Introduction

When a newly discovered phenomenon is named there is, initially at least, agreement about what phenomena the name refers to. Learned helplessness (Seligman, 1975; Maier and Seligman, 1976), for example, was proposed as a theoretical explanation for a psychological state induced through specific methods. Despite the possibility of some over- or under-generalization, there is usually good agreement about what our names for things refer to. Unfortunately, sometimes a name is at risk of losing its meaning because it is too often inconsistently applied or creatively (over-) extended. From the realm of cognitive psychology, this paper is about such a term: Inhibition of return (IOR).

We will begin with a description of the original and some subsequent use(s) of the term and then describe the results from a survey of experts’ understandings of the term. In Psychology surveys of experts have been used to help select an appropriate assessment tool (e.g., in forensic situations: Lally, 2003); to understand the meaning of an everyday term (e.g., wisdom: Jeste et al., 2010); and to determine the content validity of items in a behavioral instrument (e.g., sluggish cognitive tempo, Penny et al., 2009). In an effort to determine what leaders in the field of IOR research explicitly and implicitly think about the phenomenon, we conducted a survey of experts that targeted some of the ambiguities surrounding IOR. We believe our approach might serve as a model for investigators in other areas of psychological science (or other sciences) who are, like us, concerned about nomenclature.

Inhibition of return

Inhibition of return is often described as an effect wherein responses are slower to a target presented at a recently stimulated or inspected location compared to when the target is presented at a new location (Posner et al., 1985). IOR has been proposed to function as a novelty seeking mechanism (Posner and Cohen, 1984) and as a foraging facilitator (Itti & Koch, 2001; Klein and MacInnes, 1999); it has been likened to the gambler’s fallacy (Lyons et al., 2013); it has been observed in newborn human infants (Valenza et al., 1992) and in the archer fish (Gabay et al., 2013); its neural underpinnings have been explored using a wide variety of neuroimaging modalities, including ERPs (e.g., Prime and Ward, 2006), fMRI (e.g., Mayer et al., 2004), single unit recording (Dorris et al., 2002; Mirpour et al., 2009) and transcranial magnetic stimulation (e.g., van Koningsbruggen et al., 2010); and changes in its manifestation have been studied as a function of the administration of a wide variety of pharmaceuticals and the presence of a wide variety of neuropathologies.

The traditional paradigm (see Fig. 1) involves presenting a spatially uninformative cue at one of two peripheral locations followed by a target at one of the two peripheral locations (Posner and Cohen, 1984; see also, Berlucchi et al., 1981), and the effect (represented by the upward pointing arrows in Fig. 1) is typically observed when the time interval between the cue and target is longer than 250–500ms (Klein, 2000; see also Lupianez et al., 1997) though it can be seen earlier (Danziger & Kingstone, 1999). Berlucchi (2006; see also Klein and Taylor, 1994, p. 136) describes the canonical interpretation of the effect: attention is first drawn to the location of an uninformative stimulus, after a short period attention abandons that location and “develops a bias against returning to it” (pp. 1065).

Fig. 1
figure 1

Prototypical methods (A, B) and results (C) from a Posner cuing paradigm using uninformative peripheral cues. (A) Sequence of events begins with a display consisting of a central and two peripheral boxes. First a brief cue (brightening of one of the peripheral boxes) and then a target (stimulus inside one of the boxes) are presented with the interval between their onsets (cue-target onset asynchrony, or CTOA) varied. Observers are instructed to remain fixated on the central box and to make a simple, speeded, detection response when the target’s appearance is detected. Not illustrated, to discourage anticipatory responses no target is presented following some cues (catch trials). (B) The four possible sequences of cues and targets results in two types of trial: those with targets presented at the cued and uncued locations. Because the cues are uninformative these 4 possible sequences are equiprobable. (C) Typical pattern of results Presumed to reflect the capture of attention by the cue, detection response time is faster (represented by downward pointing arrows) for cued than uncued targets when the CTOA is short. In contrast at longer CTOAs the relation reverses and cued RT is slower than uncued RT (represented by upward pointing arrows). This is the prototypical IOR effect

