Location-cued visual selection—Placeholder dots improve target identification

Abstract

Visual cues help to select a target and attract attention to it. In the present study, a 50-ms exogenous cue was presented to select one of 80 tilted lines, and attention effects at various delays were measured as the time observers needed to identify this target. Like in earlier detection studies, there was a transient cuing effect; targets presented soon after the cue (delays of 50–300 ms) were identified particularly fast. This benefit was followed by a continuous decay of performance toward longer delays (measured up to 5 s), at which the necessary presentation time to identify the target was strongly increased. The decay was substantially reduced when placeholder dots were shown during the delay, at subsequent line positions. The simple presentation of a structured background in the form of random dots did not have this effect. When the presentation times for constant performance were taken to compute the presumed strength of underlying neural responses, the effect of placeholders was seen as a nearly constant addition to the cued target signals, with an additional transient peak about 100 ms after cue (and placeholders) onset.

Introduction

Since the early cuing experiments by Eriksen, Posner, and colleagues (Eriksen & Hoffman, 1973; Posner, 1980; Jonides, 1981; Posner & Cohen, 1984), it has become a widely accepted model that cues presented shortly before a stimulus attract or capture perceptual resources and make the cued stimulus more quickly detected and better recognized. The phenomenon described as orienting and cued visual attention has been confirmed in numerous studies, and the underlying dynamics and spatial properties of cuing effects have been extensively studied. The model also includes an inbuilt component of competition; when attention is shifted to one location in a scene, it must likely be withdrawn from other locations (Braun & Julesz, 1998; Beck & Kastner, 2009). 

This short and generalized description underlines the double function of visual cuing in detection and discrimination tasks. Cues help to select the target and simultaneously facilitate its detection and identification. While selection was often not an important factor in the early cuing experiments (which sometimes used the same single target at one of two to four possible locations), selection and target identification became predominant aspects in visual search, where cues were sometimes used to quickly find the target (Nakayama & Mackeben, 1989; Nothdurft, 2002, 2006). Various cue properties, including salience from feature contrast, have been tested to distinguish fast and slow visual search (Wolfe & Horowitz, 2004; Nothdurft, 1993, 2015; Gao, Mahadevan, & Vasconcelos, 2008), but in many of these experiments the dynamics of the cuing process itself were not further investigated (but see, e.g., Zenon, Ben Hamed, Duhamel, & Olivier, 2008). In studies that did look at the dynamics of target discrimination, the data have revealed interesting modulations in performance and eventually the speed of target processing after target-near cues (e.g., Cheal & Lyon, 1994; Nothdurft, 2002; Jefferies & Di Lollo, 2015). An initial fast improvement of target discrimination is followed by a gradual performance drop at longer delays, at which observers reveal increasing difficulty in identifying shortly presented targets. In certain properties, this modulation corresponds to reaction-time (RT) variations in the early cuing experiments, when cued and noncued targets had to be detected (Posner, 1980; Posner & Cohen, 1984) or discriminated (Eriksen & Hoffman, 1973; Jonides, 1981). While RT generally decreased with increasing delays, there was a short benefit, with shorter RT to cued than uncued targets, up to about 250 ms after the cue. This benefit was often followed by a period with faster responses to uncued than cued targets, from about 250–300 ms onward (Posner, 1980; Posner & Cohen, 1984). This reversal of preferences was only seen when attention had not been held at the cued location but had in between been shifted to other cues at other locations (Posner & Cohen, 1984); it was therefore described as inhibition of return (IOR; Posner, Rafal, Choate, & Vaughan, 1985; for a detailed and careful review, see Klein, 2000). 

In recent studies on “cued visual selection” (Nothdurft, 2017a, 2017b), I used test patterns with 80 different items and asked observers to identify a single, cued target among them. When test patterns were shown 100–200 ms after the cue, targets needed shorter presentations for identification than when test patterns were presented simultaneously with the cue or at very long delays. For a given target duration, therefore, performance was strongly modulated with the cue–target delay, as has also been found in previous studies (e.g., Cheal & Lyon, 1994). Reversing the sequence of cues and targets by presenting the cue in an already visible test pattern allowed me to separate target responses from the (cued) selection process. Dynamic variations of target identification with the same cue presented at different delays could then be compared with the (presumed) neural responses to target onset. Unfortunately, these detailed measurements were made only with the reversed order (target-cue sequences), not with the classical cuing paradigm in which the cues come first and are (later) followed by the target. 

One goal of the present study was thus to complete the earlier measurements with classical cue-target sequences. Another goal came from observations that target identification could strongly deteriorate with increasing cue–target delays (Nakayama & Mackeben, 1989; Cheal & Lyon, 1994; Nothdurft, 2002), particularly at very long delays (Nothdurft, 2017a). While it is known that exogenous (“automatic”) cuing effects last much shorter than endogenous (“deliberate”) cueing effects (Müller & Rabbit, 1989; Cheal & Lyon, 1994), I wondered if the decay in performance might not have partly been due to observers' difficulty keeping the early cues aligned with the later test patterns if the visual field was merely empty during the delay. I thought it would be interesting to see if the apparent drop of target identification at large delays could not be notably reduced if the potential item locations in test patterns were continuously indicated by visual markers (“placeholders”). 

The effects of placeholders in attention are manifold and often dramatic. They may act as “anchors” that help direct attention to certain locations or regions (Jefferies & Di Lollo, 2015), to objects (Egly, Driver, & Rafal, 1994), or to groups of objects (Dodd & Pratt, 2005). They improve the accuracy of gazed-at locations (Wiese, Zwickel, & Müller, 2013) and generally shape the focus of spatial attention before and during eye movements (Lisi, Cavanagh, & Zorzi, 2015; Puntiroli, Kerzel, & Born, 2018). What appears surprising, however, is the reported differential effect of placeholders on facilitatory and inhibitory attentional modulations. While they generally seem to facilitate the attraction and alignment of attention, several researchers who have observed strong IOR in test conditions with placeholders failed to see that when the placeholders were removed (e.g., Birmingham & Pratt, 2005; Pratt & Chasteen, 2007; Jefferies & Di Lollo, 2015; Taylor, Chan, Bennett, & Pratt, 2015). Furthermore, while attentional facilitation seems to spread with a gradient-like profile from the center of a cue (Downing & Pinker, 1985; Shulman, Wilson, & Sheehy, 1985; Shulman, Sheehy, & Wilson, 1986)—preferentially within but also across framed objects (Egly et al., 1994; Nothdurft, 2016a)—the spread of inhibition appears to be blocked by placeholder frames (Taylor et al., 2015). As a matter of fact, even some of the early cuing experiments (which revealed attentional benefits and costs and, under certain circumstances, IOR; Posner & Cohen, 1984) used placeholder boxes which were eventually cued and in which the later targets were presented. 

