WolfeLab

Previous serial and conjunction search tasks have produced valuable information about mean reaction times and slopes. However, with only several hundred trials per subject, spread over several set sizes and divided between target-present and target-absent trials, there is only mediocre information about the distribution of the data. To address this issue, we have collected 4000 trials from each of 10 subjects in a classic conjunction search task and another 4000 trials from each of 10 different subjects in a serial search task. We chose a search for a "2" among "5"s as our serial task. These stimuli share the same features, but differ in organization. More specifically, the horizontal components of the 2s and the 5s remained the same, but the positions of the vertical features differ, resulting in a change in the locations of the apertures. In the conjunction task, Ss searched for a red vertical bar in the presence of green vertical and red horizontal bars. In this case, the target shares one feature with each of the distractors. The details for the stimuli and method are presented below, and these 2 data sets are available for downloading by clicking on the appropriate search type at the bottom of the page. We hope that these data can be a test bed for models of visual search. If you use the data, please tell your readers where you got it. We would be grateful to know about any interesting facts that you might dig out of these 80,000 trials. Please send us a note (wolfe@search.bwh.harvard.edu). Thank you to AFOSR for suppporting these studies.

Both sets of stimuli were created using MATLAB and the Brainard/Pelli Psych Toolbox. A central fixation point in the shape of a white cross appeared in the center of the screen before stimulus onset for both experiments. At the viewing distance of 57.4 cm, a 22.5° square region of a Macintosh computer screen was divided into an invisible 5 x 5 array of cells. Each stimulus was placed at a random location within a cell. For the serial search, the stimuli were white 2s(target) and 5s (distractor) and each subtended a visual angle of 1.5° x 2.7°. For the conjunction search, targets were red vertical bars among red horizontal and green vertical bars. Each bar subtended 3.5° x 1 ° visual angle.

Data from 10 Ss was collected separately for the serial and the conjunction searches. The method remained the same for both experiments. The task was to press one key if the target was present and another if the target was absent. The target was present on 50% of the trials. Subjects were instructed to respond as quickly as possible while minimizing errors. Subjects were tested in 12 blocks of 30 practice and 300 experimental trials and 1 block of 30 practice and 400 experimental trials for a grand total of 4000 experimental trials and 390 practice trials. Reaction time was measured at set sizes 3, 6, 12, and 18. Subjects were given feedback after each trial.

DATA SETS
The 2 data sets available for downloading have 8 labeled fields each. The columns are labeled as subject, cond report, trialdigit, set size, target present, error, message, and reaction time in msec, respectively.

Cond report refers to the type of search and remains uniform over each data set. Serial is referred to as ?2 vs. 5? and conjunction as ?RVvRHGV.?
Trialdigit shows the order in which the subjects completed each trial,
Set size refers to the total number of items on the screen (3, 6, 12, or 18)
Target present field is 1 when the target was present and 0 when the target was absent.
Error field is 0 when the Ss responded correctly and 1 for incorrect responses.
Message explains the outcome of the trials (and reproduces the information given to the subject). Specifically, trials are labeled as HIT (target present correct trials), TNEG (True negative -trial absent correct trials), MISS (target present incorrect trials) and FA (False alarm - target absent incorrect trials).

Practice trials are not included in this dataset. The number of trials for each subject may deviate slightly from 4000 because the subjects could take breaks by pressing the space bar instead of responding. These trials were not recorded and not replaced. Otherwise, these are unfiltered data. For example, we have not defined as errors any of the very long RTs or very short RTs. You may make your own decisions about these matters.

WHAT CAN 1,000,000 TRIALS TELL US ABOUT VISUAL SEARCH?

DATA SET [PDF] [EXCEL]

What is in the data set?
The slopes come from experiments with the following general characteristics. Observers were told to look for a specific target item during a block of 300-500 trials. A target was present on 50% of the trials. The display was visible until the subject responded. Subjects were asked to respond as quickly and as accurately as possible. Error rates (which will not be presented) were almost always less than 10%, with the majority less than 5%. Three or more set sizes were randomly intermixed during a block of trials. Set sizes ranged varied widely across experiments. In a standard experiment, ten subjects were tested, yielding ten target-present and ten target-absent slopes. Some studies had fewer subjects, others more, but ten is the modal value. Subjects were generally young (18-30). All had normal or corrected to normal acuity and could pass the Ishihara color test. Testing was binocular. An average subject would have had limited practice in visual search tasks (several hundred trials).

If you decide to work with these data, we would be interested to hear what you are up to. Please drop a note to wolfe@search.bwh.harvard.edu.