Experiments with two stimuli S¹and S² and two responses suggest the existence of a stage of processing that cannot be shared between two concurrent tasks. Wide-spread support has been found for the hypothesis that response selection for Task₂ is postponed when the S¹ to S² stimulus onset asynchrony (SOA) is short (Pashler, 1994a). At short SOAs, manipulations which impact Task₂ processing prior to response selection (e.g., degradation of stimulus quality) have little effect on Task₂ response times (RTs). On the other hand, manipulations which are thought to impact response selection or execution (e.g., Stroop interference) always impact Task₂ RTs. There is, however, one particularly compelling demonstration that appears to be inconsistent with the response selection bottleneck hypothesis: Karlin and Kestenbaum (1968) report that the RT difference between detection (i.e., 1-choice) and 2-choice discrimination dramatically decreases with decreasing SOA. Given that the primary difference between detection and discrimination is believed to be at response selection, their result may indicate a processing bottleneck at response execution (Keele, 1973). We fail to replicate the Karlin and Kestenbaum result in two substantive replications of Karlin and Kestenbaum's tasks and procedures. In the single experiment in which Karlin and Kestenbaum's result is replicated, a simple response execution bottleneck account is ruled out by the stability of the difference between 2-choice and 3-choice discrimination times across SOA. Two additional experiments demonstrate that response preparation and task strategy do not substantially contribute to the attenuation of response selection-level effects with decreasing SOA.

When human observers are asked to perform two simple tasks in rapid succession, one often observes a striking cost that cost grows larger as time interval between the two tasks is made shorter. The possibility that the cost on second task performance reflects a fundamental limitation of human information processing continues to sustain interest in understanding dual-task interference. Such research is interesting both for its theoretical import and for its contribution to the pragmatics of designing time-critical man-machine systems, such as high-speed aircraft.

In this article we examine the role of some variables thought to affect response selection in simple dual-task experimental situations. In particular, we examine the effects of varying the number of response alternatives in the second of two tasks. This empirical issue has been at the center of some controversy concerning the nature of dual-task interference. A clarification is important because it is highly relevant to a class of models (postponement models) that has been considered by many (e.g., see Welford, 1980; Bertelson, 1966; Smith, 1967a; Pashler, 1994a,b, for reviews) as the most promising class of models for PRP phenomena (see Meyer, Kieras, Lauber, Schumacher, Glass, Zurbriggen, Gmeindl, & Apfelblat, 1995, for an opposing view). In the following sections, we explain why this issue is central to theorizing about dual-task interference and we follow these sections with a series of experiments that address the issue.

Consider two simple tasks, Task¹ and Task², each of which requires a separate speeded response to a separate stimulus (e.g., a tone, S¹ , followed by the visual presentation of a letter, S²). The time taken to respond to Task² (RT_{2}) is virtually always longer than if Task² had been performed in isolation. Of greater interest, RT² increases sharply as the stimulus onset asynchrony (SOA) between S¹ and S² is decreased. This SOA-dependent slowing, which is evident in all of the RT² results shown in Figure₁, is often referred to as the PRP effect (psychological refractory period; Telford, 1931; Craik, 1947), or dual-task interference (Pashler, 1994a). The interaction between the SOA manipulation and other variables will be addressed later.

The most influential accounts of the PRP effect have been postponement accounts. The key feature of postponement accounts is the postulation of a processing bottleneck that prevents concurrent processing for two different tasks at one or more stages in the processing sequence (Bertelson, 1966; Pashler, 1994a; Smith, 1967a; Welford, 1952, 1959, 1980). The introduction of a delay in processing between early Task² processing and Task² bottleneck processing is represented in Figure 2 by the horizontal dashed lines in the short SOA condition. Processing for Task² resumes when the bottleneck stage becomes available (when it is no longer busy with Task¹). Thus, Task² is slowed by the length of time processing is delayed while waiting for the bottleneck to be freed from Task¹ processing.

Pre-bottleneck processing differences may be absorbed into the slack time while continuation of Task² processing waits for the bottleneck stage to become available. At the short SOA, RT² will be modulated by Task¹ bottleneck processing rather than Task² pre-bottleneck processing. One consequence of this shift is that manipulations that affect Task² pre-bottleneck processing will have a reduced impact on RT². Figure₂ illustrates the decoupling of pre-bottleneck processing differences on RT² at short SOAs. Prolongation of a pre-bottleneck stage of processing, (represented by a shaded extension of stage 2A Figure₂), has no impact on RT². However, at long SOAs, this variable will have have its full effect because there is no period of waiting for a busy bottleneck stage. Thus, in a factorial experiment with a pre-bottleneck manipulation crossed with SOA, we expect an interaction in which the pre-bottleneck manipulation will have a larger effect at longer SOAs than at shorter SOAs (see Figure₁, upper panel).

Changes in the duration of the bottleneck stage itself, or of post-bottleneck stages will not be absorbed into the slack time. In Figure₂, the effect of a post-bottleneck processing manipulation is represented by the addition of the shaded box following the bottleneck stage. Post-bottleneck processing differences will always be fully reflected in RT², even at the short SOA. Thus, we expect additive effects of such a manipulation in combination with SOA in a factorial design experiment (see Figure₁, middle panel).

A wide range of manipulations thought to affect perceptual encoding (PE) have a decreasing influence with decreasing SOA. For instance, as is illustrated in the upper panel of Figure₁, Pashler and Johnston (1989) demonstrate that reducing the contrast between a visual stimulus and the background has a moderate effect on letter identification RT²s at the long SOA but a very small effect at the short SOA. De Jong (1993) and Van Selst and Jolicoeur (1994a) provide similar demonstrations. The effect of shape distortion on letter identification (``A'' versus ``H'') is also eliminated with decreasing SOA (Johnston, McCann, Remington, 1995). These findings suggest a processing bottleneck that occurs after PE (the presumed locus of effect of the contrast reduction manipulation; Sternberg, 1969) and after stimulus classification (the presumed locus of effect of the letter distortion manipulation).

Furthermore, several experimental manipulations thought to affect the difficulty of response selection are not affected by SOA. As is illustrated in Figure₁, the amount of Stroop interference does not vary as a function of SOA (Fagot Pashler, 1992). This demonstration is important because Stroop interference is thought to reflect difficulties in response selection (MacLeod, 1991). Other manipulations, which are also thought to impact response selection, are similarly unaffected by SOA (e.g., high versus low S-R compatibility: Broadbent Gregory, 1967; McCann Johnston, 1992; mirror versus normal version of alphanumeric characters in a mirror-normal task: Van Selst Jolicoeur, 1994a). These findings provide strong evidence consistent with a processing bottleneck at, or before, the response selection stage.

Although the evidence presented so far suggests a bottleneck at response selection, other evidence has led some researchers to argue for a postponement model with a bottleneck at response execution (RE) (Keele, 1973; Keele Neill, 1978; Logan Burkell, 1986; McLeod, 1977b; Netick Klapp, 1994). The strong version of the RE bottleneck hypothesis is that there is only one bottleneck in processing and that the bottleneck is at RE (Keele, 1973; Keele Neill, 1978).

The most compelling empirical evidence in favor of the strong RE bottleneck hypothesis is Karlin and Kestenbaum's (1968) report that the difference in RT² between simple RT (SRT) and two-alternative forced-choice (2AFC) conditions decreases with decreasing SOA. Their critical results are presented in the bottom panel of Figure₁. Task¹ was a 2AFC visual digit discrimination (the digit ``1'' or ``2''). Task² was a SRT or 2AFC tone discrimination (600 or 3000 Hz tones presented for 35 ms). Across the SOAs (out to 1150 ms) the 2AFC-SRT RT² difference decreased from 81 ms to 27 ms. This result is important because it is believed that the 2AFC-SRT RT difference largely reflects differences at RS (Donders, 1868; Fletcher Rabbit, 1978; Sternberg, 1969).

The Karlin and Kestenbaum result appears squarely inconsistent with many of the previously discussed investigations of dual-task slowing. On the one hand, the effects of variables believed to influence the duration of RS have generally been found to be additive with SOA (e.g., Pashler Johnston, 1989), suggesting a bottleneck at (or before) RS. In addition, large PRP effects obtain for both go and no-go trials when Task¹ is a go/no-go task (Bertelson and Tisseyre, 1969; De Jong, 1993), and for those trials in which Task¹ is a catch trial (Comstock, 1973; De Jong, 1993; Gottsdanker, 1979; Keele, 1973). None of these findings are consistent with a processing bottleneck at RE.

A resolution of this empirical discrepancy is important because it bears directly on the viability of postponement accounts of dual-task slowing, and particularly on the hypothesis that there is a bottleneck at RS. The principal goal of this article is to reconcile these apparently inconsistent results and to examine the concomitant theoretical implications for accounts of dual-task slowing.

Figure 2. Generic postponement model. Timing diagram of how information processing for temporally overlapping simple tasks is hypothesized to be affected by manipulations of SOA under dual-task situations. Task1 processing is always unaffected but processing for Task2 may be delayed if the bottleneck stage has not yet been completed for Task1 before it is required for Task2. (RT1 = response time for Task1; see text for further details).