The shape of placeholder objects might be important, however. In experiments where targets had to be identified rather than detected, I have noticed a strong interference between the shapes of cue and targets (Nothdurft, 2002). Also, the size of the cue strongly affects target detection time (Eriksen & St. James, 1986; Benso, Turatto, Mascetti, & Umiltá, 1998). This influenced me in my own work to move from circles (or rectangular boxes) to four-dot cues, which were then optimized for minimal interference with the target (Nothdurft, 2016b). It is likely that a similar interference might occur when placeholders are boxes. Since I wanted placeholders only to indicate the potential target locations during the delay between cues and targets, I decided to use placeholder dots that virtually disappeared when the items were shown (Figure 1). 

Figure 1

 

Experimental setup. (A) Schematic trial procedure: After presentation of a central fixation cross (bottom) a single four-dot cue was shown to indicate which line in a later-presented test pattern was the target and had to be identified. The delay between cue and test-pattern onsets and the test-pattern presentation time (until the mask) were varied. In the main experiment (Experiment 1), two test series with different stimulus variants were compared; in one (sequence on the left-hand side; “no placeholders”), the screen remained empty after the cue until the test pattern occurred, and in the other (sequence on the right-hand side; “with placeholders”), the positions of subsequent lines were indicated by dots during that time. (B) Examples of stimulus patterns: (left) a pattern with placeholder dots, here with the four-dot cue (which disappeared after 50 ms) superimposed; (middle) a typical test pattern made of 80 oblique lines, one of which was previously cued and thus selected as the target; and (right) a masking pattern, here with the possible target locations indicated—the dashed lines were not visible in experiment. Luminance settings are not exactly reproduced, and lines of fixation crosses are widened for better visibility.

Thus, the main intention of the present study was to measure the dynamics of attention effects at cued locations and to see if these would change when subsequent target locations were marked by placeholder dots. The aim was to measure target identification, not reaction times. Since I did not want placeholders to interfere with the selection process itself, I used 80 identical placeholder dots for all items in the test pattern; target selection was always provided by the initial cue. Particular effort was made to measure performance variations in sufficient temporal resolution that a certain performance level could later be selected from each data set. All these properties in combination are unique and differ in one or several aspects from previous visual-cuing studies. 

Methods

Overview

Experiments were performed in a dimly lit room under binocular fixation. Test patterns were regular arrays of randomly tilted lines; one of these lines (the target) was cued and observers had to report its orientation. Two major parameters were systematically varied in the tests: the delay between the onsets of the cue and the test pattern, and the presentation time of the test pattern until all lines were masked (target duration). Delays were varied between 0 and 5,000 ms; the range and variety of tested target durations were individually adjusted to observers to provide several measures between 50% (chance level) and 100% accuracy. These data were later fitted with cumulative response functions to estimate the target duration at 75% correct, Δt₇₅, for each tested cue–target delay. 

There were three major variants in stimulus presentation (Figure 1). In variant A (no placeholders), no stimulus (other than the fixation cross) was shown between the cue and the later test pattern. In variant B (with placeholders), all item locations in the test pattern were indicated by single dots, which appeared together with the cue and remained visible until the test pattern occurred. The placeholder patterns themselves did not provide any hint about the selected target. In Experiment 2, a third stimulus variant was tested in which the entire screen was covered with static random dots while the no-placeholder condition was shown (variant C, structured background; Figure 7A). Random-dot patterns were shown continuously during each trial to see if the presence of any other structured background might also help to improve performance. 

Stimuli

All stimuli were generated with DOS VGA techniques on a 15-in. ultrahigh-resolution monitor (Ergo-View 15; Sigma Designs Inc., Fremont, CA). The viewing distance was 67 ± 1.5 cm. Distance variations were due to head-size differences between observers, who had their heads conveniently leaned against the wall (Nothdurft, 2017b); for each observer, the viewing distance was constant in all experiments. The refresh rate of the monitor was 60 Hz, resulting in a temporal resolution of 16.7 ms between subsequent monitor patterns. 

Test patterns were 9 × 9 regular arrays with 80 oblique lines (Figure 1B), each 0.8 deg × 0.2 deg, which were individually and randomly tilted to the left or right (+45° or −45°); line orientations were randomly assigned in every new trial. The raster width was 1.8 deg, so that the entire stimulus pattern covered about 15 deg × 15 deg of the visual field. The central item in each pattern was spared and replaced by a green fixation cross (0.25 deg × 0.25 deg). One line of the test pattern was cued and served as the target in that trial; target locations were randomly selected and refreshed in every new presentation. Cues were arrangements of four dots (four-dot cues), each 0.2 deg × 0.2 deg, located around the target at 0.6 deg distance from the target center in the four oblique directions. After presentation, test patterns were masked. In the masking patterns, each previous item was replaced by a cross obtained from the superimposition of both possible line orientations at that location. To avoid interference from crowding (Nothdurft, 2017a) and limited attentional resolution (Intriligator & Cavanagh, 2001), possible target locations were restricted to 12 positions near the fixation cross; the restriction (corresponding to the near sample in Nothdurft, 2017a) is indicated in Figure 1B. Observers were not informed about the restriction. Placeholder patterns were regular 9 × 9 arrays (without the central position) of small rhomboids (0.2 deg × 0.2 deg) representing the centers of subsequent lines and then fully covered by these. These placeholder “dots” were identical for all items and did not give any hints about the cued location or the following line orientation. Figure 1B shows examples of a test pattern (including the target), the placeholder pattern (here with the cue superimposed), and the masking pattern (with possible target locations surrounded by a dashed line that was not visible in the experiment). Cues were always shown for 50 ms and then disappeared; masking patterns were shown for 500 ms. 

Random-dot patterns in Experiment 2 (structured background; Figure 7A) were made of 4,000 dots (each 0.07 deg × 0.07 deg) all over the stimulus pattern. They were switched on together with the fixation cross (1 s before the cue) and remained visible until the stimulus pattern disappeared (500 ms after mask onset). Statistically, random dots covered about 8.6% of the relevant stimulus area on the screen. The random-dot background was locally overwritten by all stimulus figures (i.e., cues, lines, masks) so that these appeared clear, compact, and not “randomized”. 

All stimuli except the (green) fixation cross were white on a dark background. Approximate luminance settings at the 60-Hz repetition rate were about 7 cd/m² for test patterns, 17 cd/m² for masks and placeholders, and over 70 cd/m² for cues, all presented on a screen background of about 1 cd/m². Random dots were shown at about 5 cd/m². 

