Perception & Psychophysics 1995,57 (5),715-723

Negative priming depends on ease ofselection

ERIC RUTHRUFF University ofCalifornia, San Diego, La Jolla, California and JEFF MILLER University ofOtago, Dunedin, New Zealand

Negative priming effects have been offered as evidence that distractor stimuli are identified. We conducted two experiments to determine ifsuch effects occur even when it is easy to discriminate target from distractor stimuli. In Experiment 1,we found the usual negative priming effect when tar­ get and distractor positions varied from trial to trial, but not when these positions remained fixed. Experiment 2 extended these results to a situation where the ease of selection varied only in the prime display. These findings argue that irrelevant inputs can be filtered out prior to stimulus iden­ tification under certain circumstances and therefore pose problems for strict late selection theories.

A central and as yet unresolved issue in the field ofse­ 1974; Stroop, 1935). For instance, when subjects are asked lective attention concerns the point at which relevant to name the color in which a string of letters is printed, stimuli begin to be processed more extensively than ir­ naming latencies are very long ifthe letters happen to spell relevant ones. According to strict late selection theories a conflicting color name (Stroop, 1935).This phenomenon, (e.g., Deutsch & Deutsch, 1963), all stimuli are identi­ known as the Stroop effect, suggests that words are identi­ fied regardless oftheir relevance to the task at hand, but fied even when they are not relevant to the current task. only a subset ofthese are selected for input to later pro­ Although the identities ofirrelevant stimuli often in­ cesses (e.g., response selection). Early selection theo­ terfere with the concurrent processing of relevant stim­ ries, on the other hand, assert that unselected stimuli are uli, these interference effects have been shown to dimin­ "filtered out" at an early stage of processing, prior to ish in experimental conditions that facilitate selection of complete stimulus identification. The strongest early se­ relevant stimuli. The Stroop effect, for instance, is greatly lection theories (e.g., Broadbent, 1958) propose that un­ reduced when the object whose color is to be named is selected stimuli are discarded prior to identification; physically separated from the irrelevant color word (Kahn­ however, in weaker versions ofsuch theories, unselected eman & Henik, 1981). Similarly, Yantis and Johnston stimuli are attenuated rather than discarded (e.g., Treis­ (1990; but see Miller, 1991) were able to virtually elim­ man, 1960). In either case, early selection theories pre­ inate another type ofinterference effect by using condi­ dict that relevant stimuli are more likely to be identified tions designed to encourage optimal attentional focus­ than are irrelevant stimuli (which we will call selective ing. These findings support the view that irrelevant identification), whereas strict late selection theories pre­ objects can be filtered out at an early stage ofperception dict unselective identification. (i.e., selective identification is possible) but that the suc­ To determine whether selective identification is possible cess ofthe filter depends upon many factors, including in selective attention tasks, many researchers have ex­ the ease ofdiscriminating relevant from irrelevant stim­ amined the fate of irrelevant stimuli. The results indicate uli. It should be noted that this view is quite compatible that such stimuli are processed semantically (i.e., are iden­ with evidence that subjects can identify multiple objects tified) in many experimental settings (e.g., Corteen & simultaneously (e.g., Shiffrin & Gardner, 1972), because Dunn, 1974; Corteen & Wood, 1972; Eriksen & Eriksen, the claim that identification can be selective does not imply that it must be selective. Theorists favoring the view that selection occurs after During the course ofthis research, the first author received support identification, however, have pointed out that the iden­ from a National Science Foundation Graduate Fellowship and Public tification of distractors need not necessarily result in Health Service Training Grant 2T32-MHI4268. The research was also interference effects. Therefore, it is logically possible supported by National Institute of Mental Health Grant PHS-40733. We would like to thank Cassandra Fink and Ian Fisher for assistance that irrelevant stimuli are being identified even when in­ in testing subjects, and Gordon Logan, Cathleen Moore, Toby Mord­ terference effects are absent (e.g., Allport, Tipper, & koff, Janice Murray, Hal Pashler, Doug Rohrer, and two anonymous re­ Chmiel, 1985; Driver & Tipper, 1989). Allport et al. and viewers for useful comments on earlier versions of the manuscript. others have strengthened this position considerably by Correspondence concerning this article should be addressed to showing that a different measure can sometimes demon­ E. Ruthruff, Department of Psychology, UCSD, La Jolla, CA 92093­ 0109, or to 1. Miller, Department ofPsychology, University of Otago, strate semantic processing even when irrelevant stimuli P.O. Box 56, Dunedin, New Zealand. produce no interference.