Experiment₁ is a replication and extension of the SRT and 2AFC Task² conditions reported by Karlin and Kestenbaum (Figure 1 top panel). Task¹ is a 2AFC tone-frequency discrimination. Task² is a SRT, 2AFC, or 3AFC color discrimination. The inclusion of the 3AFC condition allows us to make more precise predictions based on bottleneck models that postulate a bottleneck either at response selection (RS) or at response execution (RE).

Suppose there is only a bottleneck at RE. The RT² differences between SRT, 2AFC, and 3AFC tasks is thought to primarily reflect differences at RS. Since RS precedes the bottleneck at RE, the difference between 2AFC and SRT should decrease as SOA is shortened (Karlin Kestenbaum's principal result). The difference between 3AFC and 2AFC should also decrease.

If, on the other hand, the only bottleneck is at response selection (RS), (Pashler Johnston, 1989), 2AFC-SRT X SOA should be additive. Furthermore, 3AFC-2AFC X SOA should also be additive. This prediction is based on the assumption that manipulations of the number of Task² response alternatives (SRT vs 2AFC vs 3AFC) affects only the duration of the RS stage. There are, however, several reasons why RS-bottleneck models might predict underadditivity. Detection (SRT) and discrimination (2AFC, 3AFC) require different amounts of perceptual processing (Fletcher Rabbit, 1978; Sternberg, 1969; Van Selst, 1995). Van Selst (1995) demonstrated a larger RT² difference at the long SOA across stimulus contrast for 2AFC and 3AFC letter discrimination (47 and 50 ms respectively) than for an analogous SRT task (26 ms). These differences are consistent with the hypothesis that the stimuli must undergo more PE² processing when there is a need to discriminate between 2 or 3 patterns than when a discrimination is not required (i.e., when detection is sufficient). We refer to this possibility as the ``perceptual hypothesis.'' This hypothesis leads us to expect some attenuation (only partial) of the 2AFC-SRT RT² difference with decreasing SOA, but to expect the 3AFC-2AFC RT² difference to remain fairly stable across SOA.

The same predictions can be derived from De Jong's (1993) ``multiple-bottleneck'' model, illustrated in Figure 3. De Jong postulates a processing bottleneck at RS and a refractory period at RE. The RS processing bottleneck constrains processing as described earlier. The refractory period at RE imposes a further limit in the system. According to De Jong (1993), after the initiation of one motor movement, there is a motor refractory period that prevents the initiation of another motor movement. The motor refractory period is thought be about 200 ms (De Jong, 1993; Kahneman, 1973). Given this refractory period, RE² may be delayed if it occurs in close temporal proximity to RE¹. The only manipulations that affect RS that will be influenced by the RE bottleneck are for those trials in which PE² and RS² are extremely short. On these trials, RE² is attempted during the refractory period following RE¹. In the present experiment, only the SRT condition is likely to be influenced by the RE bottleneck. If SRT RS² is fast enough, the initiation of RE² may be limited by the motor refractory period following RE¹. In the 2AFC condition, however, the increased duration of RS² should delay the onset of RE² such that it no longer coincides with the refractory period following RE¹. In this experiment, the multiple-bottleneck hypothesis leads to the prediction that, due to the differential influence of the motor refractory period on SRT vs 2AFC RT², the 2AFC-SRT RT² difference should attenuate with decreasing SOA. In contrast, the 3AFC-2AFC RT² difference should be unaffected by SOA. Thus, De Jong's (1993) multiple-bottleneck hypothesis may be able to account for both the additivity of many manipulations affecting response selection and the interaction between the 2AFC-SRT difference and SOA (reported by Karlin and Kestenbaum).

There are two additional hypotheses that predict the 2AFC-SRT difference to decrease with decreasing SOA, and the 3AFC-2AFC difference to remain the same across SOA. McCann and Johnston (1992) argue that the time required for RE for a known response decreases with increased preparation time. As SRT preparation time is increased by lengthening the SOA between S¹ and S², RE² for the SRT condition will decrease. RE² for the 2AFC and 3AFC conditions will be unaffected by SOA because no particular response can be prepared given that the response is not known until S² is presented. Thus the ``response preparation hypothesis'' leads to the prediction that, as SOA is increased, the 3AFC-2AFC RT² difference will remain constant, whereas the 2AFC-SRT RT² difference will increase because SRT responses become increasingly well-prepared as time passes.

Another non postponement account of the 2AFC-SRT X SOA interaction is the ``anticipation hypothesis.'' According to this hypothesis, subjects in the SRT condition sometimes anticipate S<sup>2</sup> and respond when they expect to see it rather than in response to it. Let us call such responses ``anticipations.'' Anticipations are hypothesized to be more likely as SOA is increased, because the conditional probability that S² will occur at a particular time increases as time goes on (given that we, as did Karlin Kestenbaum, use a fix set of SOAs). Anticipations will tend to reduce the mean RT². Because the frequency of anticipations increases as SOA increases, the mean RT² in the SRT condition becomes increasingly small as SOA increases and this produces a divergence of the SRT and 2AFC functions.

Because the response is not known in the 2AFC and 3AFC conditions, anticipations are not likely. If they occur, they often lead to errors and these RTs are not included in the mean RT² for these conditions. Thus, the impact of anticipations on RT² is likely to be larger in the SRT condition than in either the 2AFC or 3AFC condition.

The anticipation hypothesis will be evaluated using analyses in which we categorize errors as either anticipation or production errors. Neither Karlin and Kestenbaum (1968) nor De Jong (1993) provide a sufficiently detailed breakdown of error rates across conditions to allow us to evaluate the potential impact of anticipations on their results.

Subjects
Twenty-four undergraduates (12 males) at the University of Waterloo were recruited from a paid subject pool. Ages ranged from 19 to₂4 (Median = 20). All subjects were naive as to the purpose of the experiment and reported normal or corrected-to-normal vision.

Stimuli and Apparatus
The stimulus for Task¹ was a pure tone of 300 or 900 Hz presented for 100 ms. The stimulus for Task² was a square color patch, 2.4\D X 2.4\D visual angle (V.A.), of red, green, or blue, presented at the center of the computer screen for an unlimited duration. The background was gray. Two button boxes (labeled ``high'' and ``low'') were used to collect tone responses and three button boxes (with appropriately colored stickers) were used to collect color responses. Both the tone and color patch for each trial were presented using an Amiga 1020 color monitor controlled by an Amiga 1000 microcomputer.

Procedure
Subjects were aware that both RTs and error rates were recorded and that it was important to be both quick and accurate. Subjects responded to the tone with their left hand, and to the color patch with their right hand.

RTs, production errors, and anticipation errors for Task¹ and Task² were examined separately using ANOVAs in which number of Task² response alternatives (SRT, 2AFC, 3AFC) and SOA (100, 400, 750, 1100 ms) were within-subject factors. Following Karlin and Kestenbaum (1968), all trials associated with responses within 100 ms of the onset of the task-relevant stimulus were treated as anticipation errors. Production errors consisted of those non-anticipation trials in which the subject chose an incorrect response alternative (i.e., pushed the wrong button or failed to respond to the tasks in the correct order). RTs for any trial with an error (production or anticipation) on either task were not included in the RT analyses.

The most salient feature of the results is the large RT² increase with decreasing SOA, F(2, 41)=193.6, p<.001. Increases in the number or response alternatives also led to an increase in RT², F(2, 46)=40.4, p<.001.

Error rates
The sharp increase in the number of anticipation errors at the longer SOAs in the SRT condition was responsible for the highly significant interaction between number of Task² response alternatives (SRT, 2AFC, 3AFC) and SOA, F(3,₁38)=56.9, p<.001. There were no production error effects, all Fs <1.27, all ps >.29.

The Tone Task (Task¹)

Response Times
Figure 5 shows the mean RT¹ for each SOA and number of Task² response alternatives collapsed across Task¹ response. There was an increase in RT¹ both with increasing number of Task² response alternatives, F(2, 46)=13.5, p<.001, and with decreasing SOA, F(3, 69)=4.0, p<.02. The interaction between number of Task² response alternatives and SOA, F(6,₁38)=3.9, p<.002, reflects the decrease in the RT¹ difference across number of Task² alternatives with decreasing SOA.

Error Rates
There was a main effect of number of Task² alternatives on Task¹ production errors, F(2, 46)=12.8, p<.001. More Task¹ production errors were made in the SRT condition than in the 2AFC and 3AFC conditions, F(1,₂3)=16.7, p<.001 (see Figure 5). Task¹ production errors generally increased with decreasing SOA between the presentation of the tone and the color patch, producing a main effect of SOA, F(3, 69)=4.5, p<.01. There were no other important production error effects, and only a minimal number of Task¹ anticipations (less than 0.3% in any cell).

We consider here three results that seem the most important. Firstly, the 3AFC-2AFC RT² difference was constant across SOA. If a processing bottleneck at RE was responsible for the PRP effect, the 3AFC-2AFC difference would have decreased with decreasing SOA. Thus, these results rule out models with a single bottleneck at RE.

Secondly, the 2AFC-SRT difference decreased as SOA was reduced. However, the rate of anticipations in the SRT condition increased sharply as SOA was increased. Thus, the pattern of results is consistent with the predictions of the anticipation hypothesis.