Procedures

The sequence of pattern presentations is shown in Figure 1A. Trials started with a 1-s presentation of the fixation cross before the cue was shown (50 ms). After a certain delay (0–5 s), the test pattern occurred for a variable duration and then was replaced by the mask (500 ms). Thereafter the entire pattern disappeared (except for the fixation cross), and subjects were asked to indicate the tilt of the cued line. After a short break (about 1 s), the fixation cross was briefly extinguished and a new trial began. This general procedure was the same in all variants of the experiment, but there were differences as to what was shown beyond these patterns in the course of a trial. In variant A (no placeholders), no further stimuli were shown. In variant B (with placeholders), placeholder dots were switched on together with the cue and remained visible until the line pattern occurred (Figure 1B). In variant C (structured background), a random-dot background was shown together with the fixation cross and remained visible until the entire pattern disappeared (except for the fixation cross; Figure 7A). 

Responses were made on a computer keyboard (German layout) in a modified two-alternative forced-choice task (Nothdurft, 2017b); response keys were selected to match the perceived tilts (left-hand “<” key for targets tilted to the left, right-hand “-” key for targets tilted to the right). Response times were not measured, and subjects could take time for responding. 

In Experiment 1, tests were blocked for the different experimental variants (with or without placeholders) and for constant cue–target delays; the various blocks were run in interleaved sequences. Individual runs covered five to 10 repetitions of trials, with different target durations at the same delay in same experimental conditions. At the very beginning, the variations of presentation times within each run were relatively coarse, to find the appropriate testing range of each observer in which performance changed from 50% to 100%. For all later tests, intermediate target durations were added to measure performance variations with better resolution. Experiment 2 (variant C, structured background) was tested at the end of the study; the blocking of test conditions in that experiment was different and will be described later. In all experiments, the different runs were intermingled and several times repeated to generate a continuous and reliable database with 30–60 repetitions of every test condition. These data were fitted with cumulative functions to evaluate the target presentation time for 75% correct responses (Δt₇₅ values). For reasons to be described later, the standard cumulative Gaussian (in percent correct responses)—y = 50 + 25 ⋅ (1 + erf[(x − a₀)/(Display Formula\(\def\upalpha{\unicode[Times]{x3B1}}\)\(\def\upbeta{\unicode[Times]{x3B2}}\)\(\def\upgamma{\unicode[Times]{x3B3}}\)\(\def\updelta{\unicode[Times]{x3B4}}\)\(\def\upvarepsilon{\unicode[Times]{x3B5}}\)\(\def\upzeta{\unicode[Times]{x3B6}}\)\(\def\upeta{\unicode[Times]{x3B7}}\)\(\def\uptheta{\unicode[Times]{x3B8}}\)\(\def\upiota{\unicode[Times]{x3B9}}\)\(\def\upkappa{\unicode[Times]{x3BA}}\)\(\def\uplambda{\unicode[Times]{x3BB}}\)\(\def\upmu{\unicode[Times]{x3BC}}\)\(\def\upnu{\unicode[Times]{x3BD}}\)\(\def\upxi{\unicode[Times]{x3BE}}\)\(\def\upomicron{\unicode[Times]{x3BF}}\)\(\def\uppi{\unicode[Times]{x3C0}}\)\(\def\uprho{\unicode[Times]{x3C1}}\)\(\def\upsigma{\unicode[Times]{x3C3}}\)\(\def\uptau{\unicode[Times]{x3C4}}\)\(\def\upupsilon{\unicode[Times]{x3C5}}\)\(\def\upphi{\unicode[Times]{x3C6}}\)\(\def\upchi{\unicode[Times]{x3C7}}\)\(\def\uppsy{\unicode[Times]{x3C8}}\)\(\def\upomega{\unicode[Times]{x3C9}}\)\(\def\bialpha{\boldsymbol{\alpha}}\)\(\def\bibeta{\boldsymbol{\beta}}\)\(\def\bigamma{\boldsymbol{\gamma}}\)\(\def\bidelta{\boldsymbol{\delta}}\)\(\def\bivarepsilon{\boldsymbol{\varepsilon}}\)\(\def\bizeta{\boldsymbol{\zeta}}\)\(\def\bieta{\boldsymbol{\eta}}\)\(\def\bitheta{\boldsymbol{\theta}}\)\(\def\biiota{\boldsymbol{\iota}}\)\(\def\bikappa{\boldsymbol{\kappa}}\)\(\def\bilambda{\boldsymbol{\lambda}}\)\(\def\bimu{\boldsymbol{\mu}}\)\(\def\binu{\boldsymbol{\nu}}\)\(\def\bixi{\boldsymbol{\xi}}\)\(\def\biomicron{\boldsymbol{\micron}}\)\(\def\bipi{\boldsymbol{\pi}}\)\(\def\birho{\boldsymbol{\rho}}\)\(\def\bisigma{\boldsymbol{\sigma}}\)\(\def\bitau{\boldsymbol{\tau}}\)\(\def\biupsilon{\boldsymbol{\upsilon}}\)\(\def\biphi{\boldsymbol{\phi}}\)\(\def\bichi{\boldsymbol{\chi}}\)\(\def\bipsy{\boldsymbol{\psy}}\)\(\def\biomega{\boldsymbol{\omega}}\)\(\def\bupalpha{\unicode[Times]{x1D6C2}}\)\(\def\bupbeta{\unicode[Times]{x1D6C3}}\)\(\def\bupgamma{\unicode[Times]{x1D6C4}}\)\(\def\bupdelta{\unicode[Times]{x1D6C5}}\)\(\def\bupepsilon{\unicode[Times]{x1D6C6}}\)\(\def\bupvarepsilon{\unicode[Times]{x1D6DC}}\)\(\def\bupzeta{\unicode[Times]{x1D6C7}}\)\(\def\bupeta{\unicode[Times]{x1D6C8}}\)\(\def\buptheta{\unicode[Times]{x1D6C9}}\)\(\def\bupiota{\unicode[Times]{x1D6CA}}\)\(\def\bupkappa{\unicode[Times]{x1D6CB}}\)\(\def\buplambda{\unicode[Times]{x1D6CC}}\)\(\def\bupmu{\unicode[Times]{x1D6CD}}\)\(\def\bupnu{\unicode[Times]{x1D6CE}}\)\(\def\bupxi{\unicode[Times]{x1D6CF}}\)\(\def\bupomicron{\unicode[Times]{x1D6D0}}\)\(\def\buppi{\unicode[Times]{x1D6D1}}\)\(\def\buprho{\unicode[Times]{x1D6D2}}\)\(\def\bupsigma{\unicode[Times]{x1D6D4}}\)\(\def\buptau{\unicode[Times]{x1D6D5}}\)\(\def\bupupsilon{\unicode[Times]{x1D6D6}}\)\(\def\bupphi{\unicode[Times]{x1D6D7}}\)\(\def\bupchi{\unicode[Times]{x1D6D8}}\)\(\def\buppsy{\unicode[Times]{x1D6D9}}\)\(\def\bupomega{\unicode[Times]{x1D6DA}}\)\(\def\bupvartheta{\unicode[Times]{x1D6DD}}\)\(\def\bGamma{\bf{\Gamma}}\)\(\def\bDelta{\bf{\Delta}}\)\(\def\bTheta{\bf{\Theta}}\)\(\def\bLambda{\bf{\Lambda}}\)\(\def\bXi{\bf{\Xi}}\)\(\def\bPi{\bf{\Pi}}\)\(\def\bSigma{\bf{\Sigma}}\)\(\def\bUpsilon{\bf{\Upsilon}}\)\(\def\bPhi{\bf{\Phi}}\)\(\def\bPsi{\bf{\Psi}}\)\(\def\bOmega{\bf{\Omega}}\)\(\def\iGamma{\unicode[Times]{x1D6E4}}\)\(\def\iDelta{\unicode[Times]{x1D6E5}}\)\(\def\iTheta{\unicode[Times]{x1D6E9}}\)\(\def\iLambda{\unicode[Times]{x1D6EC}}\)\(\def\iXi{\unicode[Times]{x1D6EF}}\)\(\def\iPi{\unicode[Times]{x1D6F1}}\)\(\def\iSigma{\unicode[Times]{x1D6F4}}\)\(\def\iUpsilon{\unicode[Times]{x1D6F6}}\)\(\def\iPhi{\unicode[Times]{x1D6F7}}\)\(\def\iPsi{\unicode[Times]{x1D6F9}}\)\(\def\iOmega{\unicode[Times]{x1D6FA}}\)\(\def\biGamma{\unicode[Times]{x1D71E}}\)\(\def\biDelta{\unicode[Times]{x1D71F}}\)\(\def\biTheta{\unicode[Times]{x1D723}}\)\(\def\biLambda{\unicode[Times]{x1D726}}\)\(\def\biXi{\unicode[Times]{x1D729}}\)\(\def\biPi{\unicode[Times]{x1D72B}}\)\(\def\biSigma{\unicode[Times]{x1D72E}}\)\(\def\biUpsilon{\unicode[Times]{x1D730}}\)\(\def\biPhi{\unicode[Times]{x1D731}}\)\(\def\biPsi{\unicode[Times]{x1D733}}\)\(\def\biOmega{\unicode[Times]{x1D734}}\)\(\sqrt 2 \) · a₁)]), with fit parameters a₀ and a₁ for the center and width of the underlying Gaussian—was sometimes replaced by a general cumulative Gaussian function, y = a₂ + a₃ ⋅ 0.5 ⋅ (1 + erf[(x – a₀)/(Display Formula\(\sqrt 2 \) ⋅ a₁)]), in which also the level, a₂, and the amplitude, a₃, were fitted to the data. With good resolution and all repetitions, each such fit is based on 300–500 stimulus presentations at any given cue delay for each observer. Fits typically revealed r² (coefficients of determination) on the order of 0.9 or more, often 0.98, with certain exceptions also to be discussed later. Experiments were carried out in sessions of up to 2 hr, each covering several test runs. 

