Atten Percept Psychophys (2015) 77:150–159 DOI 10.3758/s13414-014-0762-8
Perceptual failures in the selection and identification of low-prevalence targets in relative prevalence visual search Hayward J. Godwin & Tamaryn Menneer & Charlotte A. Riggs & Kyle R. Cave & Nick Donnelly
Published online: 12 September 2014 # The Psychonomic Society, Inc. 2014
Abstract Previous research has shown that during visual search tasks target prevalence (the proportion of trials in which a target appears) influences both the probability that a target will be detected, and the speed at which participants will quit searching and provide an ‘absent’ response. When prevalence is low (e.g., target presented on 2 % of trials), participants are less likely to detect the target than when prevalence is higher (e.g., 50 % of trials). In the present set of experiments, we examined perceptual failures to detect low prevalence targets in visual search. We used a relative prevalence search task in order to be able to present an overall 50 % target prevalence and thereby prevent the results being accounted for by early quitting behavior. Participants searched for two targets, one of which appeared on 45 % of trials and another that appeared on 5 % of trials, leaving overall target prevalence at 50 %. In the first experiment, participants searched for two dissimilar targets; in the second experiment, participants searched for two similar targets. Overall, the results supported the notion that a reduction in prevalence primarily influenced perceptual failures of identification, rather than of selection. Together, these experiments add to a growing body of research exploring how and why observers fail to detect low prevalence targets, especially in real-world tasks in which some targets are more likely to appear than others.
Keywords Visual search . Relative prevalence . Eye movements
H. J. Godwin (*) : T. Menneer : C. A. Riggs : K. R. Cave : N. Donnelly School of Psychology, University of Southampton, Highfield, Southampton, Hampshire SO17 1BJ, UK e-mail: [email protected]
Introduction A vital component of visual search is the perceptual selection and identification of objects that could be the target (Schwark, MacDonald, Sandry, & Dolgov, 2013; Wolfe, Cave, & Franzel, 1989; Wolfe, 2007). Previous research examining the eye movement behavior of radiographers searching Xrays for tumors has explored how this perceptual selection and identification process can result in the failure to detect targets (Nodine & Kundel, 1987). Observers can fail to select (i.e., directly fixate) a target, thereby leading them to fail to detect that target. In addition, merely fixating a target does not guarantee that it will be detected: observers can also fail to identify a fixated target, and incorrectly decide that the object is not a target (see also Cain, Adamo, & Mitroff, 2013).1 In the present set of experiments, we explored how failures to perceptually select and identify objects in visual search are modulated by the prevalence of a search target. Target prevalence refers to the proportion of trials in which that target is presented, and previous work (Wolfe, Horowitz, & Kenner, 2005) has demonstrated that, when prevalence is low (e.g., a target is presented on 2 % of trials), participants are less likely to detect a target than when prevalence is higher (50 % of trials). Early accounts of the prevalence effect suggested that it was the result of motor priming of target-absent responses (Fleck & Mitroff, 2007), though this was discounted as a complete explanation after it was demonstrated that the prevalence effect was still present when motor priming was controlled for (Godwin, Menneer, Cave, Helman, et al., 2010; Wolfe et al., 2007).
1 In the radiographic literature, failures of identification have also been referred to separately as recognition and decision-making errors (e.g., see Nodine & Kundel, 1987), but since they both involve the failure to identify a target object, or components of a target object, we have grouped them together here.
