[page 18↓]

3  Empirical Evidence for and against Direct Coding

As outlined above, compatibility effects are commonly attributed to a stage in information processing known as the ‘response selection stage’. More specifically, the assumption is that some sort of match between (features of) the response representations that are used to control responses on the one hand, and (features of) the stimuli (à stimulus-response compatibility), response effects (à response-effect compatibility), or responses on a simultaneously performed task (à response-response compatibility), on the other hand, leads to priming of the corresponding response, which is beneficial when the correct response is primed, but leads to response competition when this is not the case. According to this logic, investigating which match or compatibility relations contribute to compatibility effects under which instruction conditions allows conclusions about the cognitive codes that are used to control instructed responding.

Therefore, this chapter provides a review of findings regarding the impact of response instructions on several compatibility effects involving spatially organized keypress responses. The focus will be on two broad classes of compatibility effects that bear most directly on the experiments presented in Chapter 4 and Chapter 5.

The first class (Chapter 3.1) deals with a variety of SRC effects. The SRC effects reviewed include (i) the left-right prevalence effect observed with two-dimensional spatial S‑R mappings (Section 3.1.1), and (ii) spatial and non-spatial SRC effects observed when irrelevant stimulus attributes overlap with (features of) spatially organized responses, that is, variations of the Simon effect and the manual Stroop effect (see below, sections 3.1.2-3.1.4). Whereas instruction manipulations in the first two sections (i.e., Section 3.1.1-3.1.2) mainly involve an emphasis on different spatial aspects of the response array, the majority of findings reviewed in Sections 3.1.3 and 3.1.4 are concerned with non-spatial response instructions and response coding.

The second class of compatibility effects to be reviewed (Section 3.2) involves inter-task consistency effects obtained in dual task studies that require compatible viz. incompatible responses on two concurrently performed tasks.

The main question throughout this literature review will be whether variations of response instructions affect the type, size, or direction of observed compatibility effects. According to both versions of the direct coding hypothesis, it should be expected that variations [page 19↓]in response instructions affect response coding, and hence, how responses are accessed. Consequently, one would expect that compatibility effects can be observed with respect to the instructed dimension even when the overlapping stimulus- (or concurrent response-) attribute is task irrelevant. It will be argued that especially irrelevant non-spatial effects (i.e., symbolic Simon-type effects) provide convincing evidence for the direct coding hypothesis because they cannot (easily) be explained in terms of translation efficiency. The strong version of the direct coding hypothesis furthermore predicts that instructed codes are weighed more strongly than uninstructed codes. Accordingly, instruction manipulations (e.g., non-spatial vs. spatial response instructions) should lead to variations in the direction or size of a given spatial compatibility effect.

In contrast, the spatial coding hypothesis predicts instruction independent spatial compatibility effects (especially Simon-type effects) and a lack of irrelevant effects for instructed non-spatial response dimensions.

3.1 Response Instructions and Stimulus-Response Compatibility

As described in Chapter 2, stimulus-response compatibility refers to systematic variations in choice RT and error likelihood that depend on the relations between stimulus and response sets. Reaction times are shorter when there is correspondence between a stimulus attribute and features of the response (representation) than when there is not. In the following, I will review SRC phenomena obtained with spatially organized choice reactions (mostly bilateral keypress responses) that have been subject to instruction manipulations.

3.1.1 Right-Left Prevalence in Two-Dimensional Stimulus-Response Mappings

Right-left prevalence refers to the finding, first reported by Nicoletti and Umiltà (1984, 1985; Nicoletti, Umiltà, Tressoldi, & Marzi, 1988; for a summary, see Umiltà & Nicoletti, 1990), that under situations where stimuli and responses simultaneously overlap on the horizontal and the vertical dimension, compatibility on the horizontal dimension dominates vertical compatibility, regardless of instructions.

The prototypical task used to investigate this effect is the two-dimensional spatial mapping task. In this task, both horizontal and vertical compatibility are varied orthogonally such that a particular response is vertically and/or horizontally compatible or incompatible with the stimulus. For example, in one subtask, keys on the top-left and the bottom-right have to be pressed in response to stimuli on the top-left and bottom-right side respectively, yielding hori[page 20↓]zontal and vertical compatibility. In another subtask, the same keys have to be pressed in response to stimuli on the top-right and bottom-left respectively, yielding vertical compatibility and horizontal incompatibility (see Table 1 for examples of conditions in a paradigmatic experiment).

Instructions in this task have been varied by emphasizing either the vertical or the horizontal dimension. Vertical instructions refer to stimuli as well as responses only in terms of their vertical position (e.g., “press the upper key in response to the bottom stimulus”), without mentioning the horizontal dimension. Conversely, horizontal instructions only refer to the horizontal dimension (e.g., “press the left key in response to a left stimulus”).

Table 1: Sample stimuli and responses, and the resulting compatibility relations in four subtasks of the two-dimensional spatial mapping task.

Task/

Stimulus 1

Response 1

Stimulus 2

Response 2

Compatibility

Block

    

Vertical

Horizontal

1

Top left

Top left

Bottom right

Bottom right

+

+

2

Top right

Top left

Bottom left

Bottom right

+

-

3

Bottom left

Top left

Top right

Bottom right

-

+

4

Bottom right

Top left

Top left

Bottom right

-

-

The weak version of the direct coding hypothesis does not discriminate between implicit and explicit overlap. Accordingly, it predicts that responses are coded with respect to both dimensions, without the possibility to weigh vertical and horizontal response codes. Therefore, symmetrical horizontal and vertical effects, both on the instructed dimensions (i.e., vertical effects under vertical instructions and horizontal effects under horizontal instructions) and the “irrelevant” dimensions (i.e., vertical effects under horizontal instructions and horizontal effects under vertical instructions) should be expected, although the effects on the instructed dimensions may be larger than on the uninstructed dimensions. This is so because both the conditional and the direct route can be expected to contribute to the former, but only the direct route would be responsible for the latter effect.

In contrast, the strong version of the direct coding hypothesis predicts that response codes can be weighed and used as instructed. Consequently, symmetric vertical and horizontal instructed effects (i.e., vertical | vertical and horizontal | horizontal) should be observed. Uninstructed (irrelevant) effects (i.e., vertical | horizontal and horizontal | vertical) should be negligible for both dimensions because the response codes corresponding to the uninstructed dimension can be assumed to be less strongly weighed.


[page 21↓]

The predictions of the spatial coding hypothesis are more ambiguous. This is so because none of the dual route models (see Chapter 2.2) adhering to the spatial coding view explicitly distinguish between the vertical and the horizontal spatial dimension. Because both dimensions are spatial, the spatial coding hypothesis is therefore consistent with the view that both dimensions are used for coding. In this case, the predictions of the spatial coding hypothesis would be identical with those of the weak version of the direct coding hypothesis (see above).

On the other hand, in a strong interpretation of spatial coding as satisfying constraint setting, which holds that responses are coded in terms of features that allow to discriminate all possible responses in a given task context, one dimension is sufficient for response coding. In the interpretation of spatial coding favored here, only one dimension (e.g., the left/right dimension) is used for response coding. Generally, one would therefore expect compatibility effects primarily for the ‘default’ dimension (e.g., the horizontal dimension), independently of instructions. However, in this task there is overlap regarding both the instructed and the uninstructed dimensions. Therefore, one might expect translation-based compatibility effects for the other (e.g., vertical) dimension as well when this dimension is task relevant (e.g., under vertical instructions), but not when the default dimension is instructed (e.g., under horizontal instructions). In contrast, a compatibility effect for the default dimension (e.g., horizontal effects) should be observed under both horizontal and vertical instructions, although the latter might be somewhat smaller. Because only the conditional route leads to vertical effects, but both the direct and conditional route can be expected to contribute to horizontal effects, the horizontal effect should be generally larger than the vertical effect.

The majority of results obtained with this task generally seem to support the spatial coding hypothesis (in its one-dimensional interpretation). That is, most of the studies that included both, a vertical and a horizontal instruction condition, and that required bimanual responses found what Vu, Proctor, and Pick (2000) termed ‘weak horizontal prevalence’ (e.g., Hommel, 1996a, Exp. 1A; Vu & Proctor, 2001, Exp. 2). That is, a substantial vertical compatibility effect shows up under vertical instructions, although the vertical effect under vertical instructions is usually smaller than the horizontal compatibility effect under horizontal instructions. Moreover, horizontal effects tend to be obtained under vertical instructions, whereas most often no vertical effect is observed under horizontal instructions. Therefore, the overall horizontal effect is typically larger than the vertical effect. The horizontal effect is usually also modified by instructions such that more pronounced horizontal effects can be [page 22↓]observed under horizontal than under vertical instructions, possibly indicating the additional contribution of the conditional route under horizontal instructions.

These findings suggest that instructions modify behavior, but that they are less effective in influencing (i.e., overriding) left-right coding than top-bottom coding, at least when stimuli and/or bilateral responses can easily be discriminated on the left-right dimension (see Vu & Proctor, 2001, 2002, for manipulations of the salience, that is, discriminability, of the two dimensions). They contradict both versions of the direct coding hypothesis according to which symmetric effects would have been expected.

However, Hommel (1996a) noted that horizontal prevalence effects are open for alternative explanations, namely (a) recoding of instructions, and (b) logical recoding.

The first of these explanations is based on the observation that, in the two-dimensional mapping task, the correct response is always redundantly signaled by both spatial stimulus attributes. That is, within a given subtask, the correlation between each stimulus and response dimension is either +1 or -1 (see Table 1). Hence, subjects may have ignored vertical instructions in some cases, basing their responses on horizontal stimulus codes instead. This explanation seems reasonable given that some studies by Nicoletti and colleagues (e.g., Nicoletti & Umiltà, 1984, Exp. 4; 1985, Exp. 1; also see Vu & Proctor, 2001, Exp. 3, normal instructions) reported (a) larger horizontal than vertical effects under vertical instructions (in fact, vertical effects were small to nonexistent), and (b) overall RTs under vertical instructions that closely resembled RTs under a purely (one-dimensional) horizontal mapping rather than those obtained for a purely vertical mapping. Furthermore, when Vu et al. (2000, Exp. 4) compared normal vertical instructions with exclusion instructions that encouraged subjects to exclusively rely on the vertical dimension, the horizontal | vertical effect of 57 ms under normal vertical instructions was reduced to 8 ms under exclusion instructions, whereas the vertical | vertical effects were 19 ms and 79 ms under normal and exclusion instructions, respectively (no horizontal instruction conditions were included). This finding presumably implies that participants were more likely to use vertical codes under vertical exclusion instructions (but see Hommel, 1996a, Exp. 1A, for slightly different findings; Vu et al., 2000, for an alternative interpretation).

Further evidence for participants’ use of some sort of recoding strategy – albeit a somewhat different one – stems from underadditive interactions between horizontal and vertical compatibility sometimes observed under vertical instructions (e.g., Hommel, 1996a, [page 23↓]Exp. 1A; Vu et al., 2000, Exp. 1A). That is, with vertical incompatibility, horizontal incompatibility often leads to faster responses than horizontal compatibility, indicating that the benefit of spatial S‑R compatibility decreases, and can even be inverted when there is incompatibility on the other dimension.

As Hommel (1996a) noted, this finding is similar to an interaction of spatial and non-spatial dimensions first observed in a study by Hedge and Marsh (1975; see Chapter 3.1.4 below). Hedge and Marsh’s participants always responded to colored (red vs. green) stimuli that randomly appeared on the left or the right side of the screen by pressing lateralized response keys that were labeled by colors (i.e., red vs. green). In the direct mapping condition, participants were instructed to press the key of the corresponding color (e.g., the red key in response to a red stimulus), whereas the reversed mapping required opposite color responses (e.g., green key to red stimulus). Although the spatial dimension was irrelevant, a spatial effect was observed for both mappings. However, whereas responses were faster when stimulus position and response position overlapped (i.e., in the spatially compatible condition) with the direct mapping, under the reversed mapping spatially compatible responses were slower than incompatible responses.