All tasks in the present study were performed under fixation, which was regularly checked with a video camera placed above the monitor and focused upon the observer's eyes. Camera recordings were displayed on a distant monitor and inspected online by the experimenter. Checks were frequently made during the first session for every subject and regularly repeated in later sessions. All subjects quickly learned to perform the task without moving their gaze. The test series also included many conditions with short cue–target delays (<200 ms), in which there was no benefit from moving the eyes anyway (see Fischer et al., 1993). However, since shifting the gaze might have been tempting, particularly during long cue–target delays, observers were frequently reminded to keep their gaze strictly on the fixation cross and were instructed to skip a trial if their eyes had moved away, by pressing a different key. This was occasionally but very rarely the case. There was no automatic rejection of a trial, however, when a gaze shift occurred. Observers performed all tasks in a kind of relaxed fixation mode, in which the gaze was “parked” on the fixation cross and remained there even during long cuing delays, as was repeatedly confirmed by inspection of the camera monitor. 

Subjects

Analysis in this article is based on data from four observers (one female). Three of them (20–23 years) were students at Göttingen University and were paid for the time they spent in experiment; the fourth observer was the author (69–70 years). All subjects had normal or corrected-to-normal visual acuity and, except the author, were unaware of the aim of the experiments. They all had previously carried out other experiments with similar cued target identification. 

Results

All experiments in this study measured the speed of target identification at various delays after the cue. In the main part (Experiment 1), two stimulus variants were compared, one in which the screen remained empty between the cue and the later target (no placeholders) and one in which small dots (placeholders) were shown at all item locations until the test pattern occurred (cf. Figure 1). In Experiment 2 a stimulus variant was tested in which, instead of placeholders, a structured background of random dots all over the pattern was shown. The aim of all experiments was to document dynamic variations of cued (attentional) performance with increasing cue–target delays and to investigate the influence of visual placeholders upon these dynamics. 

Experiment 1: Dynamic variations of cued attention and the effect of placeholders

It has been previously reported that cues not only accelerate, or perhaps slow down, the reaction time to subsequent targets (e.g., Posner & Cohen, 1984) but also improve or deteriorate the visibility (identifiability) of the cued target (e.g., Eriksen & Hoffman, 1973; Handy, Jha, & Mangun, 1999). Benefits and costs depend on the delay between cue and target. 

Figure 2 shows mean performance data for all four observers at selected cue–target delays. With few exceptions (open circles), data points represent conditions that were tested with all observers; the data from single observers were partly better resolved and included performance measurements at intermediate target presentation times. Black data points and curves plot the target identification rates in no-placeholder conditions. At each delay, performance increased with increasing presentation time, but performance increments slowed down with increasing delays. When test patterns were onset together with the cue (delay 0 ms) or shortly thereafter, targets were identified faster than when test patterns were largely delayed. To visualize the variations, data points for each individual observer at each delay were fitted with cumulative Gaussian functions from which the target durations for 75% correct responses (Δt₇₅ values) were taken. 