Thirdly, the 121.5 ms residual 2AFC-SRT RT² difference at the shortest SOA is substantially larger than that reported by Karlin and Kestenbaum (27 ms). It is also larger than the 58 ms 2AFC-SRT residual RT² difference reported by De Jong (1993, Experiment₂). The large residual RT² difference is consistent with the possibility that a substantial portion of the decrease in the 2AFC-SRT RT² difference with decreasing SOA could have been produced by the attenuation of SRT versus 2AFC processing differences at PE or at stimulus categorization (Johnston, McCann, Remington, 1995). Thus, the pattern of results is consistent with the perceptual hypothesis.

The anticipation hypothesis, the response preparation hypothesis (McCann Johnston, 1992), and the perceptual hypothesis may account for the observed attenuation of the 2AFC-SRT RT² difference in Experiment₁. These hypotheses are considered further in subsequent experiments.

EXPERIMENT 2

The results of Experiment₁ are consistent with many models postulating a general processing bottleneck at RS in which SRT processing introduces some unusual aspect to the operation of the system (e.g., by allowing for increased response preparation with increasing SOA for the SRT trials [McCann Johnston, 1992], by increasing the bias to produce ``anticipation'' responses, or by allowing a secondary bottleneck to affect performance on SRT trials [De Jong, 1993]). This type of interpretation contrasts sharply with those based on a straightforward `postponement' analysis (Keele, 1973; Pashler, 1989; Schweickert Boggs, 1984) of Karlin and Kestenbaum's critical findings. To address the possibility that differences in experimental procedure were responsible for the empirical differences between our results and those reported by Karlin and Kestenbaum (1968), a close replication of Karlin and Kestenbaum's critical conditions was undertaken.

In Experiment₁ Task¹ was a tone discrimination and Task² was a color discrimination. Karlin and Kestenbaum (1968), meanwhile, used a visual 2AFC Task¹, and a SRT or 2AFC auditory Task². In Experiment₂, Karlin and Kestenbaum's tasks are used.

Subjects
Three female volunteers and one of the authors (MVS) participated as subjects (ages were 22, 24, 26, and 36 years). All had normal or corrected-to-normal vision and all had participated in previous dual-task experiments (as had Karlin and Kestenbaum's subjects).

Apparatus and Stimuli
The stimuli used for the digit task were the digits ``1'' and ``2,'' drawn in Helvetica font, presented on a Commodore 1084 color monitor for 50 ms at a viewing distance of 76 cm. The ``1'' was 1.0\D V.A. high by .34\D V.A. wide, and the ``2'' was 1.0\D V.A. high by .68\D V.A. wide. The characters were always upright and presented in white (56 cd/m^2) against a dark background (.05 cd/m^2). The stimuli used for the tone task were clearly audible 600 Hz and 3000 Hz tones presented for 100 ms. The presentation of the stimuli and recording of responses was controlled by an Amiga 2500 microcomputer.

To examine the effects of practice, two separate sets of analyses were performed. One set, with session as a factor, included the data from all nine experimental sessions for each of the SRT and 2AFC conditions (days 1 through 18); a second set included only the data from the last session of each number of Task² response alternatives condition (days 17 and₁8). The analyses produced largely parallel results. Mean overall RT² decreased by almost 120 ms from days 1 and 2 to days 17 and 18, F(8,₂4)=12.7, p<.001. Mean overall RT¹ decreased by almost 60 ms from days 1 and₂ to days 17 and₁8. Initially (days 1 and 2), RT¹ was 17 ms slower in the 100 ms SOA condition than at the longer SOAs. This may have been responsible for the only significant interaction involving session---the interaction between session and SOA on RT¹, F(32, 96)=2.00, p<.01.

Days 17 and 18 Only

RTs, production errors, and anticipation errors for Task¹ and Task² were examined separately using ANOVAs in which number of Task² response alternatives (SRT, 2AFC) and SOA (100, 200, 400, 800, 1200 ms) were within-subject factors. Outlier elimination identified 2.46% of the correct RTs for the digit task and 2.05% of the correct RTs for the tone task as outliers.

The Tone Task (Task²)

Response Times. As is shown in Figure 6, RT² increased with decreasing SOA, producing a main effect of SOA, F(4,₁2)=286.8, p<.001. There was also a main effect of number of Task² alternatives due to slower 2AFC RT²s than SRT RT²s, F(1, 3)=19.9, p<.025. The approximately 200 ms 2AFC-SRT RT² difference was unaffected by SOA, F(4,₁2)=1.27, p >.33.

Production Errors. There were no SRT production errors. There was a non-significant trend for more 2AFC production errors with decreasing SOA (see Figure 6), F(4,₁2)=1.37, p >.30 (analysis of 2AFC trials only). Anticipation Errors. There were no 2AFC anticipation errors. SRT anticipation errors increased from 0.0% at the shortest two SOAs to 3.98% at the 1200 ms SOA. This increase produced a main effect of SOA, F(1, 3)=29.4, p<.02 (analysis of SRT trials only). The Digit Task (Task¹) As is shown in Figure 7, RT¹ was unaffected by SOA, F<1. 2AFC RT¹s were marginally slower than SRT RT¹s, F(1, 3)=6.5, p<.09. This trend did not interact with SOA, F(4,₁2)=1.46, ns. There were no significant production error effects, all Fs <1, and no tone task anticipation errors.

A large 2AFC-SRT RT² difference obtained at each of the SOAs ({\Xbar}= 211.5 ms across subjects, SOA, and all experimental sessions; {\Xbar}= 190.7 ms for Days 17 and₁8, see Figure 6). The most important result is that this 2AFC-SRT RT² difference showed no sign of decreasing with decreasing SOA. The failure to replicate Karlin and Kestenbaum's critical result is not due to task differences since the experimental tasks used in Experiment₂ were comparable to those used by Karlin and Kestenbaum (1968). The failure is not due to differential practice effects because, in Experiment₂, the 2AFC-SRT RT² difference was the same across SOA for the entire range of practice reported by Karlin and Kestenbaum (3--8 days of practice within each number of Task² alternatives condition).

The SRT and 2AFC RTs in both Experiments 1 and₂ were slower than those reported for comparable conditions by Karlin and Kestenbaum (1968). Typically, both RT¹ and RT² under dual-task conditions are inflated relative to single-task conditions (Pashler, 1984). In contrast, the RTs reported by Karlin and Kestenbaum are comparable to those expected for single-task performance (e.g., Hick, 1952).

Karlin and Kestenbaum's data treatment may have contributed to both the apparent lack of RT² inflation and to the reported decrease in the SRT-2AFC RT² difference with decreasing SOA. Karlin and Kestenbaum eliminated trials {\it ``in which either RT was less than 100 or more than 100 ms longer than the next longest RT''} {\rm (Karlin Kestenbaum, 1968, p169)}. This procedure will eliminate more long RTs as the density of the observations within the cell decreases. This tendency has three important consequences. First, the post-outlier mean of the more sparsely populated error-prone cells will be underestimated relative to less error-prone cells (Van Selst, 1995). The impact will be particularly pronounced in the SRT condition at the longer SOAs, potentially leading to an overestimation of the main effect of SOA on SRT. Second, as SOA decreases, the variability of RT² increases (Van Selst, 1995). This increase may lead to underestimation of the main effect of SOA on RT². The third is that 2AFC RTs are more variable than SRT RTs. Because of this, the 2AFC-SRT RT² difference will be underestimated.

The data from Experiment₂ was used in a direct comparison of the effect of Karlin and Kestenbaum's outlier procedure and our outlier procedure. Their procedure reduced the 2AFC-SRT RT² difference by 20 ms at the 1200 SOA and increased the amount of observed attenuation of the 2AFC-SRT RT² difference by 11 ms. Although these differences cannot fully explain the empirical discrepancy, their outlier procedure is suspect and may have contributed to the observed discrepancy.

EXPERIMENT 3

Experiment 3 was a second attempt to replicate closely Karlin and Kestenbaum's critical finding of the decrease in the 2AFC-SRT RT² difference with decreasing SOA. This second attempt was motivated by concerns over comparing results across small numbers of subjects and the previous dual-task experience of the subjects in Experiment₂.

If subject's task strategies play a large role in determining bottleneck-like outcomes (Meyer Kieras, 1995), it is possible that the previous dual-task experiences of the subjects carried forward into Experiment₂. In one or more of the previous dual-task experiments performed by the subjects in Experiment₂, R¹ had been strongly encouraged to precede R². Meyer and Kieras (1995) argue that this may lead subjects to adopt a strategy of serial Task¹:Task² processing rather than fully parallel Task¹:Task² processing of which they (arguably) would otherwise be capable (but see Pashler Johnston, 1989). To explore this possibility Experiment₂ was repeated, but only naive subjects were used. In addition, since in Experiment₂ practice showed no signs of interacting with SOA nor number of Task² alternatives, only a single experimental session was used. The task and instructions from Experiment₂ were (again) comparable to those used by Karlin and Kestenbaum (1968).

