Hybrid foraging search: Searching for multiple instances of multiple types of target

Abstract

This paper introduces the "hybrid foraging" paradigm. In typical visual search tasks, observers search for one instance of one target among distractors. In hybrid search, observers search through visual displays for one instance of any of several types of target held in memory. In foraging search, observers collect multiple instances of a single target type from visual displays. Combining these paradigms, in hybrid foraging tasks observers search visual displays for multiple instances of any of several types of target (as might be the case in searching the kitchen for dinner ingredients or an X-ray for different pathologies). In the present experiment, observers held 8-64 target objects in memory. They viewed displays of 60-105 randomly moving photographs of objects and used the computer mouse to collect multiple targets before choosing to move to the next display. Rather than selecting at random among available targets, observers tended to collect items in runs of one target type. Reaction time (RT) data indicate searching again for the same item is more efficient than searching for any other targets, held in memory. Observers were trying to maximize collection rate. As a result, and consistent with optimal foraging theory, they tended to leave 25-33% of targets uncollected when moving to the next screen/patch. The pattern of RTs shows that while observers were collecting a target item, they had already begun searching memory and the visual display for additional targets, making the hybrid foraging task a useful way to investigate the interaction of visual and memory search.

Keywords: Attention; Human search; Hybrid foraging; Hybrid search; Multiple targets; Visual search.

Fig. 1

Hybrid foraging: How would one go about finding multiple instances of each of the four targets at the top of the figure?

Fig. 2

An example of a hybrid foraging display used in this experiment. This example is a still screen shot from what otherwise would be a moving display.

Fig. 3

Miss rates as a function of memory set size and visual set size. Error bars show ±1 SEM.

Fig. 4

Average number of target types where no instances of that target were selected in the patch. Error bars show ±1 SEM.

Fig. 5

Average rate of acquiring targets (targets/s) as a function of proportion of items left on the screen (miss rate). Lines are best-fit linear regressions.

Fig. 6

Instantaneous (data points) and average (horizontal lines) rates of return for each of four memory set sizes. Note that the x-axis is “Reverse Click” Clicks are aligned to the end of the trial, such that Reverse Click 0 is the click on the “next” button. Click 1 is the last item collected, and so on, backwards in time.

Fig. 7

Instantaneous rate plotted against average rate for every observer at each memory set size for the final click on the image (Reverse Click 1) and the two preceding clicks. The diagonal line indicates equality between the instantaneous rate and average rate.

Fig. 8

Distributions of ‘runs’ where the same target is picked repeatedly. Purple (filled symbols) are empirical data. Green (outlined symbols) are simulated data, based on an assumption of random sampling among targets. Each line is a different memory set size, but they overlap extensively. Error bars, where visible, are ±1 SEM. Arrows indicate run lengths with a significant difference between observed and simulated data, with the arrow indicating direction of the deviation from random sampling. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Fig. 9

RT as a function of position in a run of selections of the same target type. Each curve represents the average RT for runs of a different length (1–8). Horizontal line is the average RT over all conditions. Data are averaged over memory set size and observers. Error bars are ±1 SEM.

Fig. 10

RT as a function of log2 of the memory set size for four types of collection events. See text for details. Regression lines are fit to the 3 lower memory set sizes and used to predict the RT value for set size 64. Open symbols show linear predictions. Outlined shaded symbols show log predictions. For graphing purposes, we actually plot predicted data points for a hypothetical memory set size of 73.5 to offset them from the actual data.

Fig. 11

RT × effective visual set size (total items/current targets) as a function of memory set size (MSS) and type of collection event (Run or Switch). Error bars are ±1 SEM. Filled, green symbols show data from Runs. Black-outlined symbols show Switch events. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)