Figure 2

 

Mean performance accuracy at selected cue–target delays, for tests with (red rhomboids) and without (black circles) placeholder dots. At every delay, test-pattern presentation time was systematically varied to measure performance variations between chance (50%) and perfect identification (100%). Filled symbols represent the means (and standard error of the mean) of conditions tested with all observers, open symbols the means of conditions tested with three observers when the fourth observer had already reached 100% correct. Data for individual observers as well as the mean data curves shown here were fitted with cumulative Gaussian functions to find the presentation time for 75% correct responses (Δt₇₅). Test-pattern onset was always at Δt = 0; for delays ≥ 50 ms, the relative timing of cues fell outside the graphs. At each delay, performance ratings increase with increasing target presentation time. The increase is faster at short than at long delays, and generally much faster with (red curves) than without (black curves) placeholder dots.

The averaged curves from different observers showed similar variations of Δt₇₅ values with the cuing delay as did curves obtained from the mean data in Figure 2, which are plotted in Figure 3 (black data curves). Apparently, there is a small dip soon after the cue, at delays of 50–300 ms, where the presentation time required to see half of the targets (Δt₇₅) was slightly reduced (to 50 ms at a delay of 50 ms, and to an averaged 59 ± 3.9 ms in the delay range of 50–200 ms) compared to targets cued at test-pattern onset (78 ms at delay 0 ms); in the present averages, however, this latter difference did not reach significance. For large delays, the necessary presentation time continuously increased up to values of more than 200 ms at a delay of 5,000 ms. 

Figure 3

 

Required presentation time for 75% correct responses (Δt₇₅), for the full test range (left) and enlarged presentation of short delays (right). The Δt₇₅ values were computed from cumulative Gaussian functions as shown in Figure 2. There is a general increase with increasing cue–target delays, and performance with placeholders is generally better (requiring shorter presentation times) than with no placeholders. Beyond the continuous variation (sketched by a dashed line in the right-hand graph) there are small modulations—for example, faster performance accuracy at delays between 50 and 300 ms. Averages of Δt₇₅ curves from individual observers look almost identical, except that values (and standard error of the mean) in both curves increase more steeply at very long delays. This is mainly due to an increasing error rate for one observer with long cue–target delays (cf. text and Figure 4B).

Figure 4

 

Performance variations in the no-placeholders condition (A, B) Accuracy measures of all four observers at two exemplary delays. Even though each observer revealed the same characteristic variations (increasing performance with increasing presentation time, slower increases at longer delays), individual Δt₇₅ values were reached at slightly different presentation times. Note that the temporal resolution of test series was individually adjusted to observers.

It is interesting to compare these data with the performance variations in placeholder conditions (red data curves in Figures 2 and 3). Performance generally increased much faster, leading to notably smaller Δt₇₅ values when placeholder dots indicated the future item locations during the delay (Figure 2). Note that placeholder dots occurred at cue onset and disappeared when the line pattern was shown; thus, there were no placeholders at delay 0 ms. But already at delay 50 ms (when cues and placeholder dots disappeared with test-pattern onset), performance in target identification tended to be slightly improved with placeholders. This difference became much more pronounced when the cue–target delay was further increased (Figure 2). The presence of placeholder dots generally reduced the necessary presentation time to identify targets, not only at short delays but also—and apparently much more strongly—at delays from 300 ms onwards. The difference between conditions with and without placeholders appeared to continuously grow over the entire tested range. 

It would be fair to mention the individual variations between observers (Figure 4). At delay 0 ms (Figure 4A), for example, one observer (filled triangles) required particularly short presentation times to correctly identify the targets (80% correct at 33 ms); the other observers needed slightly longer presentations (67–100 ms) for the same performance level. The variations between observers generally increased at longer delays, particularly in no-placeholder conditions (cf. Figure 4B); some observers then also began to make mistakes. For one observer in Figure 4B (open rhomboids), for example, performance accuracy never reached 100%, not even with target durations of 1,000 ms (not shown). This was the reason for using a generalized cumulative Gaussian function instead of the standard normal one, so that the level and amplitude could also be fitted to the data (see Methods). In curves like the one illustrated in Figure 4B (rhomboids), the general cumulative function revealed better fits and increased the coefficients of fit determination (from, here, r² = 0.47 with the standard Gaussian to r² = 0.74 with the generalized cumulative function). Performance curves reaching 100% accuracy were about equally well fitted by standard and generalized cumulative functions. 

Despite the individual variations, however, all observers showed the same characteristic changes in accuracy, with steep slopes at short delays and a need for notably increased target durations at long delays. And all revealed a similar strong improvement of target identification speed when placeholder dots were shown. The characteristic modulations in Figure 3 were seen in all individual Δt₇₅ curves. 

Statistics

As already seen in Figure 2, mean performance accuracy was notably different between conditions with and without placeholders. From delay 100 ms on, all red data points lie well above the black data points, except when both curves reach the 50% or 100% level; differences often exceeded the standard error of the mean. The statistical significance of this difference was analyzed using the Wilcoxon signed-rank test. After zero differences were removed there were N ≥ 86 data pairs from each observer (same delay, same target duration; different stimulus conditions), and N = 370 data pairs for the entire data sample. For such a large N, the sampling distribution converges to a normal distribution and the test statistic W can be transformed into a z score. The corresponding z scores are |z| ≥ 5.71 for individual observers and |z| = 12.62 for the entire sample, and thus large enough to reject the null hypothesis that data samples from conditions with and without placeholders might be identical (p < 0.001). This analysis included all delays from 50 ms onwards (there were no placeholder conditions at delay 0 ms). 

These systematic, consistent differences also resulted in distinct and clearly shifted Δt₇₅ curves for conditions with and without placeholders (cf. Figure 3). In fact, only three of the N = 48 data pairs (N = 12 per observer) revealed a larger Δt₇₅ value in the with-placeholders condition (red) than in the no-placeholders condition (black), two at delay 50 ms; in all other pairs, targets in the with-placeholders condition were detected faster. For individual observers, the Wilcoxon signed-rank test gave test statistics of |W| ≤ 3, which is much smaller than the critical value W_crit (n = 12; α = 0.005) = 7, so that the null hypothesis can be rejected (p < 0.005). For the entire sample (all observers; N = 48), the values are (large N) |z| = 5.94, p < 0.001. 

However, the modulation of Δt₇₅ values with increasing cue–target delays was also significant. The Mann–Whitney U test applied to performance data with delays of 50–300 ms (n₁ = 20) versus performance data with delays of 750–5,000 ms (n₂ = 20) in all observers revealed U = 23 for the no-placeholders condition and U = 26 for the with-placeholders condition; both values are smaller than the critical U_crit(α = 0.01; n₁ = 20; n₂ = 20) = 114; the differences are thus statistically significant (p < 0.01). 

However, the significance of differences on the other side of the apparent dip in Figure 3—that is, single differences between target presentations in synchrony with the cue (delay 0 ms) and target presentations at short delays (50–300 ms)—could not be established statistically in the present sample. With the given variability between observers (cf. Figure 4A), comparisons of single measures (one data point per observer at delay 0 ms) had to be based on a much larger data sample. 

