Running head: IDEOMOTOR COMPATIBILITY

For 30 years, Greenwald and Shulman’s (1973) psychological refractory period (PRP) study has been cited as evidence for perfect timesharing with two ideomotor (IM) compatible tasks. Recently, Lien, Proctor, and Allen (2002) failed to replicate their results and concluded that the use of two IM compatible tasks is neither necessary nor sufficient to eliminate the PRP effect. Greenwald (2003) suggested that Lien et al.’s non-replication may have been due to the use of (1) a non-IM compatible task, (2) varied trial spacing, and/or (3) inappropriate instructions. We argue that the first two factors are not critical and that instructions merely affect the criterion for speed versus accuracy. In each of his experiments, dual-task costs were evident on RT or error rates. Furthermore, the small dual-task costs in Greenwald’s study are consistent with a bottleneck model. Thus, his study does not provide evidence that IM compatible tasks enable perfect timesharing. Ideomotor Compatibility and PRP 3 Still No Evidence for Perfect Timesharing with Two Ideomotor Compatible Tasks: An Observation on Greenwald (2003) The psychological refractory period (PRP) effect, the slowing of response time (RT) for the second of two tasks at short stimulus onset asynchronies (SOAs), has often been attributed to an inability to perform central operations (e.g., response selection) for two tasks at the same time (see Lien & Proctor, 2002; Pashler, 1984; Pashler & Johnston, 1998, for reviews). Many studies have examined whether the central bottleneck can be bypassed, thus greatly reducing or eliminating the PRP effect. Results have shown that the PRP effect is remarkably robust and sometimes remains even after extensive practice (e.g., Ruthruff, Johnston, & Van Selst, 2001; Van Selst, Ruthruff, & Johnston, 1999) or with highly compatible stimulus and response sets (e.g., Brebner, 1977; Smith, 1967). In one important exception, however, Greenwald and Shulman (1973) argued that central operations can be bypassed and the PRP effect can be eliminated (i.e., allowing perfect timesharing) with two ideomotor (IM) compatible tasks. They defined IM compatibility as situations in which the “stimulus resembles sensory feedback from the response” (p. 70). According to their IM compatibility theory, response codes for the IM compatible tasks can be activated directly and thus bypass the limited-capacity central stage. Lien, Proctor, and Allen (2002) In the three decades since Greenwald and Shulman’s (1973) study, their conclusion has been widely cited, but rarely questioned. Lien, Proctor, and Allen (2002), however, recently questioned Greenwald and Shulman’s conclusion for three major reasons. First, Greenwald and Shulman’s conclusion was oversimplified. When RTs were averaged over the two tasks, the PRP effect was relatively small. However, when the PRP effect was measured in the standard way (short-SOA Task-2 RT minus long-SOA Task-2 RT), as in most PRP studies, results from Greenwald and Shulman’s two experiments were actually in conflict: A significant PRP effect of Ideomotor Compatibility and PRP 4 89 ms was observed with two IM tasks in Experiment 1 but little or no PRP effect was observed in Experiment 2. Greenwald and Shulman identified the instructions, “most often the 2 signals [S1 and S2] on each trial would be simultaneous” (p. 73), as being the crucial methodological factor differentiating their Experiment 2 from Experiment 1. However, their final conclusion that, “the PRP effect is eliminated when both tasks are IM compatible” (p. 74), did not acknowledge the importance of particular instructions. Second, Lien et al. failed to replicate Greenwald and Shulman’s results. All four of Lien et al.’s experiments showed a significant PRP effect with two IM compatible tasks, even when the method of Greenwald and Shulman’s Experiment 2 was directly replicated. Third, Greenwald and Shulman’s IM compatibility theory is inconsistent with the finding that the PRP effect is evident when only one of the two tasks is IM compatible. Greenwald and Shulman proposed that dual-task interference is primarily due to the overloading of central operations and that these central operations are bypassed for IM compatible tasks. If so, the IM compatibility theory seems to imply that the PRP effect should be absent whenever either task alone is IM compatible, even if the other is not. Contrary to this implication, Greenwald and Shulman’s own experiments, as well as Lien et al.’s replications, unambiguously showed a PRP effect when one task was IM compatible and the other was not. Greenwald (2003) Recently, Greenwald (2003) reiterated his position that perfect timesharing does occur with two IM