[page 63↓]

4  Dual Task Experiments

As noted in Chapter 3.3, the first goal of the set of experiments presented in this chapter was to extend the findings of inter-task consistency effects (e.g., Hommel, 1998; Lien & Proctor, 2000) to more abstract response dimensions, that is, to response instructions that do not encourage spatial coding. If participants indeed arbitrarily code their responses on two concurrently performed tasks as instructed – as predicted by (both versions of) the direct coding hypothesis – then R‑R forward and backward compatibility of the kind observed by Hommel (1998) should generalize to more abstract response dimensions. Thus, instructions in the present Experiments 2 and 3 suggested color coding of manual responses instead of spatial coding. Manual responses cannot reasonably be assumed to be pre-defined with respect to color prior to instruction because no training phase with consistent color-effects was administered.

The second, related goal of the present set of experiments was to explore whether R‑R (backward) compatibility effects and their reduction for non-overlapping tasks (Logan & Schulkind, 2000) depend on the overlap of instructed response representations. As discussed in Chapter 3.2 and 3.3, the experiments reported by Logan and colleagues are inconclusive with respect to whether overlap in instructed response categories is required for inter-task consistency effects to occur.

To address these questions, I used the experimental dual-task procedure of Hommel (1998, Experiment 1) involving a verbal and a manual task with arbitrary S‑R mappings that had to be performed in close succession. I manipulated the overlap in response representations by varying the instructions of the manual keypress responses. More specifically, I instructed the left and the right response keys on the manual task either as left and right keys (Exp. 1) or as blue and green keys (Exp. 2 and Exp. 3), and required either “left” vs. “right” verbalizations (Exp. 1 and Exp. 3) or “blue” vs. “green” verbalizations (Exp. 2) on the verbal task. If indeed instructed response labels affect response coding, then one would expect that participants code their manual responses in terms of location under left/right instructions (Exp. 1), but that, under color instruction, conceptual color codes become part of the response representations and are or can be used in the control of manual responding (Exp. 2 and 3). This should lead to (forward and backward) compatibility effects on both the verbal and the manual task whenever the verbal task requires the same conceptual codes as the manual task, that is, when both the verbal and the manual task are coded in terms of location (Exp. 1) or color (Exp. 2). [page 64↓]In contrast, if response coding is restricted to spatial coding, as suggested by the spatial coding hypothesis, cross-task compatibility should be restricted to the spatial dimension (Experiment 1).

In addition, if such effects depend on the amount of overlap of task relevant (instructed) response dimensions, then they should be eliminated or extremely reduced when instructed response dimensions differ across tasks, that is, when the verbal task requires location coding, but instructions suggest color coding of left and right keypresses on the manual task (Exp. 3). In contrast, both the weak coding hypothesis that assumes comparable effects for implicit (uninstructed) as for explicit (instructed) overlap and the spatial coding hypothesis predict spatial inter-task consistency effects in Experiment 3 similar to those in Experiment 1.

For each experiment, additional practice analyses including the practice block were carried out. Although similar analyses performed by Hommel (1998) suggest that practice does not affect the size of backward (and forward) compatibility under spatial response instructions, these analyses were included here to assess whether subjects (spatially) re-code non-spatially instructed responses after practice. If so, one might expect potential color effects in Experiment 2 to decrease with practice, and a location-based effect to build up after some amount of practice in Experiment 3.

4.1 Experiment 1

Experiment 1 was a conceptual replication of Hommel (1998, Experiment 1). Participants were asked to perform a verbal and a manual task in close succession. As in the Hommel experiment, stimuli were univalent and were arbitrarily mapped to left and right verbalizations viz. keypresses in a 1:1 fashion. Experiment 1 differed from Hommel (1998, Exp. 1) in two aspects. First, in the present experiment the verbal task was the primary task, and the manual task was the secondary task. The order of the tasks was reversed to allow assessment of backward effects from the manual task. As has been argued in Chapter 3.2, backward compatibility effects not only show that the overlapping codes are part of the Task 2 response representation, but indicate that these codes are actively used to access and guide responding on the secondary task. Second, I used non-integrated stimuli for the two tasks, namely tone stimuli and geometric form stimuli, which were presented asynchronously with a SOA of 50 ms. This was done to reduce the probability that participants recode the instructed S‑R rules by assigning one response code to an integrated stimulus representation on compatible trials (e.g., red H <> right).


[page 65↓]

I expected to obtain R‑R compatibility effects between the responses on both tasks. Hence, both verbal and manual responses were expected to be faster on response-compatible (e.g., a verbal “left” response followed by a left keypress) than on response-incompatible trials (e.g., a verbal “left” response followed by a right keypress).

4.1.1 Method

Participants

Forty-seven undergraduate students (28 female, 19 male, mean age = 24 years) at Humboldt University, Berlin, participated for partial fulfillment of course credit. All subjects had normal or corrected to normal vision. Thirteen subjects of the total sample were excluded from the data analyses according to pre-defined criteria (see Section Data Analysis for a description of the final sample).

Apparatus and Stimuli

The experiment was controlled by Pentium II computers with SoundBlaster 16 Audio cards that were connected to external speakers, headphones with microphones attached to the headphones and a standard (German) keyboard. Stimulus presentation and response recording was controlled by a modified version of the Runword software, a freeware provided by C.T. Kello. Runword runs in DOS mode on IBM compatible PCs with (ISA compatible) SoundBlaster 16 Audio Cards.

Two different tone stimuli were used in the verbal task. Tone stimuli were generated by converting two different Windows wav-files (krt08.wav and schl05.wav, cut down to 50 ms duration each) into voc-files. Krt08 is essentially a squeak tone, whereas schl05 can best be described as a snap tone. Tone stimuli were displayed by Runword via speaker output. The speakers were located to the left and the right of the screen, and tones were simultaneously presented through both speakers. Volume was adjusted individually before the experiment started.

Squares and circles served as visual stimuli on the manual task. The diameter of the circle was 3 cm, as was the length of the sides of the square. Visual stimuli appeared as black frames against a white background plate of 10 x 8 cm (height x width) size at the screen center of 17 inch monitors. The viewing distance was approximately 50 cm.