715 Copyright 1995 Psychonomic Society, Inc. 716 RUTHRUFF AND MILLER

In one experiment, Allport et al. (1985) asked subjects asked subjects to count the number ofred items in a dis­ to name a line drawing of a target object, displayed in play while ignoring the black items. Because both red red, while ignoring a distractor drawing displayed in and black items could appear in any of the eight possi­ green. In a difficult-selection condition, the locations of ble stimulus locations, it seems unlikely that selection the target and distractor drawings varied randomly from was especially easy. Fox (1994, 1995) precued target lo­ trial to trial, whereas in an easy-selection condition, the cation 150 msec before stimulus onset, but the cue nei­ target always appeared at fixation. They found little or ther significantly reduced overall RT nor completely no interference in the easy-selection condition-that is, eliminated distractor-based interference effects, so it the semantic category ofthe distractor item had little ef­ probably did not allow efficient selection of the relevant fect on the time to name a simultaneously presented tar­ stimulus. Perhaps the most convincing evidence ofneg­ get. They did find, however, that subjects were slower to ative priming in conditions thought to favor efficient se­ name a target picture when it was semantically related to lection comes from the easy-selection condition studied the distractor picture displayed on the preceding trial. To by Allport et al. (1985), in which the target always ap­ explain this slowing, which was termed negative prim­ peared at fixation. In this condition, however, the dis­ ing, they suggested that the distractor on the preceding tractor location varied randomly from trial to trial, and trial was identified and then suppressed by inhibitory it is possible that the putative early filter can prevent the mechanisms. So when the subsequent target was seman­ identification of irrelevant objects much more effec­ tically related to this distractor, a certain amount ofextra tively when the locations of these objects are known in time was required to overcome the residual inhibtion. advance. Ifso, then selection might have been further fa­ The most important conclusion was that irrelevant pic­ cilitated by using predictable distractor locations. tures were identified, as evidenced by negative prim­ In sum, because previous experiments may not have ing, even though identification did not result in interfer­ done enough to facilitate selection, and because of the ence. From this conclusion, Allport et al. argued that theoretical importance of negative priming effects, the previous reports of negligible or reduced interference present experiments were designed as stronger tests of (e.g., Eriksen & Eriksen, 1974; Francolini & Egeth, the claim that negative priming persists even when se­ 1980; Kahneman & Henik, 1981; Kahneman & Treis­ lection is easy. Ifselective identification is possible, then man, 1984) do not unambiguously support early selec­ negative priming effects should diminish greatly and tion. Similar findings ofnegative priming in conditions perhaps even disappear under more ideal conditions for producing small or negligible interference effects have selection. But if the identification of irrelevant stimuli since been reported by Driver and Tipper (1989) and cannot be prevented, then priming effects (negative or Fox (1994, 1995). facilitory) should be preserved despite our attempts to Negative priming, which has now been observed with create optimal conditions for efficient selection. a variety of stimuli and tasks (e.g., Dalrymple-Alford & Budayr, 1966; Neill, Terry, & Valdes, 1994; Tipper, EXPERIMENT! 1985), is a particularly important source ofevidence for nonselective identification because it occurs even when The basic design of Experiment 1 replicated Allport interference is greatly reduced or eliminated. Assuming et al.'s (1985, Experiments 6-8) manipulation of the un­ that the absence of interference implies optimal condi­ certainty ofthe target location and extended this manip­ tions for selection, the accompanying negative priming ulation by including conditions in which selection was appears to indicate that irrelevant stimuli are identified made even easier by using fixed distractor locations. In even under optimal selection conditions-that is, that addition, we manipulated the discriminability ofthe col­ selective identification never occurs. This assumption ors in which the target and distractor stimuli were drawn, seems reasonable but might nevertheless be unjustified. because it seemed worthwhile to try more than one ma­ Suppose interference effects disappear as soon as selec­ nipulation ofselection ease. At issue in this experiment tion is easy enough for the early filter to slow identifi­ was whether the amount of negative priming would be cation of irrelevant stimuli. Negative priming effects sensitive to either or both ofthese manipulations. might still remain under such conditions, because delayed Stimulus displays contained two colored letters: a tar­ distractor identification could cause negative priming get and a distractor. The subjects' task was to identify on the next trial even ifit were too late to cause much in­ the target letter in each display, designated by its color, terference on the current trial (Broadbent & Gathercole, by rapidly pressing one of four response keys. A single 1990; Driver & Tipper, 1989). Thus, although negligible trial consisted of a prime display followed by a probe interference effects do suggest that selection is relatively display, and the negative priming effect was measured by easy, they do not necessarily indicate that selection is varying the relationship between the distractor letter in easy enough for negative priming to disappear. the prime display and the target letter in the probe dis­ In fact, each of the previous studies demonstrating play. On ignored repetition trials, the target letter in the negative priming in the absence of significant interfer­ probe display was the same letter just used as a distrac­ ence seems to have used conditions that were less than tor in the prime display; on control trials, the probe tar­ optimal for the efficient selection of relevant stimuli. get was different from the prime distractor. Examples of Driver and Tipper (1989, Experiment 1), for instance, ignored-repetition and control trials are shown in Fig- EASE OF SELECTION 717