Since its first descriptions, many papers have been published on IOR (circa January, 2014, “inhibition of return” as a topic, when submitted to the Web of Science, retrieved over 800 publications, with 260 published in the last 5 years). A variety of methods have been used in these studies to observe the phenomenon, some of them quite different from the traditional cueing paradigm used by Posner and Cohen (1984) in what is often regarded as the launching document for IOR research. With no intention of conveying a value judgment, but simply for the purpose of nomenclature, we will refer to studies of IOR that use the Posner cuing paradigm as “traditional” and all others as “non-traditional”.

Posner et al. (1985), after using the traditional paradigm to explore the inhibitory aftereffect in patients with brain damage to different orienting networks, pioneered methods that can be considered non-traditional. In one such experiment they measured the effect of an uninformative peripheral cue on two tasks in which targets were presented almost simultaneously at both the cued and uncued locations. When a non-speeded, manual temporal order judgement (Spence and Parise, 2010) was required there was no effect of the cue on the “which was first” judgment. However, when observers were instructed to simply move their eyes to the first perceived target, more movements were made away from than toward the cued location. From this they inferred that the effect of cue was to bias overt orienting (eye movements) away from the originally cued location. In another experiment from the same paper, a pair of digits was presented simultaneously to the left and right of fixation and observers were asked to make an eye movement in the direction of a centrally presented arrow toward one of these digits. After reporting the digit and returning gaze to the original fixation (in response to a cue presented there) observers were slower to signal the detection of targets at the previously fixated location. From this pattern of results they inferred that the inhibition was caused by activation of the oculomotor system and not by the asymmetric stimulation of the visual periphery that characterize cues in the traditional paradigm. It was in this paper that Posner and colleagues coined the term “inhibition of return”. Here are a few other examples of “non-traditional” studies of IOR.

Using a pair of visual search tasks modelled on those of Treisman and Gelade (1980), Klein (1988) measured reaction time to a dot-probe presented immediately after a visual search episode.The methods, hypothesized inhibitory tags, and results from this study are illustrated in Fig. 2. Observers were slower to respond to the dot-probe when it appeared at the same location as a distractor in the search array when the search task was difficult, but not when the search task was easy (the target “pops out” of the array). Because it is generally assumed that attention inspects distractors to determine if they are the target when search is difficult, Klein (1988) and others using this task (Müller & von Mühlenen, 2000, and Takeda & Yagi, 2000) attributed this pattern to IOR. Boot et al., (2008) found an IOR-like pattern involving saccadic responses to sequentially presented stimuli: participants were less likely, and slower, to fixate the target when it appeared at a previously fixated location. Welsh et al., (2005) found that participants were slower to respond to a target when it was presented at a location previously responded to by a partner participant and attributed this to “social IOR.”

Fig. 2
figure 2

Illustration of the methods (top and bottom panels), hypothesized inhibitory tags (middle panel), and results (averaged across 2 experiments and shown below the bottom panel) from the probe-following-search experiments of Klein (1988). The target (not shown) was a circle for the difficult search task and a circle with a line for the easy search task. Another pair of tasks, not shown, was used in which a circle with a gap replaced circles with lines. The strength (darker is stronger) of the hypothetical inhibitory tags in the middle panel is assumed to decay with time. The easy search task, for which the target pops out and hence the individual items need not be inspected, was used to provide a baseline. IOR was operationalized as the ON-probe cost (ON probe RT minus OFF probe RT) in the difficult search task minus the ON-probe cost in the easy search task. In Klein’s experiments plotting of the search array was terminated at the time of the search response, but it was subsequently demonstrated (for a review, see Wang and Klein, 2010) that when the array is removed so are the inhibitory tags. Hence, illustrated in the bottom panel is the method that works: probes are added to the search array