[page 66↓]

In order to record vocal responses at a standardized level, the microphone was calibrated before the experiment started. During the experiment, Runword generated voc-files for each verbal response. Verbal RTs were determined after the experiment by a software voice-key provided by Runword, applying a fixed threshold as response criterion. Response alternatives on the verbal task were “links” and “rechts,” the German words for “left” and “right.” The left and the right keys on the manual task were the ‘y’ and the ‘.’ keys on a standard German keyboard, respectively.

Design and Procedure

Each session started with a written and verbal instruction specifying the required S‑R mappings and describing the sequence of events on each trial. Tone stimuli were labeled “Knackton” and “Quietschton,” the German words for snap tone and squeak tone, and were demonstrated during instruction. Verbal responses were assigned to tone stimuli, and manual responses to form stimuli. The four different mapping combinations (of tone and form stimuli to verbal and manual responses, respectively) were approximately counterbalanced across participants. Compatibility was defined in terms of overlap in ‘response location,’ so for example, a verbal “links” response followed by a left keypress was considered compatible. After instruction, participants were required to recall the mappings correctly and then received a written reminder of the instructed mapping rules that was accessible during the entire course of the experiment.

Participants then worked through eight blocks of 32 trials each. Each block contained 8 replications of each of the four combinations of tone and form stimuli, amounting to 16 trials of each compatibility condition per block. The first block was treated as a practice block and was not considered in the main analyses, resulting in 7 experimental blocks with a total of 56 trials per stimulus combination (112 trials per compatibility condition) overall. However, the practice block was considered in the practice analyses. More specifically, for the practice analyses, two blocks each (including the practice block) were aggregated into a block-cluster, resulting in four block-clusters that provided the basis to test for practice effects. Accordingly, each block-cluster contained 16 replications of each of the four tone-form stimulus combinations, amounting to 32 trials in each compatibility condition per block-cluster and 128 compatible and incompatible trials overall.


[page 67↓]

Trials were presented in one of four different quasi-random orders that required stimulus combinations to follow each other about equally often. These quasi-random trial sequences were determined before the experiment and were counterbalanced across participants.

Each trial started with a plate saying “Fertig?” (‘Ready?’) presented for at least 1000 ms. When participants were ready to commence the trial, they pressed the space key to initiate the trial. After a 700 ms blank interval, a fixation cross appeared. Seven hundred ms later one of the tone stimuli was displayed for 50 ms while the fixation cross remained on the screen. Simultaneously with the offset of the tone, one of the form stimuli replaced the fixation cross and remained on the screen for 1200 ms. Response recording for both the verbal and the manual task was initiated at the onset of the form stimulus and terminated after 2500 ms had passed. Participants had to respond verbally to tone stimuli first, and then to press the left or right key with their left and right index fingers in response to the form stimuli. Instructions emphasized the requirement to perform the two responses in strict serial order, and participants were reminded of the required response order after the practice block.

4.1.2 Results

Data Analysis

Main Results and Additional Analyses. Data from 13 participants were excluded from the main analyses of Experiment 1 because they did not perform at a pre-experimentally determined level of performance on one or both of the tasks7. Nine participants did not comply with instructions in that they performed the manual before the verbal task on more than 10% of the trials on the seven experimental blocks. Three additional participants were excluded because they produced more than 30% errors, misses, and/or uncodable vocal responses, again across the seven experimental blocks. The latter exclusion criterion was applied in order to guarantee reliable RT results8. One additional participant had to be excluded due to recording problems of vocal responses. As a consequence, data from thirty-four undergraduate students (20 female, 14 male, mean age = 24 years) entered into the main analyses of Experiment 1.


[page 68↓]

In the reduced sample, response order errors occurred on 1.0 % of the experimental trials and were excluded from analyses. Trials on which one or both of the responses was faster than 50 ms or slower than 2000 ms accounted for 1.1% of the trials and were also excluded from further analyses, as were double error trials in which responses on both tasks were incorrect or omitted (1.0% of the trials) because double errors are difficult to interpret (cf. Schuch & Koch, submitted).

For the remaining data, median RTs for trials on which both responses were correct, and the percentage of invalid trials (PI) including errors and response omissions on only one task were computed for each task as a function of the factors under consideration in the main analyses and the additional analyses, respectively. The decision to go with median RTs instead of mean RTs was motivated by the consideration that medians are less sensitive to outliers than means unless individual raw data are trimmed (cf. Ratcliff, 1993), and data trimming for individual subjects would have led to different criteria in the main analyses and the practice analyses that included the practice block in the analysis (see below). However, analyses on (untrimmed) means that were run for reasons of comparability indicated a very similar pattern of results in all three experiments, and hence lead to the same conclusions.

Because response recording of verbal responses was initiated 50 ms after the onset of the tone stimuli (see method section, above), a constant amount of 50 ms was added to all verbal RTs before data screening, data aggregation, and analysis. This was also done in Experiments 2 and 3, as well as in the practice analyses. For all factors including more than two within-subject conditions, reported p-values are based on Greenhouse-Geisser corrected degrees of freedoms. This also applies to the practice analyses and the other experiments presented in this thesis.

Practice Analyses. For the practice analyses, the data, including those from the practice block (which are not considered in the main analyses), were aggregated into four block-clusters, each block-cluster containing data from two blocks. Block-clusters instead of experimental blocks were considered in order to smooth the learning curves and to avoid unreliable Block 1 estimates resulting from high error rates in the first (practice) block.

According to the pre-experimentally defined exclusion criteria described above, two additional participants had to be excluded from the practice analysis because they produced more than 30% invalid trials overall when the practice block was included, leaving N = 32 participants (19 female, 13 male, mean age = 23.6 years) in the practice analyses.


[page 69↓]

In the final sample, response order errors accounted for 1.1% of the trials and were excluded from the analysis. Trials with responses faster than 50 ms or slower than 2000 ms occurred on 1.3% of the trials and were excluded, too, as were double error trials in which responses on both tasks were incorrect or omitted (1.6% of the trials).

For the remaining data, median RTs for trials on which both responses were correct, and PIs (again only considering trials with errors or omissions on one task) were computed for each task as a function of block-cluster (1-4), task (verbal, manual), and compatibility (compatible, incompatible).

Main Results

For the standard analyses, median RTs for trials on which both responses were correct, and PIs including errors and response omissions on only one task were computed for each task as a function of compatibility between verbal and manual responses. Table 5 shows the group means of the individual median RTs and of the accuracy data across experimental blocks for the reduced sample (N = 34).