Procedure. The subjects were instructed to indicate the identity o• DislracforTarget ofeach target letter by pressing the "n,' "m,' ",", or "" key with the fingers of their dominant hand. The mapping of the four stimulus Ignored letters to the four response keys was randomly selected for each sub­ Control Repetition ject. The subjects were told to respond as quickly as possible with­ out making many mistakes. They were also given strict instructions to maintain their gaze at the fixation point throughout each trial. Targets and distractors were differentiated by their colors. Half Prime ~ ~s of the subjects discriminated between green and purple letters, Display S which were highly discriminable, and the other halfdiscriminated between white and light blue letters, which were less discrim­ inable. Selective identification predicts that the amount of nega­ Probe II tive priming will be smaller when selection is easy-that is, when Display Oll the colors are highly discriminable. The assignment of colors to X target and distractor stimuli was counterbalanced across subjects. Figure 1. Examples of control and ignored-repetition trials of Ex­ The ease of selection was also manipulated by varying the loca­ periment 1. In the actual stimulus displays,target and distractor let­ tion uncertainty of the target and distractor stimuli, as in Allport ters were distinguished by their colors; here, however, targets are et al. (1985). There were three levels of location uncertainty, ma­ shown as filled letters and distractors as unf"IIIed letters. nipulated between subjects. For one third ofthe subjects, the target and distractor positions varied unpredictably from trial to trial (the difficult-selection condition). In this condition, the target and dis­ tractor letters were always positioned along diagonally opposite cor­ ure 1.On the basis of previous results (e.g., Allport et al., ners ofan imaginary square centered about the fixation point. For 1985), responses on ignored-repetition trials were ex­ another third of the subjects, the target was always presented at fix­ pected to be slower than those on control trials, and the ation, but the distractor appeared randomly in one offour comers of measure ofprimary interest (i.e., the amount of negative an imaginary square centered about fixation (the easy-selection priming) was defined as the difference in probe display condition). For the final third ofthe subjects, the target always ap­ peared at fixation, and the distractor always appeared at the same reaction time (RT) between the ignored-repetition and corner of the imaginary square on every trial within a block (the control conditions. As is common in these designs, we very-easy-selection condition);' For this group, the distractor posi­ also included attended-repetition trials, in which the tion varied randomly between blocks. In order to hold roughly con­ probe target was identical to the prime target. stant the degree oflateral masking between the target and distractor Following Allport et al. (1985), we manipulated the letters, the centers ofthe target and distractor letters were placed ap­ ease of selection by varying the subject's uncertainty proximately lAO apart in all three selection conditions. The letters in the prime and probe displays were chosen as follows. concerning the locations ofthe target and distractor let­ On each trial, one ofthe four stimulus letters was randomly selected ters. For one group ofsubjects, the position of the target to be the target in the probe display, with the restriction that each let­ and distractor letters varied randomly from trial to trial. ter was chosen equally often within a block of trials. The distractor This is a replication ofthe difficult-selection condition in the probe display was then chosen randomly from the remaining ofAllport et al. For another group ofsubjects, the target three letters. The letters ofthe prime display were then selected with always appeared at fixation, but the distractor location the following restrictions, depending upon the experimental condi­ varied randomly from trial to trial, replicating Allport tion. In the control condition, the target and distractor stimuli in the prime display were chosen so that they were different from each et al.'s easy-selection condition. A third group of sub­ other and from the letters in the probe display. On the ignored­ jects was tested in a previously unexamined condition in repetition trials, the prime distractor was identical to the probe dis­ which the target and distractor locations remained con­ play target, and the prime target was different from the letters in the stant throughout a block oftrials. We refer to this as the ­ probe display (see Figure I). On the attended-repetition trials, the very-easy-selection condition, because target selection prime target was identical to the probe target, and the prime distrac­ could be even easier than in the easy-selection condition tor was different from the letters in the probe display. Within each block, there were twice as many control trials as studied by Allport et al., especially ifthe early filter can ignored-repetition and attended-repetition trials, so that the prob­ more easily prevent the identification of distractors ability ofthe prime display target being the same as the probe dis­ whose locations are known in advance. play target was .25. Similarly, the probability that the prime dis­ tractor would be identical to the probe target was .25. Note that Method these are the same probabilities ofoccurrence as if the four stim­ Subjects. Eighty-four undergraduates (44 males, 40 females) at uus letters were chosen completely at random. the University ofCalifornia, San Diego, primarily 18-24 years old, Each trial began with the presentation of a fixation cross for participated to fulfill a class requirement. All subjects reported 700 msec in the center of the display, followed by a blank screen having normal or corrected-to-normal visual acuity. for 300 msec. Next, the letters ofthe prime display were shown for Stimuli. Stimuli were the uppercase letters S, 0, X, and I, cho­ 150 msec. Once the subject had responded to the prime display, the sen because of their high interstimulus discriminability. Letters fixation cross was displayed again for 200 msec. After a blank pe­ were created by drawing straight line segments between points in riod of300 msec, the probe letters were displayed for 150 msec. If a 5 (horizontal) X 7 (vertical) array. They were displayed on an the subject responded incorrectly to the prime and/or the probe, an NEC multi-sync monitor connected to an IBM-PC-compatible error message was displayed for 2 sec immediately following the computer and subtended approximately 0.60 horizontally and 1.20 probe response; no feedback was given for correct responses. The vertically from a typical viewing distance of50 em. Stimuli were next trial began 1.5 sec later. displayed in green, purple, white, or light blue, corresponding to The subjects completed six blocks of64 trials in a single session EGAIVGA mode palette numbers 10, 13, 15, and II, respectively. lasting approximately 40 min. The first block was preceded by 32 718 RUTHRUFF AND MILLER