compatible tasks. He suggested that Lien et al.’s (2002) non-replication of Greenwald and Shulman (1973) might be due to one or more of three differences in procedure. He also conducted two new experiments, which he interpreted as evidence for one of the procedural differences being crucial for enabling perfect timesharing. Below, we first describe the three procedural differences cited by Greenwald and explain why it is unlikely that they Ideomotor Compatibility and PRP 5 played a critical role in Lien et al.’s non-replication. We then discuss the results of Greenwald’s two new experiments and argue that they still do not provide strong evidence for “perfect timesharing” with two IM compatible tasks. Three Procedural Differences Between Greenwald and Shulman (1973) and Lien et al. (2002) Greenwald (2003) contended that three procedural details might explain the deviations from perfect timesharing observed in Lien et al.’s (2002) experiments. First, he argued that Lien et al.’s task – a left/right movement of the joystick to a left/right arrow – may not have been IM compatible. Because participants grasped the joystick handle with their dominant hand and placed their other hand on the base of the joystick to stabilize it, he argued that, “the nondominant hand’s role in this coordination opposed the IM-compatible direction-plus-position cue”. We argue, however, that response selection for the joystick movement in Lien et al.’s study involved selecting a left-right action made by the dominant hand; the non-dominant hand merely prevented any movement of the apparatus. Studies of stimulus-response compatibility effects have consistently found that responses are coded in terms of response goals (e.g., move the joystick left or right), defined by task instructions, rather than effectors, such as hands (e.g., Guiard, 1983; Hommel, 1993). Thus, regardless of whether, or how, the two hands were coordinated in Lien et al.’s study, the response goal was simply to move the joystick to the direction that corresponded to the arrow direction/position. Moreover, when Lien, McCann, Ruthruff, and Proctor (2003) used an immoveable joystick, so that only the dominant hand response was involved, a substantial PRP effect with two IM compatible tasks was still found. Hence, Lien et al.’s (2002) non-replication was not due to their participants’ stabilizing the joystick with the non-dominant hand. Second, Greenwald (2003) claimed that, “unlike other studies that have obtained perfect Ideomotor Compatibility and PRP 6 timesharing, LP&A’s [Lien, Proctor, and Allen’s] procedure did not use regularly spaced trials” (p. 6), which “may have prevented their subjects from preparing optimally for the timeshared task combinations” (p. 7). However, we see no reason why it is necessary to use a fixed time between stimulus presentations. In fact, this procedure has the potentially negative consequence that a relatively slow response on one trial will result in a relatively short preparation interval for the next trial. Arguably, it makes more sense to use a constant response-stimulus-interval (RSI), so that preparation time will be constant. Indeed, the latter procedure has been adopted by the overwhelming majority of modern cognitive psychology experiments, including Lien et al. (2002), where a constant RSI of 2 seconds was used. Third, Greenwald (2003) objected that the instructions used by Lien et al. (2002) were different from those of Greenwald and Shulman (1973). Greenwald and Shulman contended that the crucial factor distinguishing their Experiment 2 from Experiment 1 was the instructions “The Ss [subjects] were informed that most often the 2 signals on each trial would be simultaneous and were not given any expectation that one signal might reliably precede the other” (p. 73). Lien et al. were aware of this point and thus attempted to duplicate Greenwald and Shulman’s instructions as closely as possible. Nevertheless, Greenwald suggested that Lien et al. omitted a crucial instruction that was not mentioned in Greenwald and Shulman’s methods: “To the best of the author’s memory, however, G&S’s instructions for their Experiment 2 not only stressed the simultaneous occurrence of the stimuli but also encouraged subjects to respond both rapidly and simultaneously to the simultaneous stimuli” (p. 7). Given that these instructions were not mentioned in the original article, Lien et al. can hardly be held accountable for failing to replicate them. Moreover, we argue that Greenwald’s new experiments, summarized below, do not provide strong evidence that these instructions produce perfect timesharing. Ideomotor Compatibility and PRP 7 Greenwald’s (2003) New Experiments To further support his IM compatibility