One influential account for this crossover effect is the logical recoding account (e.g., De Jong et al., 1994; Hedge & Marsh, 1975). The basic tenet of this account (in one of its interpretations) is that participants recode the relevant and – inadvertently – also the irrelevant (spatial) stimulus attribute in order to form the response code. Under the direct mapping, so the assumption, some sort of “same” operation is applied to both stimulus dimensions, leading to faster responses when both color and position correspond with the required response (e.g., applying a “same” operation to a red, left stimulus leads to faster responses when the red key is located on the left). In contrast, under the reversed mapping condition a “respond opposite” rule is formed that transforms the values on both stimulus dimensions into their opposite. Consequently, a left stimulus attribute now primes a right response, leading to faster responses in the spatially incompatible than the compatible condition.

Similarly, under vertical instructions, a ‘same’ transformation may be applied in vertically compatible subtasks that leave the horizontal stimulus attributes unchanged. With vertical incompatibility however, participants may apply a ‘reversal’ transformation to both vertical and horizontal stimulus position, leading to slower responses on horizontally compatible than incompatible subtasks.


[page 24↓]

Thus, it seems possible that the right-left prevalence effect is at least partially due to a strategic bias that either leads people to ignore instructions (i.e., use of the left-right stimulus and/or response dimension only), or to logical recoding of stimuli under vertical instructions. Moreover, because the relevant stimulus attribute in these studies is always position, and because the irrelevant stimulus attribute is correlated with response location, it is impossible to determine whether this bias primarily refers to stimulus or response coding, or to some sort of interaction of strategically biased stimulus and response processing. Evidence against a response coding bias has been provided by studies that observed vertical compatibility effects of typical size (i.e., comparable to horizontal effects) in one-dimensional vertical mappings, both when stimulus position was task-relevant (e.g., Vu et al., 2000, Exp. 1B) and, more importantly, when stimulus position was task-irrelevant (i.e., in vertical Simon tasks, e.g., Stürmer, Leuthold, Soetens, Schröter, & Sommer, 2002, Exp. 1), suggesting that, in principle, left and right responses (i.e., effectors) can be coded as up and down.

Similarly, Proctor, Vu, and Nicoletti (2002) recently reported symmetric horizontal and vertical compatibility effects when both vertical and horizontal stimulus position were task irrelevant in a two-dimensional version of the Hedge and Marsh task. In the Proctor et al. (2002) study, colored stimuli that randomly appeared at the upper left and lower right or lower left and upper right side of the screen were assigned to colored response keys of the same color (direct mapping condition) that were again defined - but not instructed - by the two spatial dimensions. This finding again questions an interpretation of horizontal prevalence in the two-dimensional mapping task in terms of obligatory left-right coding of responses.

Taken together, the (weak) horizontal prevalence effect obtained in most studies using the two-dimensional mapping task seems generally more consistent with the spatial coding hypothesis in its one-dimensional interpretation than with both versions of the direct coding hypothesis. As would be expected if both the direct and the conditional route contributed to horizontal effects, but only the conditional route were responsible for vertical effects, horizontal effects under vertical instructions tend to be larger than vertical effects under vertical instructions, and horizontal effects typically occur even under vertical instructions, whereas no irrelevant vertical effects show up.

However, the sometimes observed interaction between vertical and horizontal compatibility as well as the fact that both stimulus dimensions signal the correct response in this task, [page 25↓]suggest that the prevalence effect might at least in part be due to a strategic bias to recode vertical instructions (i.e., mappings). The Proctor et al. (2002) finding of symmetric horizontal and vertical Simon effects when both dimensions are task irrelevant (i.e., when mappings are instructed in terms of color) furthermore questions an interpretation of horizontal prevalence as indicating a bias in response coding. More likely, such a bias refers to stimulus coding or some sort of interaction of (strategic) stimulus and response processing.

Therefore, a stronger test of an underlying response coding bias would be provided by experiments in which a non-spatial stimulus attribute (e.g., color) were task relevant and top-bottom as well as right-left spatial stimulus positions varied randomly. If the horizontal compatibility effect under horizontal response instructions (e.g., press the left vs. right key) and across instruction conditions would still be larger than the vertical compatibility effect under both vertical response instructions (e.g., press the top vs. bottom key) and overall, then this would provide strong evidence for obligatory left-right (spatial) coding. In contrast, if the effects were symmetric and modulated by instruction, then this would provide evidence for the strong version of the direct coding hypothesis. Finally, according to the weak direct coding hypothesis, symmetric vertical and horizontal Simon effects should be observed in such an experiment, regardless of response instructions. To my knowledge, such a study does not yet exist, though.

In the next section, I will review a recent debate on the impact of response instructions on a variant of the Simon task that requires responding with hands in the standard position as compared to a condition with crossed hands. As in the two-dimensional mapping task, different instructions emphasize different spatial aspects of the response ‘array’. However, in the Simon task, stimulus position is uncorrelated with response location, rendering stimulus position truly task irrelevant.

3.1.2 Stimulus-Hand Correspondence and the Simon Effect

In Simon-type tasks non-spatial stimulus attributes such as pitch of tone or the shape of a visual stimulus are mapped to spatially organized responses. Stimulus position varies randomly from trial to trial, and hence is irrelevant for selecting the appropriate responses (usually bilateral keypresses). Nevertheless, the usual finding – known as the Simon effect – is that responding is faster when stimulus and response position correspond. Typically, left and right keys are pressed with left and right effectors, respectively. Thus, effector and key position (and effector position) correspond, implying that this task cannot discriminate between [page 26↓]location (i.e., spatial) and effector (i.e., anatomical) coding. This confounding can be avoided by requiring participants to respond with crossed hands, such that the left key is operated by the right hand, and the right key is operated by the left hand (see Table 2 for stimulus-response compatibility conditions as a function of hand position). Thus, comparing normal and crossed hands conditions allows conclusions about whether responses are (primarily) coded spatially or anatomically.

Table 2: Stimulus-response compatibility conditions realized by the tasks with uncrossed vs. crossed hands.

Mapping condition

Stimulus-response position compatibility

 

compatible

incompatible

uncrossed

+/+

-/-

crossed

+/-

-/+

Note. + and - indicate compatibility and incompatibility between stimulus position and response position / response hand, respectively.

Deriving specific predictions from the hypotheses and models outlined in Chapter 2 is difficult because even coding accounts of the first generation (e.g., Wallace, 1971), the more so the dual route models reviewed in Chapter 2, are based on the results typically obtained with the crossed-hands task.

That is, a location based Simon effect is usually observed even when hands are crossed. More specifically, the usual finding with Simon tasks comparing standard vs. crossed hands responses (e.g., Roswarski & Proctor, 2000, 2003a; Simon, Hinrichs, & Craft, 1970; Wallace, 1971) is that responses are fastest for the position-compatible uncrossed condition (see +/+ condition in Table 2) in which correspondence exists for all three compatibility relations (i.e., stimulus-key, stimulus-hand, and hand-key), and slowest for the position-incompatible crossed condition (see -/+ condition in Table 2) in which neither stimulus and key position nor key position and hand (but stimulus position and hand) correspond, leading to overall faster responses with uncrossed as compared to crossed hands. More importantly, location-based Simon effects of comparable size usually show under both uncrossed (i.e., +/+ vs. -/- comparison) and crossed (i.e., +/- vs. -/+ comparison) hands conditions. Grouped according to stimulus-hand compatibility and stimulus-response location compatibility (without including hand condition as a factor), the same pattern of results can also be described as showing a main effect of position compatibility (i.e., +/+ and +/- vs. -/+ and -/-) that is qualified by an interaction of position compatibility and stimulus-hand compatibility. Usually, no overall stimulus-hand compatibility main effect is observed.


[page 27↓]

This pattern of finding led most researchers to propose that responses are primarily spatially coded, without denying some (minor) impact (on the size of the effect, but not on its direction) of anatomical coding. More specifically, most coding accounts seem to adhere to some sort of hierarchical response coding view such as the one proposed by Heister et al. (1990). According to Heister et al., responses are first and foremost coded and accessed by spatial (key) location codes. Whenever the spatial dimension allows discrimination of the responses, effector coding is assumed either (a) to simply contribute to a lesser degree (e.g., Hommel, 1993a; Worringham & Kerr, 2000), or (b) only to be included in a later stage of response selection or execution (i.e., when the response location codes are ‘translated’ into specific motor programs, leading to some sort of response-response (R-R) (in)compatibility; cf. Roswarski & Proctor, 2003a).

Recently however, Wascher et al. (2001) noted that position coding of responses is usually suggested by instructions and task demands in general. Specifically, standard instructions emphasize key positions by instructing subjects to press a particular (left or right) key in response to specific stimuli. Moreover, the stimulus-key mapping usually remains constant when hand condition is varied within subjects, whereas the stimulus-hand relations change. As would be assumed by the strong version of the direct coding hypothesis according to which response codes can be weighed as instructed, Wascher et al. argued that finger instructions (e.g., “respond with the left finger to the letter A”), combined with constant stimulus-to-finger mapping across hand conditions (implying varied stimulus-to-key mapping), should encourage subjects to more strongly rely on anatomical finger coding. Hence, a reduced or even reversed position-based effect should be observed in the crossed-hands condition under hand instructions. In contrast, such an outcome would not be expected by the spatial coding hypothesis, according to which the spatial (position-based) Simon effect under crossed-hands conditions should be largely unaffected by instructions.

Recent studies by Wascher et al. (2001) as well as Roswarski and Proctor (2003a) that compared hand and key instructions for crossed and uncrossed hands conditions speak to this issue. When using visual stimuli, Wascher et al. (Exp. 1) found no interaction between instructions, hand condition, and position compatibility. That is, the location-based Simon effect for crossed hands were 32 and 24 ms under key and hand instructions, respectively, and no overall hand-based Simon effect showed (i.e., conditions +/+ and -/+ as compared to conditions -/- and +/-, see Table 2). However, with auditory stimuli (Exp. 2), the expected out[page 28↓]come was obtained. That is, the position-based Simon effect in the crossed-hands condition was extremely reduced (3 ms) under finger instructions as compared to key instructions (36 ms), and an overall hand-based Simon effect of 19 ms was obtained. This finding seems at least partially consistent with the intentional weighing hypothesis in that finger instructions in the Wascher et al. experiment were able to reduce (but not reverse) the impact of position compatibility for auditory stimuli.

However, Roswarski and Proctor (2003a) were not able to replicate the Wascher et al. results. More specifically, they did replicate the lack of instructional influence with visual stimuli (Exp. 2) also observed by Wascher and colleagues (Exp. 1), but did not find a reduction of the position-based Simon effect for the crossed-hand condition under finger instructions with auditory stimuli (at least not in RTs, see their Exp. 3). Only when they substantially increased the number of trials and included session (practice) as a factor in the analysis of the results of their Experiment 4 (using finger instructions only), they observed a reduction of the position-based Simon effect with crossed hands in the second as compared to the first session (27 ms vs. 49 ms, respectively), whereas the Simon effect remained constant across sessions for the uncrossed hand condition. Roswarski and Proctor (2003a) concluded that spatial (location) coding is the default regardless of instruction, but that extensive practice in their Experiment 4 as well as in the Wascher et al. study, combined with finger instructions, may have encouraged stronger reliance on finger coding after practice. It remains unclear though, why this ‘practice effect’ was restricted to the finger instruction group, and why it did not occur with visual stimuli in the Wascher et al. (2001) study.

