Visual search task – Theory, implementation & analysis, ready to go for your own project

What we’ll cover

  • What is the visual search task is and how can you use it in your own research?
  • How to run a visual search task in Testable starting from our ready-made template
  • How to collect data and interpret the results

If you never worked with Testable before, you can check out our 3-minute introduction here


What is the visual search task and how can you use it in your own research?

We see our world not as a collection of individual lines, colours and shapes, but as a coherent visual scene with complex objects that we understand. But how do our brains accomplish that? Do we first perceive all the basic features and then bind them as a whole, or do we perceive the whole before we can actively search for more detail if needed? In 1980, Treisman and Gelade developed the visual search task (VST) as a simple and replicable way to test both theories.

In this task, participants need to find a target in a scene of distractors. Usually the target has a particular shape and colour (like a red triangle 🔺 ) that has to be spotted among distractors hat have could have a different color (blue triangles) a different shape (red squares) or both (blue squares).

Screenshot of the visual search task in Testable
Screenshot of the visual search task

The hallmark effect measured by the VST is that targets ‘pop-out’ of a visual scene if they can be identified unambiguously by a single feature. If the target is the only red object in a scene, it can be instantly and effortlessly spotted, no matter how many distracting blue squares and triangles surround it. This is the reason why signal colours, such as a bright yellow safety vest, work so well: They capture our attention even in the most crowded visual scene.

The process by which we can recognise a single stand-out feature in our visual scene is called feature search. The insight about the superiority of feature search has been invaluable to marketeers. They can use experiments very similar to the VST to develop product packaging and branding that ‘pops out’ from a range of competitors.

When the target shares its features with the distractors (e.g. there might be blue triangles and red squares in the scene) how quickly we can spot it depends on the busyness on the visual scene. It is almost as if we have to check each object, one by one, for the right combination of features. This effortful mental process is called conjunction search.

This combination of findings suggests that we have automatic and rapid access to “lower level” visual features, allowing us to process our entire visual scene at once. More complex, “higher level” objects that require the binding together of multiple features, however, require our selective attention and can only be processed one at a time. Although we are aware of our visual world as a meaningful ‘whole’ it seems that our brains need to puzzle it together from its constituent parts.

The visual search task allows us to test how our attention is detects features in a visual scene.

How does the Tesatble Visual Search Task work?

At the beginning of the task participants are shown a target object that they will need to find in a scene, crowded by other, distracting shapes, much like a simplified game of “Where’s Wally”. In our example the target is a red triangle (🔺) surrounded by scattered distractors, which could have a different color (blue triangles), a different shape (red squares) or both (blue squares).

It is the participant’s task to use the keyboard to indicate if the target is present (‘y’) or absent (‘n’) in the scene. Typically, three aspects of the scene are varied from trial to trial:

  1. Target present or absent: The triangle could either be or not among the objects in the scene or not, each scenario 50% likely in this version of the experiment.
  2. Size of the scene: The total number of items in the visual scene (including the target, if present) can either be 4, 8, 16 or 32.
  3. Features are unique or shared:
    1. Feature search: Distractors all share one property different from the target → e.g. all are blue, or all a squares.
    2. Conjunction search: Distractors can have any combination of properties (except for the target) → red squares and blue triangles could exist together in the same scene.

This results in 16 unique trial conditions, which each have 5 visual scene examples associated with them. This amounts to a total of 80 trials in our demo experiment and should be sufficient to observe a meaningful effect.

In feature search trials it is enough to spot one of the target’s characteristics (either red colour, or triangle shape) to distinguish it from the distractors. In conjunction trials, one must look for the combination of red color and triangle shape, as there are other red and triangular distractors around.

In each trial we measure if the participant correctly indicated whether the target was among the distractors (accuracy), and how quickly they did so (response time).

Illustration of the visual search task conditions where targets are absent or not and where search must be conjunctive or not.
Small grid conditions

Illustration of visual search task condition with a large grid size.
Large grid conditions

Run this experiment in Testable from our ready-made template

We have created a template for the VST in Testable for you that you can access from our Library. It is set up and ready to go and you can start collecting data straight away by sending the experiment link to your participant. Experiments in Testable will run in every browser, which makes it very easy to collect data both in the lab as well as online.

How to customise key parameters and play around with different options

Experiments in Testable are fully customisable and you will not need to write a single line of code to edit them. The heart of each experiment is what we call the trial file. The trial file contains all information that Testable needs to run the experiment in a simple spreadsheet, that you can edit with any spreadsheet editor you like, such as Google Sheets, Excel or Testable’s in-built preview editor.

To change any part of your experiment, you only need to change the values in the trial file.

Once you have made your changes, you can save and upload the modified trial file to your experiment’s trial file section.

Selection of columns from the Testable experiment trial file of the Visual Search Task
Visual search task trial file

Collect data by sending the experiment link to your participants

After importing this template to your library, you can collect data for your experiment by sharing the unique experiment link (i.e. tstbl.co/xxx-xxx) with your participants. Once participants complete the experiment, their results will appear in the ‘Results’ section of your experiment.

Working with results from the visual search Task

This version of the VST automatically measures a participants’ response time (RT column) and accuracy (correct column) for indicating whether the target was present or not in a trial.

In the trial file we have defined three columns called condition1, condition2 and condition3 that each code one dimension that the stimuli were varied on in this task. condition1 tells us if it was present or not, condition2 tells us if feature search (target can be identified by a single feature) or conjunction search (target can only be identified using a combination of features) and condition3 codes the number of objects in a scene (4-32). We added these columns manually added to the trial file and they have no bearing on the logic of the experiment. But they do come handy for our analysis, as we can now easily group the response times and accuracies for each level of experimental conditions and calculate their mean values.

There are two key effects that you should observe when running the VST:

  • Conjunction search trials should have significantly longer response times compared to feature search trials.
  • Response times are longer for trials with more objects, but only for conjunction search and not for feature search, where response times should be unaffected by the number of objects.

In technical language you would be looking for a main effect of search type, which means that overall conjunction search takes longer than feature search. In addition, you are looking for an interaction between search type and array size. That means that the response times you can expect for the different array sizes depends on the search type required for that trial. This interaction effect allows us to conclude that single features can ‘pop out’ in crowded a visual scene. But once features need to be combined, objects are processed one by one and with conscious attention, which takes longer and is more effortful.

Illustrative diagram of the typical effect observed in a visual search task
2

Once you have collected data from multiple participants, you can also use Testable’s ‘wide format’ feature, that allows you to automatically collate all individual result files into a single file. In wide format results every participant’s data is represented as one row in the data file. This makes it easily compatible with statistical analysis packages like R or SPSS where you can assess the statistical significance of any differences you may find.

Reference list:

Treisman, A. M., & Gelade, G. (1980). A feature-integration theory of attention. Cognitive Psychology, 12(1), 97–136. doi:10.1016/0010-0285(80)90005-5

Jansson, C., Marlow, N., & Bristow, M. (2004). The influence of colour on visual search times in cluttered environments. Journal of Marketing Communications, 10(3), 183-193.