theory, Greenwald (2003) presented two new experiments, each of which used two IM compatible tasks. Experiment 1 included single-task blocks, 0-ms SOA dual-task blocks, and 1000-ms SOA dual-task blocks with a fixed trial spacing (2 seconds for single-task blocks and 0-ms SOA dual-task blocks; 3 seconds for 1000-ms SOA dual-task blocks). To demonstrate the importance of instructions, he compared the Greenwald and Shulman (GS) condition, which stressed speed and simultaneous responding (i.e., “YOU ARE TO MAKE TWO RESPONSES AT THE SAME TIME”), with the Lien, Proctor, and Allen (LPA) condition, which emphasized speed and accuracy equally (i.e., “respond to each task as quickly and accurately as you can”). We have summarized his results in Table 1. In both instruction conditions, RTs were slower in the 0-ms SOA dual-task blocks than in the single-task blocks. Comparing the performance between the 0-ms and 1000-ms SOA dual-task blocks, however, Greenwald claimed that results from the GS condition replicated Greenwald and Shulman’s (1973) earlier finding of “perfect timesharing”, and results from the LPA condition replicated Lien et al.’s (2002) finding of a substantial PRP effect. Greenwald (2003) further argued that neither the pure single-task blocks nor the 1000-ms SOA dual-task blocks in Experiment 1 constitute appropriate control conditions for measuring dual-task interference. He contended that the extra response preparation for the dual-task trials compared to the single-task trials “constitutes a burden that will increase latencies”, and that the switching from Task 1 to Task 2 in the 1000-ms SOA dual-task blocks would “produce slowed responding on the second task” (for an opposite view, see Lien, Schweickert, & Proctor, 2003). In Experiment 2, therefore, he used a mixed-task control condition, in which the two kinds of single-task trials (Task 1 alone or Task 2 alone) were mixed randomly within blocks. In this Ideomotor Compatibility and PRP 8 control condition, each relevant stimulus was accompanied by an irrelevant “accessory” stimulus in the other modality (a click accompanied the visual Task 1 and three asterisks accompanied the auditory Task 2). Experiment 2 adopted (approximately) the GS instructions of Experiment 1 and included conditions where two non-IM compatible tasks were used. For the two IMcompatible tasks condition, RTs were faster in the 0-ms SOA dual-task blocks (which Greenwald called “timeshared blocks”) than in the mixed-task control blocks. For the two non-IM compatible tasks condition, however, RTs were slower in the 0-ms SOA dual-task blocks than in the mixed-task control blocks. Greenwald concluded that these results confirm perfect timesharing with two IM compatible tasks. We argue that these two experiments still do not provide strong evidence for perfect timesharing with two IM compatible tasks for reasons to be discussed below. First, although Greenwald (2003) suggested that proper instructions are the key to enabling perfect timesharing, we argue that changing instructions merely shifts speed-accuracy criteria. Second, although the PRP effect was absent on RT in the GS condition of Experiment 1, a PRP effect was present on error rates. Finally, we argue that relatively small PRP effects do not necessarily indicate that the central bottleneck has been bypassed. Are Instructions the Key that Unlocks the Door to Perfect Timesharing? Greenwald (2003) suggested that specific instructions are necessary to enable perfect timesharing. We argue, however, that the instruction differences between the GS and LPA conditions might merely have shifted speed-accuracy criteria. At the 0-ms SOA dual-task blocks, for instance, Task 1 RT was faster in the GS condition (291 ms) than in the LPA condition (350 ms), but, the error rate was 4 times higher in the GS condition (3.57%) than in the LPA condition (0.88%). Similarly, Task 2 RT was faster in the GS condition (410 ms) than in the LPA condition (534 Ideomotor Compatibility and PRP 9 ms), but, the error rate was much higher in the GS condition (13.41%) than in the LPA condition (10.96%). The likely presence of a speed-accuracy criterion shift between the GS and LPA conditions at the 0-ms SOA dual-task blocks has two implications. First, if there is a bottleneck, then the speedup in RTs should reduce the PRP effect (see below for detailed discussion of this point). Thus, the corresponding reduction of the PRP effect in the GS condition is not surprising. Second, when speed is emphasized, participants might attempt to maintain a constant speed across blocks, causing any differences in difficulty between blocks to show up primarily on error rate. This situation is opposite to what is normally assumed for the traditional RT