Many researchers who explore IOR using the Posner cueing paradigm (Lupianez et al., 2006; Wolf et al., 2009) incorporate non-traditional IOR ideas and papers into their introductions or discussions (e.g., Dodd & Pratt, 2007; Lupianez et al., 1997; Tipper, Weaver, Jerreat & Burak, 1994). The converse is equally true (e.g., Klein and MacInnes, 1999; Ogawa, Takeda & Yagi, 2002). These practices suggests that many researchers implicitly endorse the notion that traditional and non-traditional observations of IOR are observations of the same phenomenon. A possible consequence of such an assumption is that any effect labeled as IOR would be considered part of the IOR puzzle. In contrast, several researchers have suggested that the term IOR should be confined to those effects observed under very specific circumstances, or that there are different forms of IOR depending on the specific conditions under which the effects are generated. Berlucchi (2006), for example, took the position that IOR effects observed in traditional cue-target paradigms are different from the IOR-like effects observed in tasks involving saccades in static displays. Similarly, in trying to provide a habituation-based interpretation of the effect, Dukewich (2009) limited IOR to effects generated by repetitive stimulation using uninformative (non-predictive) cues.Footnote 1 Despite these examples, the zeitgeist in the IOR literature has been the lumping together, under the umbrella term “IOR,” all effects that seem to slow responding to previously inspected or stimulated locations. Consequently, non-spatial (Francis and Milliken, 2003; Mondor et al., 1998; Morgan and Tipper, 2007), multi-modal (Poliakoff et al., 2002), memorial (Johnson et al., 2013), motor/reaching (Cowper-Smith et al., 2013) and other forms of IOR have been proposed and explored.

Whether, when generated and measured under such a diverse set of conditions, all inhibitory aftereffects that have been called “IOR” are due to the same processes or mechanisms should be carefully examined. If these phenomena are not the same, then scientists trying to determine the mechanisms, components, consequences and neural underpinnings of “IOR” are setting themselves up to fail. Such an effort would be akin to asking what soup is made of: It depends on the kind of soup in question. It seems that implicit inconsistencies may be obstructing the creation of a coherent theoretical framework – no theory or framework will be satisfactory to the majority of IOR researchers if they all have different understandings of what IOR actually is.

Survey of IOR experts

The survey results do not tell us anything about IOR per se, only what experts in IOR think about the phenomenon. We asked researchers: 1) what they thought was meant by causes, effects, mechanisms and components of IOR; 2) which criteria they used for identifying IOR; 3) about their views on potential explanations for the effect; and 4) about their views on non-spatial forms of IOR, and IOR in visual search. Because we wanted to keep the focus of this observation on the implications of the survey results rather than highlighting the methods and procedures for data collection, our presentation of the survey methods will be brief (see Appendices A & B for a more detailed description of the methods). We have also made the data set available in the online Supplementary data.

We selected experts in the field of IOR research by searching for “inhibition of return” in the Web of Science database. Results were screened for those publications with the highest numbers of citations, and the authors of those publications were further screened for the number of IOR publications in the database. In the end, 63 researchers with at least 4 IOR publications in the database were contacted to complete the survey; 37 researchers accepted and completed the survey (during the period from July 18th, 2011 to August 31st 2011). A more detailed description of the selection process is described in Appendix A.

Criteria for identifying IOR

One of the first questions we asked our experts was what criteria researchers felt were necessary for an effect to warrant the label“IOR”. Table 1 presents the proportion [p(yes)] and number [n] of experts selectingeach criterion, as well as the similarity matrix scores (frequency of co-endorsements) for the different criteria. The most frequently endorsed criterion is slower responses to previously stimulated or inspected location, with 0.92 proportion of the surveyed experts selecting it. Endorsed by such a large proportion of experts one might be tempted to identify this as a defining feature. It is perhaps worth noting that the way we wrote this criterion it implies one effect (slowed responding) and two possible causes (stimulation and inspection).