Table 5: Mean Median Reaction Times (RT, in ms) and % Invalid (PI) for Verbal (Primary) and Manual (Secondary) Responses as a Function of Response-Response Compatibility in Experiment 1.

Note: a Columns labeled Δ indicate effect sizes of the compatibility effects (incompatible minus compatible). bNumbers in parentheses represent results based on n = 33 participants, excluding the compatibility effect outlier (see text for details).

RT. Median RTs were submitted to an analysis of variance (ANOVA) with task (verbal, manual) and response compatibility (compatible, incompatible) as within-subjects factors. This analysis yielded a significant main effect of task, F(1,33) = 329.26, p < .001, MSe = 15,048.23, indicating faster responses on the verbal task. More important for the present purposes, the main effect of compatibility also reached significance, F(1,33) = 6.61, p < .05, MSe = 11,451.85, whereas the interaction between task and compatibility did not, F(1,33) < 1, MSe = 2,936.96, suggesting that the verbal (primary) and the manual (secondary) tasks were similarly affected by the compatibility relation between responses. However, planned comparisons that tested the effect of compatibility separately for each task, showed that compatibility only reached significance in the analysis of manual responses, [page 70↓] F(1,33) = 9.19, p < .01, MSe = 8,011.04. It just missed significance for the verbal task, F(1,33) = 3.75, p < .07, MSe = 20,766.58. A closer inspection of the verbal data revealed that this outcome was due to a single participant who showed a particularly large compatibility effect on the verbal task (> 800 ms). Eliminating this participant from the analyses numerically reduced the compatibility effects for verbal and manual responses to 25 ms and 37 ms (see Table 5, numbers in parentheses), respectively, but now the verbal effect also reached significance, F(1,32) = 6.8, p < .05, MSe = 1,503.68, while all other effects (including those in the omnibus analysis) remained qualitatively the same.

PI. The ANOVA on PIs9 yielded only a significant main effect of compatibility, F(1,33) = 13.95, p < .001, MSe = 11.72; the main effect of task and the interaction between task and compatibility were not significant (both F’s < 1). Planned comparisons that tested the compatibility effect separately for each task revealed a significant compatibility effect for manual responses, F(1,33) = 6.4, p < .05, MSe = 21.16, and a significant effect for verbal responses, F(1,33) = 15.3, p < .001, MSe = 12.69.

Additional Analyses

In order to assess whether the backward compatibility effect on the verbal task depended on response grouping on a certain number of trials, that is, on withholding R1 until R2 is selected, I additionally assessed whether the compatibility effects depended on the lag between R1 and R2. If the compatibility effects were due to response grouping, then they should be particularly pronounced at short inter-response intervals (IRIs) and decrease with increasing IRI. Moreover, verbal RTs should be slowed at short as compared to long IRIs. For each participant and compatibility condition, I therefore determined IRI quintiles and calculated median RTs for verbal and manual responses for each quintile. Figure 3 shows the group means of the medians for each IRI quintile.

Of particular interest for an interpretation of the backward effect is how verbal responses behave across IRIs. As is clear from inspection of Figure 3, verbal RTs remain relatively constant across IRIs. More importantly, the size of the compatibility effect in verbal responses seems unaffected by IRI.


[page 71↓]

Figure 3: Mean median verbal and manual reaction times (RTs) as a function of inter-response-interval (IRI) and compatibility between verbal and manual responses in Experiment 1.

Mean median IRI quintiles were 191, 279, 354, 435, and 573 ms for compatible trials, and 220, 295, 359, 456, and 598 ms for incompatible trials.

These conclusions were supported by a two-way ANOVA that produced a significant main effect of compatibility, F(1,33) = 7.45, p < .05, MSe = 9,238.59; the main effect of IRI quintile and the interaction between IRI and compatibility did not reach significance (both F’s < 1). The analogous analysis on manual responses that was run for reasons of comparability, showed that manual RTs increased with IRI quintile, F(4,132) = 156.83, p < .001, MSe = 9,100.94, and also revealed a significant main effect of compatibility, F(1,33) = 11.25, p < .01, MSe = 16,226.32. However, the interaction of quintile and compatibility did not reach significance, F(4,132) < 1, MSe = 3,645.05, implying that the size of the manual compatibility effect was independent of IRI.

In addition, in order to gain insight into the temporal dynamics of the compatibility effects, distribution analyses were carried out on the RT data. To this end, RT quintiles were determined for each participant, task, and compatibility condition, and median RTs were computed accordingly. Figure 4 shows the resulting group means of the individual medians for each RT quintile and condition. Note that averaging the means of medians across quintiles leads to higher mean median scores per condition than those presented in Table 5. Whereas such a difference would not be expected in analyses on means, it is not particularly surprising [page 72↓]with respect to medians. This is so because medians are more sensitive to the positive skewness of RT distributions than are means (i.e., they respond less strongly to data in the slow end tail of the RT distribution), leading to higher overall RT estimates in the quintile analysis that “weighs” responses in the slow end tail of the distribution more heavily. Therefore, the tabled values (i.e., mean median overall RTs) roughly correspond to the mean medians of the third RT quintile of each condition, instead of the means across quintiles.

Figure 4: Mean median verbal and manual reaction time quintiles (RT quintiles) as a function of compatibility between verbal (primary) and manual (secondary) responses in Experiment 1.

The calculated median RTs were entered into a 2 (task) x 2 (compatibility) x 5 (RT quintile) ANOVA. Aside from producing a trivial main effect (F(1,132) = 195.57, p < .01, MSe = 22,563.13), quintile participated in both two-way interactions, F(4,132) = 13.92, p < .01, MSe = 5,600.32 for the interaction of quintile and task, and F(4,132) = 6.03, p < .01, MSe = 2,181.77, for the interaction of RT quintile and compatibility. The three-way-interaction of task, quintile, and compatibility did not reach significance, F(4,132) = 1.35, p > .26, MSe = 1,244.16, indicating that the compatibility effects on the two tasks similarly increased with increasing RT.