T land rKl"'~' the distractors facilitated selection ofthe target stimuli. ....u An analysis oferror rates revealed no significant effects. .! 20 The most important result is that the RT difference be­ tween the control and ignored-repetition conditions (i.e., ~ the negative priming effect) depended significantly on E location uncertainty [F(2,78) = 7.83,p < .01], as shown e,.. 10 in Figure 2. In addition, separate analyses within each of ~ the three selection conditions indicated that there was ClII significant negative priming in the difficult- and easy­ Z 0 selection conditions (p < .00 I) but not in the very-easy­ selection condition (p > .20). Furthermore, the amount of negative priming in the very-easy-selection condition was D l lllellil hi)' v ry fll)' significantly less than that ofthe easy-selection condition Locati on Unc rta ny (p < .05) and that ofthe difficult-selection condition (p < .05). The comparison between the easy- and very-easy­ Figure 2. Experiment1: Mean negative primingeffect, with the as­ sociated standard error, as a function oflocation uncertainty andthe selection conditions is especially important because the discriminability of the target and distractor colors. displays in these two conditions were identical except for the fact that, in the very-easy-selection condition, the subjects could anticipate the location of the distractor practice trials, and subsequent blocks were preceded by 3 warm­ letter. Thus, the negative priming effect was reduced or up trials. The subjects were allowed to take a short break at the eliminated when selection was especially easy, as pre­ middle and end of each block. dicted by the hypothesis ofselective identification. An analysis including both easy- and difficult-selection Results and Discussion conditions, but excluding the very-easy-selection condi­ Trials in which an error was made to the prime or tion, revealed a significant negative priming effect (p < probe display were excluded from analyses of RT, as .00 I) that did not vary significantly between the two were trials in which the subject responded in less than selection conditions [F(1,52) = 1.53,p> .2]. Thus, the re­ 200 msec or more than 1,500 msec. The resulting mean sults ofExperiment I replicate the results from the diffi­ RTs to probe displays in the control, ignored-repetition, cult- and easy-selection conditions ofAllport et a1. (1985). and attended-repetition trials were 679, 689, and It is the new condition, with fixed distractor locations, that 553 msec, respectively, and the corresponding percent­ seems primarily responsible for the decrease in the nega­ ages ofcorrect response (PCs) were 96.5,96.6, and 98.9. tive priming effect. This suggests that advance knowledge Negative priming. Figure 2 shows the amount of ofdistractor location is important, which is consistent with negative priming as a function of location uncertainty models in which selective attention operates largely by and the discriminability ofthe target and distractor col­ preventing the identification of distractors-rather than ors. Table I shows the mean error rates. simply enhancing the identification oftargets (cf. Shiu & Repeated measures analyses of variance (ANOVAs) Pashler, 1994}-and in which filtering is more effective were conducted on mean RTs and PCs to the probe dis­ when distractor locations are known in advance. play in the ignored-repetition and control conditions.I The negative priming effect did not depend on the There was a small but significant negative priming effect discriminability of the target and distractor colors on RT: responses were 10 msec slower, on average, in the [F(1,78) < I]. This result cannot be interpreted as evi­ ignored-repetition condition than in the control condi­ dence that negative priming is independent of selection tion [F(1,78) = 24.52,p< .001]. MeanRT also depended ease, however, because color discriminability did not on location uncertainty [F(2,78) = 7.26, p < .01], with produce a significant main effect. mean RTs of 734, 695, and 624 msec in the difficult-, Facilitation. Similar ANOVAswere conducted on mean easy-, and very-easy-selection conditions, respectively. probe RTs and error rates in the control and attended­ Responses in the very-easy-selection condition were repetition conditions. As expected, the subjects were faster significantly faster than those in the easy-selection con­ [F(1,78) = 404.57,p<.001] and more accurate [F(I,78) = dition (p < .01), suggesting that fixing the position of 84.32,p < .001] in the attended-repetition condition than in

Table 1 Mean Error Rates for Experiment 1 as a Function of Trial Type, Location Uncertainty, and Color Discriminability Color Discrimination Location Difficult Location Easy Location Very Easy Trial Type Difficult Easy Difficult Easy Difficult Easy Attended repetition 0.7 1.6 0.8 0.6 1.9 1.0 Ignored repetition 2.5 3.2 3.0 3.0 5.2 3.6 Control 3.2 3.1 3.4 2.8 4.9 3.8 EASE OF SELECTION 719

the control condition. Unlike the negative priming effect, the ease of selection in the probe display. According to however, this difference did not depend on location uncer­ this explanation, distractor letters were identified and tainty [F(2,78) < 1], nor did it depend on color discrim­ inhibited in all conditions, but, for some reason, this in­ inability [F(1,78) = 1.l4,p>.2]. hibition only produced negative priming when selection Prime displays. Mean prime display RT depended on ofthe probe target was relatively difficult. location uncertainty [F(2,78) = 6.65, p < .01]; mean There is, in fact, some evidence that the magnitude of RTs were 727, 691, and 627 msec in the difficult-, easy-, negative priming depends on characteristics ofthe probe and very-easy-selection conditions, respectively. As was display. Allport et al. (1985, Experiment 9), for example, the case for probe displays, mean RTs in the very-easy­ found that negative priming was observed only when the selection condition were significantly faster than those probe display contained a distractor stimulus. Moore in the easy-selection condition (p < .05). (1994) extended this finding by demonstrating that neg­ Interference? The lack of negative priming in the ative priming is generally not observed whenever it can very-easy-selection condition suggests that distractor be easily determined that the probe display contains no letters were not identified. If this is the case, then one conflicting information. So, although displays in the would certainly expect interference effects to disappear very-easy-selection condition ofExperiment 1 did con­ as well. The present design did not allow us to measure tain distractors, it is still conceivable that negative prim­ interference effects, however, so we tested this predic­ ing was eliminated because they were easy to discrimi­ tion by running a control experiment with 48 new sub­ nate from targets. jects equally divided between the difficult- and very­ The purpose ofExperiment 2, therefore, was to repli­ easy-selection conditions. The stimulus set in the cate Experiment 1 using a design where the ease oftar­ control experiment included an additional letter that get selection on the probe display was held constant. sometimes appeared as the distractor but never as the Once this was done, any differences across conditions target. Because this neutral stimulus was not associated could be attributed unambiguously to differential pro­ with a particular response, it should produce less in­ cessing ofprime displays. terference, if identified, than would the other letters in In both prime and probe displays, stimulus letters (E, F, the stimulus set (cf. Eriksen & Eriksen, 1974). Thus, if G, and H) were drawn by removing line segments from distractor letters are