When searching for a target at a cued location in a limited amount of time, an obvious search strategy is to direct your first saccade to that location. Even when the probability of the target being at that location drops between 100% and 50%, sending your eyes there before you look elsewhere still sounds like a good idea. Araujo, Kowler, and Pavel (2001, Vis. Res.) found that, counterintuitively, most participants (Ps) routinely made initial saccades to low-probability locations despite a resulting accuracy cost. Why would this indirect scan path be a good idea at all? When information about target location is given prior to a brief (∼500 ms) presentation of the search display, Ps can set attentional weights for each spatial location and plan saccades accordingly. If a two-saccade path is anticipated, it may actually be cheaper in terms of planning and enacting saccades to look from the low- to the high-probability location: the eyes would be more automatically drawn to the high probability location from the low probability side than in the converse situation. The present experiments explore the conditions under which the indirect (Pragmatic) path is chosen over the Direct path. Ps viewed 2 clusters of items for a brief time (250, 500 ms) and reported if the target was normal or reversed. Their eye movements were recorded with an ISCAN eyetracker. Before each trial Ps were told target identity, the locations of the 2 clusters, and each cluster's probability of containing the target (Definite, Equal, or Weighted). We varied the complexity of items (digital numbers, objects) and background (blank, indoor scene). Under these conditions, contrary to those used by Araujo et al., very few Ps used the Pragmatic path and error rates were low. Further variations of display complexity and task constraints will be presented to examine the modulation of the choice between a Direct and a Pragmatic scan path in the rapid goal-directed exploration of a visual scene.

In visual search tasks, observers look for targets in displays containing distractors. Likelihood that targets will be missed varies with target prevalence, the frequency with which targets are presented across trials. Miss error rates are much higher at low target prevalence (1%-2%) than at high prevalence (50%). Unfortunately, low prevalence is characteristic of important search tasks such as airport security and medical screening where miss errors are dangerous. A series of experiments show this prevalence effect is very robust. In signal detection terms, the prevalence effect can be explained as a criterion shift and not a change in sensitivity. Several efforts to induce observers to adopt a better criterion fail. However, a regime of brief retraining periods with high prevalence and full feedback allows observers to hold a good criterion during periods of low prevalence with no feedback.

Errors in spotting key targets soar alarmingly if they appear only infrequently during screening. Visual searches can be vitally important: looking for knives in luggage or tumours in mammograms, for instance. These are visual searches for rare targets, and a study comparing performance on high- and low-prevalence versions of an artificial baggage-screening task shows that we are disturbingly bad at it. If observers looking for ‘tools’ among objects drawn from other categories do not find what they are looking for fairly frequently, they often fail to notice it when it does appear. Our society relies on accurate performance in visual screening tasks — for example, to detect knives in luggage or tumours in mammograms. These are visual searches for rare targets. We show here that target rarity leads to disturbingly inaccurate performance in target detection: if observers do not find what they are looking for fairly frequently, they often fail to notice it when it does appear.

Our society relies on accurate performance in visual screening tasks--for example, to detect knives in luggage or tumours in mammograms. These are visual searches for rare targets. We show here that target rarity leads to disturbingly inaccurate performance in target detection: if observers do not find what they are looking for fairly frequently, they often fail to notice it when it does appear.

In laboratory visual search tasks, targets are typically presented on 50% of trials. However, in many important real-world search tasks (e.g X-ray screening at airports, surveillance, routine screening in radiology), target-present trials are rare. Miss errors on these tasks can have serious consequences. We mimicked this situation in the laboratory by having Os search for targets (tools) amongst other objects (not tools) and varying the percentage of target-present trials. When targets were present on only 1% of 2000 trials, Os missed 42% of targets, far more than the 6% missed when the same targets were present on 50% of trials. In order to help real-world searchers avoid these catastrophic miss rates, we need to understand how the structure of the task influences error rates. Models of target-absent trials propose that that Os set quitting criteria based on implicit and explicit feedback about their performance. They search longer after an error and terminate unsuccessful searches more quickly after accurate responses. These adaptive search termination rules become maladaptive when targets are rare. In our experiments, Os came to terminate target absent trials with average RTs that were shorter than the average time needed to find targets on target-present trials. Can we ameliorate this situation? Again, we asked Os to search for tools among other objects. As in the first experiment, the critical target tool (for example, a drill) only appeared on 1% of the trials. Other tools were targets on 49% of the trials. Under these conditions, in which Os were responding “yes” about as often as “no”, the error rate for critical rare targets dropped to 21% - a substantial improvement though far from ideal. Keeping Os' search termination criteria properly calibrated may lead to major increases in accuracy on tasks where accuracy really counts.

In most visual search tasks, target identity is blocked (i.e. Os search for the same item on each trial). If target identity changes on each trial and Os are shown what to search for just prior to each trial, how long does it take to configure the visual system to search for the target? Our prior work has shown that a cue is maximally effective 200 msec after onset if it is an exact copy of the target (e.g. picture of dog #1 cues target picture of dog #1). However, if the cue indicates the type or category (e.g. dog #1 cues dog #2) its maximum effectiveness takes longer to develop. Moreover, such cues are never as effective as an exact cue (Kenner and Wolfe, VSS 03). What happens if the cue is a modification of an exact cue? That is, when does an exact cue become an inexact “type” cue? Os searched an array of photorealistic objects for a different target on every trial. Picture cues preceded the search array by 50, 100, 200, 400, or 800 msec. Cues could be rotated (45, 135, 225, or 315 deg), shrunken (50% smaller), achromatic, or flipped about the vertical axis. Performance was compared to performance using exact cues and to performance in a blocked condition with an unchanging target on all trials. Rotated, shrunken, and flipped cues produced the same results as exact cues. Achromatic cues produced greater error rates and slower RTs but this failed to reach statistical significance. Optimal cueing does not require a precise match between cue and target. It is invariant over substantial changes in size, orientation, and reflection. Only a color change produced a hint of an effect. When searching for “this particular dog”, any of a wide range of views of “this dog” will cue search more effectively than any view of any other dog. Transportation Security Administration