[page 73↓]

Practice

Although additional analyses performed by Hommel (1998) suggest that practice does not affect the size of spatial backward (and forward) compatibility, additional analyses were performed on the Experiment 1 data to assess whether this was also true for the present experiment. To this end, median RTs for trials on which both responses were correct, and the percentage of invalid trials on only one task were computed for each task as a function of block-cluster (1-4), task (verbal, manual), and compatibility (compatible, incompatible) on the basis of the data of the remaining thirty-two participants (see Section Data Analysis)10, the group means for which are presented in Table 6. Note that the means in the columns of Table 6 labeled “overall” do not correspond to the values presented in Table 5 because (a) the data from the practice block are included, (b) the data are based on a reduced sample, and (c) because aggregating medians in a two-step fashion (i.e., first on block-clusters and then across block-clusters) can lead to different overall means even if the same data are included.

Table 6: Mean Median Reaction Times (RT, in ms) and % Invalid (PI) for Verbal (Primary) and Manual (Secondary) Responses as a Function of Response-Response Compatibility and Block Cluster in Experiment 1.

  

Block Cluster

  

1

2

3

4

 

overall

Response

R1-R2

RT

PI

RT

PI

RT

PI

RT

PI

 

RT

PI

             

verbal

compatible

638

4.6

592

2.1

556

1.2

540

1.7

 

582

2.4

incompatible

657

8.7

626

4.4

588

3.4

556

3.9

 

607

5.1

 

Δ

19

4.1

34

2.3

32

2.2

16

2.2

 

25

2.7

             

manual

compatible

1069

6.5

973

2.3

919

2.1

893

2.1

 

963

3.3

incompatible

1118

8.3

1007

5.5

960

2.9

917

3.0

 

1000

4.9

 

Δ

49

1.8

34

3.2

41

0.8

24

0.9

 

37

1.6

Note: Rows labeled Δ indicate effect sizes of the compatibility effects. The two rightmost columns contain the means across blocks.

RT. Median RTs were submitted to an ANOVA with task (verbal, manual), response compatibility (compatible, incompatible), and block-cluster (1-4) as within-subjects factors. This analysis again yielded a significant main effect of task, F(1,31) = 359.91, p < .001, MSe = 53,464.38, indicating faster responses on the verbal task. The main effect of block-cluster was also significant, F(3,93) = 34.44, p < .001, MSe = 14,552.6, as was the interaction [page 74↓]between task and block-cluster, F(3,93) = 11.19, p < .001, MSe = 4,584.19, implying that, although both tasks benefited from practice, the reduction of RT with practice was more pronounced for the manual task. As in the analysis without block-cluster as a factor (see Section Main Results, above), the main effect of compatibility was significant, F(1,31) = 15.33, p < .001, MSe = 8,051.67, whereas the interaction of task and compatibility was not, F(1,31) = 1.29, p > .26, MSe = 3,483.72.

More importantly, block-cluster did not interact with compatibility. That is, neither the interaction of block-cluster and compatibility (F(3,93) < 1, MSe = 5,048.3), nor the three-way interaction of block-cluster, compatibility and task (F(3,93) < 1, MSe = 1,326.99) reached significance, indicating that practice, while speeding up overall RTs, did not affect the compatibility effects. The latter conclusion receives support from separate analyses for each task that did not yield significant block-cluster x compatibility interactions either (both F’s < 1).

PI. The ANOVA on PIs only yielded significant main effects of compatibility, F(1,31) = 10.26.95, p < .01, MSe = 59.57, and of block-cluster, F(3,93) = 23.31, p < .001, MSe = 24.95. Block-cluster did not interact with any other variable (all F’s ≤ 1). As in the RT analysis, the interaction of task and compatibility did not reach significance (F(1,31) = 1.6, p > .15, MSe = 22.08), but unlike RTs there was no overall difference between tasks regarding PIs (F(1,31) < 1, MSe = 72.08). Aside from this relatively minor discrepancy, the error results closely resemble the RT results.

4.1.3 Discussion

In Experiment 1, I found that both primary and secondary task performance was influenced by the compatibility relation between the responses on a verbal and a concurrently performed manual task. More specifically, both verbal (primary) and manual (secondary) responses were faster and less error prone when the same response code was required on the two tasks (e.g., when both responses were left) than when correct responding required different spatial codes (i.e., when one response was left and the other was right). Hence, I was able to replicate the Hommel (1998) findings, although, in my Experiment 1, the size of the compatibility effects on the two tasks was numerically smaller than in the Hommel experiment. Possibly, the reduced size of the compatibility effects was due to the use of non-integrated stimuli and to asynchronous stimulus presentation. The former may have reduced the likelihood of stimulus- or response recoding into integrated rules on compatible trials, whereas the latter may have reduced the overlap in S‑R processing of the two tasks.


[page 75↓]

A second difference between the Hommel experiment and my Experiment 1 (but see Experiment 2 below) is that in the former, but not in the latter experiment, the Task 2 (forward) effect was significantly larger than the backward effect. Whereas Hommel (1998) explained his finding by some additional contribution from meaning based forward-priming due to less-than-perfect response-reset, Hommel and Eglau (2002) noted that in many studies (using a slightly different paradigm) “compatibility effects on the secondary task commonly reflect little more than mere propagation of the effect from the primary task” (p. 272). At present, I do not have a good explanation for why some studies find comparable effects on both tasks while others do not (but see Hommel & Eglau, 2002, for an attempt). However, because my primary concern is with backward compatibility, this question does not appear to be of primary importance.

Apart from these differences, the present results are similar to those obtained by Hommel (1998), in that the backward compatibility effect was more pronounced for long verbal RTs, indicating that compatibility effects are larger when the overlap between S1-R1 and S2-R2 processing is enhanced. However, the effect in verbal responses did not depend on IRI, implying that response grouping cannot explain the backward compatibility effect.

Moreover, the effects did not depend on practice, that is, they remained relatively constant across blocks although overall RTs decreased with practice. This finding presumably implies that – whatever changes regarding processes involved in translation (e.g., shortcuts or instance based processing) lead to more efficient translation with practice – the codes responsible for the compatibility effects remain part of the response representations.

Taken together, these results have three important implications. First, when responses on two concurrently performed tasks are instructed in terms of location, then the same abstract conceptual (left and right) codes mediate responding on the two tasks. Second, S‑R translation for the two tasks must have overlapped in time to some degree. Third, response coding remains largely unaffected by practice, at least within the range of practice used in the present experiment.

However, Experiment 1 cannot discriminate between the different hypotheses regarding how instructions influence response coding. As discussed earlier (see Chapter 3.2), this is so because the instructions in my Experiment 1, just like the instructions by Hommel (1998), Lien and Proctor (2000), and by Koch and Prinz (submitted) encouraged spatial coding on the [page 76↓]two tasks. Therefore the results are consistent with both, the direct and the obligatory spatial coding hypothesis.