Reconstruction of hypothetical response strength

With certain simplifications it is possible to relate the Δt₇₅ values at different delays to the strength of underlying signals in the visual system. If we assume that the growing identification rates with increasing presentation time (Figure 2), at every delay, result from an accumulation of neural signals, and if we further assume that constant performance levels (e.g., 75% correct) were reached when neural responses accumulated up to a similar level, we may reconstruct the relative strength of neural signals at every cue–target delay (cf. Nothdurft, 2017a). The principle is that stronger responses should require shorter periods of signal accumulation, and hence shorter presentation times, than weak responses. Formally, constant performance accuracy (e.g., 75% correct) should be proportional to the strength of the underlying signal at the according delay (“del”), integrated (accumulated) over the required presentation time Δt₇₅ at this delay. If we use mean signals, <signal>(del), averaged over the corresponding presentation time Δt₇₅, the integration becomes a simple multiplication:  

\begin{equation}{accuracy}\sim {\lt\! {\it signal} \gt} \left( {{\it{del}}} \right)\cdot\Delta {{\it t}_{75}}\left( {{\it{del}}} \right){\rm {,}}\end{equation}

or, with an unknown proportionality factor c,  

\begin{equation}{accuracy} = c\cdot {\lt\! {\it signal} \gt} \left( {{\it{del}}} \right)\cdot\Delta {{\it t}_{75}}\left( {{\it{del}}} \right){\rm {,}}\end{equation}

which can then be solved for the (averaged) signal strength:  

\begin{equation} {\lt\! {\it signal} \gt} \left( {{\it{del}}} \right) = {\it{accuracy}}/\left( {{\it c}\cdot\Delta {{\it t}_{75}}\left( {{\it{del}}} \right)} \right){\rm {.}}\end{equation}

 

Since performance accuracy was constant (75%) in Figure 3, the underlying signals at various delays should thus be proportional to the inverse of the measured presentation times:  

\begin{equation}{\lt\! {\it signal} \gt} \left( {{\it{del}}} \right)\sim 1/\Delta {{\it t}_{75}}\left( {{\it{del}}} \right){\rm {.}}\end{equation}

 

This is plotted in Figure 5. We cannot give the signal strengths in absolute values, since we do not know how big the neural signal had to be to generate a certain level of accuracy in an observer. Similar computations in another study (Nothdurft, 2017a) have shown that the required signals may strongly vary between observers. This is also suggested from the data in Figure 4, where some observers needed shorter presentation times to reach 75% accuracy than others, at the same delays. 

Figure 5

 

Inverse (1/Δt₇₅) plots of performance variations in Figure 3. With certain simplifications, these plots represent the relative strength of (neural) signals leading to a constant rating performance of 75%. For details, see text. (A) Presumed signal strengths for trials with and without placeholder dots. (B) Differences between curves in (A) directly showing the signal improvement with placeholder dots. As in Figure 3, graphs on the left-hand side show the data from the full test range and graphs on the right-hand side an enlarged copy of it at short delays. Notice that curves in (A) run almost in parallel, producing a merely flat difference curve in (B); that is, placeholder dots generated a nearly constant overall improvement of rating performance from short to long cue–target delays. In addition, there was a specific signal enhancement from placeholder dots with a peak at 100 ms.

Interestingly, the curves in Figure 5A appear to be merely shifted, as if the presentation of placeholder dots constantly increased the strengths of underlying signals at all delays. Different from the original data curves in Figure 3, where the performance differences seemed to increase toward long delays, the underlying signals were apparently less strongly modulated, with differences merely constant over various cue–target delays. This is due to the fact that adding a constant increment to a big signal will not shorten accumulation time as much as adding the same increment to a relatively weak signal. If the accumulation time for a reliable response is already short, it will only be a little reduced when the signal is increased; but if the required accumulation time to reach that response level is very long, increasing the signal strength should shorten it considerably. When the signal differences between conditions with and without placeholder dots are computed, the resulting curve is nearly flat, except for a strong peak at 100 ms indicating a facilitation effect upon target discrimination by the placeholder dots. 

Given the variability between observers (Figure 4), I wondered if this peak would show up with each observer, and if so, at which delay. For that, the individual Δt₇₅ curves of each observer were inverted and the presumed signal differences between conditions with and without placeholders were calculated (Figure 6). Despite some variability, all curves reveal a peak at delay 100 ms. Altogether, this indicates that the addition of placeholder dots during the cue–target delays generated a constant increase of the underlying signals with an additional peak around 100 ms, which was seen in each individual observer. 

Figure 6

 

Presumed signal improvements from placeholder dots in individual observers. Data as in the right-hand graph of Figure 5B are now given for each individual observer. Each curve has a peak at delay 100 ms.

Figure 7

 

Experiment 2: Presentation of random-dot backgrounds instead of placeholders. In Experiment 2, a new stimulus variant (with random dots; A) was introduced and compared with the standard tests of Experiment 1 (no placeholders, here labeled “no background,” and “placeholder dots”; B). Tests were performed by all observers at two delays (500 and 2,500 ms) with a constant presentation time of 150 ms; the mean ratings and standard error of the mean are shown in (B). For tests with and without placeholder dots on the normal background, the data reproduce the findings already seen in Experiment 1 (cf. Figure 2). The replacement of no background with random dots did not improve performance.

The nearly constant improvement of target identification by placeholder dots calls for the exclusion of potential artifacts which might have generally improved target identification after placeholder patterns with no relationship to cued attention. While I could not think of any such sensory improvement, I noticed as an observer in these experiments that, particularly after long delays, lines in the test patterns appeared dimmer in their centers after previous placeholder presentations. This can be explained by adaptation effects from placeholder dots under good fixation. It is unlikely that such a deterioration of the test stimulus should have systematically improved target identification. This was confirmed in two test runs with accordingly deteriorated lines, which indeed reduced performance in target identification. 

Experiment 2: No improvement with a random-dot background

The strong improvement of target identification by placeholder dots raised the question whether the presence of any structured background during the delays might have a similar effect on performance. Since the differences between conditions with and without placeholders were particularly strong at very long cue–target delays, there was an early suspicion that subjects might have simply lost the spatial alignment of cues and targets in the empty screen. Could the display of another structured background during that time also help to keep the cues aligned with the subsequent test patterns? To address this question, the blank (no-placeholders) condition was repeated on backgrounds made of static random dots (RD) that were shown during the entire trial. 