paradigm, where it is assumed that participants try to maintain a constant error rate across conditions, and the effect of processing difficulty show up primarily on RT (see Pachella, 1974; Wickelgren, 1977). Was the PRP Effect Eliminated? In Greenwald’s (2003) Experiment 1, the GS instructions in the 0-ms SOA dual-task blocks emphasized speed and simultaneous responding, which might have led to the understanding that accuracy is not important. Furthermore, only one of the two responses was collected on each trial in the 0-ms SOA dual-task blocks (due to software restrictions, see his footnote 4), and then the feedback message “error” was based only on that response. Thus, when participants made an error to one task, they often did not receive the error message. The frequent absence of error feedback, combined with strong speed emphasis, might have led participants to respond quickly at the cost of accuracy. Note that this software restriction did not affect the 1000-ms SOA blocks, where feedback was based on both responses within a trial. In addition, the GS instructions emphasized simultaneous responding for the two tasks in the 0-ms, but not the 1000-ms, SOA dual-task blocks; these instructions Ideomotor Compatibility and PRP 10 might have further increased the emphasis on speed. These arguments suggest that the participants’ bias toward response speed was higher in the 0-ms SOA dual-task blocks than in the 1000-ms SOA dual-task blocks. Given the likelihood of a speed-accuracy tradeoff between blocks, both RT and error data should be considered in determining whether the PRP effect has been eliminated with two IM compatible tasks. Although there was no PRP effect on RT in the GS condition of Experiment 1, Task 1 error rate was 3 times higher in the 0-ms SOA dual-task blocks (3.57%) than in the 1000ms SOA dual-task blocks (1.17%). Similarly, Task 2 error rate was much higher in the 0-ms SOA dual-task blocks (13.41%) than in the 1000-ms SOA dual-task blocks (10.31%). Even with Greenwald and Shulman’s (1973) measurement of the PRP effect, averaging the data for Task 1 and Task 2, the PRP effect on error rate was statistically significant in the GS condition. Greenwald (2003) downplayed these effects by noting that they were smaller and non-significant early in the experiment. Nevertheless, the error rate was higher in the 0-ms SOA dual-task blocks than in the 1000-ms SOA dual-task blocks for all four phases of the experiment (see his Table 1). A PRP effect on error rates might also have occurred in Experiment 2 of Greenwald (2003). Because of the lack of detailed error data, however, we are not able to evaluate this possibility. Furthermore, in contrast to Greenwald’s view, we argue that his mixed-task blocks do not provide an appropriate control condition for measuring dual-task interference. In his mixed-task blocks, there was uncertainty prior to each trial about what task would need to be performed. Also, because stimuli were presented in both modalities in the mixed-task blocks, participants had to determine which stimulus was relevant and which one was irrelevant and inhibition of the irrelevant stimulus might have been needed. Meanwhile, there was no task Ideomotor Compatibility and PRP 11 uncertainty in the dual-task blocks and no need to inhibit an irrelevant stimulus. These advantages might have cancelled out a real disadvantage of having to perform two tasks at the same time. In fact, if one instead compares the dual-task blocks to the single-task blocks, there was significant dual-task interference for the visual-manual IM compatible task in Experiment 2. Consequently, neither experiment in Greenwald’s study provides unambiguous support for “perfect timesharing” with two IM compatible tasks. Was the Central Bottleneck Bypassed? Do the small dual-task costs observed by Greenwald (2003) support the key assumption of IM compatibility theory that “the IM compatible tasks could bypass a limited-capacity response selection process [central bottleneck]” (p. 3; see also Greenwald & Shulman, 1973)? It has previously been noted that small dual-task costs can, under some circumstances, be obtained even when the central bottleneck still exists (e.g., Byrne & Anderson, 2001; Ruthruff, Johnston, Van Selst, Whitsell, & Remington, 2003). According to the central bottleneck model, the predicted PRP effect at the 0-ms SOA is given by 1A+1B-2A (where 1A and 1B refer to the perceptual and central processing stage of Task 1, respectively, and 2A refers to the perceptual processing stage of Task 2; Van Selst et al., 1999). Thus, there are two conditions leading to little or no PRP effect: (a) short stage durations for 1A and 1B, and (b) a long stage duration for 2A. Short stage durations