identified, we expected faster re­ figure eights (Yantis & Jonides, 1984). The subjects were sponses to targets accompanied by the neutral distractor told to respond on the basis ofthe identityofthe purple let­ than to targets accompanied by one ofthe other response­ ter, while ignoring the green letter. To control the ease of assigned letters as a distractor. This effect was obtained selection on prime trials, we varied the stimulus onset with difficult selection (35±1l msecj' but not with asynchrony (SOA)of the color and identityinformation. In very easy selection (1 ± 6 msec). In the very-easy­ a "simultaneous" condition (SOA = 0 msec), the two types selection condition, then, distractors produced neither of information were presented together (i.e., gray figure negative priming nor interference. Although it is still eights turned into colored letters). In three "identity-first" logically possible that the distractors are nonetheless conditions, the figure eights turned into gray letters, which identified in this condition, as would be revealed by then changed to purple and green after SOAsof 100,200, some other measure, the most parsimonious explanation or 300 msec. In three "color-first" conditions,the gray fig­ of these results is simply that the distractors were not ure eights changed to purple and green 100, 200, or identified. Thus, the overall conclusion from Experi­ 300 msec before the letters were revealed. Because it was ment I is that selective identification is possible, at least possible for the subjects to move their eyes before the let­ under suitable conditions. ters were revealed at the longest SOAs in the color-first condition, the stimuli were presented at approximatelythe EXPERIMENT 2 same locations (i.e., overlapping) so that eye movements would not be particularly useful. The results ofExperiment I indicate that the amount When the colors are presented before the letters, stim­ ofnegative priming is reduced or even eliminated when ulus selection should be relatively easy, because the sub­ it is especially easy to discriminate targets from distrac­ ject can "focus on" the location of the purple figure tors, contrary to the conclusion ofAllport et al. (1985). eight and ignore the green figure eight. Conversely, se­ These results are important because they suggest that the lection should be difficult when the colors are presented identification of irrelevant stimuli can be prevented after the letters, because the subject initially does not under these conditions. know which letter identity is the target. To the extent that However, one might criticize Experiment 1 because selective identification is possible when it is easy to se­ the difficult-, easy-, and very-easy-selection conditions lect target stimuli, we should observe the greatest nega­ differed not only in the ease with which the target could tive priming effects in the identity-first conditions and be selected in the prime display but also in the ease with the smallest negative priming effects in the color-first which the target could be selected on the probe display. conditions. Alternatively, if selective identification is It is conceivable that the lack ofnegative priming in the impossible, then distractors will always be identified, very-easy-selection condition ofExperiment 1 was due and priming should be substantial regardless ofthe order not to successful filtering ofthe prime distractor but to in which color and letter information are presented. 720 RUTHRUFF AND MILLER

In summary, whereas ease of selection was manipu­ The subjects completed eight blocks of 56 trials in a single ses­ lated in both prime and probe displays in Experiment I, sion. The first block was preceded by 12 practice trials, and sub­ ease ofselection was manipulated only on the prime dis­ sequent blocks were preceded by 2 warm-up trials. The entire ses­ plays in Experiment 2. This was accomplished by reli­ sion lasted approximately 45 min. ably and accurately precuing the locations of the prime display targets and distractors on a proportion ofthe tri­ Results and Discussion als (i.e., in the color-first conditions). Note that because Mean RTs to probe displays in the control, ignored­ both target and distractor locations were precued, this repetition, and attended-repetition trials were 748, 754, condition most resembles the very-easy-selection con­ and 6 I0 msec, respectively, and the corresponding PCs dition of Experiment I, in which negative priming was were 97.2, 96.5, and 99.2. virtually eliminated. Because ease of selection on the Negative priming. Figure 3 shows the mean probe dis­ probe display was held constant across conditions, elim­ play RTs and error rates in the ignored-repetition and con­ ination ofnegative priming in the precued conditions of trol conditions as a function ofSOA. Analyses ofprobe Experiment 2 could be confidently attributed to differ­ display RT and PC as a function of SOA and ignored­ ential processing of distractors (i.e., selective identifi­ repetition versus control conditions yielded a significant cation) on the prime display. overall negative priming effect on PC [F(I,29) = 4.87, P < .05], but not on RT [F(I,29) = 2.04, P > .1]. Method Although the overall negative priming effect on RT Except where noted, the methods were identical to those ofEx­ failed to reach significance, the amount ofnegative prim­ periment I. ing decreased (p < .05, linear trend) as the SOA between Subjects. Thirty undergraduates (17 males, 13 females) at the color and identity information decreased-that is, nega­ University of California, San Diego, primarily 18-24 years old, tive priming decreased with increasing ease ofselection. participated to fulfill a class requirement. All subjects reported In addition, a planned comparison ofmean RTs from the having normal orcorrected-to-normal visual acuity. None had par­ two most extreme SOAs (300 msec identity-first, and ticipated in Experiment I. Stimuli. Stimuli were the letters E, F, G, and H, drawn with 300 msec color-first) revealed a significant difference in straight line segments. The letters subtended approximately 0.60 the amount of negative priming [F(I,29) = 4.25, p < horizontally and 1.20 vertically, and their centers were placed ap­ .05]. Furthermore, examination of mean RTs at each proximately 0.60 apart so that the letters were largely overlapping, SOA separately indicated that the negative priming effect with one letter centered above and to the left ofthe other. Stimuli was significant at an SOA of 300 msec in the identity­ were displayed in gray, green, or purple, corresponding to first condition [F(l,29) = 8.01, p < .01], but not at any of EGAIVGA mode palette numbers 7, 10, and 13, respectively. Procedure. The subjects were instructed to respond to the pur­ the other SOAs (p > .1). Overall, these results are com­ ple letters in the prime and probe displays, while ignoring the patible with the claim that negative priming decreases green letters. They responded to the letters E, F,G, and H by press­ with increasing ease ofselection and that negative prim­ ing the "d,' "c, "rn,' and "k" keys, respectively, using