In sum, the results on finger vs. key instruction manipulations on the Simon effect with crossed hands seem to be more compatible with the spatial coding hypothesis than with the strong version of the direct coding (i.e., intentional weighing) hypothesis. First, with relatively limited practice (as in common behavioral experiments), the Simon-effect tends to be limited to stimulus-response location correspondence, and is little affected by type of response instructions and hand position. Second, for visual stimuli, even extended practice did not modify the location-based Simon effect for the finger-instruction group in the Wascher et al. study. Thus, subjects seem to prefer location-based (left-right) response coding, regardless of how responses are referred to in the response instructions.

As noted before, it remains unclear, why finger instructions modify the auditory Simon effect, and why such an ‘instruction effect’ only shows after extended practice. Moreover, the [page 29↓]practice effect points to a more general problem inherent in all studies on the impact of instructions discussed in Chapter 3. This problem concerns the confounding of possible effects of instruction on (initial) response coding and changes in coding with practice that may or may not depend on instructions. In Chapter 5, I will present experiments that try to avoid this kind of confound.

While the findings presented in this section seem to favor the spatial (location) coding view when different spatial attributes of responses are emphasized by instruction, in the next section I will present evidence supporting arbitrary (non-spatial) response coding, both instructed and uninstructed.

3.1.3 Anticipated Action Effects

The evidence reviewed in this section is based on findings of so-called response-effect compatibility. ‘Response effect compatibility’ refers to the observation that stimuli presented after responding can come to influence responding. For example, Kunde (2001) had subjects respond to the color of centrally presented stimuli by pressing horizontally arranged keys. Pressing a key “produced” visual effects at different horizontal positions. In some conditions, the spatial relation between the manual response (i.e., the location of the finger and the key) and the visual effect was compatible, that is, the relative spatial locations of the responses and the response effects corresponded. In other conditions, however, responses and effects were incompatible. Kunde found slowed responses in the incompatible as compared to the compatible condition, even though the visual effect was presented after responding and was irrelevant for the task, indicating that anticipated action effects primed the (non-) corresponding response.

The question of interest here is whether response effect compatibility extends to more arbitrary (non-spatial) response-effects. If so, this would provide evidence for non-spatial response coding (see below).

Studies addressing this issue typically employ a two-step procedure including an acquisition phase and a test phase. In the acquisition phase, choice-responses (e.g., left and right keypresses) to arbitrary stimulus attributes are paired with some novel, arbitrary response-effect stimuli (e.g., color patches, tones of a certain pitch) that consistently succeed the responses (e.g., a left response is always followed by a red color patch, whereas a right response is always followed by a green color patch). In the subsequent test-phase the same responses (e.g., left and right keypresses) are required to imperative stimuli that often differ from those [page 30↓]used in the acquisition phase. Importantly, stimuli resembling those that (formerly) served as response-effects are now used as primes (distractors) that are presented together with the imperative stimuli. The primes either correspond or do not correspond with the previously acquired effect associated with that response. If response effects were integrated into the response representation and used to access responses in the test phase, the primes should speed up or slow down responding. This is what ought to be expected according to both versions of the direct coding hypothesis, which assume that responses can be coded in terms of non-spatial features, in addition to spatial features. Moreover, the intentional weighing hypothesis predicts that the acquired action effect codes can be weighed more strongly than spatial response codes if responses are instructed in terms of their effects. Therefore, stimulus-response-effect compatibility can even be expected to override spatial correspondence effects.

In contrast, if responses are spatially coded regardless of response-effects and instructions, the arbitrary prime-distractors that correspond to the response effects should not affect responding.

Results obtained in studies on arbitrary (acquired) response-effect compatibility seem generally more consistent with the direct coding hypothesis in that they demonstrate arbitrary response priming. For example, Beckers, De Houwer, and Eelen (2002) demonstrated action-effect learning regarding affective evaluation of electrocutaneous feedback. In a training phase, participants moved a response key up or down in response to a go-signal. One of the responses was consistently followed by aversive electrocutaneous stimulation. In the test phase, word stimuli with a positive or negative connotation had to be classified according to their grammatical category (noun or adjective) by using the same responses (and response effects) as in the practice phase. Beckers et al. found a substantial valence-based compatibility effect (also termed ‘affective Simon effect’, cf. De Houwer & Eelen, 1998) even when the category-to-response mapping was regularly switched during the test phase, indicating that response valence had become integrated into the action representation and was used to control responding.

Similarly, Hommel (submitted) showed that subjects can access manual left and right responses via color codes. In his Experiment 1, participants first had to press a left or right key in response to letter identity of letters presented in a gray frame. Upon responding, the letter stimuli turned either green or red, depending on which key had been pressed. In a later test phase, red or green frames surrounding the letter stimuli were used as distractor stimuli. [page 31↓]Hommel found that responses were faster when the color of the distractor frame and the (learned) action effects (letters turning green or red) corresponded than when they differed, implying that the color effects had become integrated into the action representation and were used to access and guide manual responses. Similar effects have also been observed with other arbitrary effect-stimulus features such as pitch of tone (Elsner & Hommel, 2001; Hommel, 1996b), letter identity (Ziessler & Nattkemper, 2002), or the semantic category membership of words (Hommel, Alonso, & Fuentes, in press).

Whereas the findings reported thus far demonstrate that irrelevant action effects come to affect behavior, Experiments 2 and 3 by Hommel (submitted) suggest that the actual use of color codes may at least in part depend on their usefulness in a particular task, and hence on intentional weighing according to task demands. Hommel combined the manual color Stroop task (i.e., requiring left and right keypresses to the color of neutral words, or of color words written in a congruent or incongruent color) with color-related action effects (i.e., color frames or color words presented upon responding). He found a reduction of the manual Stroop effect (i.e., differences in performance as a function of word-color congruency) in RTs (but not in errors) for the group trained with a compatible color-effects mapping as compared to a control group without any response effects, but no statistically reliable difference between the control group and the incompatible color-effect group. Moreover, he observed a reduced (nonsignificant) impact of color-word action effects on the Stroop effect, leading him to conclude that the use of color-related action effects in response coding is under strategic control.

Taken together, these findings suggest that, in principle, abstract, non-spatial arbitrary codes can be used to control responding, implying considerable flexibility in response coding. This is consistent with the direct coding hypothesis by suggesting that other than spatial codes can be used to code and access responses. However, with the possible exception of Hommel (submitted) who provided evidence that the actual use of such effect codes may at least partially depend on the usefulness in a particular task, and hence on intentional weighing according to task demands, none of the studies reported so far speaks to the issue of whether task demands affect response coding. More specifically, none of these studies provides evidence for a direct influence of response instructions on response coding. Instead, they demonstrate non-instructed response coding, that is, the use of codes/dimensions that may have been primed through extensive practice and/or which may have proven useful for the task at hand.


[page 32↓]

The only study I am aware of that directly manipulated response instructions with respect to action effects of laterally organized manual responses has been conducted by Hommel (1993a; but also see Wang, Proctor, & Pick, 2002, for comparable findings with wheel rotation responses in a replication and extension of Guiard, 1983). Hommel (1993a) had participants react to the pitch of tones that were randomly presented to the left or right ear by pressing either a left or a right hand key. Pressing a key flashed on a light on the opposite side of the response key (so for example a left keypress would switch on a right light and vice versa). One group of participants was instructed to press the left vs. right key in response to tone pitch. This group produced a response-location-based Simon effect, that is, responses were faster when the location of the key corresponded to the side on which the tone was presented than when stimulus and response location did not correspond.

A second group of participants was instructed in terms of the light to be switched on. For example, they were told to respond to a low tone by switching on the light on the left. Under the latter task instruction the Simon effect reversed. That is, responses were now faster when tone location and the location of the light (that had to be switched on by the contralateral hand) corresponded. This finding implies that the response descriptions given in task instructions determined response coding, such that left responses were coded as right and vice versa.

It should be noted however that the reversed Simon effect under light instructions (-30 ms) was numerically smaller than the Simon effect under key instructions (52 ms). Similarly, the Simon effect under key instructions with incompatible key-to-light mapping was numerically smaller than the Simon effect observed for a control group (77 ms) that worked under key instructions with a compatible key-to-light mapping (i.e., for them, a left keypress switched on a left light and vice versa). These variations in the size of the effects are qualitatively similar to those observed in crossed-hands studies (cf. Chapter 3.1.2, above), presumably implying that the instructed codes were weighed more strongly than uninstructed codes, but that the latter may still have been part of the action representation, and hence contributed to the size of the (reversed) Simon effect by adding and/or subtracting from the overall effect.

To summarize, research on response-effect compatibility supports two important conclusions regarding the question of how responses are or can be coded. First, consistent with the direct coding view, studies investigating the acquisition of arbitrary irrelevant action effects show that response effects such as color become integrated into action representations [page 33↓]and are or can be used to code and access responses in subsequent task performance. Second, the Hommel (1993a) results strongly suggest that instructions determine how responses are coded when salient coding alternatives are introduced by presenting clearly visible action feedback. When participants were instructed in terms of the locations of the lights turned on by a contralateral keypress, they relied more heavily on light location than key location (and anatomical hand) codes, whereas the key instruction group weighed response-location codes more strongly than light location codes.

This latter finding thus provides strong evidence in favor of the strong version of the direct coding hypothesis, which assumes that response codes can be weighed according to instructions. However, instructions in the Hommel study (as well as in the Wang et al. study) again emphasized different spatial aspects of the response array. Hence, one could argue that response instructions might have led to attentional shifts between different spatial aspects of the spatial dimension, thus affecting the hierarchy of spatial coding, but not the “prevalence” of spatial coding per se. More specifically, introducing salient lateralized action effects may have added another spatial dimension to the spatial coding hierarchy proposed by Heister et al. (1990), with key-location and effect-location being almost equally strong coding alternatives.

In the next section, I will review findings obtained with non-spatial response instructions. The majority of the studies used color instructions of responses, and originates from work on the Simon and the (manual) Stroop effect.

3.1.4 Non-Spatial Response Instructions

Compatibility effects with non-spatially instructed keypress responses have been studied with several paradigms, the majority of which instructed the buttons in terms of colors.

What would generally be expected according to the direct coding hypothesis is that compatibility effects should be observed with respect to the instructed dimension even when the overlapping stimulus attribute is task irrelevant. The strong version of the direct coding hypothesis furthermore predicts that instructed codes are weighed more strongly than uninstructed codes. Accordingly, non-spatial response instructions should reduce a given spatial compatibility effect.

In contrast, the spatial coding hypothesis predicts instruction independent spatial compatibility effects and a lack of irrelevant effects for instructed (non-spatial) response dimensions.


[page 34↓]

The paradigms involving non-spatial (color) instructions of keypress responses to be reviewed in this section include (a) the manual Stroop task, (b) the Hedge and Marsh task, and (c) color-compatibility effects with stimulus color as the irrelevant stimulus dimension. The results obtained with these tasks will be discussed in turn.

The Manual Stroop Task

The overwhelming majority of studies investigating the Stroop effect (e.g., Stroop, 1935) used verbal responses. In a typical verbal Stroop task, participants are required to name the color of incongruent color words (e.g., the word GREEN printed in red), the color of a control string (e.g., XXXX, or the word ‘TABLE’ printed in red), or the color of congruent color words (e.g., the word RED in red ink). The common finding with this task, the Stroop effect, is that it is much harder to name the color when the color is accompanied by an incongruent color word (e.g., GREEN printed in red) than to name the color of a colored control string (e.g., XXXX printed in red) or the color of a color-congruent word (e.g., RED printed in red).

This finding (for a comprehensive review, see MacLeod, 1991) has been interpreted as indicating response competition resulting from ‘direct’ response activation from the irrelevant stimulus (i.e., the color word). More specifically, some researchers (e.g., Cohen et al., 1990; Virzi & Egeth, 1985) have suggested that the irrelevant stimulus attribute (i.e., a color name) leads to response competition on incongruent trials, whereas it primes the correct response in the congruent condition.