In Experiment 2, an attempt was made to discriminate between the spatial coding hypothesis, on the one hand, and (both versions of) the direct coding hypothesis, on the other hand. This was done by instructing non-spatial response dimensions on the two tasks.

4.2 Experiment 2

In Experiment 2, left and right response keys on the manual task were no longer instructed as left vs. right, but as blue vs. green. The verbal task also required “blue” vs. “green” responses.

If color instructions lead to integration and use of color codes in manual response representations and responding, as predicted by both versions of the direct coding hypothesis, then color based R‑R compatibility effects between verbal and manual color responses should be observed (e.g., a “blue” verbalization followed by a blue keypress should be easier than, say, a “blue” verbalization followed by a green keypress). Alternatively, if manual responses are spatially coded in an instruction-independent way, no such color-effect should be obtained.

Note that instructing arbitrary S‑R mappings in terms of (response) color differs from the ‘standard’ Hedge and Marsh task (see Chapter 3.1.4) in that the relevant stimulus attribute in the latter is color, but an arbitrary stimulus attribute was mapped to color-responses in my Experiment 2. Whereas participants can perform the H&M task on the basis of a ‘same’ or ‘opposite’ rule, the use of such a simple rule is not possible for arbitrary S‑R mappings.

As has been argued in Chapter 3, the strongest evidence for arbitrary coding of manual keypress responses so far comes from the response-effect literature (cf. Chapter 3.1.3). For instance, Hommel (submitted) demonstrated that consistently presented color effects become integrated into the action representations of left and right keypress responses, and can be used to access and guide responding, as indicated by color-based S‑R compatibility effects observed when color frames served as distractors in the test phase.

In a similar vein, Koch and Kunde (in press) demonstrated that verbal color responses (i.e., verbalizing color names in response to digits) were affected by the identity of color words presented after responding. Response-effect words were either congruently colored or not colored (e.g., the word BLUE was presented in blue or grey color). Responses were faster when the color words were consistent with the response (e.g., a “green” response followed by the word GREEN) than when incompatible color words (e.g., a “green” response followed by [page 77↓]BLUE) served as response effects, and response-effect compatibility was even more pronounced when the color words were congruently colored. This finding suggests that conceptual color codes mediated verbal response selection.

The crucial differences between these experiments and the present Experiment 2 are that in the present experiment (a) there was no practice phase in which color effects might have become associated with manual responses, as was the case in the Hommel (submitted) experiment, and (b) no response-compatible or incompatible action effects were presented upon responding. Rather, I simply instructed participants to press the blue or green key in response to form stimuli. Whereas stimulus-to-color mapping remained constant throughout the experiment, color-to-(left/right) key assignment varied unpredictably from trial to trial.

4.2.1 Method

Participants

Forty-two undergraduate students (31 female, 11 male, mean age = 23.2 years) at Humboldt University, Berlin, received either € 7,- or partial course credit for participation11.

Apparatus and Stimuli

Apparatus and stimuli, as well as response keys and response recording were identical to those used in Experiment 1.

Design and Procedure

In Experiment 2, I instructed left vs. right keypresses as blue vs. green keypresses and also required blue vs. green verbalizations on the verbal task. Accordingly, participants were asked to first respond verbally to tone stimuli by saying either “blau” or “grün” (the German words for blue and green, respectively), and then to press either the blue or the green key, depending on the form stimulus. The four different stimulus-to-color mapping combinations (e.g., combination 1: squeak tone < > “green”, snap tone < > “blue”; square < > green key, circle < > blue key; combination 2: squeak tone < > “blue” …; square < > green key …) were counterbalanced across participants and remained constant throughout the experiment for in[page 78↓]dividual participants. Congruency was defined in terms of ‘response color’ (for example, a “blue” verbalization followed by a blue keypress was considered congruent).

Because the keys were not colored themselves and instruction only specified the stimulus-to-color mapping for the manual task, leaving open the color-to-(left/right) key assignment on a given trial, each trial started with the presentation of a plate that graphically depicted the color-to-key assignment for the next trial. For instance, when the plate showed a green color patch on the left and a blue color patch on the right (with pictograms of a left and a right hand beneath the color patches), then participants knew that the green key would be the key on the left and the blue key the one on the right on the forthcoming trial. The color plate remained on the screen for at least 1000 ms after which participants could trigger stimulus presentation by pressing the space key.

Color-to-key assignment varied randomly from trial to trial with the following constraints. First, in each block, each combination of tone and form stimuli appeared equally often under each color-to-key assignment. Second, each color-to-key assignment x congruency condition followed each other about equally often. Four quasi-random sequences (i.e., trial orders) fulfilling these constraints were again determined before the experiment, and were counterbalanced across participants.

The first block again served as a practice block and was not considered in the main and additional analyses. On the remaining seven experimental blocks, participants saw a total of 112 trials in each congruency condition, 28 in each stimulus combination x color-to-key assignment condition. Again, the practice block was included in the practice analyses. Hence, each block-cluster (on which the practice analyses are based) contained 16 congruent and incongruent trials under each color-to-key assignment, realized by 8 replications of each S1-S2 combination.

The rest of the procedure was identical to the procedure of Experiment 1.

4.2.2 Results

Data Analysis

Main Results and Additional Analyses. Two participants were excluded from Experiment 2 because they performed the manual task before the verbal task on more than 10% of the trials, and further six participants were excluded because they produced more than 30% errors, misses and uncodable vocal responses overall for the same reasons as in Experiment 1. [page 79↓]Additional participants were tested in their place. As in Experiment 1, the pattern of results for excluded participants was similar to that for included participants.

Hence, only data of a reduced sample consisting of thirty-four undergraduate students (26 female, 8 male, mean age = 22.7 years) were analyzed. As in Experiment 1, median RTs for trials on which both responses were correct, and PIs for trials with only one invalid response were computed for each participant as a function of the variables under consideration (see Sections Main Results and Additional Analyses, below). This was again done after excluding trials with response order errors (0.93%), those with responses faster than 50 ms or slower than 2000 ms (2.32%), as well as trials with double errors (0.83%).

Practice Analyses. Based on the exclusion criteria, two additional participants had to be excluded because they produced more than 30% invalid trials overall when the data from the practice block were included, leaving N = 32 participants (24 female, 8 male, mean age = 22.6 years) in the practice analyses.