Subjects
Eleven undergraduates (5 males) at the University of Waterloo participated to fulfill a course requirement. Ages ranged from 20 to₂7 (Median = 22). All subjects were naive as to the purpose of the experiment and all reported normal or corrected-to-normal vision.

Procedure
The procedure was identical to that used in Experiment₂ with the exception of the organization and number of experimental blocks. The single experimental session consisted of four blocks of 160 experimental trials. During the experimental session the experimental blocks alternated between SRT and 2AFC, with the type of initial block counterbalanced across subjects. Each experimental block was preceded by a set of 10 warmup trials of the same type as the experimental block. The first set of experimental trials was preceded by an additional 120 practice trials, 60 SRT and 60₂AFT. These two initial blocks of practice trials were presented in the same alternating order as the experimental blocks.

Four of the subjects did not yield useable data. Two subjects failed to switch to the SRT task from the 2AFC task, one experienced equipment failure, and the fourth produced excessively high error rates (33.7% in Task¹, 11.2% in Task²).

RTs, production errors, and anticipation errors for Task¹ and Task² were examined separately using ANOVAs in which number of Task² response alternatives (SRT, 2AFC) and SOA (100, 200, 400, 800, 1200 ms) were within-subject factors. Outlier elimination identified 2.78% of the correct RT¹s and 2.64% of the correct RT²s as outliers.

The Tone Task (Task²)

The Digit Task (Task¹

As is shown in Figure 9, RT¹ was unaffected by SOA, F<1. Responses were slower in the Task² 2AFC condition than in the Task² SRT condition, F(1, 6)=45.9, p<.001. A marginal number of Task² alternatives by SOA interaction was found, F(4,₂4)=2.34, p<.09. This interaction does not compromise any conclusions based on Task² performance (see Figures 8 and 9). There were no significant production nor anticipation error effects.

Once again, there was no significant attenuation of the 2AFC-SRT RT² difference with decreasing SOA. The difference between this result and the results reported by Karlin and Kestenbaum (1968) is not attributable to our choice of experimental tasks nor to the previous experience of the subjects. Quite simply, Karlin and Kestenbaum's experimental effect did not replicate.

EXPERIMENT 4

Karlin and Kestenbaum's critical results failed to obtain in both Experiments 2 and 3. In Experiment 4 and 5 we test the alternative hypotheses put forward to account for the (partial) underadditivity found in Experiment₁. Experiment 4 was designed to test the predictions of the response preparation hypothesis (McCann Johnston, 1992) using the design of Experiment₁ (in which Karlin and Kestenbaum's result were partially replicated).

In the current experiment, a go/no-go task was used as Task². For the go trials, it was anticipated that motor responses would be prepared in advance of stimulus presentation because, like SRTs, the action to be executed is known in advance (and it is very simple). Furthermore, we attempted to modulate the degree of response preparation by manipulating the proportion of go versus no-go trials. Go responses should yield differentially faster responses when go responses are more likely than when no-go (non-)responses are more likely. This effect should be stronger at the longer SOAs than at the shorter SOAs because of the increased time for response preparation (McCann Johnston, 1992).

As an additional benefit, using a go/no-go procedure should minimize the number of anticipation errors relative to a SRT condition because of the increased likelihood that a response in the absence of a stimulus will result in an error. In the go/no-go procedure the subject must attend to the stimulus to determine if a response is appropriate. Thus, the proportion manipulation may provide a means of evaluating the influence of differential response preparation across SOA (McCann Johnston, 1992) in the absence of the undesirable influence of anticipatory responding.

To manipulate the amount of response preparation, the proportion of go and no-go trials was manipulated. There were two conditions. In one condition, a response was to be made on 75% of the trials (the 75% go condition). In the other condition, a response was to be made on only 25% of the trials (the 25% go condition). If differential response preparation across SOA is possible, the effect of the go/nogo proportion manipulation on go trials should increase with increasing SOA. This increase should obtain because at the longer SOAs the go response should be better prepared in the 75% go condition than in the 25% go condition.

Subjects
Thirty-eight undergraduates (sixteen male) at the University of Waterloo participated to fulfill a course requirement. Ages ranged from 19 to 29 (Median = 21). All subjects were naive as to the purpose of the experiment and all reported normal or corrected-to-normal vision.

Procedure
The procedure was identical to that used in Experiment₁ with the following exceptions. First, Task² was now a go/no-go task. Second, the SOAs were changed to 100, 200, 400, and 800 ms. Third, there were now 320, rather than 288, trials in each experimental block. An untimed break was provided after every 160 trials.

The distribution of go versus no-go trials was 75% go: 25% no-go in one condition, and 25% go: 75% No-go in another condition. The subject was to respond to the color patch by pressing the `green' button (the only color task response button provided) if the color patch was green; if the color patch was red, the subject was not to execute a response to the color patch (trials timed out 2000 ms after presentation of the color patch).

Tone task RTs, and tone and color task production and anticipation errors, were examined separately using ANOVAs in which go/no-go, proportion (75% go versus 25% go), and SOA (100, 200, 400, 800 ms) were within-subject factors. No-go trials did not produce RT²s so the RT² analysis did not include go/nogo as a factor. Outlier elimination identified 1.86% of the correct RT¹s and 1.89% of the RT²s as outliers.

The Color Task (Task²)

Response Times As is shown in Figure₁0, go trial RT²s were faster in the 75% go condition than in the 25% go condition, producing a main effect of proportion, F(1, 37)=31.7, p<.001. There was an accelerating increase in RT² with decreasing SOA, producing a main effect of SOA, F(3,₁11)=256.3, p<.001. The proportion by SOA interaction was not significant, F(3,₁11)=1.90, p<.14.

Error Rates: Production Errors. More production errors were made in the 75% go condition than in the 25% go condition, producing a main effect of proportion, F(1, 37)=13.5, p<.001 (see Figure₁0). There was also a main effect of go/no-go, F(1, 37)=16.0, p<.001, produced by more incorrect go responses to no-go trials than failures to respond to go trials. The main effect of go/no-go was driven by the interaction between go/no-go and proportion, F(1, 37)=19.8, p<.001. In the 75% go condition subjects usually responded to go trials appropriately (all but 0.24%), but often mistakenly responded to no-go trials (2.89%). In the 25% go condition, subjects failed to respond to go trials (0.64%) as often as inappropriately responding to no-go trials (0.68%).

There was no main effect of SOA, F<1, although there was an interaction between SOA and go/no-go, F(3,₁11)=2.98, p<.04. More failures to respond to go trials were observed at the shorter SOAs, and more go responses to no-go trials were observed at the longer SOAs. There was a marginal interaction between SOA and proportion, F(3,₁11)=2.14, p<.10, apparently caused by the increase, with increasing SOA, in the error rate of the 75% go condition but not the 25% go condition. The three-way interaction between SOA, go/no-go, and proportion was not significant, F(3,₁11)=1.65, p>.18.

Error rates: Anticipation Errors. More anticipation errors were made in the 75% go condition (0.41%) than in the 25% go condition (0.12%), F(1, 37)=11.9, p<.002 (see Figure₁0). There was neither a main effect of go/no-go, F(1, 37)=1.81, p>.18, nor an interaction between go/no-go and proportion, F<1. Although there were no anticipation errors at the shortest two SOAs, the proportion of anticipation errors increased from 0.16% to 0.89% across the two longer SOAs, producing a main effect of SOA, F(3,₁11)=11.6, p<.001. This increase in the number of anticipation errors was more pronounced in the 75% go trials condition than the 25% go trial condition, producing a proportion by SOA interaction, F(3,₁11)=8.94, p<.001. This increase was also marginally more pronounced for go trials than for no-go trials (Figure₁0), yielding a marginal SOA by go/no-go interaction, F(3,₁11)=2.29, p<.09. The three-way interaction between SOA, go/no-go, and proportion was not significant, F<1.

The Tone Task (Task¹)

Response Times As shown in Figure₁1, subjects responded faster to the tone when the color patch was of the dominant proportion. This produced a proportion by go/no-go interaction, F(1, 37)=39.3, p<.001. Note that this pattern was only found at the three shorter SOAs. At the 800 msec SOA, subjects responded to the tone in the presence of go trials faster than in the presence of no-go trials. This produced a three-way interaction between proportion, go/no-go, and SOA, F(3,₁11)=10.7, p<.001. RTs decreased with decreasing SOA, which produced a main effect of SOA, F(3,₁11)=7.32, p<.001. No other interactions nor main effects approached significance, all other Fs <1 (with the exception of the proportion by SOA interaction, F(3,₁11)=1.49, p>.22).

Error Rates. Consistent with the RT¹ analysis subjects made fewer errors when the color patch was of the dominant color, producing an interaction between proportion and go/no-go, F(1, 37)=5.02, p<.04 (see Figure₁1). At the longer SOAs, this pattern was replaced by more accurate tone responses for go trials than for no-go trials, thus producing a three-way interaction between proportion, go/no-go, and SOA, F(3,₁11)=4.05, p<.01. There was a general increase in error rates with decreasing SOA, F(3,₁11)=19.8, p<.001, an effect that opposes the decrease observed in the RTs. No other main effect nor interaction approached significance; all Fs <1, except the go/no-go by SOA interaction, F(3,₁11)=1.43, p >.23. There were no significant anticipation error effects, all Fs <1.81, all ps >.18.