Experiment 2 was performed after Experiment 1 and covered two test series. In series I, performed by all observers of the main study, one target duration (150 ms) was tested at two delays (500 and 2,500 ms) with three different stimulus conditions (variant A, no background; variant D, random dots; and variant B, placeholder dots). In an additional test series II, performed by two observers of the study (both male, 20 and 23 years old), a complete test run on static RD backgrounds was performed at one delay (2,000 ms) with various target durations, and the target duration for 75% correct was computed from the data. 

Stimuli and procedures were identical to those before, except that RD patterns were switched on together with the fixation cross (i.e., 1 s before the cue) and remained visible until the end of the trial. Random dots were only shown in otherwise “empty” pattern regions (Figure 7A). For programming reasons, the conditions with no and RD backgrounds in test series I were intermingled in one test run and the placeholder condition was tested in a separate run, in interleaved sequence. In either run, the two tested delays were mixed. There was only one run in test series II, which covered all tested target durations from 0 to 250 ms in the RD-background condition. 

The mean performance of all four observers in test series I is shown in Figure 7B. At both delays, performance with placeholder dots was best (hatched bars), and much better than performance with no background (blank bars). These differences were statistically significant (Wilcoxon signed-rank test), W = 0, W_crit (α = 0.005, n = 8) = 0, p < 0.005, and replicate the observations of Experiment 1. Notice that the difference between conditions with and without (no background) placeholders is larger at delay 2,500 ms than at delay 500 ms, which is similarly seen in Figure 2 for delays of 500 and 2,000 ms. Performance in the RD-background condition (filled bars), however, was clearly below performance with placeholder dots (W = 0, p < 0.005) and not significantly different from that with no background (W = 6, n = 6). These findings reveal that the structured background made of static random dots did not improve performance. 

In addition, the measurements from two observers in test series II (delay 2,000 ms) revealed no improvement of target identification from superimposed random dots. For one observer, the Δt₇₅ values with no background and with the RD background almost exactly matched (103 and 106 ms, respectively); that is, there was no improvement at all from the static RD pattern in the background. With the second observer, there was seemingly a small improvement from the RD background (213 vs. 169 ms), but this improvement did not reach the performance levels observed with placeholder dots (99 ms). 

Discussion

The experiments revealed three major observations. 

First, the identification of cued targets is fast at short cue–target delays (up to 300 ms), but requires increasingly more time when the delay is extended. In the mean data of four observers, the fastest identification of tilted lines was measured at a cuing delay of 50 ms; observers then needed, on average, only 50 ms to correctly identify half of the targets. This was faster than for targets cued at their onset (78 ms at delay 0 ms; Figure 2), and much faster than for targets cued much earlier (164 ms at delay 2,000 ms; 246 ms at delay 5,000 ms). It was not easy to hold attention at the cued location for such a long time, and some observers did not reach 100% at these delays (Figure 2), not even with the longest presentation time tested (400–1,000 ms). 

Second, the presentation of placeholder dots at all items' locations enhanced performance. This was obvious at long cue–target delays, where placeholders reduced the ongoing deterioration of cuing effects. But analysis showed that underlying signals were similarly strongly enhanced by placeholder dots at short delays. In addition, placeholder dots caused their own peak of enhancement about 100 ms after the cue, which led to the fastest target identification measured in the study (25 ms at delay 100 ms; 75% performance with placeholder dots). Placeholders themselves were nonselective—that is, they did not indicate which of the items was cued. 

Third, the presentation of randomly patterned backgrounds instead of placeholder dots did not have the same effect. 

Comparison with early cuing experiments

The dynamic variations in performance, in particular the improvement at short delays, call to mind the numerous studies reporting reduced RTs when cued targets have to be detected. These studies report a cuing benefit (cued targets are detected faster than uncued targets) at short delays. The difference (about 25 ms in Posner & Cohen, 1984, figure 32.2; delay 100 ms) has the same magnitude as the Δt₇₅ differences measured in the present study (28 ms) between a target cued at onset and a target cued at the “best” delay (Figure 3). Notice, however, that in the present study observers always identified cued targets; there is no information about the identification speed of uncued lines. Under exogenous cuing, the RT benefit in Posner and Cohen's study lasted up to 200–250 ms. Thereafter, under certain circumstances, a reversal in target detection occurred and uncued targets were responded to faster (although not necessarily detected faster; cf. Gibson & Egeth, 1994; Klein, Schmidt, & Müller, 1998). The circumstances under which this reversal was observed were previous attention shifts to other locations on the screen; likely this withdrawal of attention then generated inhibition of return (Posner et al., 1985). The task in the present experiments did not require attention shifts to locations other than the ones initially cued, and thus we should not expect to see IOR (cf. Klein, 2000). Indeed, instead of an abrupt decrease in performance after 250 ms, performance diminished continuously with increasing delays from 300 ms on. The direct comparison of the present performance data with earlier cuing experiments is thus limited. But it is interesting to note that the early cuing benefit in target detection (presumably reflecting the effects of directed attention) was also seen with target selection and identification in the present study. 

A similar, but in fact much stronger improvement of performance than in RT has been observed in target identification and discrimination studies (Müller & Rabbit, 1989; Cheal & Lyon, 1994), sometimes even together with a subsequent drop-down from IOR in appropriate test conditions (Cheal, Chastain, & Lyon, 1998). Note, however, that in these studies performance was measured with a constant target presentation time, thus revealing performance variations across different graphs in Figures 2 and 4, not the modulation of Δt₇₅ values shown in Figure 3. Depending on the presentation time used, performance variations with an increasing cue–target delay might then be dramatic (like for a target duration of 100 ms in Figure 2) or moderate and visible only at very long delays (like for a target duration of 300 ms). 

None of the previous studies measured target identification up to delays of 5 s, and to my knowledge none has reported a similar performance decay with long delays as found in the present study. What might have caused this deterioration? The most likely explanation, of course, is that attention simply dispersed in time. If cued spatial attention produced a temporal, local capture or enhancement of neural resources, this enhancement should likely diminish over time, and attention effects gradually vanish, when the cue disappeared. Several studies have provided evidence that cues generate an attention gradient that spreads from a local maximum at the center of the cue to lower efficiency at more distant locations (e.g., Downing & Pinker, 1985; Shulman et al., 1985; Shulman et al., 1986). Even without active suppression, such a local attention peak should disperse, and attention effects disappear, when the attentional resources are not actively and continuously held there. Another cause might be inhibition (though not IOR) evoked by the cue and temporarily overwritten by short-lasting, early facilitation (see discussion in Taylor et al., 2015). It seems difficult, however, to explain the continuously growing decay in performance seen in the present work with inhibition lasting for 5 s (and likely more). 

