial to trial; half the trials have no target at all.

Challenge — what do slopes license?

Five claims somebody might make on the basis of search functions like these. Decide, for each, whether data of this kind support it, contradict it, or simply do not reach it.

Claim How does the evidence bear on it?

What this demonstrates

Set size matters, or does not, depending on what defines the target

When the target differs from everything else in a single property, adding items costs very little. When the target is defined only by a combination of two properties, each of which it shares with some of the distractors, adding items costs a great deal. The same eyes, the same screen, the same instruction: what changed is how the target is specified.

The slope is the measurement

A search function is a line: reaction time against number of items. Its slope, in milliseconds per item, is the quantity that distinguishes the two conditions, and it is far more informative than any single mean. A condition with a high intercept and a flat slope is doing something quite different from one with a low intercept and a steep slope, even if their average reaction times happen to match.

Absent trials are steeper, and that is informative

To say "yes" you can stop the moment you find the target. To say "no" you have to be satisfied that it is not there. If items were checked one at a time in a random order, a present trial would need about half the items on average and an absent trial all of them. That gives an absent slope about twice the present slope. Ratios near two are commonly observed. That is a striking fit, but it is not a proof: quitting rules, guessing and graded processing produce similar ratios.

Why "parallel" and "serial" are contrasts, not conclusions

The original two-stage account read a flat slope as unlimited parallel processing and a steep slope as an item-by-item scan. Both readings have been challenged for decades. Limited-capacity parallel models produce steep slopes without any scanning; signal detection accounts produce them from accumulating noise; and search efficiency turns out to vary continuously across tasks rather than falling into two types. The distinction is worth learning because it organises the findings, not because it settles them.