Our failures to replicate the decrease in the 2AFC-SRT RT² difference with decreasing SOA, as reported by Karlin and Kestenbaum, may be due to differences in task demands and instructional set (De Jong Sweet, 1994; Meyer Kieras, 1995; Meyer et al., 1995). The instructions (Experiments 1--4) and performance feedback (Experiments 1 4) may have led subjects to prepare for Task¹ to a greater extent than Karlin and Kestenbaum's subjects. It is possible that differential preparation may have led to either less parallel processing of Tasks 1 and₂ (Meyer Kieras, 1995; Meyer et al., 1995) and/or a decreased likelihood that RE² would be delayed as a result of the motor response refractory period postulated by De Jong (1993). Also contributing to the latter possibility is the presumed decrease in anticipation errors resulting from our explicit instructions to the subjects that they wait for the Task² stimulus to appear before making a Task² response. This instruction may have inadvertently slowed legitimate SRT processing beyond the postulated motor response refractory period.

Subjects
Thirty-Two undergraduates (sixteen males) at the University of Waterloo were recruited from a paid subject pool and participated in return for payment of \1.00 plus performance bonuses (total payment of \6.50--\8.50). Ages ranged from 19 to 28 (Median = 21). All subjects were naive as to the purpose of the experiment and all reported normal or corrected-to-normal vision.

Procedure
The only differences in procedure from Experiment₁ were that (a) only SRT and 2AFC trials were run, (b) a payoff matrix was used, (c) 1/9^{th} of the trials in both Task² alternative conditions were catch trials in which no stimulus was presented, and (d) the same two stimuli were used for both the SRT and the 2AFC trials for each subject (the SRT stimulus was green on 50% of the trials and blue on 50% of the trials). For the color task, the subject was provided with two buttons for the 2AFC condition (a green and a blue button), and a single button for the SRT condition.

Payment information was included with the trial by trial performance feedback. Correct responses earned one cent, incorrect responses cost one cent. If the trial was correct and the RT for the emphasized task was faster than the mean RT for that SOA, the fast RT was rewarded with an additional half cent. Anticipation errors resulted in a five cent penalty.

RTs, production errors, and anticipation errors for Task¹ and Task² were examined separately using ANOVAs in which number of Task² response alternatives (SRT, 2AFC) and SOA (100, 200, 400, 800 ms) were within-subject factors. Signal trials and catch trials were analyzed separately for the anticipation error rate analysis of Task², but were included as a fourth factor in the analysis of production errors of Task² and in all analyses of Task¹ performance. Outlier analysis identified 2.4% of the correct RT¹ Task²-signal trials, 1.8% of the correct RT¹ Task²-catch trials, and 2.5% of the correct RT² (Task² signal) trials as outliers. Note that the signal trial cells are based on 64 trials per cell whereas catch trial cells are based on only eight trials per cell. Color task RT²s and error rates are presented in Figure₁2. Tone task RT¹s and error rates are presented in for signal trials and Figure₁4 for catch trials.

Color Task (Task²)

Response Times RT² differences between SRT and 2AFC trials produced a main effect of number of Task² response alternatives, F(1, 30)=104.6, p<.001. As well, mean RT²s increased with decreasing SOA, producing a main effect of SOA, F(3, 90)=537.8, p<.001 (Figure₁2). There was no evidence of an interaction between number of Task² response alternatives and SOA, F(3, 90)=1.23, ns. Neither the main effect nor any interaction involving task emphasis were statistically significant, all Fs <1.

Error Rates: Production Errors. here were no SRT production errors. For the 2AFC trials there was a marginal main effect of SOA, F(3, 90)=2.23, p<.09 (analysis of 2AFC trials only). No other effects were significant, all other Fs <1. The marginal main effect reflects both idiosyncratic variation as well as a trend for more production errors at longer SOAs (Figure12).

Error Rates: Anticipation Errors< (signal trial analysis). The anticipation error rate increased with increasing SOA, F(3, 90)=70.3, p<.001. This increase was more pronounced for SRT trials than for 2AFC trials (see Figure₁2), yielding both an interaction between number of Task² response alternatives and SOA, F(3, 90)=74.3, p<.001, and a main effect of Task² response condition, F(1, 30)=61.2, p<.001. Task emphasis had no impact on the anticipation error rate for signal trials (all Fs <1).

Error Rates: Anticipation Errors (catch trial analysis). More anticipation errors were made to catch trials in the SRT condition than in the 2AFC condition, F(1, 30)=60.0, p<.001. This effect was marginally stronger in the Task¹ emphasis condition than in the Task² emphasis condition, F(1, 30)=3.01, p<.10 (see Figure₁2). Although there was no main effect of SOA (unsurprising since the SOA manipulation merely affected the time that had to elapse before the trial timed out), F(3, 90)=1.19, ns; idiosyncratic variations produced a significant, although meaningless, interaction between task emphasis and SOA, F(3, 90)=3.32, p<.03. No other effects were significant, all Fs <1.

Tone Task (Task¹)

Response Times As in previous experiments, when Task² was a 2AFC task, RT¹s were slower (by 54 ms) than when Task² was a SRT task, F(1, 30)=33.4, p<.001 (see Figures 13 14). Mean RT¹ was almost 100 ms slower when fast RT²s were rewarded ({\Xbar}= 524 ms) than when fast RT¹s were rewarded ({\Xbar}= 427 ms), F(1, 30)=12.4, p<.002. When fast RT¹s were rewarded, there was no signal-catch RT¹ difference ({\Xbar}= 427 ms for both; Figures 13 14). When fast RT²s were rewarded, there was a 13 ms signal-catch RT¹ difference ({\Xbar}= 517 ms for signal trials [Figure₁3] ; {\Xbar}= 530 ms for catch trials [Figure₁4]).

These RT¹ differences produced a significant interaction between task emphasis and signal/catch, F(1, 30)=4.48, p<.05, and a marginal main effect of signal/catch, F(1, 30)=3.54, p<.07. The only other effect approaching significance was the interaction between task emphasis, number of Task² response alternatives, and SOA, F(3, 90)=2.36, p<.08. This marginal interaction largely reflects the 2AFC-SRT RT¹ difference observed in the Task² emphasis condition at the shortest SOA.

Error Rates More Task¹ production errors were made when a Task² signal was present (Figure₁3) than when a Task² signal was not present (Figure₁4), F(1, 30)=6.95, p<.02. More production errors were made with decreasing SOA for signal trials and fewer production errors were made with decreasing SOA for catch trials, producing an interaction between signal/catch and SOA, F(3, 90)=3.20, p<.03. Overall, 2AFC trials were not responded to more accurately than SRT trials, F<1. The 2AFC-SRT difference was, however, greater for signal trials (3.23% versus 1.26%) than for catch trials (2.11% versus 2.05%), producing an interaction between number of Task² response alternatives and signal/catch, F(1, 30)=7.46, p<.02. Non-systematic 2AFC-SRT differences across SOA produced a a marginal SOA by Task² response condition interaction, F(3, 90)=2.6, p<.06.

Rewarding fast RT¹s yielded a slight rise in production errors with decreasing SOA. Rewarding fast RT²s yielded a slight decrease in production errors with decreasing SOA. These opposing trends produced a marginal interaction between task emphasis and SOA, F(3, 90)=2.21, p<.10. A single exceptionally high 2AFC production error rate rate (6.25%; at the long SOA of the catch trials in the Task² emphasis condition; see Figure₁4) is likely responsible for the marginal evidence of an interaction between task emphasis, SOA, and number of Task² response alternatives, F(3, 90)=2.10, p<.11; and between task emphasis, SOA, number of Task² response alternatives, and signal/catch, F(3, 90)=2.08, p<.11. There were no other production error effects (all other Fs < 1.30, ps >.25).

The analysis of Task¹ performance indicated that the task emphasis manipulation was clearly successful; subjects were almost 100 msec faster at Task¹ when speeded RT¹s were rewarded than when speeded RT²s were rewarded. The Task¹ emphasis condition also produced faster RT¹s than observed in the comparable conditions of the previous experiments (e.g., compare Figures₁3 and 14 with Figures 5 and₁1). The obvious effectiveness of the task emphasis manipulation alleviates any concerns that the efforts to minimize anticipatory responding would undermine the potential effectiveness of the task emphasis manipulation.

According to several investigators, dual-task interference is caused in large part by a bottleneck in response selection (McCann \& Johnston, 1992; Pashler, 1994b). We reexamined the effects of a manipulation thought to affect response selection --- whether one (SRT) or two responses (2AFC) have to be made --- in the second of two tasks in a dual-task paradigm.

Karlin and Kestenbaum (1968) report that the 2AFC-SRT RT₂ difference is attenuated with decreasing SOA. Keele (1973) and others (De Jong, 1993; Meyer \ Kieras, 1995) have suggested that the attenuation of the 2AFC-SRT RT₂ difference with decreasing SOA suggests a processing bottleneck at response execution. We consider six possible alternative accounts: the Anticipation Hypothesis, the Perceptual Hypothesis, the it Outlier Hypothesis, the it Response Preparation Hypothesis (McCann and Johnston, 1992), the Multiple Bottleneck Hypothesis (De Jong, 1993), and the it Task Preparation Hypothesis (De Jong, 1995; De Jong & Sweet, 1994; Meyer & Kieras, 1995; Meyer et al., 1995). Each of these hypotheses are now considered in light of the results of Experiments 1-5.