Atten Percept Psychophys (2015) 77:150–159
More recently, the prevalence effect has been explained in terms of a multiple-decision model (Wolfe & Van Wert, 2010; Wolfe et al., 2007). This model explains prevalence effects in terms of two discrete decisions. The first of these occurs at a perceptual level (Schwark et al., 2013), and relates to the object currently under examination, addressing whether that object is a target (or not). When prevalence is low, the search system becomes biased towards responding ‘no’ to this question. The second of these decisions occurs at the level of the search array as a whole, and has been modeled as the accumulation of evidence towards a quitting threshold. This is regarded as a cognitive, rather than a perceptual decision (Schwark et al., 2013). If the target has not been found when this threshold is reached, participants quit searching through the objects and respond ‘absent’ (Chun & Wolfe, 1996). Under conditions of low prevalence, the threshold for making this second decision shifts such that participants make rapid target-absent responses, often giving themselves only a short time to detect the target (Godwin, Menneer, Cave, & Donnelly, 2010; Menneer, Donnelly, Godwin, & Cave, 2010; Wolfe et al., 2007, 2005). Here, our goal was to examine how reductions in target prevalence influence the perceptual selection and identification of targets. We engaged participants in a relative prevalence search task in which participants searched for two targets. One target was presented at a higher prevalence level (45 % of trials) than a second target (5 % of trials). This type of prevalence manipulation can be distinguished from manipulation of overall prevalence, where only a single target is searched for at a given prevalence level (e.g., Wolfe et al., 2005). The first study of this type was conducted by Wolfe et al. (2005, supplementary materials) where they asked participants to search for guns, knives, and clocks. Targets were presented at prevalence levels of 44 %, 10 %, and 1 %. Overall target prevalence was set to 50 %, though multiple targets could appear in a given trial. Participants were found to be sensitive to the individual prevalence levels of the different targets, and were more likely to miss targets as their prevalence decreased (for further experiments and discussion regarding tasks where multiple targets can appear in a single trial with different prevalence levels, see Fleck, Samei, & Mitroff, 2010). For our current purposes, we presented only a single target on a given trial and maintained an overall prevalence level of 50 % firstly to equate motor priming across targetpresent and target-absent responses, and secondly to avoid the cognitive decision to quit searching early that usually accompanies low-prevalence targets. Previous studies have also explored relative prevalence search (Beanland, Lenné, & Underwood, 2014; Fleck et al., 2010; Godwin, Menneer, Cave, Helman, et al., 2010; Wolfe et al., 2007, 2005), and showed that participants were slower and less likely to detect lower-prevalence targets than higher-prevalence targets. The fact that prevalence effects for lower-prevalence targets
151
emerge in relative prevalence searches demonstrates that the effects of prevalence cannot be attributed to motor errors, nor can they be attributed to boredom or vigilance errors (Davies & Parasuraman, 1982; Wolfe et al., 2007, 2005). More recently, Mitroff and Biggs (2014) explored performance when participants searched for a wide range of prevalence levels (which they refer to as different target frequencies), including very low prevalence levels (including
Perceptual failures in the selection and identification of low-prevalence targets in relative prevalence visual search Hayward J. Godwin & Tamaryn Menneer & Charlotte A. Riggs & Kyle R. Cave & Nick Donnelly
Published online: 12 September 2014 # The Psychonomic Society, Inc. 2014
Abstract Previous research has shown that during visual search tasks target prevalence (the proportion of trials in which a target appears) influences both the probability that a target will be detected, and the speed at which participants will quit searching and provide an ‘absent’ response. When prevalence is low (e.g., target presented on 2 % of trials), participants are less likely to detect the target than when prevalence is higher (e.g., 50 % of trials). In the present set of experiments, we examined perceptual failures to detect low prevalence targets in visual search. We used a relative prevalence search task in order to be able to present an overall 50 % target prevalence and thereby prevent the results being accounted for by early quitting behavior. Participants searched for two targets, one of which appeared on 45 % of trials and another that appeared on 5 % of trials, leaving overall target prevalence at 50 %. In the first experiment, participants searched for two dissimilar targets; in the second experiment, participants searched for two similar targets. Overall, the results supported the notion that a reduction in prevalence primarily influenced perceptual failures of identification, rather than of selection. Together, these experiments add to a growing body of research exploring how and why observers fail to detect low prevalence targets, especially in real-world tasks in which some targets are more likely to appear than others.