the middle ing essentially disappears when it is especially easy to and index fingers of their left and right hands. discriminate targets from distractors. Similar analyses on Each trial began with the presentation of the figure eights for PC indicated a significant effect of negative priming on I sec, at which point line segments were removed to reveal the tar­ get and distractor letters of the prime display.The figure eights were accuracy only at an SOA of 100 msec in the identity-first initially gray on all trials. In the simultaneous condition, the gray condition [F(I,29) = 5.64,p < .05]; the overall decrease figure eights changed abruptly to purple and green letters, exactly in negative priming with increasing ease ofselection was I sec after their onsets. In the color-first conditions, the gray figure not reliable in the PC data (p > .10, linear trend). eights corresponding to the target and distractor locations turned The overall effect ofnegative priming on RT was very purple and green, respectively, 700, 800, or 900 msec after their on­ weak, except in the identity-first condition at an SOA of sets (i.e., 300, 200, or 100 msec before the letters were revealed). In 300 msec. This was somewhat surprising, because previ­ the identity-first conditions, colors were not presented until 100, 200, or 300 msec after the figure eights changed to letters. The SOA ously published experiments using the simultaneous con­ for each trial was chosen randomly, with the restriction that each of dition (O-msec SOA) have generally revealed significant the seven different SOA conditions occur equally often within a negative priming effects on RT,4 and, ifanything, the ef­ block of trials. The interval between onset of the figure eights and fects should be even larger with letters presented first by onset of the letters was held constant at I sec across all SOA condi­ 100 or 200 msec. This finding suggests that some aspect tions, so that the temporal uncertainty of the to-be-reported infor­ of the present procedure was especially effective at in­ mation (i.e., letter identity) would be equal across conditions. Prime letters remained on the screen until the subject made a re­ ducing selective identification. The most obvious candi­ sponse. After a 200-msec blank interval, the gray figure eights date is the temporal separation of color and identity in­ were presented again. This time, they remained for 300 msec, at formation in this experiment, because, in most previous which point they abruptly changed into the purple and green let­ experiments, target/distractor and identity information ters of the probe display. Thus, the SOA between color and iden­ were given simultaneously. Because the two types ofin­ tity information in the probe displays was always 0 msec. The formation were separated here, and the SOA varied ran­ probe display remained visible until the subject responded. Ifthe domly from trial to trial, the subjects in our experiment subject responded incorrectly to either the prime or the probe, an error message was displayed for 800 msec immediately following may have adopted a strategy of waiting for color infor­ the probe response; no feedback was given following correct re­ mation to arrive before fully processing letter identities, sponses. The next trial began 1.5 sec later. just as they seem to wait to find out which letter is rele- EASE OF SELECTION 721

790 ----0-- Ignored Repetition To check on the importance ofboth temporal separa­ _ Control tion and the use offigure eights, we ran a control exper­ iment (N = 23) in which color and identity information were always presented simultaneously on both prime 770 'U and probe displays. On halfofthe trials, the letters were G> III revealed by removing line segments from the figure ! eights (exactly as in the main experiment); on the other G> E 750 half of the trials, the letters appeared abruptly, without i= figure eights. This control experiment revealed a reli­ c o able (p < .001) negative priming effect of 20:::': 10 msec. 'E... Contrary to the hypothesis that figure eights prevent neg­ G> Ill: 730 ative priming, we actually observed somewhat greater negative priming with figure eights (M = 23 msec) than without (M = 17 msec), although this difference did not approach significance (p > .10). Thus, we conclude that 300 200 100 o 100 200 300 the lack ofnegative priming in the main experiment at Color-First Identity-First the O-msec SOA likely resulted from the temporal sepa­ SOA ration of color and identity information, rather than the use offigure eights. Another control experiment (N = 10) was run using only simultaneous and identity-first SOAs (l00 and 200 msec), and these SOAs yielded reliable negative priming effects (p < .05) of 37, 39, and 48 msec, re­ spectively. Thus, the extent of negative priming in the ...II... identity-first conditions seems to depend on whether Ill: color-first conditions are also tested in the experiment, which provides additional evidence that selective or non­ selective identification may be partly strategic. One possible explanation for these strategic effects is that when color information often arrives prior to the on­ set ofthe letters (as in Experiment 2), subjects adopt the strategy ofdeferring identification until the relevant ob­ ject can be selected. But when color information is never 300 200 100 o 100 200 300 presented prior to stimulus onset (i.e., there are no color­ Color-First Identity-First first conditions, as in the previously discussed control SOA experiments), then color information will never arrive Figure 3. Experiment 2: Ignored-repetition and control condition before the onset ofthe letters. Rather than always defer mean reaction time and error rate as a function of the stimulus onset asynchrony (SOA) between color andidentity information. Thestan­ identification of the letters until the color information dard errors about these means were computed using the condition arrives, subjects choose to begin processing letter iden­ (ignored repetition versus control) X SOA X subjects interaction tities immediately. This hypothesis correctly predicts term as the error term. that the negative priming resulting from distractors in the simultaneous and identity-first conditions should be greatly reduced whenever color-first conditions are also vant before fully identifying perceptually degraded let­ used in the experiment. Because letter identification ters in a bar-probe task (Mewhort, Johns, & Coble, 1991; tends to be triggered by the onset ofcolor information in Pashler, 1984). Note that the strategy of preventing the these conditions, this hypothesis also predicts that once identification ofstimulus letters that are not known to be color information arrives, it should make little differ­ relevant is perfectly consistent with the notion of selec­ ence how much earlier the letters were presented. As tive identification. On this view, the presence ofnegative discussed below, the prime data support this prediction. priming at the longest SOA in the identity-first condition Facilitation. ANOVAs were conducted on mean might reflect an inability, or an unwillingness, to indefi­ probe display RTs and PCs in the control and attended­ nitely defer identification ofall stimuli. repetition conditions. As expected, the subjects were Another possibility, suggested by a comparison ofthe much faster [F(1 ,29) = 231.97, p < .001] and more ac­ present procedure with that ofExperiment 1, is that neg­ curate [F(l,29) = 34.05, p < .001] in the attended­ ative priming was diminished because stimulus letters repetition condition. There was also a significant main were drawn by removing segments from figure eights. It effect of SOA [F(6,174) = 2.47, p < .05]; the subjects might be easier to avoid identifying offset stimuli than responded slightly faster to probe displays following onset stimuli (Yantis & Jonides, 1984). color-first prime displays. 722 RUTHRUFF AND MILLER