Alternative Hypotheses

The Anticipation Hypothesis.
As SOA is decreased, fewer SRT anticipation errors will be made. The observed decrease in the 2AFC-SRT RT₂ difference with decreasing SOA could be an artifact of this trend. In fact, no reported study, including our Experiment 1, has found a significant decrease in the 2AFC-SRT RT₂ difference with decreasing SOA which cannot potentially be attributed to SRT anticipation errors. Thus, the anticipation hypothesis is sufficient to accommodate all reported decreases in the 2AFC-SRT RT₂ difference with decreasing SOA.

The Perceptual Hypothesis
Perceptual processing differences are known to be attenuated with decreasing SOA (De~Jong, 1993; Pashler \& Johnston, 1989; Van~Selst \& Jolicoeur, 1994a). SRT and 2AFC tasks have demonstrably different perceptual processing requirements (Fletcher \& Rabbit, 1978; Sternberg, 1969; Van~Selst, 1995). Thus, when attenuation between SRT and 2AFC tasks with decreasing SOA is found, some of this attenuation could reflect the attenuation of perceptual processing differences. The perceptual hypothesis remains viable and is supported by convergent evidence.

The Outlier Hypothesis
The observed decrease in the 2AFC-SRT RT$_2$ difference with decreasing SOA could be an artifact of outlier elimination. Outlier elimination procedures are affected by the characteristics of the sample populations submitted to them (Miller, 1991; Ulrich \& Miller, 1994; Van Selst \& Jolicoeur, 1994b). As discussed in Experiment~2, the outlier procedure used by Karlin and Kestenbaum interacts with SOA-induced changes to computationally bias the obtained results towards attenuation of the 2AFC-SRT RT$_2$ difference with decreasing SOA. For this reason, the data from the Karlin and Kestenbaum study is suspect (Van Selst, 1995). Other procedures, including our own, are less affected by these changes.

The Response Preparation Hypothesis
No evidence was found to support the response preparation hypothesis. However, if increased response preparation occurs only at the longer SOAs, and only when it is certain that the prepared response will be executed, then differential response preparation could account for the results reported by Karlin and Kestenbaum (1968). This elaborated hypothesis could then account for the decrease in the 2AFC-SRT RT² difference with decreasing SOA found in Experiment₁ and the failure to find evidence of analogous decreases in Experiments 4 and 5. The findings of Experiments 2 and 3, however, are inconsistent with both the original and the revised hypothesis; the 2AFC-SRT difference did not attenuate with decreasing SOA despite the certainty of execution of the prepared response. Neither the original nor the revised response preparation hypothesis remain viable.

The Multiple Bottleneck Hypothesis
If the RS² processing requirements are minimal, both a bottleneck at RS and a motor refractory period may impinge on Task² SRT processing at the short SOAs (De Jong, 1993). In order for the motor refractory period to affect RT², R² must be attempted within the motor refractory period of R¹ (see Figure 3). The duration of the motor refractory period is hypothesized to be approximately 200 ms (De Jong, 1993; Kahneman, 1973).

Karlin and Kestenbaum's finding that the 2AFC-SRT RT² difference decreases substantially with decreasing SOA was partially replicated only in Experiment₁. When the effect of the influence of the outlier procedure and the influence of anticipation errors was minimized, Karlin and Kestenbaum's critical result was not replicated---even when using their tasks and procedures. In Experiments₂--5, the RT differences across manipulations of Task² response condition were essentially unchanged across SOA. In addition, the 3AFC-2AFC RT² difference remained stable across SOA. This result is not consistent with the hypothesis that a bottleneck following response selection impacted Task² processing.

Given the weakness of the supposed attenuation of the 2AFC-SRT RT² difference with decreasing SOA, it is perhaps surprising that such strong empirical support was found for a number of alternatives accounts of the experimental effect. To resurrect the post-selection bottleneck hypothesis, the Karlin and Kestenbaum effect will have to be re-established and these alternate hypotheses discounted. Until then, the entirety of any observed attenuation of the difference between 2AFC and SRT RT²s can likely be attributed to some combination of the attenuation of perceptual processing differences (the perceptual hypothesis), anticipatory responding (the anticipation hypothesis), and statistical biases (the outlier hypothesis).

In summary, the data from Karlin and Kestenbaum's four subjects can no longer be considered as strong evidence in favor of a processing bottleneck at RE. There is no compelling evidence that the processing bottleneck at RS can be bypassed.

The motor refractory period is equal to the IRI at the shortest SOA provided two assumptions in addition to those of the postponement models discussed earlier. The first additional assumption is that the motor refractory period will induced a delay in Task² postponement on all of the short SOA trials in the SRT condition (reflected by a slope of -1 indicating a perfect trade-off between RT² and SOA) Second, as is illustrated in Figure 14, the motor refractory period is assumed to extend from movement completion, as opposed to initiation. If not, and the motor component of RE is not of equivalent duration for SRT and 2AFC tasks, the duration of the motor refractory period will be over-estimated. The amount of over-estimation will be the amount that SRT RE is shorter than 2AFC RE.

The estimate of the motor refractory period period yielded by Equation₁ is 410 ms for the SRT condition of Experiment₁ (Experiment₁ is the only experiment in which a decrease in the 2AFC-SRT RT² difference with decreasing SOA was found. SRT responses are most likely to have been affected by the postulated motor refractory period). Equation₁ produces a value for the duration of the motor refractory period of 220 ms for the equivalent condition from Karlin and Kestenbaum's (1968) study (the 90 ms SOA condition).

The decrease in the 2AFC-SRT RT² difference with decreasing SOA in Experiment₁ is not consistent with De Jong's (1993) multiple bottleneck model. The IRIs are much too long to postulate that a motor refractory period affected the responses. The estimate of the required motor refractory period to produce the results reported in Experiment₁ (410 msec) is well in excess of the duration computed based on Karlin and Kestenbaum's (1968) experiment (220 msec) and of the 200 ms estimated duration of the refractory period suggested by De Jong (1993) and Kahneman (1973).

The longer IRIs in Experiment₁ should have decreased the likelihood of postponement due to the RE bottleneck. With this decrease in the likelihood of postponement due to the RE bottleneck, the 2AFC-SRT RT² difference in Experiment₁ should have been less affected by the postulated motor refractory period than the experiments reported by Karlin and Kestenbaum (1968) and De Jong (1993). Despite the difference in IRIs, the 2AFC-SRT RT² difference with decreasing SOA in Experiment₁ is equivalent to that reported by Karlin and Kestenbaum (1968). The similarities in the amount of attenuation (58.5 ms in Experiment₁; 54 ms reported by Karlin and Kestenbaum, 1968) despite the large IRI differences present a problem for the multiple bottleneck hypothesis.

Broadbent, D.E., Gregory, M. (1967) Psychological refractory period and the length of time required to make a decision. Proceedings of the Royal Society, 168B, 181--193.

Cohen, J. (1988). Statistical power analysis for the behavioural sciences. 2nd ed., Hillsdale, NJ: Erlbaum.

Comstock, E.M. (1973). Processing capacity in a letter-match task. Journal of Experimental Psychology, 100, 63--72.

Craik, K.W. (1947). Theory of the human operator in control systems. I. The operator as an engineering system. British Journal of Psychology, 38, 56--61.

Craik, K.W. (1948). Theory of the human operator in control systems. II. Man as an element in a control system. British Journal of Psychology, 38, 142--148.

Davis, R. (1956). The limits of the "Psychological Refractory Period." Quarterly Journal of Experimental Psychology, 8, 24--38.

Davis, R. (1957). The human operator as a single channel information system. Quarterly Journal of Experimental Psychology, 9, 119--129.

Davis, R. (1959). The role of "attention" in the psychological refractory period. Quarterly Journal of Experimental Psychology, 11, 211--220.

Davis, R. (1962). Choice reaction times and the theory of intermittency in human performance. Quarterly Journal of Experimental Psychology, 14, 157--166.

Davis, R. (1965). Expectancy and Intermittency. Quarterly Journal of Experimental Psychology, 17, 75--78.

De Jong, R., (1993). Multiple bottlenecks in overlapping task performance. Journal of Experimental Psychology: Human Perception and Performance, 19, 965--989.

De Jong, R. (1995). The role of preparation in overlapping-task performance. Quarterly Journal of Experimental Psychology, 48A, 2--25.

De Jong, R. (1995). Personal correspondence.

De Jong, R., Coles, M.G., Logan, G.D., Gratton, G. (1990). In search of the point of no return; The control of response processes. Journal of Experimental Psychology: Human Perception and Performance, 16, 164--182.

De Jong, R., Sweet, J.B. (1994). Preparatory strategies in overlapping-task performance. Perception and Psychophysics, 55, 142--151.

De Jong, R., Wierda, M., Mulder, G., Mulder, L.J.M. (1988). Use of partial stimulus information in response processing. Journal of Experimental Psychology: Human Perception and Performance, 14, 682--692.

