Item

Abstract

Psychologists have recently discovered the Internet for demonstrating visual illusions and for education. We have examined the feasibility of using the Internet for another purpose, namely large-scale data collection in visual psychophysics. Web-experiments promise access to a huge number of subjects. The technique is therefore potentially suitable for experiments that either (1) need few data per subject (e.g., one-shot recognition experiments), (2) cover a large parameter space, (3) need data from many subjects to get a significant result, (4) are designed to catalogue possible behaviors (e.g., identification of exploration strategies in navigation), or (5) have a demand for subject diversity. Some disadvantages one has to deal with are that Internet subjects are anonymous, spend only a short amount of time on an experiment, and use unknown equipment. We have implemented five web-experiments that investigate gender perception, perception of face orientation, visual encoding of scenes, canonical views, and memory for faces (http://exp.kyb.tuebingen.mpg.de/web-experiment/). In the first year more than 2000 subjects participated in one or more web-experiments. The number of subjects per experiment varied from roughly 100 to more than 1500. Some of these experiments were derived from published experiments conducted previously in our lab, which allowed us to make direct comparisons between data obtained in the lab and on the Internet. The results of these comparisons plus a further comparison with the performance of a control group (n=20) that ran the Internet experiments using a computer in our lab strongly confirmed the general validity of our web-experiment data. We conclude that web-experiments form a valuable method for accessing large groups of subjects, provided careful thought is given to the limitations of using anonymous subjects and loosely specified experimental conditions. The technique is especially well suited for performing quick pilot-studies and for validating lab-experiments using larger numbers of subjects. Experiments that require precise control of timing, color, or display characteristics (often required for low-level psychophysics) should not be considered for Internet implementation.