Table 1 Responses to the question, “Which of the following criteria do you feel are necessary for an effect to be called IOR? Check as many as you think are necessary to warrant the label ‘IOR’?”

The remaining criteria were selected in varying degrees, but none approached even 0.5 selection among participants. A shift of attention (0.43), the presentation of a target (0.43) and non-predictive cues (0.32) were the next most selected criteria. Each similarity score in Table 1 represents the number of participants who selected a criterion listed on a row given they had selected the criterion for the corresponding column. Surveyed experts tended to cluster their selections around the slowed responding to a previously stimulated location, presentation of a target, shifts of attention, and the use of non-predictive cues. Importantly, experts who endorsed one of these three criteria were not particularly likely to also endorse either of the other two (there was no pair of criteria from this group for which the agreement of the endorsing experts exceeded 50 %).

The most striking aspect of the data from Table 1 is the inconsistency of the participants’ selections. For example, no criterion we listed was selection by 100 % of our participants, not even the criteria that is most often used to define IOR (slowed responding to a previously stimulated location). The variety of responses to the open-ended question, listed at the bottom of the table, suggests that the results are not simply because we failed to include some essential criterion when designing the survey. None of the criteria we listed were completely ignored by the participants. Based on these responses, it appears that we cannot eliminate any criteria from the list. The low rate of endorsement of fixating eyes on centre (0.08 participants) might be taken to suggest that most researchers believe that the IOR generated when the eyes are allowed (or required) to move is substantially the same as the IOR generated when eye movements are forbidden (but see Berlucchi, 2006; Hilchey, Klein & Satel, 2014; Taylor and Klein, 2000).

Two perspectives: Is there one IOR or are there many IORs?

Several of our questions probed whether respondents viewed IOR as one specific phenomenon or as a rubric that was used to refer to collection of different phenomena that share a surface similarity but might have different causes and effects and might be mediated by different underlying mechanisms.

Almost 80 % of the experts surveyed reported that they had read an article claiming to be about visuospatial IOR that they thought was not actually IOR (Table 2, #1). Fromthis high rate rate of “red herring” assertions we might infer that researchers implicitly view IOR as a particular combination (or combinations) of cause, effect and possibly mechanism. Were this the case generally, then we would expect a similar proportion of experts to reject the view of IOR as rubric for similar-looking effects. Explicitly probing this assumption yielded a split: as many as 43 % of the experts agreed that IOR is an umbrella term for similar-looking effects (Table 2, #6) and conversely 57 % rejected the idea that IOR was an umbrella term. The correlation of this response with responses to the “red-herrings” in the literature question (Table 2, #1) was low. Although one might infer from this pattern an internal inconsistency, mitigating against this inference is the likelihood that some endorsers of the “umbrella” idea have encountered assertions of IOR in the literature that they view as over-generalizations of the umbrella term as they construe it. We also thought it logical to assume that researchers who endorsed the umbrella term would not limit the IOR label to only those results that could be explained by a single theory (i.e. neural implementation would not limit nomenclature). However, almost half of the experts identified IOR as an umbrella term, and almost half indicated that phenomena that could not be explained by a comprehensive theory could still be called IOR (Table 2, #14), yet there is very little overlap among two these groups. Only about half of the experts felt that a biological or cognitive explanation of IOR would have to explain all of the IOR literature (Tables 2, #7 & #8).

Table 2 Probability of participants endorsing statements about IOR and the strength of relation (Phi Coefficients; bolded values are significant) between responses to statements

There is a third perspective. A researcher could believe that IOR is an effect with certain properties that are present in some IOR-like phenomena but absent in others. However, there is little consistency among the criteria experts used to identify IOR (Table 1), with the possible exception of slowed responding to a previously inspected location. Presumably, then, there would also be very little agreement regarding which characteristics differentiate ’real’ IOR effects from imposters. Hence, how one would decide which effects to include or exclude would be idiosyncratic and relatively specific to individual researchers.

Different exemplars of IOR?