Interestingly, the Stroop effect is extremely reduced or even nonexistent when the task is changed to word reading (i.e., when the relevant attribute is the color word and the irrelevant attribute the color of the print), indicating that there is some difference in type (e.g., Barber & O’Leary, 1997; Phaf, Van der Heijden, & Hudson, 1990; Virzi & Egeth, 1985) and/or strength (e.g., Cohen et al., 1990; Lu, 1997) of the pathways used to perform the word reading viz. color naming task (see below). That is, word reading is easier than color naming and less affected by irrelevant colors than vice versa either because of unequal (lifelong) practice, and/or because words and naming responses are of a similar ‘format’ and hence processed in the same system or module.

In the manual version of the Stroop task, verbal responses are replaced with keypress responses. Responses are instructed in terms of color, and the keys are either labeled with color patches or with color-words printed in black, allowing the comparison of the subtasks [page 35↓]presented in Table 3, which result as a function of the relevant stimulus dimension (i.e., responding to color / ignoring color-names vs. responding to words / ignoring colors) and the type of the labels on the buttons (i.e., color patches vs. word-labels).

Table 3: Subtasks realized by different combinations of relevant (S_r) / irrelevant (S_i) stimulus attributes and key-label format in the manual Stroop task.

S_r

S_i

Label type

Response type

Translation

Color

Word

Color word

Translated word response

+

Color

Word

Color patch

Untranslated color response

-

Word

Color

Color patch

Translated color response

+

Word

Color

Color word

Untranslated word response

-

Note: ‘Translation’ refers to differences in format between relevant stimulus dimension and response label (terminology adopted from Sugg & McDonald, 1994).

To the extent that Stroop interference (only) measures response competition (e.g., Cohen et al., 1990) the spatial coding hypothesis predicts no manual Stroop effect. This is so because neither color words nor color patches should directly activate spatially coded responses. According to the direct coding hypothesis, on the other hand, a manual Stroop effect should be observed. Whether or not asymmetric effects should be expected for the translated versions of the two tasks (i.e., those for which the format of the relevant stimulus attribute and the key labels differ; see Table 3) and their untranslated counterparts depends on the assumptions regarding the role of perceptual and/or structural similarity. Although many models are silent or vague with respect to this question (but see Hommel, submitted) they nevertheless seem consistent with the view that perceptual and/or structural overlap leads to higher overall S‑R overlap between stimuli and responses than conceptual overlap alone. If perceptual/structural overlap enhanced overall similarity, one would expect larger manual Stroop effects in translated as opposed to untranslated tasks, that is, similar asymmetries as the one observed in the verbal version of the Stroop task. Note that the strong and the weak version of the direct coding hypothesis seem to make the same predictions regarding this task because instructions refer to the key labels that are visible throughout the experiment.

The pattern of results typically obtained with manual responses can be described as follows. First, Stroop interference shows with both, verbal and color labels on the keys, although manual Stroop effects are typically smaller than verbal effects (cf. Sharma & McKenna, 1998; [page 36↓]see Sugg & McDonald, 1994, for an overview of findings obtained with different versions of the manual Stroop task). Importantly, Stroop interference tends to be modulated by task requirements, that is, by the combination of relevant stimulus dimension and label format (see Table 3). More specifically, pronounced effects are observed for both, the translated word-response-task in which subjects are required to respond to stimulus color (and to ignore the color-word) by pressing keys labeled with color-words, and the translated color-response-task that requires responding to color-words (and ignoring the color of the print) by pressing keys with color-patch labels. In contrast, usually no interference is obtained in the untranslated word-response task that requires word-label responses to the color-words irrespective of the color of the print (Pritchatt, 1968; Sugg & McDonald, 1994).

So far, the results of the manual task are consistent with and extend (by adding a translated color response task) those obtained with the verbal task, and seem generally consistent with the direct coding hypothesis. That is, under the assumption that key responses labeled with color words are primarily coded in terms of color names, direct activation from color-word distractors to responses (e.g., RED à “red”) is stronger than direct activation from color to color names (e.g., redà “red”), leading to more interference if the strong associate serves as distractor (i.e., in the translated word response task).

Similarly, the effect obtained in the translated color response task can be explained by assuming that key responses are primarily coded in terms of color or color concepts when keys are labeled with color patches. Interestingly however, most often significant (albeit typically smaller) Stroop interference is also observed for the untranslated color-response task that requires responding to color by pressing keys labeled by color patches (e.g., Keele, 1972; Pritchatt, 1968; Redding & Gerjets, 1977; Simon & Sudalaimuthu, 1979; Sugg & McDonald, 1994; White, 1969; but see McClain, 1983). In my view, the most plausible explanation for this outcome is that (automatic) lexical/verbalàsemantic/conceptual activation (e.g., “red” à red) is stronger than semantic/conceptualà lexical/verbal code activation (e.g., redà red à “red”; cf. Sugg & McDonald, 1994, discussion of Experiment 1). According to this account, color names automatically activate their corresponding concepts in the untranslated color-response task, leading to interference with the correct response that is also accessed via conceptual (and/or perceptual) codes. In contrast, lexical activation from conceptual codes repre[page 37↓]senting irrelevant colors in the untranslated word-response task would be weak or nonexistent, leading to less interference4.

In sum, if one assumes that Stroop interference is primarily due to response competition, as many researchers do (e.g., Cohen et al., 1990; Virzi & Egeth, 1985), then keypress responses must have been coded in terms of color, color names, and/or conceptual color codes. Color coding is induced by instructing participants to press a particular key that is labeled in a certain way. Hence, according to response-competition accounts of the manual Stroop effect, instructions (and key labels) influence response coding, thus supporting the direct coding view. Effects of label format are consistent with this view by demonstrating considerable flexibility of coding.

However, not all researchers agree with the response competition account of the Stroop effect. On the one hand, some researchers attribute the (manual) Stroop effect primarily to conceptual stimulus identification (e.g., Hasbroucq & Guiard, 1991; Kornblum & Lee 1995; Kornblum et al., 1999; Lu, 1997; Lu & Proctor, 2001), that is, to congruency viz. incongruency between the two stimulus attributes. According to this view, both verbal and color stimuli activate their corresponding concepts, thus hindering identification of the relevant stimulus or stimulus attribute in case of incongruent stimuli. Whereas ’pure’ identification explanations do not seem to be particularly well suited to account for labeling-effects, translation models are.

Translation models (e.g., Glaser & Glaser, 1989; also see Sugg & McDonald, 1994, for an adapted version of the Glaser & Glaser model) emphasize the structural relationship between the relevant stimulus type and the response type. In general, they propose that colors and words are processed in different processing modules, each of which has its own codes. So, for example, according to Glaser and Glaser (1989; also see Phaf et al., 1990), color stimuli have privileged access to semantic (conceptual) codes, whereas word stimuli predominantly activate lexical (verbal) codes. According to translation models, substantial interference occurs if (a) information has to be translated to a code in the other system, and (b) irrelevant information has privileged access to the required code. Importantly, the primary source of interference according to these models appears to be activation (competition) at some in[page 38↓]termediate (lexical or semantic) stage, not necessarily response activation per se (cf. Sugg & McDonald, 1994). For instance, left responses to red stimuli in the translated word-response task (i.e., redà ”red”à left) are slowed because the distractor word GREEN activates a competing lexical representation (i.e., GREEN à “green” à right).

Thus, stimulus identification and translation accounts seem to share the view that the primary ‘locus’ of interference concerns some intermediate stage between perceptual stimulus identification and response selection, rather than response coding and response selection itself. Deciding between response competition and stimulus identification/translation accounts of the manual Stroop effect is difficult or even impossible because there is overlap between (a) the relevant and the irrelevant stimulus dimension, (b) the relevant stimulus dimension and responses, and (c) the irrelevant stimulus dimension and responses, leading to multiple possible sources of interference.

Moreover, response-coding alternatives are not assessed in the manual Stroop task. More specifically, it could be the case that responses are coded in terms of left and right (at least in situations where only two response alternatives and horizontal key arrangements have been used), independently of instructions and key labels. The manual Stroop task does not provide a means to rule out this possibility because spatial response coding, and its susceptibility to instruction, are not assessed in this task (but see Lu and Proctor, 1995, for a review of findings obtained with variants of the spatial Stroop task that requires naming or keypress responses to positions or position words).

In sum, multiple possible sources of interference as well as the lack of measures for spatial response coding defy a strong interpretation of Stroop congruency effects in favor of the direct coding hypothesis. Rather, the results obtained with the manual Stroop task seem to be somewhat uninformative with respect to the question of whether responses are arbitrarily coded when so instructed. Some of the criticisms regarding the Stroop task have been met by studies using the Hedge and Marsh task, a task named after the researchers that developed it (Hedge & Marsh, 1975). Findings obtained with this task are reviewed in some detail in the next section because they have been interpreted as major evidence for obligatory spatial response coding (e.g., Lu & Proctor, 1995).

The Hedge and Marsh Task

In the Hedge and Marsh (H&M) task, responses are instructed in terms of color, and spatial response coding is assessed by randomly varying stimulus position. More specifically, [page 39↓]participants are required to respond to colored (e.g., red and green) stimuli that randomly appear to the left or the right by pressing lateralized response keys that are labeled with corresponding color patches (i.e., red vs. green). Hence, relevant and irrelevant stimulus dimension do not overlap in this task (but see Hasbroucq & Guiard, 1991, for a different view).

Typically, two different color-mapping conditions are compared. In the direct mapping condition, participants are either instructed to press, for example, the red key in response to the red stimulus (i.e., instructions do not explicitly mention the correspondence relationship; Hedge & Marsh, 1975; Proctor & Lu, 1999), or to press the key of the corresponding color (e.g., Hasbroucq & Guiard, 1991; Lu & Proctor, 1994; Proctor & Pick, 2003). In contrast, the reversed mapping condition requires opposite color responses (e.g., green key to red stimulus). Table 4 illustrates the resulting conditions in terms of color compatibility and (task-irrelevant) spatial compatibility as a function of position compatibility and mapping.

Table 4: Stimulus-response compatibility conditions realized by spatial compatibility under the direct (same color) and reversed (alternate color) mapping in the Hedge and Marsh (1975) task.

Mapping condition

Position compatibility

 

compatible

incompatible

direct

+/+

+/-

reversed

-/+

-/-

Note. + and - indicate compatibility and incompatibility between stimulus color and response color / stimulus position and response location, respectively.

The spatial coding hypothesis predicts a spatial compatibility effect (Simon effect) of typical size under both the direct and the reversed color mapping because irrelevant stimulus position should directly activate the correspondingly coded responses, regardless of instructions. Moreover, a color compatibility effect (i.e., faster responses under the direct than under the reversed color mapping) is also expected because a compatible color mapping should speed up translation in the conditional route.

The weak version of the direct coding hypothesis makes essentially the same predictions as the spatial coding hypothesis, albeit for different reasons. That is, if one assumes that responses in the H&M task are coded in terms of both, color and location (e.g., Zhang et al., 1999) both stimulus color and position should activate their corresponding response codes via the direct route, augmented by ‘controlled’ activation via the conditional route. Because the weak version of the direct coding hypothesis does not predict differential weighing of response codes, and because the models do not provide a means to predict the relative contribu[page 40↓]tion of the direct and indirect routes to response activation in a principled way, the H&M task therefore cannot differentiate between the spatial coding hypothesis and the weak direct coding hypothesis.

The strong version of the direct coding hypothesis also predicts a color compatibility effect. In addition, if response codes can be weighed according to instructions no or extremely reduced spatial effects should be observed under both mappings.

The pattern of results usually obtained with the H&M task can be described as follows (see Figure 2, for a summary). First, responses are much faster in the two color-compatible conditions (i.e., under the direct mapping) than in the color-incompatible conditions (i.e., under the reversed mapping). The size of the color-compatibility effects in different studies (using visual stimuli) is typically much larger than the overall Simon effect and the Simon effect under the direct mapping.