Keywords Visual search . Relative prevalence . Eye movements
H. J. Godwin (*) : T. Menneer : C. A. Riggs : K. R. Cave : N. Donnelly School of Psychology, University of Southampton, Highfield, Southampton, Hampshire SO17 1BJ, UK e-mail: [email protected]
Introduction A vital component of visual search is the perceptual selection and identification of objects that could be the target (Schwark, MacDonald, Sandry, & Dolgov, 2013; Wolfe, Cave, & Franzel, 1989; Wolfe, 2007). Previous research examining the eye movement behavior of radiographers searching Xrays for tumors has explored how this perceptual selection and identification process can result in the failure to detect targets (Nodine & Kundel, 1987). Observers can fail to select (i.e., directly fixate) a target, thereby leading them to fail to detect that target. In addition, merely fixating a target does not guarantee that it will be detected: observers can also fail to identify a fixated target, and incorrectly decide that the object is not a target (see also Cain, Adamo, & Mitroff, 2013).1 In the present set of experiments, we explored how failures to perceptually select and identify objects in visual search are modulated by the prevalence of a search target. Target prevalence refers to the proportion of trials in which that target is presented, and previous work (Wolfe, Horowitz, & Kenner, 2005) has demonstrated that, when prevalence is low (e.g., a target is presented on 2 % of trials), participants are less likely to detect a target than when prevalence is higher (50 % of trials). Early accounts of the prevalence effect suggested that it was the result of motor priming of target-absent responses (Fleck & Mitroff, 2007), though this was discounted as a complete explanation after it was demonstrated that the prevalence effect was still present when motor priming was controlled for (Godwin, Menneer, Cave, Helman, et al., 2010; Wolfe et al., 2007).
1 In the radiographic literature, failures of identification have also been referred to separately as recognition and decision-making errors (e.g., see Nodine & Kundel, 1987), but since they both involve the failure to identify a target object, or components of a target object, we have grouped them together here.
Atten Percept Psychophys (2015) 77:150–159
More recently, the prevalence effect has been explained in terms of a multiple-decision model (Wolfe & Van Wert, 2010; Wolfe et al., 2007). This model explains prevalence effects in terms of two discrete decisions. The first of these occurs at a perceptual level (Schwark et al., 2013), and relates to the object currently under examination, addressing whether that object is a target (or not). When prevalence is low, the search system becomes biased towards responding ‘no’ to this question. The second of these decisions occurs at the level of the search array as a whole, and has been modeled as the accumulation of evidence towards a quitting threshold. This is regarded as a cognitive, rather than a perceptual decision (Schwark et al., 2013). If the target has not been found when this threshold is reached, participants quit searching through the objects and respond ‘absent’ (Chun & Wolfe, 1996). Under conditions of low prevalence, the threshold for making this second decision shifts such that participants make rapid target-absent responses, often giving themselves only a short time to detect the target (Godwin, Menneer, Cave, & Donnelly, 2010; Menneer, Donnelly, Godwin, & Cave, 2010; Wolfe et al., 2007, 2005). Here, our goal was to examine how reductions in target prevalence influence the perceptual selection and identification of targets. We engaged participants in a relative prevalence search task in which participants searched for two targets. One target was presented at a higher prevalence level (45 % of trials) than a second target (5 % of trials). This type of prevalence manipulation can be distinguished from manipulation of overall prevalence, where only a single target is searched for at a given prevalence level (e.g., Wolfe et al., 2005). The first study of this type was conducted by Wolfe et al. (2005, supplementary materials) where they asked participants to search for guns, knives, and clocks. Targets were presented at prevalence levels of 44 %, 10 %, and 1 %. Overall target prevalence was set to 50 %, though multiple targets could appear in a given trial. Participants were found to be sensitive to the individual prevalence levels of the different targets, and were more likely to miss targets as their prevalence decreased (for further experiments and discussion regarding tasks where multiple targets can appear in a single trial with different prevalence levels, see Fleck, Samei, & Mitroff, 2010). For our current purposes, we presented only a single target on a given trial and maintained an overall prevalence level of 50 % firstly to equate motor priming across targetpresent and target-absent responses, and secondly to avoid the cognitive decision to quit searching early that usually accompanies low-prevalence targets. Previous studies have also explored relative prevalence search (Beanland, Lenné, & Underwood, 2014; Fleck et al., 2010; Godwin, Menneer, Cave, Helman, et al., 2010; Wolfe et al., 2007, 2005), and showed that participants were slower and less likely to detect lower-prevalence targets than higher-prevalence targets. The fact that prevalence effects for lower-prevalence targets
151
emerge in relative prevalence searches demonstrates that the effects of prevalence cannot be attributed to motor errors, nor can they be attributed to boredom or vigilance errors (Davies & Parasuraman, 1982; Wolfe et al., 2007, 2005). More recently, Mitroff and Biggs (2014) explored performance when participants searched for a wide range of prevalence levels (which they refer to as different target frequencies), including very low prevalence levels (including