Two possible exemplars of IOR would be its observation in a spatial cueing paradigm and its observation in a visual search paradigm. Spatial cueing IOR is typically generated by peripheral cues and measured by peripherally presented targets with participants instructed to maintain fixation centrally. Visual search IOR is generated, not by a cue, but by the shifts of attention and/or gaze direction that occur naturally in many visual search situations. Indeed, in most search tasks that have been used to elicit and measure IOR, eye movements are usually necessary both to complete the task and to observe the effect. At the very least, experts largely agree that IOR has more than one mechanism, and that these mechanisms are not always contributing equally to the observed effects (Table 2, #4 & #5). Given this agreement among experts, one task now is to determine which mechanisms contribute to which effects under different circumstances. Indeed, this is the direction in which IOR research is moving (e.g., Klein, 2004).

The survey revealed that some IOR researchers view these possible variants of IOR as distinct effects (Table 2, #11, #12 & #13). Almost 40 % of the experts surveyed endorsed this view. We were interested to discover that while 62 % of the experts reported that they believed these two exemplars of IOR probably shared the same mechanisms, and 65 % reported that they believed these two exemplars shared the same causes, there was considerable non-overlap between these groups as the correlation between these responses was only .47. This suggests that causes and mechanisms are not the same things in the minds of some IOR experts (see the ”Ambiguous terms” section below for more on how these terms are used by researchers). Given the relatively low correlations in Table 2, an expert could believe it probable that spatial cueing IOR and visual search IOR are the same effect(s), have different mechanisms, but share the same cause(s). (In fact, four researchers fit this profile.) A similar pattern was found in response to questions regarding spatial versus non-spatial forms of IOR. Experts surveyed were much more likely to think that non-spatial forms of IOR have the same causes than the same mechanisms (Table 2, #9 & #10). These apparent inconsistencies may be rooted in the ambiguous status of the terms.

Ambiguous terms

One of the most illuminating components of the survey (Question 2) involved having IOR experts classify concepts from the literature as a cause, mechanism, effect or component of IOR, selecting any and all categories that they thought applied to each concept irrespective of whether they supported a role for that concept in IOR. Experts also had the option of selecting none of these and I’m not familiar with that concept. Table 3 illustrates the probabilities of classification for each concept.

Table 3 Probabilities for classifying a concept as a cause, mechanism, effect or component of IOR

The goal of this question was to determine whether these terms (cause, mechanism, effect, component) are used consistently among researchers, as they are often used in the literature without the benefit of explicit definition.Footnote 2 A general trend was that concepts classified as a cause were likely to be classified as a mechanism, and vice versa. Another trend was that those concepts most likely to be classified as a cause or mechanism were the least likely to be called an effect, and vice versa (with inhibition being the sole exception to this generalization). These findings suggest that (a) researchers believe there is considerable overlap between the definitions of cause and mechanism but they are somewhat distinct, and (b) the definition of effects is very distinct from causes/mechanisms. All of the listed concepts were identified as a component with only moderate frequency (ranging from 14 % to 30 %), and no one concept was endorsed by a majority of experts as a component.

Based on these trends in classifying concepts, researchers tend to use the term mechanism to refer to anything that might slow responding to a previously stimulated or fixated location. Concepts such as sensory adaptation and inhibition tended to be classified as mechanisms. Researchers tend to use causes to refer to the environmental or procedural factors that are present and considered necessary to produce an IOR effect, like eye-movements or manual responses to a centrally-presented symbolic cue (Taylor and Klein, 1998) or repeated stimulation of a location (Dukewich, 2009). From among the concepts listed, sensitivity change, response bias and inhibition were the most frequently classified as effects.

The term component as applied to IOR is much more flexible. Each of the listed concepts was selected as a component with moderate frequency regardless whether the concept was primarily endorsed as a cause, mechanism, or effect. This suggests that when used in the context of IOR, component tends to refer to neurocognitive modules that implement either causes or effects. These neurocognitive modules have been identified with some consistency already: sensory/perceptual/input (c.f. Dukewich & Boehnke, 2008), motor/oculomotor/output (c.f. Klein and Taylor, 1994; Posner et al. 1985), and cognitive/attentional (c.f. Hunt & Kingstone, 2003; Reuter-Lorenz et al., 1996) (see also Berlucchi (2006) for a discussion of IOR components with these categories).