Second, and more importantly, a Simon effect of normal size (i.e., about 25 ms) is typically observed under the direct mapping (involving a +/+ vs. +/- comparison, see Table 4). That is, responses are faster when stimulus position and response location correspond, even though stimulus position is task irrelevant and responses instructions do not refer to key location (e.g., De Jong et al., 1994; Hedge & Marsh, 1975; Lu & Proctor, 1994; Simon, Sly, & Vilapakkam, 1981). Because the response keys in the H&M task are not instructed with reference to their location, Lu and Proctor (1995) summarized that “spatial coding of the responses influences RT even when the response alternatives are not directly defined by spatial features”, and concluded, “the Simon effect occurs whenever the response must be coded spatially.” (p. 181)

Thus, the results obtained under the direct mapping apparently support the spatial and/or the direct coding hypothesis, and seem to provide evidence against intentional weighing.

However, the picture gets more complicated if one considers the two mapping conditions in combination and the reversed mapping condition in isolation. More specifically, when both mapping conditions are considered together, the usual finding is that the two compatibility effects interact. That is, whereas responses are faster when stimulus position and response location correspond (i.e., when they are spatially compatible) with the direct mapping, under the reversed mapping spatially compatible responses are slower than incompatible responses (see Figure 2). This pattern of results is difficult to reconcile with all three response-coding hypotheses that either predicted Simon effects in the usual direction or no spatial effect at all.


[page 41↓]

Figure 2: Schematic diagram of the basic findings in the Hedge and Marsh (1975) task under the direct (left panel) and the reversed (right panel) mapping (adopted from Lu & Proctor, 1995). Note. Different fillings indicate different colors.

Nevertheless, this finding is quite robust. It has been replicated in several studies (e.g., De Jong et al., 1994; Lu & Proctor, 1994), both with the original color-to-color mapping conditions, and with color-word response labels (i.e., with a variant of the translated word-response task, see previous section). Interestingly, Simon and Sudalaimuthu (1979) observed a similar RT pattern across mapping conditions with both an untranslated and a translated color-response version of the manual Stroop task with two response alternatives (i.e., faster responses with congruent distractors under the direct mapping as opposed to faster responses with incongruent distractors under the reversed mapping; see previous section for an overview of the different versions of the Stroop task), presumably implying that the H&M and the manual Stroop task share some properties.

The mechanisms underlying the impact of mapping instructions on the direction of the Simon effect in general, and especially its reversal under the reversed mapping, have been subject to considerable debate.

Two classes of explanations tend to be most prevalent to date, namely logical recoding accounts and an explanation in terms of display-control-arrangement correspondence (DCC)5.

The basic tenet of the – arguably most widely accepted - logical recoding account (in the version of De Jong et al., 1994, and Lu & Proctor, 1994) is that participants recode the [page 42↓]relevant and – inadvertently – also the irrelevant (spatial) stimulus attribute in order to form the response code. Under the reversed mapping condition a “respond opposite” rule is formed that transforms the values on both stimulus dimensions into their opposite. Consequently, a left stimulus attribute now primes a right response, leading to faster responses in the incompatible than the compatible condition.

Hardly surprising, the assumptions regarding what leads to the normal Simon effect under the direct mapping are more heterogeneous. According to some researchers (e.g., Hedge & Marsh, 1975; Lu & Proctor, 1994), an analogous (to the reversed-mapping condition) recoding rule, in this case some sort of “same” operation, is applied to both stimulus dimensions, leading to faster responses when both color and position correspond with the required response. This assumption seems plausible, given that (a) most studies manipulated mapping within subjects, and (b) the mapping instructions of many studies emphasized the correspondence viz. noncorrespondence between relevant stimulus and response dimension (see above). On the other hand, some researchers (e.g., De Jong et al., 1994) assume that, in addition to identity transformations, direct activation of compatible responses is responsible for the Simon effect under the color-compatible mapping condition, but not for its reversal under the reversed mapping.

This view gains support from distribution analyses of the effects and from studies that tracked response activation via lateralized readiness potentials (i.e., an electrophysiological index taken to reflect motor preparation processes). These studies show early activation of the spatially compatible response regardless of mapping (e.g., De Jong et al., 1994; Valle-Inclán, 1996), suggesting that fast (but not slow) responses are influenced by direct activation, leading to an enhanced Simon effect for the fast responses under the direct mapping, and to a reduced reversed Simon effect for faster responses under the reversed mapping (e.g., De Jong et al., 1994; but see Zhang & Kornblum, 1997, for a critique of this interpretation).

In contrast, the DCC account initially proposed by Simon et al. (1981) attributes the results to display-control arrangement correspondence. Such correspondence exists when the location of the color stimulus corresponds to the location of the same-color response-label (control). DCC is invariant across mappings. Consequently, DCC under the reversed mapping is present when stimulus location and response location are incongruent. So, for example, a red stimulus that requires a green/right response can be spatially aligned with the red/left re[page 43↓]sponse label when presented to the left, but not when presented to the right, leading to faster responses on spatially (S-R) incompatible trials.

Evidence in support of this view comes from two task modifications of the original H&M task. In one (e.g., Proctor & Pick, 2003, Exp. 1 and 2; Simon et al., 1981, Exp. 3), relevant and irrelevant stimuli are presented in different modalities. For example, color-responses are required to centrally presented color-stimuli, whereas irrelevant location information comes from tone stimuli that are randomly presented to the left or the right ear. With this arrangement, usually no spatial effect (i.e., no Simon-reversal) shows under the reversed mapping. Second, Proctor and Pick (2003) noted that, in the majority of studies using the H&M task, color labels were clearly visible during target processing. That is, most often they are presented below the stimuli in the lower part of the screen. When Proctor and Pick (Exp. 2) used key-labels that were not visible during stimulus processing, the Simon-reversal under the reversed mapping was absent, presumably implying that participants (perceptually) aligned stimuli and responses in the studies under visible-label conditions (but see De Jong et al., 1994, who found a reversal in the translated word-response task; also see Zhang, 2000, for critical findings).

Taken together, the Simon effect under the direct mapping and the results from distribution analyses (as well as from electrophysiological recording) obtained with the H&M task seem generally more consistent with the view that responses are spatially coded, either with (as assumed with by the weak version of the direct coding hypothesis) or without (as assumed by the strict spatial response coding view) color response codes contributing to the color compatibility effect. That is, the results seem to provide evidence against the strong direct coding hypothesis.

However, the origin of the reversal of the Simon effect under the reversed color mapping, and the origin of the normal Simon effect under the direct mapping is still unclear. That is, (non-) correspondence instructions as well as clearly visible response labels may have induced a bias to either logically recode stimuli (and/or responses) in terms of same/opposite rules, or use a perceptual matching strategy as proposed by the DCC hypothesis. Moreover, because color-overlap only existed in the task-relevant dimension, the color effect cannot unambiguously be interpreted as indicating color-coding of responses.

A stronger test of color coding of responses as well as intentional weighing would be provided by experiments that (a) investigated the influence of task-irrelevant stimulus-color [page 44↓]on responses that are instructed in terms of color while avoiding S‑S correspondence and assessing spatial coding, and/or (b) studied instructional modulation of (irrelevant) spatial stimulus-response compatibility by instructing responses spatially vs. non-spatially. It seems as if relatively few studies chose either of these approaches. Those I am aware of are reviewed in the next section. Again, irrelevant color effects would be expected according to the direct coding hypothesis. Moreover, the intentional weighing hypothesis predicts reduced to nonexistent spatial effects under color (non-spatial) instructions of responses. In contrast, according to the spatial coding hypothesis, irrelevant stimulus effects should be restricted to the spatial dimension, and spatial effects should be unaffected by response instructions.

Response Instructions and Spatial vs. Non-Spatial Irrelevant Stimulus Effects

The evidence regarding response coding stemming from experiments that either assessed the impact of irrelevant stimulus color or investigated the spatial Simon effect under non-spatial response instructions appears to be mixed.

For example, Simon, Acosta, Mewaldt, and Speidel (1976, Exp. 2) conducted an experiment that provides evidence in favor of spatial response coding despite non-spatial response instructions. Simon et al. (1976) instructed participants to press a key of a certain color in response to the pitch of tone. Whereas a fixed-label group worked under a constant color-to-key assignment (i.e., a constant assignment of color to key-location), for other groups, the color-to-key mapping varied from trial to trial. In the varied color-to-key conditions, key labels were either presented 1s before the imperative stimulus (pitch), simultaneous with the stimulus, or after stimulus onset. Simon et al. found a significant Simon effect (i.e., a compatibility effect between irrelevant tone location and response side) both, with fixed key labels and in the varied color-to-key conditions that presented the labels prior to or simultaneous with the imperative stimulus. Because responses were instructed in terms of color (i.e., non-spatially), the Simon effect observed under these conditions seems to support the spatial coding view, and hence, the strict spatial coding hypothesis and/or the weak version of the direct coding hypothesis according to which implicit and explicit S‑R overlap lead to comparable direct route activation.

However, although responses must have been coded spatially to some extent (otherwise no Simon effect would have been observed), the question of whether response coding is modulated by instructions remains unanswered by the Simon et al. (1976) results because the experiment did not include a condition with spatial response instructions. That is, the experi[page 45↓]ment does not allow a comparison of the size of the Simon effect under spatial vs. non-spatial response instructions. Moreover, participants in the Simon et al. (1976) experiment fixated the response arrangement during auditory stimulus processing, possibly encouraging some subjects to spatially re-code their responses (and hence, to partially ignore instructions), thus leading to the Simon effect.

Some evidence in favor of the spatial re-coding interpretation stems from a comparison between the fixed-label group on the one hand, and the 1-sec prior and simultaneous label groups on the other hand. Whereas the Simon effect in the fixed-label group was about 60 ms (a standard effect size with lateralized auditory stimuli; cf. Proctor & Pick, 2003), the Simon effect in the two varied-mapping groups was (marginally, but see sample sizes) reduced to 36 ms. Interestingly, the Simon effect in the 1-sec prior group and the simultaneous-label group was comparable even though overall RT level in the latter group was much higher, indicating that overall RT differences were not responsible for the difference in effect sizes between the latter two groups and the fixed-label group. Therefore, it seems possible that the fixed-label group recoded the instructions in terms of response location during the course of the experiments, leading to a more pronounced Simon effect.

Stronger evidence for the prevalence of spatial response coding has been provided by Brebner (1979) who switched the relevant and irrelevant dimensions in the H&M task (see previous section). More specifically, Brebner used a paradigm in which stimuli and responses did or did not correspond in terms of both, location and color, but unlike the original H&M task, participants were required to respond with the key on the same side (direct mapping) or the opposite side (reversed mapping) of the stimulus regardless of stimulus (and key) color. With this task, Brebner neither found a color-compatibility effect under the direct location-mapping, nor a reversed color effect under the reversed location-mapping. This result suggests that the irrelevant color dimension can be ignored more easily than irrelevant location with respect to whatever transformations or processes are required to perform the task under the different mappings.

However, Brebner (1979) has been criticized by Kornblum et al. (1990, p. 267) on methodological grounds. Moreover, results of studies that also used two-dimensional color-space arrangements, but instead instructed S‑R mappings cross-dimensionally, lead to different conclusions with respect to the impact of instructions on response coding.


