Elsevier

Acta Psychologica

Volume 152, October 2014, Pages 158-165
Acta Psychologica

Rare, but obviously there: Effects of target frequency and salience on visual search accuracy

https://doi.org/10.1016/j.actpsy.2014.08.005Get rights and content

Highlights

  • We compared how target frequency and salience affected visual search accuracy.

  • The task included 79 unique target types and millions of visual search trials.

  • Both target frequency and salience explained independent portions of accuracy.

  • Target salience could explain why some ultra-rare targets were found often.

Abstract

Accuracy can be extremely important for many visual search tasks. However, numerous factors work to undermine successful search. Several negative influences on search have been well studied, yet one potentially influential factor has gone almost entirely unexplored—namely, how is search performance affected by the likelihood that a specific target might appear? A recent study demonstrated that when specific targets appear infrequently (i.e., once in every thousand trials) they were, on average, not often found. Even so, some infrequently appearing targets were actually found quite often, suggesting that the targets' frequency is not the only factor at play. Here, we investigated whether salience (i.e., the extent to which an item stands out during search) could explain why some infrequent targets are easily found whereas others are almost never found. Using the mobile application Airport Scanner, we assessed how individual target frequency and salience interacted in a visual search task that included a wide array of targets and millions of trials. Target frequency and salience were both significant predictors of search accuracy, although target frequency explained more of the accuracy variance. Further, when examining only the rarest target items (those that appeared on less than 0.15% of all trials), there was a significant relationship between salience and accuracy such that less salient items were less likely to be found. Beyond implications for search theory, these data suggest significant vulnerability for real-world searches that involve targets that are both infrequent and hard-to-spot.

Introduction

Visual search—the act of finding targets among distractors—is a common activity conducted countless times every day; people regularly look for specific messages in their E-mail inbox, scan restaurant menus for their favorite meals, and look for their cars in a crowded parking lot. While accurately and efficiently completing such common visual searches is desirable, other search scenarios place a much higher priority on accuracy. For example, airport security screening and radiology demand high search accuracy as their outcomes can have life-or-death consequences. Unfortunately, a variety of factors can negatively influence search accuracy, and thus it is important to understand, and overcome, these influences.

Recent evidence has examined the influence of target prevalence—the likelihood of any target appearing during search—on visual search accuracy. While many laboratory-based search tasks employ a relatively high frequency rate (e.g., most have a target present on 50% of the trials), many real-world searches—such as airport security screening and radiological cancer screening—are rare-target searches in which targets are only present on a very small percentage of trials. For example, the cancer rate in mammography is estimated at less than 5 cancers per 1000 examinations, or approximately 0.5% of cases examined (NCI, 2009). Searchers rarely encounter what they are trying to find when target prevalence is so low, and previous research has suggested that search accuracy is much lower for rarely appearing targets versus frequently appearing targets (e.g., Godwin et al., 2010, Hon et al., 2013, Menneer et al., 2010, Rich et al., 2008, Wolfe and Van Wert, 2010, Wolfe et al., 2005, Wolfe et al., 2007; but see also Fleck & Mitroff, 2007). This effect has been demonstrated in cancer screening (Evans et al., 2013, Evans et al., 2011) and for newly trained airport baggage screeners (Wolfe, Brunelli, Rubinstein, & Horowitz, 2013).

While target prevalence has been the focus of recent investigations, a related influence has gone largely unstudied. Namely, distinct from how often any target might appear (i.e., target prevalence), there is also variability across visual search tasks in how often a specific target might appear. That is, when there are multiple possible targets that can appear in a search environment, some of those targets may be present relatively more often than others regardless of overall target prevalence. For example, it is rare for any contraband item to appear in an airport X-ray image (i.e., there is a low target prevalence rate), but among these targets, some items (e.g., water bottles), are more likely to appear than others (e.g., hand grenades). In this scenario, a water bottle would have a higher frequency rate than a hand grenade. We refer to this particular issue, the likelihood of a specific target appearing during a search as individual target frequency (ITF), and note that it is distinct from target prevalence—the likelihood of any target appearing during a given search.