for 1A and 1B are likely to occur when Task 1 is easy (e.g., with IM compatible tasks), as reflected in short RTs. Previous studies have provided evidence that when Task 1 is sufficiently short, the central bottleneck can become “latent”, producing little or no interference (Van Selst et al., 1999; Ruthruff et al., 2003). In addition, a small PRP effect is likely to occur when stimulus identification for Task 2 is timeconsuming, resulting in a relatively long duration for stage 2A. Both of the conditions (short 1A and 1B; long 2A) that enable the bottleneck to produce a Ideomotor Compatibility and PRP 12 small PRP effect seem to be present in the GS condition of Greenwald (2003). As a concrete example, the bottleneck model could explain the small PRP effect as follows. In the 1000-ms SOA dual-task blocks, the visual-manual task was presented as Task 1 and the auditory-vocal task as Task 2. Given that RTs were much faster for the visual-manual task (291 ms) than the auditory-vocal task (410 ms) in the 0-ms SOA dual-task blocks, it is reasonable to assume that participants performed central operations for two tasks in the same order as in the 1000-ms SOA dual-task blocks. Suppose, as seems reasonable, that the average combined duration of stages 1A and 1B is about 180 ms (leaving 111 ms for response execution). Further, suppose that the duration of stage 2A is about 200 ms. This estimate seems reasonable given that it took 200 ms to fully present the auditory sound (“A” or “B”) and that mean Task 2 RT was 410 ms. Under these durations, stage 1B would tend to finish before stage 2A; thus, this bottleneck stages would usually not conflict and little PRP effect should be expected. Although one could argue about the exact values of the relevant stage durations, it is clear that under certain plausible conditions little PRP effect would occur. Given that a plausible bottleneck model can account for Greenwald’s data, one cannot conclude that the bottleneck was bypassed, or in Greenwald’s words, “evaded or minimized” (p. 3), with two IM compatible tasks. Conclusions We argue that two of the three procedural differences (non-IM compatible task and variable trial spacing) suggested by Greenwald (2003) are unlikely to be responsible for Lien et al.’s (2002) non-replication. In fact, Greenwald’s Experiment 1 used two-IM compatible tasks and fixed trial spacing, yet confirmed Lien et al.’s finding of a PRP effect with the LPA instructions. This finding also provides converging evidence that the use of two IM compatible tasks is not sufficient to eliminate the PRP effect. Even Greenwald agrees that perfect Ideomotor Compatibility and PRP 13 timesharing does not occur unless the instructions stress speed and simultaneous responding. We agree that instructions matter, but only because they cause a shift in the criterion for speed versus accuracy. Furthermore, we argue that the small PRP effects in Greenwald’s study do not necessarily indicate that the central bottleneck was bypassed with two IM compatible tasks. We describe how a simple and plausible central bottleneck model can easily explain small PRP effects when Task 1 RT is unusually short. Some insight into the present debate is provided by Greenwald, Pratkanis, Leippe, and Baumgardner’s (1986) arguments against theory-based research strategies as contrasted with result-centered research strategies. They argued, “Confirmation bias is an expectable product of theory-centered research strategies” (p. 216), noting that “researchers display confirmation bias when they persevere by revising procedures until obtaining a theory-predicted result” (p. 216). According to Greenwald et al., “This strategy produces findings that are overgeneralized in avoidable ways” (p. 216), because theorists overlook the experimental modifications needed to produce the theory-predicted result. Consequently, they noted, “Although the conclusions from such research need to be qualified by reference to the tried-and-abandoned procedures, those conclusions are often stated only in the more general terms of the guiding theory” (p. 220). We contend that overgeneralization with respect to stating conclusions in the general terms of IM compatibility theory occurred with respect to Greenwald and Shulman’s (1973) original findings. Such overgeneralization will continue to occur if the central message of Greenwald’s (2003) study, as implied in the Conclusion of his article, is that the “perfect timesharing” prediction of IM compatibility theory has been confirmed. IM compatibility is only a subset of the conditions that reduce dual-task interference, quite possibly for reasons very different from those suggested by IM compatibility theory. Ideomotor Compatibility and PRP 14