Dimond, S.J. (1966). Facilitation of performance through the use of the timing system. Journal of Experimental Psychology, 71, 181--183.

Donders, F.C. Over de snelheid van psychische processen. (On the speed of mental processes.) Onderzoekingen degaan in het physiologisch Laboratorium der Ugtrechtsche Hoogeschool, 1868--69, Tweede reeks, 11, 92--130. Translated by W.G. Koster, in W.G. Koster (ed.), Attention and Performance II, Acta Psychologica, 1969, 30, 412--431.

Dutta, A., Walker, B.N. (1995). Persistence of the PRP effect: Evaluating the response-selection bottleneck. Poster presented at the 36th Annual Meeting of the Psychonomic Society, Los Angeles, November, 1995.

Duncan, J. (1980). The demonstration of capacity limits. Cognitive Psychology, 12, 75--96.

Fagot, C., Pashler, H. (1992). Making two responses to a single object: Implications for the central attentional bottleneck. Journal of Experimental Psychology: Human Perception and Performance, 18, 1058--1079.

Fisher, S. (1975). The microstructure of dual task interaction. 1. The patterning of main-task responses within secondary task intervals. Perception, 4, 267--290.

Fisher, S. (1975b). The microstructure of dual task interaction. 2. The effect of task instructions on attentional allocation and a model of attention-switching. Perception, 4, 459--474.

Fletcher, B.C., Rabbit, P.M.A. (1978). The changing pattern of perceptual analytic strategies and response selection with practice in a two-choice reaction time task. Quarterly Journal of Experimental Psychology, 30, 417--427.

Gladstones, W.H., Regan, M.A., Lee, R.B. (1989). Division of attention: The single-channel hypothesis revisited. Quarterly Journal of Experimental Psychology, 41A, 1--17.

Gopher, D., Brickner, M., Navon, D. (1982). Different difficulty manipulations interact differently with task emphasis: Evidence for multiple resources. Journal of Experimental Psychology, 71, 181--183.

Gottsdanker, R. (1966). The effect of superseding signals. Quarterly Journal of Experimental Psychology, 18, 236--249.

Gottsdanker, R. (1975). The attaining and maintaining of preparation. In P.M.A. Rabbitt S. Dornic (Eds.) Attention and Performance V. NY: Academic Press.

Gottsdanker, R. (1979). A psychological refractory period or an unprepared period? Journal of Experimental Psychology: Human Perception and Performance, 5, 208--215.

Gottsdanker, R. (1980). The ubiquitous role of preparation. In G.E. Stelmach and J. Requin (Eds.), Tutorials in Motor Behavior, (pp 315--371). Amsterdam: North-Holland.

Gottsdanker, R., Broadbent, L., Van Sant, C. (1963). Reaction time to single and to first signals. Journal of Experimental Psychology, 66, 163--167.

Gottsdanker, R., Perkins, T., Afthab, J. (1986). Studying reaction times with non-aging intervals: An effective procedure. Behavior Research Methods, Instruments, and computers, 18, 287--292.

Gottsdanker, R. Stelmach, G.E. (1971). The persistence of psychological refractoriness. Journal of Motor Behavior, 3, 301--312.

Gottsdanker, R. Way, T.C. (1966). Varied and constant intersignal intervals in psychological refractoriness. Journal of Experimental Psychology, 72, 792--804.

Greenwald, A. (1972). On doing two things at once: Time sharing as a function of ideomotor compatibility. Journal of Experimental Psychology, 94, 52--57.

Greenwald, A. Shulman, H.G. (1973) On doing two things at once: II. Elimination of the psychological refractory period effect. Journal of Experimental Psychology, 101, 70--76.

Herman, L.M. (1965). Study of the single channel hypothesis and input regulation within a continuous, simultaneous task situation. Quarterly Journal of Experimental Psychology, 17, 37--46.

Hick, W.E. (1948). The discontinuous function of the human operator in pursuit tasks. Quarterly Journal of Experimental Psychology, 1, 36--51.

Hick, W.E. (1952). On the rate gain of information. Quarterly Journal of Experimental Psychology, 4, 11--26.

Hick, W.E. Welford, A.T. (1956). Comment on the paper ``Central inhibition: some refractory observations'' by Alick Elithorn and Catherine Lawrence, Quarterly Journal of Experimental Psychology, 8, 39--41.

Hicks, R.E. (1975). Intrahemispheric response competition between vocal and unimanual in a normal adult human male. Journal of Comparative and Physiological Psychology, 89, 50--60.

Hoffman, J. E., Nelson, B. (1981). Spatial selectivity in visual search. Perception Psychophysics, 30, 283--290.

Hoffman, J.E., Nelson, B., Houck, M.R. (1983). The role of attentional resources in automatic detection. Cognitive Psychology, 51, 379--410.

Hommel, B. (1995). S-R compatibility without response uncertainty. Poster presented at the 36th Annual Meeting of the Psychonomic Society, Los Angeles, November, 1995.

Irvy, R.B., Franz, E.A., Kingstone, A., Johnston, J.C. (1994) The PRP effect in a split-brain patient: Response Uncoupling despite normal interference. Paper presented at the 35th Annual Meeting of the Psychonomic Society, St. Louis, November, 1994.

Johnston, J.C., Delgado, D.F. (1993). Bypassing the single-channel bottleneck in dual-task performance. paper presented to the 34th annual meeting of the Psychonomic Society, Washington, November, 1993.

Johnston, J.C., McCann, R.S., Remington, R.W. (1995). Chronometric evidence for two types of attention. Psychological Science, 6, 365--369.

Jolicoeur, P., Ullman, S., MacKay, M. (1986). Curve Tracing: A possible basic operation in the perception of spatial relations. Memory and Cognition, 14, 129--140. .

Jolicoeur, P., Van Selst, M. (1994). [Monte Carlo simulations testing temporary exclusion of the most extreme value when using the modified recursive outlier elimination procedure with moving criterion.] (unpublished raw data).

Kahneman, D. (1973). Attention and effort. Englewood Criffs, NJ: Prentice-Hall.

Karlin, L. (1965). Comments on ``expectancy and intermittency.'' Quarterly Journal of Experimental Psychology, 17, 351.

Karlin Kestenbaum (1968). Effect of number of alternative on the psychological refractory period. Quarterly Journal of Experimental Psychology, 20, 167--178.

Kay, H. Weiss, A.D. (1961). Relationship between simple and choice reaction time. Nature, 191, 790--791. [cited in Bertelson, 1966]

Kerr, B. (1983). Preplanning for aimed movements: Disruptions from a preliminary task. Journal of Experimental Psychology: Human Perception and Performance, 9, 596--606. ABSTRACT

Keele, S.W. (1973). Attention and Human Performance. Goodyear Publishing Company, Inc. Pacific Palisades, California.

Keele, S.W. Neill, W.T. (1978). Mechanisms of Attention in Handbook of Perception IX. Goodyear Publishing Company, Inc. Pacific Palisades, California.

Kieras, D.E., Meyer, D.E. (1994). The EPIC architecture for modeling human information processing and performance: A brief introduction. University of Michigan technical report-94/ONR-EPIC-1.

Kleiss, J.A., Lane, D.M. (1986). Locus of persistence of capacity limitations in visual information processing. Journal of Experimental Psychology: Human Perception and Performance, 12, 200--210.

Logan, G. (1978). Attention in character classification tasks: Evidence for the automaticity of component stages. Journal of Experimental Psychology: General, 107, 1--36.

Logan, G. (1979). On the use of a concurrent memory load to measure attention and automaticity. Journal of Experimental Psychology: Human Perception and Performance, 5, 189--207.

Logan, G.D., Burkell, J. (1986). Dependence and independence in responding to double stimulation: A comparison of stop, change, and dual-task paradigms. Journal of Experimental Psychology: Human Perception and Performance, 12, 549--563.

Lauber, E.J., Schumacher, E.H., Glass, J., Zubriggen, E., Kieras, D.E. Meyer, D.E. (1994). Adaptive PRP effects: Evidence of flexible attention to action. Paper presented at the 35th Annual Meeting of the Psychonomic Society, St. Louis, November, 1994.

Logan, G.D., Zbrodoff, N.J. Fostey, A.R. (1983). Costs and benefits of strategy construction in a speeded discrimination task. Memory and Cognition, 11, 485--493.

MacLeod, C.M. (1991). Half a century of research on the Stroop effect: an integrative review. Psychological Bulletin, 109, 163--203.

McCann, R.S., Folk, C.L., Johnston, J.C. (1992). The role of spatial attention in visual word processing. Journal of Experimental Psychology: Human Perception and Performance, 18, 1015---1029.

McCann, R.S., Folk, C.L., Remington, R.W. (1991). When tasks collide: How does preparing for one task affect performance on another? paper presented to the 32nd annual meeting of the Psychonomic Society, San Francisco, November, 1991.

McCann, R.S., Johnston, J.C. (1992). Locus of the single-channel bottleneck in dual-task interference Journal of Experimental Psychology: Human Perception and Performance, 18, 471-484.