[page 46↓]

For instance, Hasbroucq and Guiard (1991) included two cross-dimensional mappings in their Experiment 2. In the color-to-position mapping, subjects were required to press the left or right key (color-labels of keys varied randomly from trial to trial) in response to stimulus color, and to ignore (randomly varied) stimulus position (for example, the instruction was “respond to a red stimulus by pressing the left key”, thereby making stimulus position task irrelevant). Similarly, in their position-to-color mapping, subjects were required to press a key of a specific color in response to the location of a stimulus, with stimulus color being task irrelevant (e.g., “respond green to a left stimulus”). Again, color-labels randomly changed from trial to trial. Hasbroucq and Guiard (1991) found comparable irrelevant stimulus color and stimulus position compatibility effects (both about 50 ms) in the position-to-color and the color-to-position mappings, respectively. Moreover, they did not find an effect of irrelevant overlap between the relevant stimulus dimension and the irrelevant response dimension, at least not for the position-to-color mapping (i.e., no effect of spatial correspondence between relevant stimulus position and – randomly varying – irrelevant response location). This finding indicates that the instructed response dimension determines whether a stimulus-response compatibility effect occurs, thus supporting the strong version of the direct coding hypothesis. That is, these results support the view that non-spatial response instructions reduce the weights of spatial response codes, and hence the effect of (irrelevant) spatial stimulus attributes on responding.

Similarly, Smith and Brebner (1983) also manipulated mapping across dimensions. In their experiment, the (irrelevant) color compatibility effect in the position-to-color condition was even larger than position compatibility effect in the color-to-position condition, and a color effect even showed in the color-to-position condition, again seemingly supporting the strong version of the direct coding hypothesis.

However, both the Hasbroucq and Guiard (1991) and the Smith and Brebner (1983) study found slowed color-responses as compared to location-responses, possibly indicating an additional translation step. Moreover, both studies have been (e.g., Hommel, 1995), or can be, criticized for methodological reasons as well. For example, Hommel (1995) noted that participants in Hasbroucq and Guiard’s Experiment 2 had to work under several mapping requirements in succession (i.e., color-to-color, position-to-position, color-to-position, and position-to-color), possibly leading to carry-over effects or specific strategies induced by previous task requirements. Similarly, Smith and Brebner’s participants were required to work under [page 47↓]both of two variants of either the color-to-position or the position-to-color mapping in succession. For example, in one variant of the position-to-color mapping subjects had to press the key of the same color as the stimulus when the stimulus appeared on the left, but to press the alternate color button when the stimulus appeared on the right. In the other variant however, the same/alternate assignment was reversed for the same subjects (i.e., they were now required to respond with the opposite-color button to a left stimulus, and to make same-color responses to right stimuli). This design presumably also led to pronounced inter-task transfer effects.

In sum, the studies reported in this section are highly inconclusive with regard to the main question of interest in this thesis, namely the question of whether response instructions determine how responses are coded. On the one hand, experiments with cross-dimensional mappings show symmetric (irrelevant stimulus-) color and position effects that depend on response instructions, apparently supporting the intentional weighing hypothesis, that is, the strong version of the direct coding hypothesis. However, these studies can be criticized because they used within-subjects manipulations of instructions, and because slowed responses on color-response tasks question the comparability of the position-to-color and the color-to-position task (but see Simon et al., 1976, who did not find an effect of overall RT level on the Simon effect in the two varied-label groups).

On the other hand, both the findings by Brebner (1979) and by Simon et al. (1976) apparently support the view that responses are spatially coded, regardless of response instructions. However, the Brebner (1979) study has been criticized for its lack of appropriate control conditions (cf. Kornblum et al., 1990). Simon et al. (1976), on the other hand, did not directly compare spatial and non-spatial response instructions, and their results with fixed vs. varied label-to-key mapping indicate that non-spatial response instructions (combined with varied labels) reduced the Simon effect observed in the fixed-label group who possibly (re-) coded their responses spatially. Therefore, their results cannot conclusively rule out the intentional weighing hypothesis (i.e., the strong direct coding hypothesis) either.

In Chapter 5, I present experiments that directly compare the Simon effect under spatial vs. non-spatial response instructions in order to more fully assess whether non-spatial response instructions reduce the Simon effect with spatial response instructions.


[page 48↓]

3.1.5  Summary

Stimulus-response compatibility effects are commonly attributed to a stage in information processing known as the ‘response selection stage’. More specifically, dual route models assume that they result from response competition induced by response activation via a conditional (controlled) route that depends on mapping instructions, and a direct route that solely depends on overlap between stimulus and response codes. Accordingly, investigating which match or compatibility relations contribute to compatibility effects under which response instruction conditions allows conclusions about the cognitive codes that are used to control instructed responding.

Therefore, Chapter 3.1 provided a review of findings regarding the impact of (response) instructions on several SRC effects involving spatially organized keypress responses. This review was guided by the question of whether the results obtained with several tasks provide evidence for or against the coding hypotheses identified in Chapter 2.

According to the direct coding hypothesis, it was expected that variations in response instructions affect response coding, and hence, how responses are or can be accessed. Whereas the weak version of the direct coding hypothesis does not discriminate between implicit and explicit (instructed) overlap, the strong version assumes that response codes can be weighed according to instructions. Consequently, both versions predict compatibility effects resulting from stimulus-overlap with instructed (including non-spatial) response dimensions even when the overlapping stimulus attribute is task irrelevant. However, only the strong version of the direct coding hypothesis assumes that instruction manipulations should lead to variations in the direction or size of a given spatial compatibility effect. In contrast, the spatial coding hypothesis predicted (irrelevant) spatial compatibility effects of comparable size and in the same direction regardless of the specific contents of the response instructions. Other than spatial compatibility effects should be restricted to the task-relevant dimension, that is, they should be attributable to translation efficiency in the conditional route.

Given these criteria to decide between the different views, the general conclusion I arrived at has been that the findings obtained with the tasks reviewed in Chapter 3.1 provide highly inconsistent and/or ambiguous evidence for and against all hypotheses.

Sections 3.1.1 – 3.1.2 were mainly concerned with instruction manipulations according to which different instruction conditions emphasized different spatial aspects of the stimulus- and/or response array.


[page 49↓]

With the two-dimensional spatial mapping task (see Chapter 3.1.1) a weak left-right prevalence effect is typically observed. That is, although horizontal and vertical compatibility effects tend to be modulated by type of instruction (i.e., vertical vs. horizontal instructions) in most studies, the overall horizontal compatibility effect is usually larger and less reduced by vertical instructions, indicating a dominance of left-right over top-bottom coding, and hence, apparently providing evidence in favor of the spatial coding hypothesis (in its one-dimensional interpretation).

However, it has been argued that, in this task, the two stimulus-dimensions (i.e., horizontal and vertical position) always redundantly signal the correct response, thus inviting different types of strategic recoding biases (i.e., re-interpretations of instructions). Moreover, because the relevant stimulus attribute in this task is always position, it cannot be determined whether this bias refers to stimulus and/or response coding, rendering this task less than optimal to answer the question of whether response instructions determine response coding.

The findings regarding the impact of response location vs. finger instructions in auditory and visual Simon tasks requiring crossed-hands responses (see Chapter 3.1.2) are inconclusive as well. Whereas Wascher et al. (2001) found an impact of response instructions on the pattern of hand-based vs. location-based Simon effects in the auditory task, Roswarski and Proctor (2003a) only found a small impact of anatomical coding under finger instructions after considerable practice. Moreover, instruction effects were negligible for visual tasks in both studies. Thus, although the results favor hierarchical spatial coding (with location-coding on the top of the hierarchy; e.g., Heister et al., 1990), and hence tend to speak against the intentional weighing (i.e., the strong direct coding) hypothesis, the reason for (a) the different pattern of results observed across stimulus modalities and (b) the effect of practice on response (re-?) coding in the auditory task remains unclear.

The strongest evidence in favor of intentional weighing of response codes (i.e., the strong version of the direct coding hypothesis) has been provided by Hommel (1993a) who instructed responses either in terms of response location or in terms of the location of contralaterally presented response effects (see Chapter 3.1.3). Hommel found that the response-location based Simon effect was partially reversed for the effect-instruction group (e.g., left keypress responses were now faster when the stimulus and the response effect appeared on the right, that is, when the stimulus appeared at a response-location incompatible position), suggesting that instructed response features were weighed more strongly than uninstructed [page 50↓]response features. However, instructions in the Hommel (1993a) study emphasized different spatial aspects of the response array, thus possibly affecting the hierarchy of spatial coding, but not necessarily the prevalence of spatial coding per se.

Therefore, it has been argued that SRC effects resulting from non-spatial irrelevant stimulus overlap with the instructed response dimension would provide more stringent evidence for the direct coding hypotheses.

Evidence regarding non-spatial response coding as a consequence of non-spatial response instructions and labels (Chapter 3.1.4) seems ambiguous as well, though.

On the one hand, color-compatibility effects obtained with manual versions of the Stroop task and with the H&M task are consistent with the interpretation that responses were at least partially coded in terms of color (or color names).

On the other hand, in the manual Stroop task, spatial response coding has not been assessed, and researchers do not agree regarding the source(s) of interference in the Stroop task. More specifically, (conceptual) stimulus identification and/or interference at some intermediate translation stage cannot be ruled out as explanations because there is not only overlap between (irrelevant) stimulus attributes and responses, but also between the relevant and the irrelevant stimulus dimensions.

In contrast to the manual Stroop task, the H&M task avoids overlap between the relevant (color) and the irrelevant (position) stimulus attributes, and does provide a means to measure spatial response coding. However, the results obtained with this task are partially (in)consistent with all three hypotheses. First, the color compatibility effect does not differentiate between the alternative hypotheses because overlap exists regarding the task-relevant dimension. Second, whereas the (location-based) Simon effect typically observed under the direct mapping seems to indicate spatial response coding, and hence to support both the spatial and/or the weak direct coding hypothesis, the origin of the Simon effect in this task and its reversal under the reversed-color mapping are unclear. More specifically, the pattern of results is also consistent with an explanation in terms of a (strategic) bias to either perceptually align stimuli and response labels, as proposed by the DCC account, or to logically recode the mappings into “same” and “opposite” rules that are inadvertently applied to the irrelevant (spatial) dimension as well.

Studies that investigated irrelevant color effects with cross-dimensional instructions in the H&M task (Hasbroucq & Guiard, 1991; Smith & Brebner, 1983) seem generally better [page 51↓]suited to assess response coding in terms of color. These studies found a symmetric influence of irrelevant stimulus position and irrelevant stimulus color that depended on response instructions. Whereas these results seem to favor the strong direct coding hypothesis, both studies using this type of mapping manipulation have been severely criticized on methodological grounds, again defying any firm conclusions.

Hence, it seems as if the strongest evidence in favor of arbitrary (i.e., non-spatial) response coding to date has been provided by experiments on response-effect compatibility (see Chapter 3.1.3). These studies demonstrate that arbitrary response effects such as color, pitch of tone, or affective valence, become integrated into the action representation, and are or can be used to code and access responses during subsequent task performance. However, none of these (arbitrary response-effect) studies provides evidence for or against a direct impact of response instructions. Rather, they demonstrate non-instructed response coding, that is, the use of codes that may have been primed by practice and/or that may have proven useful for the task at hand.

Finally, Simon et al.’s (1976, see Chapter 3.1.4) finding of a Simon effect under conditions where stimulus attributes (i.e., pitch of tone) were arbitrarily mapped to color-responses (i.e., when responses were non-spatially instructed) indicates that responses are at least partially coded in terms of location in their experiment, apparently contradicting the strong version of the direct coding hypothesis. However, their study did not include a condition with spatial response instructions. It seems possible that the Simon effect under spatial response instructions would be larger than under non-spatial response instructions. Therefore, intentional weighing of response codes according to instructions, and hence, a reduced impact of spatial response codes under non-spatial response instructions, cannot be precluded.