In the remaining data, response order errors occurred on 0.9% of the trials. These trials were excluded from the analyses, as were double errors (1.7% of the trials) and trials with responses that were faster than 50 ms or slower than 2000 ms on one or both tasks (2.4%). As in Experiment 1, median RTs for completely correct trials and PIs for trials with invalid responses on only one task were computed as a function of task, block-cluster (consisting of two blocks each, including the practice block), and R‑R compatibility, to assess whether practice modifies the compatibility effects.

Main Results

Based on the data of N = 34 participants, median RTs and PIs were computed for each participant, task, and compatibility condition (see Table 7 for group means). The pattern of results can be described as follows. Manual responses seem to be slower and more error prone than verbal responses. More important for the present purposes, R‑R compatibility affects the speed of responses as well as the likelihood to make an error or to omit a response on both tasks, although the compatibility effect seems to be somewhat larger on the (secondary) manual task.


[page 80↓]

Table 7: Mean Median Reaction Times (RT, in ms) and % Invalid (PI) for Verbal (Primary) and Manual (Secondary) Responses as a Function of Response-Response Compatibility in Experiment 2.

 

Compatible

 

Incompatible

 

Δ

         

Response

RT

PI

 

RT

PI

 

RT

PI

         

Verbal

656

2.0

 

692

4.0

 

36

2.0

Manual

1039

4.0

 

1114

6.5

 

75

2.5

Note: Columns labeled Δ indicate effect sizes of the compatibility effects (incompatible minus compatible).

These observations were supported by the ANOVA results.

RT. The omnibus ANOVA including task and compatibility as within-subjects factors revealed highly significant main effects of task, F(1,33) = 291.77, p < .001, MSe = 18, 840.94, and compatibility, F(1,33) = 24.52, p < .001, MSe = 4,234.04, as well as a highly significant interaction between task and compatibility, F(1,33) = 21.69, p < .001, MSe = 585.72. However, planned comparisons that tested the compatibility effects separately for the two tasks showed that compatibility was significant for both manual responses, F(1,33) = 26.92, p < .001, MSe = 7,036.67, and verbal responses, F(1,33) = 16.8, p < .001, MSe = 2,620.84.

PI. The PI analysis yielded significant main effects of task, F(1,33) = 10.91, p < .01, MSe = 15.16, and compatibility, F(1,33) = 18.83, p < .001, MSe = 8.99; the interaction between task and compatibility did not reach significance, F(1,33) < 1, MSe = 6.56. Planned comparisons of the compatibility conditions on the two tasks again revealed significant effects for both manual, F(1,33) = 11.54, p < .01, MSe = 18.72, and verbal responses, F(1,33) = 10.36, p < .01, MSe = 12.39.

Additional Analyses

To assess whether the backward compatibility effect depended on IRI, I again determined the IRI quintiles for each participant, task, and compatibility condition. Figure 5 shows the mean median RTs per task, compatibility condition, and IRI.


[page 81↓]

Figure 5: Mean median verbal and manual RTs as a function of IRI and compatibility between verbal and manual responses in Experiment 2.

Mean median IRI quintiles were 234, 314, 372, 437, and 581 ms for compatible trials, and 261, 338, 398, 475, and 610 ms for incompatible trials.

Again, verbal RTs remained relatively constant across IRIs, and the size of the compatibility effect for verbal responses was unaffected by IRI, as confirmed by a two-way ANOVA that yielded a highly significant effect of compatibility, F(1,33) = 17.7, p < .001, MSe = 7,282.61, but no effect of IRI quintile, F(4,132) < 1, MSe = 7,811.09, and no significant interaction between compatibility and IRI quintile, F(4,132) < 1, MSe = 2,722.36. As in Experiment 1, manual RTs increased with increasing IRI, F(4,132) = 154.39, p < .001, MSe = 9,327.74, and also revealed a relatively constant compatibility effect across IRI, as reflected by a significant main effect of compatibility, F(1,33) = 32.71, p < .001, MSe = 12,135.17, and a nonsignificant interaction between IRI and compatibility, F(4,132) < 1, MSe = 3,208.14.

Distribution analyses on RT quintiles (see Figure 6), including task, compatibility, and RT quintile as within subjects factors, again revealed that quintile was significant, F(4,132) = 429.72, p < .001, MSe = 11,271.06, and that it interacted with task, F(4,132) = 38.88, p < .001, MSe = 2,662.94, as well as with compatibility, F(4,132) = 15.92, p < .001, MSe = 1,463.62. The three-way-interaction between RT quintile, task, and compati[page 82↓]bility did not reach significance, F(4,132) < 1, MSe = 504.13, again indicating that the compatibility effects on both tasks increased as response speed decreased.

Figure 6: Mean median verbal and manual RT quintiles as a function of compatibility between verbal (primary) and manual (secondary) responses in Experiment 2.

Practice

As in Experiment 1, in order to assess the impact of practice on the compatibility effects, median RTs for trials on which both responses were correct and PIs for trials on which only one response was invalid were computed for the sample of N = 32 participants (see Section Data Analysis, above, for a description of the remaining sample) as a function of compatibility, task, and block-cluster. The group means of the aggregated data are shown in Table 8.


[page 83↓]

Table 8: Mean Median Reaction Times (RT, in ms) and % Invalid (PI) for Verbal (Primary) and Manual (Secondary) Responses as a Function of Response-Response Compatibility and Block Cluster in Experiment 2.

  

Block Cluster

  

1

2

3

4

 

overall

Response

R1-R2

RT

PI

RT

PI

RT

PI

RT

PI

 

RT

PI

             

verbal

compatible

688

4.0

650

2.3

637

1.2

635

1.8

 

653

2.3

incompatible

724

6.3

688

3.4

685

3.6

668

3.1

 

691

4.1

 

Δ

36

2.3

38

1.1

48

2.4

33

1.3

 

38

1.8

             

manual

compatible

1124

10.6

1057

4.1

1016

3.5

997

2.9

 

1048

5.3

incompatible

1172

11.3

1134

5.5

1102

6.6

1075

6.3

 

1121

7.4

 

Δ

48

0.7

77

1.4

86

3.1

78

3.4

 

73

2.1

Note: Rows labeled Δ indicate effect sizes of the compatibility effects. The two rightmost columns contain the overall means across block means.