McLeod, P. (1977a). A dual task response modality effect: support for multiprocessor models of attention. Quarterly Journal of Experimental Psychology, 29, 651--667.

McLeod, P. (1977b). Parallel processing and the psychological refractory period. Acta Psychologica, 41, 381-391.

Meyer, D.E., Kieras, D.E. (1992). The PRP effect: Central bottleneck, perceptual-motor limitations, or task strategies? paper presented to the 33nd annual meeting of the Psychonomic Society, St. Louis, November, 1992.

Meyer, D.E., Kieras, D.E. (1994). EPIC computational models of psychological refractory-period effects in human multiple-task performance. University of Michigan technical report-94/ONR-EPIC-2.

Meyer, D.E., Kieras, D.E. (1995). A computational theory of human multiple-task performance: the EPIC architecture and strategic response-deferment model. Psychological Review, in press.

Meyer, D. E., Kieras, D. E., Lauber, E., Schumacher, E. H., Glass, J., Zurbriggen, E., Gmeindl, L., Apfelblat, D. (1995). Adaptive executive control: Flexible multiple-task performance without pervasive immutable response-selection bottlenecks. Acta Psychologica\/, 90, 163--190.

Meyer, D.E., Yantis, S., Osman, A., Smith, J.E.K. (1984). Discrete versus continuous models of response preparation: A reaction time analysis. In S. Kornblum J. Requin (Eds.) Preparatory states and processes. Hillsdale, NJ: Erlbaum.

Miller, J. (1988). Discrete and discontinuous models of human information processing: Theoretical distinctions and empirical results. Acta Psychologica, 67, 191--257.

Miller, J. (1991). Reaction time analysis with outlier exclusion: Bias varies with sample size. The Quarterly Journal of Experimental Psychology, 43A, 907--912.

Mowbray, G.H., Rhoades, M.V. (1959). On the reduction of choice reaction times with practice. The Quarterly Journal of Experimental Psychology, 14, 1--36. difference between 2 and 4 choice disappears with practice

Netick, A., Klapp, S.T. (1994). Hesitations in manual tracking: A single-channel limit in response programming. Journal of Experimental Psychology: Human Perception and Performance, 20, 766--782.

Oldfield, S.R. (1976). Hemispheric and attentional effects on the transmission of somatosensory information. Unpublished thesis. Australian National University, Canberra. Cited in Welford (1980).

Osmond, A., Moore, C. (1990). The effects of dual-task interference on movement-related brain potentials. paper presented to the 31st annual meeting of the Psychonomic Society, New Orleans, November, 1990.

Page, D.J., Ellis, J.G. (1993). Combining letter-matching and tone discrimination: Evidence of automatic perceptual and spatial congruity effects. Journal of Experimental Psychology: Human Perception and Performance, 46A, 161--180.

Pashler, H. (1984). Processing stages in overlapping tasks: Evidence for a central bottleneck. Journal of Experimental Psychology: Human Perception and Performance, 10, 358--377.

Pashler, H. (1989). Dissociations and dependencies between speed and accuracy: evidence for a two-component theory of divided attention in simple tasks. Cognitive Psychology, 21, 469--514.

Pashler, H. (1990). Do response modality effects support multiprocessor models of divided attention. Journal of Experimental Psychology: Human Perception and Performance, 16, 826--842.

This ``Tech report'' style document w. the yellow cover University of California, San Diego, Center for Human information processing Tech report 133. Feb. 1991. is the same as...

Pashler, H. (1991). Shifting visual attention and selecting motor responses: Distinct attentional mechanisms. Journal of Experimental Psychology: Human Perception and Performance, 17, 1023--1040.

Pashler, H. (1992). Dual-task interference and elementary mental mechanisms. In D.E. Meyer and S. Kornblum (eds.), Attention and Performance XIV. Cambridge, MA: MIT press.

Pashler, H. (1992). Attentional limits in doing two tasks at the same time. Current Directions in Psychological Science, 1(2), 44--48.

Pashler, H. (1994). Overlapping mental operations in serial performance with preview. Quarterly Journal of Experimental Psychology, 47A, 161--191.

Pashler, H. (1994). Comment on McLeod and Hume, Overlapping mental operations in serial performance with preview: Typing. Quarterly Journal of Experimental Psychology, 47A, 201--205.

Pashler, H. (1994a). Dual-Task interference in Simple Tasks: Data and Theory Psychological Bulletin, 116, 220--244.

Pashler, H. (1994b). Graded Capacity-Sharing in Dual-Task Interference? Journal of Experimental Psychology: Human Perception and Performance, 20, 330--342.

Pashler, H., Carrier, M., Hoffman, J. (1993). Saccadic eye movements and dual-task interference. Quarterly Journal of Experimental Psychology, 46A, 51--82.

Pashler, H., Johnston, J.C. (1989). Chronometric evidence for central postponement in temporally overlapping tasks. Quarterly Journal of Experimental Psychology, 41A, 19--45.

Pashler, H., O'Brien, S. (1993). Dual-task interference and the cerebral hemispheres. Journal of Experimental Psychology: Human Perception and Performance, 19, 315--330.

Prinzmetal, W., Banks, W.P. (1983). Perceptual capacity limits in visual search. Bulletin of the Psychonomic Society, 21, 263--366.

Reynolds, D. (1964). Effects of double stimulation: Temporary inhibition of response. Psychological Bulletin, 62, 333--347.

Riggs, L.A. (1965). Electrophysiology of Vision. In Graham, C.H. (Ed.), Vision and Visual Perception (pp. 81--131). New York, NY: John Wiley Sons.

Schvaneveldt, R.W. (1969). Effects of complexity in simultaneous reaction time tasks. Journal of Experimental Psychology, b

Schweickert, R., Boggs, G.J. (1984). Models of central capacity and concurrency. Journal of Mathematical Psychology, 28, 223--281.

Smith, E.E. (1968). Choice Reaction Time: An Analysis of the major theoretical positions. Psychological Bulletin, 69, 77--109.

Smith, M.C. (1967a). Theories of the psychological refractory period. Psychological Bulletin, 67, 202--213.

Smith, M.C. (1967b). The psychological refractory period as a function of performance on a first task. Quarterly Journal of Experimental Psychology, 19, 350--352.

Smith, M.C. (1969). The effect of varying information on the psychological refractory period. Acta Psychologica, 30, 220--231.

Sternberg, S. (1969). The discovery of processing stages: Extensions of Donders' method. Acta Psychologica, 30, 276--315.

Stevens, S.S., Volkman, J., Newman, E.B. (1937). A scale for the measurement of the psychological magnitude of pitch. Journal of the Acoustical Society of America, 8, 185--190.

Telford, C.W. (1931). Refractory phase of voluntary and associative responses. Journal of Experimental Psychology, 14, 1--35.

Tolkmitt, F.J., Lunn, S.E. (1973). Psychological refractory period and the effect of stimulus discriminability. American Journal of Psychology, 86, 103--113.

Van Selst, M. (1995). Karlin and Kestenbaum's results do not require a dual-task bottleneck. Unpublished doctoral dissertation, University of Waterloo, Waterloo, Ontario, Canada

Van Selst, M., Besner, D., Jolicoeur, P. (1995). Word Frequency affects processing prior to the dual-task bottleneck. unpublished manuscript.

Van Selst, M., Johnston, J.C. (1996, July). Multiple alternative Go/No-Go task performance is consistent with a central dual-task processing bottleneck. Paper presented at the 3rd Annual Meeting of the Cognitive Science Association for Interdisciplinary Learning, Hood River, OR.

Van Selst, M., \Jolicoeur, P. (1994a). Can mental rotation occur before the dual-task bottleneck? Journal of Experimental Psychology: Human Perception and Performance, 20, 905--921.

Van Selst, M., Jolicoeur, P. (1994b). A solution to the effect of sample size on outlier elimination. Quarterly Journal of Experimental Psychology, 47A, 631--650.

Van Selst, M., Jolicoeur, P. (1995). [Stroop interference for rotated but not upright items is attenuated with increasing task overlap]. Unpublished raw data.

Vince, M.A. (1948a). The intermittency of control movements and the psychological refractory period. British Journal of Psychology, 38, 149--157.

Vince, M.A. (1948b). Corrective movements in a pursuit task. Quarterly Journal of Experimental Psychology, 1, 85--103.

Welford, A.T. (1952). The `psychological refractory period' and the timing of high-speed performance --- A review and theory. British Journal of Psychology, 43, 2--19.

Welford, A.T. (1959). Evidence of a single-channel decision mechanism limiting performance in a serial reaction task. Quarterly Journal of Experimental Psychology, 11, 193--210.

Welford, A.T. (1967). Single-channel operation in the brain. Acta Psychologica, 27, 5--22.

Welford, A. T. (1980). The single-channel hypothesis. In A.T. Welford (Ed.), Reaction time (pp. 215--252). New York: Academic Press.

Wickens, C.D. (1983). Processing resources in attention, dual task performance, and work-load assessment. In R. Parasuraman R. Davies (Eds.), Varieties of Attention. New York: Academic Press.

Williams, L.R.T. (1974). Effects of number of alternatives on the psychological refractoriness of an extended movement. Journal of Motor Behavior, 6, 227--234.