The experiments presented in the empirical part of the thesis (Chapters 4 and 5) attempt to assess more directly inhowfar the response labels used in verbal task instructions determine response coding. More specifically, they extend the evidence for arbitrary response coding obtained in studies on response-effect compatibility by addressing whether arbitrary (non-spatial) response coding occurs as a function of response instruction (Chapter 4), and they extend the Simon et al. (1976) results by directly comparing (irrelevant) spatial correspondence effects under spatial vs. non-spatial response instructions (Chapters 4 and 5). Moreover, in the experiments presented in Chapter 5, an attempt was made to avoid the confound between instructions and practice present in all prior studies so far (see Roswarski & Proctor, [page 52↓]2003a, for results indicating an impact of practice on coding) by introducing new imperative stimuli (and mapping instructions) on each trial.

Whereas the experiments reported in Chapter 5 rely on a Simon-type task and hence – methodologically speaking – directly relate to the findings reviewed in Chapter 3.1 above, the experiments presented in Chapter 4 used a dual-task paradigm that involved consistent viz. inconsistent responses on the two tasks (i.e., R‑R compatibility).

Therefore, before proceeding to the empirical part of this thesis, Chapter 3.2 provides a review of findings concerning inter-task consistency effects similar to those presented in Chapter 4.

3.2 Response Instructions and Cross-Task Compatibility

‘Inter-task consistency’ or ‘cross-task compatibility’ (CTC) refer to the finding obtained in dual task studies, which require participants to simultaneously perform two tasks, that responding on either task is easier when the mapping for the other task requires consistent rather than inconsistent responses (see Lien & Proctor, 2002, for a recent review).

For example, Hommel (1998, Experiment 1) had participants perform a manual (primary) and a verbal (secondary) task in response to centrally presented visual stimuli. The (primary) manual task was to press a left or a right key in response to the color of colored letters, whereas the verbal (secondary) task required saying either “left” or “right,” depending on letter identity. Hommel found R‑R compatibility effects on both tasks, that is, both manual and verbal responses were faster when response ‘locations’ corresponded (e.g., faster responses when a left keypress was followed by a “left” as opposed to a “right” verbal response) than when they did not correspond.

This result has several important implications. First, response selection for the two tasks must have overlapped in time to produce backward (R2-R1) compatibility effects from Task 2 to Task 1. This finding contradicts strong response-selection-bottleneck accounts of the ‘psychological refractory period’ (PRP) effect, that is, the finding that R2 responses are particularly delayed when S1-S2 stimulus-onset asynchrony (SOA) is short. More specifically this finding is inconsistent with the proposal that a response selection bottleneck leads to complete postponement of S2-R2 translation until R1 selection has finished (see Lien & Proctor, 2002; Pashler, 1994, for comprehensive descriptions of the PRP effect and bottleneck as well as alternative accounts thereof). Rather, it indicates that responses on the two tasks were acti[page 53↓]vated automatically and in parallel even though the S‑R mappings for both tasks were arbitrary.

Second, the finding of inter-response effects between physically dissimilar responses such as pressing a key and saying a word indicates that participants used overlapping codes when accessing their verbal and manual responses. Under the assumption that cross-task compatibility in the Hommel experiment reflects parallel activation of responses (i.e., response codes) the results suggest that relatively abstract conceptual representations of space were used to code both verbal and manual responses (see General Discussion of Experiments 1-3, Chapter 4, for a more detailed discussion). Moreover, whereas forward R‑R compatibility effects can be explained by some sort of (meaning based) automatic priming and thus might only indicate that the overlapping code had been part of the response representation of the secondary task, the finding of forward and backward compatibility effects suggests that these codes were not only part of both response representations, but that they were actively used to access and guide responding on the secondary task.

Inasmuch as such inter-task consistency effects reflect response-related processes, the nature of such effects and their susceptibility to response instructions again allow inferences about the codes involved in response activation and selection. If one generalizes the response coding hypotheses derived above to dual-task performance (i.e., by assuming some sort or R‑R priming when response codes for the two tasks overlap)6, the spatial coding hypothesis would again predict only spatial R‑R compatibility effects that should occur independently of the response labels used in manual task instructions. In contrast, according to the direct coding hypothesis, one would expect cross-task compatibility for instructed (non-spatial) response dimensions as well. Moreover, the strong version of the direct coding hypothesis again predicts that cross-task compatibility should primarily be observed for the instructed response-dimension, regardless of whether it is spatial or not.

To anticipate the conclusions drawn in this section, the results on inter-task consistency effects so far seem to support the view that such effects depend on the instructed S‑R map [page 54↓] pings. However, they do not support a strong interpretation in terms of instructional impact on response coding.

For instance, findings by Lien and Proctor (2000) as well as Koch and Prinz (submitted) corroborate the conclusion that CTC effects as those found by Hommel (1998) are primarily based on parallel response code activation, and extend the Hommel results by showing that the direction of such effects depends on the instructed S‑R mapping.

Lien and Proctor demonstrated that, at short SOAs, response selection on an arbitrarily mapped primary task is influenced by correct R2 activation even when R2 is mapped to S2 in a spatially incompatible way (e.g., when Task 2 required left responses to right arrows). For example, a left response to the letter “x” on the primary task was faster when it was accompanied by an arrow pointing to the right (requiring a left response on Task 2) than when it was followed by a left arrow, although the R2-R1 compatibility effects tended to be numerically smaller under the reversed than under the direct mapping (i.e., when responses on Task 2 were compatibly mapped to arrow direction). Furthermore, Lien and Proctor observed that R1 selection was affected by irrelevant arrow (S2) position in a similar way as R2 was. That is, there was a small backward “Simon” effect under the direct Task 2 (arrow) mapping, whereas the irrelevant location effect was slightly reversed under the reversed mapping. These results suggest that Task 2 responses instead of Task 2 stimuli primed Task 1 responses, and that it is primarily the instructed Task 2 mapping that contributes to inter-task consistency effects.

Similarly, Koch and Prinz (submitted, see also Koch & Prinz, 2002) who used a somewhat different methodology and varied the encoding instructions of an unspeeded perceptual identification task were able to show that the direction of CTC effects between the identification task and a nested, but logically independent choice reaction task depended on the encoding instructions of the perceptual identification task. In their study, one group of participants was instructed to report the starting point of the movement of a moving target stimulus, whereas another group was required to report the endpoint of the movement. Hence, the correct answer to a target moving to the left was “left” for the endpoint group, but “right” for the starting point group. Koch and Prinz found that choice task responses (speeded left/right finger movements to predetermined keys in response to the color of letter stimuli presented before the moving target) were faster when the direction of the finger movement corresponded to the movement aspects that had been emphasized by the instruction.


[page 55↓]

Thus, the findings obtained by Hommel (1998), Koch and Prinz (submitted), and Lien and Proctor (2000) indicate automatic response activation and inter-response priming that follows instructed S‑R mappings. However, although these results show an impact of mapping instructions, they cannot conclusively differentiate between the different coding hypotheses. This is so because they did not directly manipulate response instructions. That is, responses were always instructed in terms of left and right, thus strongly encouraging spatial response coding. Accordingly, all three hypotheses, including the spatial coding hypothesis, are consistent with the results. Therefore, demonstrations of inter-task compatibility for other than spatial response dimensions, as well as studies that investigated whether consistency effects are restricted to the instructed response dimension appear to be better suited to address the question of the impact of instructions on response coding.

Logan and colleagues (Logan & Schulkind, 2000; see also Logan & Gordon, 2001) extended the findings regarding CTC effects reviewed so far by generalizing forward and backward consistency effects to non-spatial stimulus and/or response dimensions. Moreover, they were able to show that the occurrence of such inter-task consistency effects depends on the overlap of instructed categorizations.

For example, participants in Logan and Schulkind (2000, Experiment 2) had to categorize two numbers that were presented with varying SOAs. In some sessions, participants had to classify the numbers according to the same categories, that is, they either had to judge magnitude or parity on both tasks. In other sessions however, the categorization task varied from Task 1 to Task 2 (i.e., from magnitude judgments on Task 1 to parity judgments on Task 2 or vice versa).

Logan and Schulkind observed forward as well as backward category matching effects over a wide range of SOAs when the same categorization was required on the two tasks (e.g., responses were faster when both tasks required parity judgments and the numbers presented on Task 1 and Task 2 were both odd as opposed to one being odd and the other being even). Interestingly, however, no consistency effects (neither forward nor backward) were obtained when the two tasks required different stimulus categorizations (i.e., when one task required magnitude judgments while the other required parity judgments) although stimuli could still be classified according to both categories. So, for example, presenting the digit “4” (for Task 2) did interfere with Task 1 categorization of the digit “8” as being larger than “5” when task-[page 56↓]2 also required a magnitude judgment (i.e., a “smaller” response), but it did not interfere with Task 1 performance when it required a parity judgment (i.e., an “even” response).

Logan and Gordon (2001) interpret these and similar results as indicating that parallel activation or retrieval of response-relevant information depends on the amount of overlap of task-relevant response sets. One interpretation of this account (cf. Chapter 2.1) holds that responses are coded in terms of what they signal (e.g., a left response as meaning ‘odd’; for a similar view, see e.g., Meiran, 2000). According to this view, the Logan and Schulkind results appear to support the strong version of the direct coding hypotheses.

However, in the experiments described by Logan and colleagues the two tasks were mapped onto different hands, and the contribution of responding at corresponding vs. noncorresponding relative response locations to the category matching effect (is it negligible, additive, or does it interact?) has not been assessed. Moreover, stimuli were bivalent in that each stimulus provided evidence for categorizations on both tasks (e.g., the stimulus ‘7’ is both larger than 5 and odd). Hence, the contribution of stimulus-related processes, such as semantic priming (category-category priming) and stimulus-categorization processes to their findings is probably substantial. That is, it remains unclear to what degree, if at all, response representations were responsible for their results (but see Schuch & Koch, submitted; Watter & Logan, 2001, for promising attempts to disentangle stimulus- and response-related processes in the magnitude/parity judgment task).

Taken together, the findings reviewed in this section allow several conclusions with respect to response coding. First, backward compatibility of the kind observed by Hommel (1998), Lien and Proctor (2000), as well as Logan and colleagues, seems to be well suited to examine the actual use of specific response codes in Task 2 performance. In this regard, backward compatibility extends findings of forward compatibility (and, in a sense, also findings on stimulus-response compatibility, see Chapter 3.1) that may only show that the overlapping codes have been part of the Task 2 action representations.

Second, inter-task consistency effects indicate that conceptual – possibly arbitrary (e.g., Logan & Schulkind, 2000) – codes can be used to access physically dissimilar responses such as verbal and manual responses, or left-right responses with different hands.

Third, the results obtained by Koch and Prinz, Lien and Proctor, as well as by Logan and colleagues suggest that inter-task consistency depends on task demands, that is, on the instructed mapping rules. However, they do not support a strong interpretation in terms of an [page 57↓]instructional impact on response coding because either (a) only spatial response instructions were used for both tasks (Koch & Prinz, submitted; Lien & Proctor, 2000), or (b) because spatial response coding has not been assessed (e.g., Logan & Schulkind, 2000). It has been argued that the latter does not allow firm conclusions in terms of arbitrary response coding, whereas the former does not conclusively differentiate between the alternative coding hypotheses.

A more straightforward way to address the question of whether response instructions directly influence response coding of spatially organized responses (i.e., left-right responses on a manual task), would be to use univalent stimuli and arbitrary S‑R mappings (as, for example, in the Hommel, 1998, study) and to vary the response labels used in the instructions of a manual (secondary) task. On the one hand, such an approach allows less ambiguous conclusions as to whether inter-task consistency effects generalize to more abstract (non-spatial, e.g., color) response instructions, and hence, non-spatial response codes when a concurrently performed (e.g., verbal) task also requires this type of code (e.g., color codes). This would provide evidence in favor of direct coding in general.