RT. The median RTs were submitted to an ANOVA with task (verbal, manual) and response compatibility (compatible, incompatible), and block-cluster (1-4) as within-subjects factors. This analysis yielded a significant main effect of task, F(1,31) = 294.74, p < .001, MSe = 73,994.27, indicating faster responses on the verbal task. The main effect of block-cluster was also significant, F(3,93) = 10.74, p < .001, MSe = 15,450.31, as was the interaction between task and block-cluster, F(3,93) = 4.75, p < .05, MSe = 4,648.41, again implying that, although both tasks benefited from practice, the reduction of RT with practice was more pronounced for the manual task. As in the analysis without block-cluster as a factor (see main results), the main effect of compatibility, as well as the interaction of task and compatibility, was significant (F(1,31) = 36.45, p < .001, MSe = 10,801.08; and F(1,31) = 22.18, p < .001, MSe = 1,664.95, for the compatibility main effect and the interaction, respectively).

More importantly, block-cluster again did not interact with compatibility. That is, neither the interaction of block-cluster and compatibility, F(3,93) < 1, MSe = 4,795.41, nor the three-way interaction of block-cluster, compatibility, and task, F(3,93) = 1.9, p > .14, MSe = 932.45, reached significance, indicating that practice, while speeding up overall RTs, did not affect the compatibility effects. The latter conclusion was confirmed by separate ANOVAs on each task that did not show reliable interactions of block-cluster and compatibility (both p’s > .28).

PI. As is evident from Table 8, PIs tended to follow the RT pattern. This was confirmed by the PI ANOVA, which yielded significant main effects of task, F(1,31) = 17.97, p < .001, MSe = 71.21, block-cluster, F(3,93) = 36.22, p < .001, MSe = 16.8, and compatibility, [page 84↓] F(1,31) = 13.87, p < .001, MSe = 35.66, as well as a significant interaction between task and block cluster, F(3,93) = 4.45, p < .05, MSe = 24.03, that signals a more pronounced reduction in errors with practice for manual responses.

Neither the task x compatibility interaction, F(1,31) < 1, MSe = 25.24, nor the interactions including compatibility and block-cluster reached significance (F(3,93) = 1.22, p > .3, MSe = 14.44; and F(3,93) = 1.03, p > .37, MSe = 17.19, for the two-way and the three-way interaction including task, respectively).

4.2.3 Discussion

When color was the relevant response dimension for the verbal and the manual task, there were again forward and backward compatibility effects of remarkable size, implying that color codes were used to control responses on the two tasks. This demonstration of color based congruency effects for manual left and right keypresses that had not been defined with respect to color prior to instruction extends studies on response-effect compatibility (see Chapter 3.1.3) that demonstrated arbitrary response coding after consistent effect-stimulus presentation.

Moreover, the IRI quintile analyses again showed that the backward compatibility effect was independent of response grouping, but rather seemed to result from parallel response activation on the two tasks that shared a common response dimension. Furthermore, neither the forward nor the backward effect was affected by practice, although overall RT level decreased with practice, again indicating that (relatively limited) practice does not lead to changes in the codes responsible for the effects. As was the case in Experiment 1, speed of responding did influence the effects. That is, the compatibility effects increased with increasing RT level, adding support to the view that enhanced processing overlap between the two tasks leads to increased crosstalk between the tasks.

Unlike Experiment 1, however, the compatibility effect in Experiment 2 significantly increased from verbal (primary) to manual (secondary) responses. Possibly, meaning-based R1-R2 priming across response modes due to a less-than-perfect reset of R1 (cf. Logan & Gordon, 2001) contributed to the compatibility effect on the manual task over and above parallel code activation (see Chapter 4.1.3, for a discussion of the lack of this interaction in Experiment 1).

In sum, the color-based backward compatibility effect suggests that instructed color codes were used to access and guide manual responses. This result extends findings on re[page 85↓]sponse-effect compatibility demonstrating that arbitrary action effects become included into response representations with practice. Moreover, because the instructed color codes do not contain any spatial information, the compatibility effects in Experiment 2 provide stronger evidence for an impact of response instructions on behavioral control (i.e., the direct-coding hypothesis) than the dual-task experiments that used left/right response instructions.

One argument that might be raised against this conclusion is that color label-to-key assignment varied unsystematically from trial to trial, thus giving “instructed” color coding an unfair advantage. Varied color-to-key assignment does not appear to be a necessary condition for the effects to occur, however, given preliminary results from twenty-three participants in a current replication experiment with constant color-to-key assignment. More specifically, in this replication experiment, subjects saw the same color plate (i.e., the same color-to-key assignment) before each trial, all other things being equal to the Experiment 2 described above. The results obtained so far indicate that substantial backward as well as forward compatibility effects also emerge with constant color-to-key assignment.

Another argument that can be raised against an interpretation of Experiment 2 as supporting the direct coding hypothesis is that, as in the Experiments by Logan and colleagues, as well as in many studies involving color-label responses (cf. Chapter 3), spatial response coding of manual responses has not been assessed. That is, Experiment 2 cannot decide whether color instructions can override spatial response coding. Experiment 3 addresses this issue.

4.3 Experiment 3

The goal of Experiment 3 was to assess the status of spatial response coding of manual responses under non-spatial response instructions. More specifically, Experiment 3 is concerned with the question of whether CTC effects depend on overlap of instructed response dimensions, and hence, whether non-spatial response instructions can reduce spatial effects or not.

To address these questions, Experiment 3 again required left and right keypresses that were instructed as blue vs. green keypresses on the manual task. This time however, the verbal task required “left” vs. “right” responses, hence spatial coding. Consequently, responses on the two tasks still overlapped regarding “location” (either responses being left or right), but no longer overlapped with respect to instructed response dimensions (color vs. location).

Both the spatial coding hypothesis and the weak version of the direct coding hypothesis predict a location-based congruency effect because they assume that instructions cannot over[page 86↓]ride spatial response coding. If, however, response codes can be weighed according to instructions, as implied by the strong version of the direct coding hypothesis, then we should expect reduced (location based) R‑R compatibility effects in Experiment 3.

Aside from response instructions, Experiment 3 was identical to Experiment 2.

4.3.1 Method

Participants

Overall, fifty-one undergraduate students (41 female, 10 male, mean age = 23.1 years) at Humboldt University, Berlin, received either € 7,- or partial course credit for participation (see Section Data Analysis, below, for a description of the final sample).

