Visual Search Test

What the test asks

A red apple among green ones is found about as fast however many apples there are. One particular combination of features, or a shape that differs only in how its parts are arranged, takes longer with every item you add. How much longer is the search slope, in milliseconds per item, and this page measures it for you, separately for displays with a target and displays without one.

• Feature. An orange upright bar among blue upright bars. Colour alone gives it away.
• Conjunction. An orange upright bar among blue upright bars and orange flat bars. Neither colour nor orientation is enough, only the two together.
• 2 among 5s (optional). A digital 2 among digital 5s, built from the same strokes in another arrangement.

Each condition has 128 trials: 3, 6, 12 or 18 items, each with a target on half of the trials and without one on the other half, 16 trials for each of those 8 cells, in random order. A cross appears for 500 ms, then the display stays until you answer, at most 4 s (8 s for the 2 among 5s). The first run of a condition starts with 8 practice trials with feedback, which are not counted. Feature and conjunction together are 256 trials, about 6 minutes.

The items sit in a 6 × 5 grid of invisible cells, each item at a random position inside its cell, so they never touch. Wolfe, Palmer and Horowitz used the same idea with a 5 × 5 grid.

Things to try

• Compare the target-present lines of feature and conjunction in the chart: is the feature line flat? The difference between them is the classic result.
• Compare present and absent in the conjunction condition. If the absent slope is about twice the present slope, what does that suggest about how you search?
• Add the 2 among 5s. Here colour cannot guide you at all, and the published slopes are the steepest of the three.
• Do one run on a keyboard and one by touch, then compare the slopes, not the times: the hand movement adds to every answer, so it should mostly change the height of the lines.
• Repeat the conjunction condition on several days and see in History whether your slope changes by more than the width of its interval.

How the result is calculated

For each number of items the page takes the median time of your correct answers, separately for displays with and without a target. Answers under 200 ms and trials with no answer within the limit are left out of the times, the same limits as Wolfe and colleagues used. The slope is the straight line through those four medians:

Here N is the number of items, M(N) the median time in ms for that number, b the search slope in ms per item and a the intercept in ms, the time the line gives for no items at all: seeing, deciding and pressing, with no searching. A number of items with fewer than 3 usable answers is left out of the line.

The 95% interval is a percentile bootstrap inside each number of items: the page draws 2000 new sets of times from yours, with replacement, keeps the plan of 3, 6, 12 and 18 items, fits the line each time and keeps the middle 95% of the slopes. For a simulated participant with a true slope of 25 ms per item the interval was typically about 14 ms per item wide, or about 20 when the times also get more variable with more items (simulation of 1 November 2026), so a difference of a few ms per item between two runs means little.

The ratio is the absent slope divided by the present slope. It is shown only when the interval of the present slope lies clearly above 0: with a nearly flat present slope the ratio can come out as any number at all. Misses and false alarms are given for each number of items: a miss is a target you did not report, a false alarm a target you reported when there was none. If you miss more targets in large displays, those slow searches drop out of the medians and the present slope comes out too flat, and the result says so.

Compared with published slopes

These values come from other colours, shapes and equipment, so read them as landmarks.

Wolfe, Palmer and Horowitz (2010) collected 4000 trials from each of 9 or 10 observers per task, with 3, 6, 12 and 18 items and a target on half of the trials. Their summary of what such tasks typically give is in words: a feature search near zero, colour with orientation around 10 ms per item, a 2 among 5s 30 or more.

Wolfe's 1998 distribution has a single peak. There is no gap between fast "parallel" and slow "serial" searches, so this page does not sort your result into either kind. The median absent-to-present ratio was 2.24, and Wolfe and colleagues describe the ratio in the conjunction and 2 among 5s tasks as somewhat greater than 2, though in their data the difference from 2 was not statistically reliable.

Why there is no percentile

No openly available study was found that tests how well one person's slope repeats from one session to the next. The published slopes also come from other colours, shapes and screens. A rank among other people would claim more than one run can show, so History compares you with your own runs on the same input and a similar refresh rate.

Where the task comes from

Anne Treisman and Garry Gelade (1980) proposed that separate features such as colour and orientation are registered across the visual field at once, while combining them into one object takes focused attention, moved from item to item. In their Experiment I the target was a green T among brown Ts and green Xs, or in the feature condition a blue letter or an S. They offered the roughly 1 to 2 ratio between present and absent slopes as a sign of a serial search that stops when it finds the target.

Jeremy Wolfe's 1998 analysis of 2,500 sessions showed a continuum instead of two kinds of search, and Wolfe, Palmer and Horowitz note that parallel models can produce the same ratio. They call a colour and orientation search quite efficient, but clearly less efficient than a feature search.

Limits

• Colours and shapes. Wolfe and colleagues used red and green; this page uses blue and orange so that red-green colour blindness does not break the task. The bars are not matched for brightness.
• Touch. A finger moving to the button adds time and variation to every answer. A keyboard is better; with touch, compare only with your own touch runs.
• A large screen. The further apart the items are, the more you have to move your eyes, and that adds to the slope. On a PC the display is at most 640 px wide for this reason.
• Few trials. 16 trials for each cell give a rough slope, not a fine one. Read the interval and several runs.
• Your device. The screen, the keyboard and the browser add a delay to every time. Added equally to all numbers of items it raises the line without changing its slope. Below 50 Hz or with uneven frames, which remote desktop and power-saving modes often cause, the page warns you first and offers Run anyway: the run is marked less precise and kept apart in History.
• 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 of vision, attention disorders or neglect, and your slope says nothing about your health.

Controls

• Keyboard: J = target present, F = no target. Enter or Space on the focused button goes on; Esc stops a run.
• Touch and mouse: the Yes and No 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.

Sources

• Treisman, A. M., & Gelade, G. (1980). A feature-integration theory of attention. Cognitive Psychology, 12, 97–136. doi:10.1016/0010-0285(80)90005-5
• Wolfe, J. M. (1998). What can 1 million trials tell us about visual search? Psychological Science, 9, 33–39. doi:10.1111/1467-9280.00006
• Wolfe, J. M., Palmer, E. M., & Horowitz, T. S. (2010). Reaction time distributions constrain models of visual search. Vision Research, 50, 1304–1311. Full text (PMC2891283)