On the other hand, such an approach also allows testing whether instructed response codes can override spatial response coding by requiring spatial coding on a primary (e.g., verbal) task and non-spatial coding of (spatially organized) responses on the secondary task. Whereas both the spatial and the weak direct coding hypothesis predict that the spatial backward-compatibility effects from a manual keypress task should be unaffected by response instructions, the strong version of the direct coding hypothesis predicts reduced spatial (forward and) backward effects under non-spatial manual response instructions.

This rationale is exactly the logic underlying the experiments in Chapter 4 that will be presented after summarizing the general aims of the study.

3.3 Summary and Aims of Study

Chapter 3 provided a review of findings that speak to the main question of this thesis, namely whether or not the specific response labels (e.g., “left” and “right”) given in simple binary choice task instructions involving spatially organized keypress responses (e.g., “when you see a square on the screen, press the left key; when you see a circle, press the right key”) determine how such a task is performed, that is how responses are coded and selected.

To this end, an overview of findings on the impact of response instructions on a number of different stimulus-response and inter-task compatibility phenomena involving spatially [page 58↓]organized keypress responses has been provided. The main question throughout this literature review was whether variations of response instructions affect the type, size or direction of observed compatibility effects. More specifically, according to the direct coding hypothesis derived in Chapter 2, it was expected that variations in response instructions affect response coding, and hence, how responses are accessed. Whereas the weak version of the direct coding hypothesis does not discriminate between implicit and explicit (instructed) overlap, the strong version assumes that response codes can be weighed according to instructions. Consequently, both versions predict compatibility effects resulting from overlap on the instructed dimension even when the instructed response dimension is non-spatial and the overlapping stimulus or response attribute is task irrelevant. However, only according to the strong direct coding view should instruction manipulations lead to variations in the direction or size of a given spatial (implicit) compatibility effect.

In contrast, the spatial coding hypothesis predicted (irrelevant) spatial compatibility effects of comparable size and in the same direction regardless of the specific contents of the response instructions. If other than spatial compatibility effects occurred, they should be attributable to translation efficiency in the conditional route.

The main conclusion derived from both, the stimulus-response and the response-response compatibility literature has been that the results are largely ambiguous with respect to the question of how instructions influence response coding for several reasons.

First, findings obtained with a wide variety of paradigms suggest that instructed S‑R mapping affects how a task is performed. For example, an impact of task demands on task performance has been observed with the Hedge and Marsh task (direct vs. reversed mapping), the two-dimensional mapping task (vertical vs. horizontal stimulus position to response location mapping), different versions of the manual Stroop task (translated vs. untranslated versions of the color-response and the word-response task), and in dual task studies with overlapping viz. non-overlapping responses on the two tasks (e.g., direct vs. reversed Task 2 mapping in the Lien & Proctor, 2002, experiments).

However, these findings cannot unambiguously be attributed to an effect of response instructions on response coding for several reasons. One reason is that, in most of these tasks, there was not only overlap between the irrelevant stimulus dimension and the instructed vs. uninstructed response dimensions, but there was also overlap between the relevant stimulus dimension and the (un)instructed response dimension (in the H&M task, the two-dimensional [page 59↓]mapping task and the manual Stroop task), and, in some cases, also between the relevant and the irrelevant stimulus dimensions (in the manual Stroop task). Moreover, several of the tasks that indicate an impact of irrelevant non-spatial information when responses are instructed non-spatially (i.e., the manual Stroop task and the experiments by Logan and colleagues) did not assess spatial coding.

It has been argued that these tasks do not permit firm conclusions as to the sources (i.e., the locus) of interference. More specifically it cannot be precluded that non-spatial compatibility effects were due to some intermediate translation (or stimulus-recoding) stage that transforms stimuli onto left/right (i.e., spatially coded) responses. Therefore, they cannot firmly rule out the spatial coding hypothesis.

The dual task studies that showed inter-task consistency effects involving left-right responses to depend on instructed S‑R mapping (e.g., Koch & Prinz, submitted; Lien & Proctor, 2000), on the other hand, are uninformative because they did not vary response instructions or labels (i.e., they only used spatial response instructions). Because all three hypotheses make the same predictions regarding spatial response instructions, these results do not directly speak to the issue of how response instructions affect response coding.

Second, those studies that varied response instructions directly and studied the impact of irrelevant stimulus attributes as a function of response instructions lead to inconsistent conclusions that depend on the task (and the instructions) used. Regarding instruction manipulations that emphasized different spatial aspects of the response array, results on anatomical vs. location instructions provide inconsistent results that seem at least slightly more consistent with the spatial coding hypothesis, that is, with the view that location-based coding dominates regardless of instructions. In contrast, instructions that either emphasized response location or salient contralateral response effects (e.g., Hommel, 1993a) support the strong version of the direct coding hypothesis by showing that instructed codes were weighed more strongly than spatial (key location) codes. Participants in the contralateral-effect instruction-group must have coded their left and right location responses as right and left, respectively, in order to produce a reversed Simon effect. While the Hommel (1993a) result is probably the strongest evidence in favor of intentional weighing of codes obtained so far, response instructions for both the response location and the response-effect group referred to the spatial (i.e., left/right) dimension.


[page 60↓]

However, studies that either investigated the impact of irrelevant stimulus location on non-spatially instructed responses for arbitrary S‑R mappings (Simon et al., 1976), or that studied the impact of irrelevant stimulus information that overlapped with the instructed response dimension with cross-dimensional mappings in the two-dimensional color-location task (Hasbroucq & Guiard, 1991; Smith & Brebner, 1983) have been criticized for either a lack of appropriate control conditions (i.e., the lack of a condition with spatial response instructions in the Simon et al., 1976, experiment), or for methodological reasons (i.e., within-subjects manipulations of instructions).

Therefore, the results reviewed in Chapter 3 must be considered inconclusive with respect to direct viz. spatial response coding, at least where non-spatial response instructions are concerned.

The experiments presented in Chapters 4 and 5 extend the existing findings in several regards. They are similar to the Hommel (1993a) approach in varying response instructions on otherwise identical (or at least very similar) tasks. However, unlike Hommel, response instructions in my experiments did not emphasize different spatial aspects of the response array and did not present stimulus-compatible viz. –incompatible response-effect stimuli.

More specifically, the general logic underlying the experiments was to vary response instructions for manual (left and right) keypress responses to arbitrary stimulus attributes. This was done by instructing the response keys as either left vs. right keys (spatial instructions) or as blue vs. green keys (color instructions). If participants arbitrarily code and access their responses as instructed, response instructions should (co-) determine how responding is controlled.

By using color instructions and labels I rely on the findings provided by experiments on response-effect compatibility, which demonstrate that – in principle – manual keypress responses can be arbitrarily coded and assessed. However, unlike studies on response effect compatibility, participants in my experiments were not presented with (i.e., trained on) arbitrary color effects, but instead were provided with color-labels before each trial (Chapter 4) or before each block of trials (Chapter 5).

Two different experimental approaches were used to assess whether response instructions determine response coding. Both rely on the compatibility logic outlined in Chapters 2 and 3 and used arbitrary relevant S‑R mappings in order to avoid simultaneous overlap on more than one dimension.


[page 61↓]

In one set of experiments (Experiments 1-3, Chapter 4), a dual task methodology similar to that used by Hommel (1998, Experiment 1) was employed. More specifically, in addition to a (secondary) manual keypress task with varied response instructions, participants had to perform a (primary) verbal task that either required “left” vs. “right” or “blue vs. green” concurrent verbalizations. Experiment 1 was a conceptual replication of the Hommel (1998) experiment, in which responses on both tasks were instructed in terms of left and right. This experiment served as a ‘baseline’ for Experiments 2 and 3. In Experiment 2, both verbal and manual responses were instructed in terms of color. If instructed response codes are or can be used in manual response coding, as assumed by both versions of the direct coding hypothesis, then (forward and) backward compatibility effects should generalize to the color dimension. Experiment 3 was designed to differentiate between the strong version of the direct coding hypothesis, on the one hand, and the spatial and weak direct coding hypothesis, on the other hand. This was done by again instructing manual responses in terms of color, but requiring location coding (i.e., “left” and “right” responses) on the verbal task. According to both the weak version of the direct coding hypothesis and the spatial coding hypothesis, non-spatial response instructions should not affect spatial coding of manual responses. Hence, spatial R‑R compatibility effects similar to those in Experiment 1 should be observed. In contrast, the strong version of direct coding hypothesis predicts that instructed codes are weighed more strongly. Therefore, reduced spatial inter-task effects should be obtained.

By instructing the responses on both tasks differently, Experiment 3 can also be considered an extension of the Logan and Schulkind (2000) Experiment 2 that demonstrated the importance of instructing overlapping (response-relevant) stimulus categories on both tasks.

Experiments 4-5 (Chapter 5) extend the results obtained with the dual task approach by employing the same response-instruction logic to a Simon-like task similar to that used by Hommel (1995; 1996c), in which left and right keypress responses were arbitrarily mapped to centrally presented stimuli (letter identity). Irrelevant spatial information was provided by go/no-go signals (vertical or horizontal bars) at different locations, the orientation of which indicated whether the prepared response was to be executed or not. Hence, the Experiments presented in Chapter 5 extend the Simon et al. (1976) study by directly comparing the Simon effect (i.e., the impact of irrelevant spatial stimulus information) under spatial vs. non-spatial response instructions. Whereas both the ‘spatial only’ and the weak version of the direct coding hypothesis again predict that the spatial (Simon) effect should be largely unaffected by [page 62↓]response instructions, the strong direct coding (i.e., intentional weighing) hypothesis predicts that the Simon effect should be reduced under non-spatial as compared to spatial response instructions.

In Experiments 1-3 (Chapter 4), mapping instructions were stated only once at the beginning of the experiments. Therefore, practice analyses were carried out to assess whether participants recoded instructed responses during the course of the experiments. In contrast, in Experiments 4 and 5 (Chapter 5) a procedural modification was introduced that allowed to de-confound the effects of instructions and practice present in all experiments so far (at least those that I am aware of). This was done by instructing new S‑R mappings (i.e., new letter-response pairings) on each trial.


Footnotes and Endnotes

4 A prominent alternative interpretation for untranslated color-effects states that stimuli and/or responses in the untranslated color-response task are re-coded into lexical representations, essentially transforming the untranslated color-response task into a doubly-translated word-response task (see, e.g., Hommel, submitted; Sugg & McDonald, 1994, for discussions). However, this explanation neither seems particularly plausible nor does it receive consistent empirical support.

5 The stimulus identification account proposed by Hasbroucq and Guiard (1991) is omitted here because it seems as if it has been successully rejected, both on theoretical and empirical grounds (for a comprehensive discussion, see Lu & Proctor, 1995). Moreover, it can be considered similar to the DCC hypothesis in some ways.

6 Strictly speaking, R‑R effects arising from overlap between responses on two concurrently performed, mostly arbitrarily mapped tasks, are not within the scope of explanation covered by current dual route models. However, they can easily be adapted by allowing (a) relatively fast development of direct S‑R links between conceptually dissimilar stimuli and responses with practice (e.g., Hommel & Eglau, 2002; Proctor & Lu, 1999) leading to direct response activation whenever the response-associated stimulus is presented, and/or (b) strong automatic and parallel response activation via conditional routes (e.g., Lien & Proctor, 2002; Tagliabue et al., 2000; see Hommel, 2000, for a comprehensive discussion of different types of automaticity).



© Die inhaltliche Zusammenstellung und Aufmachung dieser Publikation sowie die elektronische Verarbeitung sind urheberrechtlich geschützt. Jede Verwertung, die nicht ausdrücklich vom Urheberrechtsgesetz zugelassen ist, bedarf der vorherigen Zustimmung. Das gilt insbesondere für die Vervielfältigung, die Bearbeitung und Einspeicherung und Verarbeitung in elektronische Systeme.
DiML DTD Version 3.0Zertifizierter Dokumentenserver
der Humboldt-Universität zu Berlin
HTML generated:
02.09.2004