Conclusions

In a 2000 review of IOR Klein addressed the topic of nomenclature hopefully: “As our knowledge of the characteristics of IOR and its neural implementation grows, there will hopefully be increasing agreement on the use of the term” (Klein, 2000, p. 145). Our survey reveals that Klein was overly optimistic: 14 years later there remains a surprising variety of views regarding the causes and effects of IOR, the mechanisms that link them, and consequently the range of phenomena that have been given this label.

Greenwald (2012) recently published an articled entitled, ”There is nothing so theoretical as a good method”. Greenwald describes theoretical debates in cognitive and social psychology that never seem to resolve – with an average age of 44 years in the literature. He also notes a substantial bias in the allocation of Nobel Science Prizes to methodological rather than theoretical contributions to their fields. There are two things to consider based on Greenwald’s paper; (1) theoretical debates are difficult to resolve and (2) methods are easier to value than theory. Greenwald does not explicitly speculate as to why methods appear to be valued over theory. However, he does suggest that researchers involved in theoretical debates collaborate to identify commonly acceptable empirical findings that are germane to the debate. The purpose of collaboration would be to resolve debate boundaries and to help develop inter-translations between theories – differences in conceptual language among theories that, given similar empirical predictions, may not be semantically distinct. Herein may lie the reason methods are easier to advance than theory: methods are defined operationally and explicitly, while conceptual arguments that contribute to theory are often defined implicitly, or not at all.

We see the heterogeneous and idiosyncratic nature of researcher’s views and definitions of IOR as problematic. Our purpose here is not to advance a particular view of the causes, effects and underlying mechanisms for IOR. Rather we end with some recommendations that we believe will move the field away from covert semantic problems in the literature and provide a foundation for inter-translations that might help to resolve controversies around theory (Greenwald, 2012).

One way to minimize the ambiguities our survey has identified is for researchers to be as explicit as possible in defining the IOR that they are exploring. The challenge is to be explicit and precise about the context of any given experiment to avoid over-generalizing, both in terms of previous research and new results. For example, researchers ought to be cautious when generalizing data from spatial-cueing to visual search, and vice versa.

The variability among researchers in their views of IOR indicates that current theories and explanations of IOR are unlikely to explain everything in the IOR literature. This variability might be interpreted to imply that no such pressure really exists to explain everything, since there is so little agreement on what IOR is. Thus, for theoretical frameworks that are attempting to explain IOR or a variant of it, authors should be explicit about what phenomena are covered by their theory. Researchers attempting to validate different explanations or theories related to IOR simply need to define the parameters that limit their proposals. We believe that to the extent such a recommendation is heeded, increasing agreement on the usage of the term will follow.

A few researchers have already adopted such a careful approach. Consider, for example, Berlucchi’s (2006) commentary about the future direction of IOR research. Berlucchi was very specific about the kind IOR he was addressing at any given point; he very carefully noted when he was referring to IOR caused by peripheral spatial cues followed by peripheral spatial targets, and when he was referring to IOR caused by saccades in static displays. He specifically suggests that IOR generated in a traditional spatial cueing paradigm involving “changes in light energy” (pp.1071) has a different mechanism than IOR generated by saccadic shifts in static displays. He discusses some of the consequences of peripheral cueing besides IOR, including changes in sensitivity. Finally, his commentary discusses components in a consistent and coherent manner, suggesting that some IOR effects have sensory and attentional components, and some forms have a motor/oculomotor component. One may or may not agree with the content of Berlucchi’s (2006) IOR commentary, but its specificity, consistency, and careful avoidance of over-generalizing make it a good model for IOR researchers and for researchers in other fields.