Table 2 postidentification) selection mechanism-perhaps in­ Mean Reaction Time to Prime Displays in Experiment 2 volving inhibitory mechanisms-may be used to pre­ as a Function of SOA vent responses to distractors. SOA The present conclusion contradicts several previous re­ Color-First Identity-First ports that negative priming does not decrease with in­ 300 200 100 0 100 200 300 creasing ease ofselection (e.g., Allport et al. 1985; Driver Reaction time 762 770 778 836 799 789 799 & Tipper, 1989; Fox, 1994, 1995). Perhaps the starkest Error rate 1.4 1.4 2.6 2.2 2.9 2.4 3.2 empirical contradiction concerning negative priming is between the results ofthe present Experiment 2 and those of Fox (1994, 1995), who used a similar paradigm to Prime displays. Mean error rates and RTs to the study the effects of selection ease on negative priming. prime displays, measured from the moment both color She presented a precue at or near the location ofthe tar­ and identity information were available for processing, get letter 150msec before that letter appeared, or no precue are given in Table 2 as a function of SOA. The effect of at all, and measured both negative priming and interfer­ SOA on RT was significant overall [F(6,174) = 17.18, ence effects produced by distractors. Although interfer­ P < .001]. In addition, the simultaneous condition was ence was reduced by precues, negative priming was at reliably slower (p < .05) than each of the other condi­ least as large with them as without them, suggesting that tions, indicating that subjects are helped by a preview of distractor letters are still identified even when target lo­ either color or identity information (i.e., information cation is cued in advance. The effectiveness ofFox's pre­ available before the RT clock starts). Furthermore, for cues can be questioned, however, because (I) the main ef­ every SOA, the RT benefit ofcolor preview was reliably fect ofcuing on RT was small and not statistically reliable, larger than that ofan identity preview ofthe same dura­ and (2) cuing did not entirely eliminate distractor-based tion (p < .05), indicating that it is especially helpful to interference. Thus, it is doubtful that Fox's cues were as know the target and distractor locations in advance of effective as possible at encouraging selection, and a sim­ their identities. Because ofthe relatively small benefit of ple reconciliation with the present Experiment 2 is that identity preview, it seems unlikely that the subjects took our precues were more effective. The failures to observe advantage ofthe preview by fully processing letter iden­ decreased negative priming under conditions thought to tities (cf. Pashler, 1984). facilitate selection reported by Allport et al. (1985) and Driver and Tipper (1989) can also be criticized on the GENERAL DISCUSSION grounds that conditions were not truly optimal for effi­ cient selection oftarget stimuli. Moreover, Kramer, Hum­ The results of the present experiments indicate that phrey, Larish, Logan, and Strayer (1994) have also found negative priming is reduced in conditions that facilitate that negative priming is greatly reduced when target loca­ selection and, in fact, can disappear completely when tion is cued in advance. selection is especially easy. In Experiment 1,we found re­ The present results, however, do not contradict pre­ liable negative priming effects when both target and dis­ vious findings that distractors can produce negative prim­ tractor positions varied from trial to trial (the difficult­ ing even when they do not produce much interference selection condition) and when only the distractor posi­ (e.g., Allport et al., 1985; Driver & Tipper, 1989; Fox, tion varied from trial to trial (the easy-selection condi­ 1994, 1995). One possible explanation of this apparent tion), but there was no evidence of negative priming dissociation is that, under some conditions, distractors are when both target and distractor positions were fixed attenuated effectively enough to slow their rate ofidenti­ throughout a block of trials (the very-easy-selection fication relative to that oftarget stimuli. They would thus condition). The latter finding, in combination with the be identified too late to produce much interference on the absence of interference effects, suggests that distractor current trial, yet soon enough to produce negative prim­ stimuli were not identified on prime trials when selec­ ing on the next trial (Broadbent & Gathercole, 1990).This tion was sufficiently easy, as predicted by the hypothe­ explanation can be extended quite naturally to account for sis that selection can occur prior to identification. Ex­ the disappearance ofnegative priming in the very-easy­ periment 2 extended the results of Experiment I to a selection condition, by assuming that this condition al­ situation in which ease of selection varied only in the lows even greater distractor attenuation. prime display, eliminating alternative explanations in One might attempt to preserve the idea that selection which the absence of negative priming is attributed to occurs subsequent to identification by arguing that ir­ the ease ofselecting the target in the probe display. relevant stimuli are always identified, but, for some ex­ The lack ofnegative priming argues that irrelevant in­ traneous reason, they sometimes fail to cause interfer­ puts can sometimes be filtered out prior to stimulus ence and negative priming when selection is very easy. identification and therefore poses a problem for strict This possibility cannot be denied, but it does not seem late selection theories. It appears that distractors are very likely. If a distractor is fully identified, and the identified, resulting in interference, negative priming, or same letter reappears as the target 0.5 sec later, one both, only when filtering has not been completely suc­ would expect either especially slow responses (i.e., neg­ cessful. Naturally, when the filter does fail, a late (i.e., ative priming, as is typically found) or very fast responses EASE OF SELECTION 723