We recently demonstrated that visual search accuracy can be dramatically impacted by ITF rates (Mitroff & Biggs, 2014). In our previous study, we assessed data from the mobile application Airport Scanner (Kedlin Co.; https://www.airportscannergame.com) to examine search accuracy for 78 unique targets that appeared throughout millions of trials. This immense dataset provided a means to examine the influences of ITF on accuracy across a range of frequency rates (from 0.08% to 3.70%), and for extraordinarily low frequency rates (thirty items had an ITF rate below 0.15%). The evidence showed a strong logarithmic relationship between ITF and search accuracy, with relatively accurate search performance above 1% ITF and a substantial decline in accuracy below 1% ITF. The thirty targets with an ITF rate below 0.15% (i.e., each of the items appeared less than 15 times out of every 10,000 trials), which were referred to as “ultra-rare” items, had an average detection rate of 27%. This was relatively low compared to an average detection rate of 92% for the targets with an ITF rate above 1%. However, some of the ultra-rare items were often found (accuracy rates of approximately 75%) while others were almost never found (accuracy rates below 10%). This variability is intriguing, and it is important to understand why there would be such substantial differences in accuracy for ultra-rare targets. If ultra-rare items are generally harder to find, but some are actually quite easy to find, then determining what drives this variability can inform both search theory and practical implementations for real-world searches that include ultra-rare targets.

The most straightforward possible explanation for why some ultra-rare items may be found more often than others is that they “stand out more.” Salience—how much an item stands out in a display—is a common concept in attention studies: “high-salience” is used to describe items that readily stand out, whereas “low-salience” is used to describe items that do not stand out (e.g., Parkhurst et al., 2002, Treue, 2003). Notably, differences in target salience are known to have robust influences on both search speed and search accuracy. For example, a singleton—an item that differs from the rest of the display on a single, basic feature dimension—often can be found very quickly despite many homogenous distractors in the display (Treisman & Gelade, 1980). The important point is that even very similar items could vary in how well they stand out.

Here we asked a simple question: Could salience explain why some ultra-rare targets are found more often than others? Although ITF and salience are both potent contributors to visual search accuracy, they have yet to be directly compared. Highly salient targets might be found very quickly and with high accuracy, but does that also mean a searcher would miss an infrequently appearing target no matter how prominent it was in the display? While this is a relatively straightforward question, it is not easily answered in a typical laboratory-based study as it requires a range of target frequencies and salience levels. With access to a remarkably large dataset from the mobile application Airport Scanner, here we investigated the roles of ITF and salience on visual search performance across 79 different targets that varied in both ITF and salience.

Section snippets

Overview

All data reported here came from anonymous gameplay data recorded in accordance with the terms and conditions of the standard Apple User Agreement and those provided by Kedlin Co (https://www.airportscannergame.com). Players voluntarily consented to the terms and conditions upon installing the Airport Scanner application, and Kedlin Co. made the data available to our research team for analysis. Approval for research use was obtained from the Duke University Institutional Review Board.

Below we

Results

For all regression analyses, we assessed outliers based upon a Cook's D greater than 1 (Cook & Weisberg, 1982), but no data points were trimmed due to this criterion. Additionally, we assessed collinearity through the variance inflation factor (VIF) when two factors were present in analyses, although we never observed a VIF above 10—which indicates that collinearity was not an issue for any of our regression models (O'Brien, 2007).

General discussion

The purpose of the present investigation was to examine how individual target frequency (ITF) and salience influenced visual search accuracy. Specifically, could salience explain why some infrequently appearing targets are found with high accuracy while others are seldom found? Although assessing both factors simultaneously presents an intractable problem for laboratory-based research, we utilized “big data” from the mobile application Airport Scanner to assess search accuracy for targets with

References (40)

  • A. Bolfing et al.

    How image based factors and human factors contribute to threat detection performance in X-ray aviation security screening

  • Breast Cancer Surveillance Consortium

    Cancer rate (per 1,000 examinations) and cancer detection rate (per 1,000 examinations) for 1,960,150 screening mammography examinations from 2002 to 2006 by age

  • M.S. Cain et al.

    A taxonomy of errors in multiple-target visual search

    Visual Cognition

    (2013)
  • M.S. Cain et al.

    Anticipatory anxiety hinders detection of a second target in dual-target search

    Psychological Science

    (2011)
  • R.D. Cook et al.

    Residuals and influence in regression

    (1982)
  • S. Eltiti et al.

    Selective target processing: Perceptual load or distractor salience?

    Perception & Psychophysics

    (2005)
  • K.K. Evans et al.

    If you don't find it often, you often don't find it: Why some cancers are missed in breast cancer screening

    PLoS ONE

    (2013)
  • K.K. Evans et al.

    Prevalence of abnormalities influences cytologists' error rates in screening for cervical cancer

    Archives of Pathology & Laboratory Medicine

    (2011)
  • M.S. Fleck et al.

    Rare targets are rarely missed in correctable search

    Psychological Science

    (2007)
  • M.S. Fleck et al.

    Generalized “satisfaction of search”: Adverse influences on dual target search accuracy

    Journal of Experimental Psychology: Applied

    (2010)
  • Cited by (0)

    We thank Ben Sharpe, Thomas Liljetoft, and Kedlin Company for access to the Airport Scanner data, and we thank the members of the Duke Visual Cognition lab who provided helpful comments during manuscript preparation.

    View full text