Selection, and different modes of attention

Cued visual selection is based on two factors: selection of the correct target and analysis of its properties. Localization of cued or otherwise salient targets is often faster, and requires shorter presentations, than their identification (e.g., Nothdurft, 2002, 2006). There is no indication that selection should suffer from long delays if observers can concentrate on the task and are not simultaneously involved in other obligations. It is unlikely that the difficulty in identifying targets after long cuing delays was due to a short-term visual memory loss. Indeed, if observers had “forgotten” about the target location after these delays, they might have frequently identified wrong items. But that should then have also been the case in conditions with placeholder dots, since the indication of selected locations happened early and was identical in both conditions. In addition, a loss of short-term memory should have increased the error rate (as was indeed found with one observer) but should not have increased the time needed for correct target identifications. With long delays, however, the task in the present study required more concentration than with short delays, in which targets were identified almost automatically. 

Many cuing studies have distinguished different modes associated with different properties and different durations of attention effects (Jonides, 1981; Müller & Rabbit, 1989). Exogenous (“reflexive,” “automatic”) cuing as in the present study attracts attention fast and for shorter periods of time (maxima at 100–300 ms) than endogenous (“voluntary,” “directed”) cuing (maxima at 150–1,000 ms; Cheal & Lyon, 1994; Benso et al., 1998; see also Huang et al., 2017). Thus it is unlikely that targets, after a 5-s delay, were still identified from exogenously cued attention. But even at this delay, targets were still selected and often correctly identified. Some observers showed an increased error rate and did not reach 100% accuracy at long delays (Figure 4), but all still revealed slowly increasing performance with increasing presentation time, up to the 75% level and above. If they had simply guessed on the target after the 5-s delay, performance should have remained at 50%. Thus, selection and attention had never fully disappeared during the long delays. It seems likely, however, that observers switched from exogenous to endogenous control to keep the attentional focus at the cued location, although performance variations (Figure 3) do not indicate such a discontinuity. Interestingly, the presence of placeholders between cues and targets facilitated the task and strongly increased performance even at the longest delays tested. But even with placeholder dots, there was a decay in performance toward very long delays (Figure 4). 

Eye movements

A critical issue in experiments with long cue–target delays is the possible occurrence of eye movements. All observers quickly and successfully learned to perform the task under fixation, without looking to targets or cued target locations. The absence of eye movements was regularly checked in early sessions and in critical runs (test series with particularly long delays), but trials were not automatically disregarded when observers moved their eyes. However, there are two arguments why I think eye movements did not affect the data. First, looking at the cued location and thus foveating the target should have strongly increased performance and dramatically reduced the necessary presentation time to identify the target. Therefore, with eye movements there should be no differences between long and very long delays, and the Δt₇₅ values at these delays should generally be much smaller than measured. Second, shifting the gaze to a cued location requires time (Fischer et al., 1993). Observers could only have gained an advantage from it when the target was still visible after the shift—that is, for 200 ms or more after the cue (see also Remington, 1980). This was not the case with short delays. In addition to the regular inspections of observers' performances, therefore, the continuous performance variations from short to long delays and the actually ongoing increase of required presentation times over various long delays make it unlikely that the performance data suffered from eye movements. 

The presentation of placeholder dots might have created a particular temptation to move the eyes to the cued dot. This should have improved performance, as is indeed seen in the data. On the same arguments as before, however, this improvement was unlikely due to eye movements with these patterns. Improvements caused by gaze shifts should become visible from delays of 150–200 ms on, but were in fact already seen at delay 50 ms. And they should lead to a constant performance up to very long delays, whereas performance continuously deteriorated with increasing delays, even with placeholder dot patterns. 

The role of placeholders

The presentation of placeholder dots between cues and test patterns had a dramatic effect on the speed of target identification. The difference was already significant at a delay of 100 ms, and apparently even increased toward very long delays (Figures 2 and 3). While this would be in agreement with the initial assumption that placeholder dots might help observers keep cues and test patterns aligned, a similarly strong improvement at very short delays, as found in the reconstruction of underlying signals (Figure 5), would be counterintuitive in this model. It also seems unlikely that attention already vanished at such short delays and needed to be recollected with placeholders. It is more likely that placeholder dots (onset simultaneously with the cue) had an additional cuing effect upon items, which produced a short performance increase at 100 ms (Figure 5). It is not clear whether this additional cuing effect was restricted to the target or occurred with all items in the pattern. But only the target (with the additional four-dot cue) was selected for identification. Together with the additional placeholder dot at this location, the cuing effect might have been strengthened and target identification accelerated. 

It is intriguing that random-dot backgrounds did not provide a similar link between cues and later-shown targets. This may, of course, be due to the fact that the random-dot backgrounds did not provide the salient landmarks which might have helped observers remember the exact cue location. The regular array of 80 placeholder dots also lacked such landmarks but was better resolved to allow a single landmark to increase the local salience distribution in a cued subregion of the pattern (cf. Intriligator & Cavanagh, 2001), which the random-dot background could not do. 

Placeholder peculiarities

Some findings in the present study are quite different from the observations in other studies which also have used placeholders in cuing experiments. Many of those (including some of the early cuing experiments by Posner and colleagues) used placeholder frames (e.g., Posner, 1980; Posner & Cohen, 1984), which constitute visual objects and thus may have a particular affinity to IOR (Jordan & Tipper, 1998; Leek, Reppa, & Tipper, 2003). If we consider that the reported prerequisites for IOR (intervening shifts of attention to other locations; Posner, 1980; Klein, 2000) did not exist in the present study, the different findings are perhaps not too surprising. The different shapes of placeholder dots used here and placeholder frames used in many other studies should also be important and might account for many of these differences. 

This becomes evident when comparing the present data with a recent study by Taylor et al., 2015). In a careful and detailed investigation, those authors measured reaction times in target detection with the aim of estimating the spatiotemporal profile of cued attention. In the absence of placeholder frames, the variations in space and time looked rather continuous and consistent, except that cuing effects around the cue (which itself was a frame box) were inhibitory. However, when the authors displayed (altogether five) placeholder frames, the spatial profiles became discontinuous around these frames and the characteristic performance pattern of IOR was observed. 

This underlines the general complexity of cuing studies. Not only explicit cues but also the onset of placeholders and test patterns must be considered as potential, exogenous triggers for spatial attention. In the present study (and many other works), all these items represented highly salient stimuli which, even if not globally selecting the target, might have locally modulated the distribution of spatial attention evoked by the cue. 

Acknowledgments

I thank all observers for their continuous patience in these experiments. I am also grateful to Uwe Mattler for discussions about statistical analysis. Support from the Max Planck Institute for Biophysical Chemistry is gratefully acknowledged. 

Commercial relationships: none. 

Corresponding author: Hans-Christoph Nothdurft. 

Email: . 

Address: Visual Perception Laboratory, Göttingen, Germany.