(facilitory priming) to the latter target. Specifically, neg­ tention. In M. Kubovy & 1.R. Pomerantz (Eds.), Perceptual organi­ ative priming would be expected ifthe initial distractor zation (pp. 181-211). Hillsdale, NJ: Erlbaum. representation were inhibited in some fashion (e.g., All­ KAHNEMAN, D., & ThEISMAN, A. M. (1984). Changing views ofatten­ tion and automaticity. In R. Parasuraman & D. R. Davies (Eds.), Va­ port et aI., 1985), and facilitory priming would be ex­ rieties ofattention (pp. 29-61). New York: Academic Press. pected ifit were not (e.g., Allport et aI., 1985, Experi­ KRAMER, A. E, HUMPHREY, D. G., LARISH, 1.E, LoGAN, G. D., & STRAYER, ments 3-4; Posner, 1978). Yet we observed essentially D. L. (1994). Aging and inhibition: Beyond a unitary view of in­ no priming ofeither type in the present experiments when hibitory processing in attention. Psychology & Aging, 9, 491-512. MEWHORT, D. J. K, JOHNS, E. E., & COBLE, S. (1991). Early and late selection was very easy. It is still logically possible that selection in partial report: Evidence from degraded displays. Per­ distractors were identified and that the resulting facili­ ception & Psychophysics, 50, 258-266. tation and inhibition happened to cancel each other out. MILLER, J. O. (1991). The flanker compatibility effect as a function of Rather than invoke such a coincidence to explain the re­ visual angle, attentional focus, visual transients, and perceptual sults, however, it seems more parsimonious at the pre­ load: A search for boundary conditions. Perception & Psycho­ physics, 49, 270-288. sent time to conclude that distractors were simply not MOORE, C. M. (1994). Negative priming depends on probe-trial con­ identified when selection was very easy. flict: Where has all the inhibition gone? Perception & Psycho­ Although the present results support claims that early physics, 56, 133-147. filtering of irrelevant inputs is possible, they also pro­ NEILL,W. T., TERRY, K M., & VALDES, L. A. (1994). Negative priming vide further evidence that the filter is not always per­ without probe selection. Psychonomic Bulletin & Review, 1, 119-121. PASHLER, H. E. (1984). Evidence against late selection: Stimulus qual­ fectly successful (e.g., in the easy-selection condition of ity effects in previewed displays. Journal ofExperimental Psychol­ Experiment 1). Thus, it remains a possibility that, much ogy: Human Perception & Performance, 10,429-448. ofthe time, attentional selection will occur after identi­ POSNER, M. I. (1978). Chronometric explorations ofmind. Hillsdale, fication-that is, "late." Even in this case, however, NJ: Erlbaum. SHIFFRIN, R. M., & GARDNER, G. T. (1972). Visual processing capac­ early filtering might play an important role by reducing ity and attentional control. Journal ofExperimental Psychology, the activations of irrelevant stimuli, thus making it eas­ 93,72-82. ier for late selection mechanisms to exclude them. SHIU,L.-P., & PASHLER, H. E. (1994). Negligible effects ofspatial pre­ cuing on identification ofsingle digits. Journal ofExperimental Psy­ REFERENCES chology: Human Perception & Performance, 20, 1037-1054. STROOP, J. R. (1935). Studies ofinterference in serial verbal reactions. ALLPORT, D. A., TIpPER, S. P., & CHMIEL, N. R. J. (1985). Perceptual Journal ofExperimental Psychology, 18, 643-662. integration and postcategorical filtering. In M. I. Posner & O. S. M. TIpPER, S. P. (1985). The negative priming effect: Inhibitory priming by Marin (Eds.), Attention and performance XI (pp. 107-132). Hills­ ignored objects. Quarterly Journal of Experimental Psychology, dale, NJ: Erlbaum. 37A,571-590. BROADBENT, D. E. (1958). Perception and communication. London: ThEiSMAN, A. M. (1960). Contextual cues in selective listening. Quar­ Pergamon Press. terly Journal ofExperimental Psychology, 12, 242-248. BROADBENT, D. E., & GATHERCOLE, S. E. (1990). The processing ofnon­ YANTIS, S., & JOHNSTON, J. C. (1990). On the locus ofvisual selection: target words: Semantic or not? Quarterly Journal ofExperimental Evidence from focussed attention tasks. Journal of Experimental Psychology, 42A, 3-37. Psychology: Human Perception & Performance, 16, 135-149. CORTEEN, R. S., & DUNN, D. (1974). Shock-associated words in a YANTIS, S., & JONIDES, J. (1984). Abrupt visual onsets and selective at­ nonattended message: A test for momentary awareness. Journal of tention: Evidence from visual search. Journal ofExperimental Psy­ Experimental Psychology, 102, 1143-1144. chology: Human Perception & Performance, 10, 601-621. CORTEEN, R. S., & WOOD, B. (1972). Autonomic responses to shock as­ sociated words in an unattended channel. Journal ofExperimental NOTES Psychology, 97, 308-313. DALRYMPLE-ALFORD, E. C., & BUDAYR, B. (1966). Examination of 1. Because subjects knew where the target and distractor letters some aspects of the Stroop colour-word test. Perceptual & Motor would appear, it would have been possible for them to fixate to the side Skills, 23,1211-1214. of the target letter away from the distractor, so that the image of the DEUTSCH, J. A., & DEUTSCH, D. (1963). Attention: Some theoretical distractor letter would fall farther into the periphery ofthe retina. Of considerations. Psychological Review, 70, 80-90. course, this would also place the target farther into the periphery, so it DRIVER, J., & TIpPER, S. P. (1989). On the nonselectivity of"selective" is not clear how much could be gained from such a strategy. In any seeing: Contrasts between interference and priming in selective at­ case, this was not possible in the other conditions, and so would rep­ tention. Journal ofExperimental Psychology: Human Perception & resent a possible confound. To prevent subjects from using this strat­ Performance, 15, 304-314. egy, they were given strict instructions to always maintain fixation at ERIKSEN, B. A., & ERIKSEN, C. W. (1974). Effects ofnoise letters upon the center ofthe screen, and no subject reported failures to do so dur­ the identification ofa target letter in a nonsearch task. Perception & ing the postexperiment debriefing. Psychophysics, 16, 143-149. 2. In all of the analyses reported in this article, p values have been Fox, E. (1994). Interference and negative priming from ignored adjusted using Greenhouse and Geisser's (1959) correction where ap­ distractors: The role of selection difficulty. Perception & Psy­ propriate. Comparable analyses have also been conducted using indi­ chophysics, 56, 565-574. vidual subject median RTs, and these yielded similar results. Fox, E. (1995). Precueing target location reduces interference but not 3. All confidence intervals reported in this article are 95% confi­ negative priming from visual distractors. Quarterly Journal ofEx­ dence intervals. perimental Psychology, 48A, 26-40. 4. On the other hand, it is quite possible that failures to observe neg­ FRANCOLINI, C. M., & EGETH, H. E. (1980). On the nonautomaticity of ative priming are equally frequent but tend not to be reported due in "automatic" activation: Evidence of selective seeing. Perception & part to a general bias against publishing nonsignificant results. Psychophysics, 27, 331-342. GREENHOUSE, S. W., & GEISSER, S. (1959). On methods in the analy­ sis ofprofile data. Psychometrika, 24, 95-112. (Manuscript received September 6, 1994; KAHNEMAN, D., & HENIK,A. (1981). Perceptual organization and at- revision accepted for publication December 12, 1994.)