Apparatus and Stimuli

Apparatus and stimuli, as well as response keys and response recording were the same as those used in the previous experiments.

Design and Procedure

The procedure was identical to Experiment 2 with the following exceptions. The verbal task required “left” vs. “right” responses, whereas left vs. right keys were again instructed as blue vs. green keys on the manual task. Accordingly, participants first responded verbally to tone stimuli by saying either “links” or “rechts,” and then pressed either the blue or the green key in response to form stimuli. Congruency was defined as in Experiment 1, that is, according to overlap regarding “location.” Thus, for example, when the blue key was the one located on the left side, then a verbal “left” response followed by a blue (left) keypress, and a verbal “right” response followed by a green (right) keypress were considered congruent.

4.3.2 Results

Data Analysis

Main Results and Additional Analyses. According to pre-experimentally defined criteria, data from seven participants were excluded from Experiment 3 because they performed the manual task before the verbal task on more than 10% of the trials, and further eight participants were excluded because they produced more than 30% errors, misses and uncodable vocal responses overall for the same reasons as in the previous experiments. Statistical analy[page 87↓]ses on the data of these participants yielded similar results as the analogous analyses of included participants (see below). Two further participants had to be excluded due to experimenter error, leaving data from N = 34 participants (26 female, 8 male, mean age = 23.2 years) in the main analysis and additional analyses of Experiment 3.

In these data, response order errors occurred on 1.4% of the trials. These trials were excluded, as were responses that were faster than 50 ms or slower than 2000 ms (2.4%), and trials with double errors (1.2%). For the remaining data, I calculated PIs and median correct RTs for each participant as in the previous experiment.

Practice Analyses. Two further participants had to be excluded according to the exclusion criteria when data from the practice block were included. One of them now produced more than 10% response order errors overall, whereas the other participant produced more than 30% errors, misses and uncodable vocal responses across blocks.

Therefore, as in the previous experiments, data from N = 32 participants (25 female, 7 male, mean age = 23.4 years) were considered in the practice analyses.

In the remaining data, response order errors occurred on 1.7% of the trials and were excluded, as were double errors (1.8%) and responses that were faster than 50 ms or slower than 2000 ms (2.7%). Median RTs for trials on which both responses were correct and PIs for trials with only one incorrect response were again computed for each participant according to task, block-cluster, and compatibility.

Main Results

Table 9 presents the group means of the individual median RTs and PIs as a function of task (response) and compatibility. As is evident from Table 9, the compatibility effects on both tasks were extremely small.

Table 9: Mean Median Reaction Times (RT, in ms) and % Invalid (PI) for Verbal (Primary) and Manual (Secondary) Responses as a Function of Response-Response Compatibility in Experiment 3.

 

Compatible

 

Incompatible

 

Δ

         

Response

RT

PI

 

RT

PI

 

RT

PI

         

Verbal

656

3.4

 

660

3.6

 

4

0.2

Manual

1143

6.5

 

1141

5.4

 

-2

-1.1

Note: Columns labeled Δ indicate effect sizes of the compatibility effects (incompatible minus compatible), whereby negative values signal faster responses or fewer errors in the incompatible condition.


[page 88↓]

RT. The RT ANOVA including task as a factor only yielded a significant main effect of task, F(1,33) = 322.97, p < .001, MSe = 24,731.89. Neither the main effect of compatibility, F(1,33) < 1, MSe = 811.75, nor the interaction between task and compatibility, F(1,33) = 1.07, p > .3, MSe = 324.42, reached significance. Nor did the compatibility effect reach significance for either task alone, as reflected by the results of planned comparisons, F(1,33) < 1, MSe = 1,639.51 for the manual task, and F(1,33) = 1.02, MSe = 632.82 for the verbal task.

PI. Similarly, the overall PI ANOVA also revealed a significant effect of task, F(1,33) = 11.08, p < .01, MSe = 18.91, whereas the main effect of compatibility, F(1,33) = 1.28, p > .25, MSe = 5.61, and the interaction between task and compatibility, F(1,33) = 2.78, p > .10, MSe = 5.06, did not reach significance. Nevertheless, planned comparisons showed that the compatibility effect for the verbal task was not significant, F(1,33) < 1, MSe = 9.02, whereas the slightly reversed effect in manual responses approached significance, F(1,33) = 3.36, p < .08, MSe = 12.32.

To assess the possibility that a speed-accuracy tradeoff might have masked the compatibility effects, I correlated the RTs and PIs for each participant, task, and compatibility condition. Whereas none of the correlations between RT and PI within each task x compatibility condition reached significance (all p’s > .2), both correlations for the verbal task were negative. Therefore, I also correlated the compatibility effects (i.e., the Δs) in RTs and PIs of each participant and task. The latter correlations revealed a significant positive relation for manual responses, r = .42, p < .05, whereas the relation for the verbal task was negative and approached significance, r = ‑.31, p < .08, suggesting that participants with large verbal RT effects tended to show small (or even reversed) PI effects and vice versa (see also practice analyses, below). However, doubly multivariate analyses of variance (MANOVAs) that simultaneously considered PI and RT as dependent variables led to similar results as the ANOVAs. That is, the compatibility effects did not reach significance, neither in the analysis including task as a factor, F(2,32) < 1, and F(2,32) = 1.47, p > . 24, for the main effect of compatibility and the interaction of task and compatibility, respectively, nor in the analyses by task, F(2,32) = 1.76, p >.18, and F(2,32) < 1, for the compatibility effect in manual and verbal responses, respectively.


[page 89↓]

Comparisons between Experiments

Despite procedural differences between the experiments reported in Chapter 4 (especially between Experiment 1, on the one hand, and Experiments 2 and 3, on the other hand), additional analyses comparing Experiment 3 with Experiments 1 and 2, respectively, were run in order to substantiate the claim that response instructions reduced the compatibility effect in Experiment 3.

The 2 (experiment) x 2 (task) x 2 (compatibility) mixed-factors ANOVA12 comparing RTs in Experiment 3 and Experiment 1 yielded significant main effects of experiment, F(1,66) = 10.88, p < .01, MSe = 108,624.43, and task, F(1,66) = 641.64, p < .001, MSe = 19,890.06, as well as a significant interaction between task and experiment, F(1,66) = 9.06, p < .01, indicating slower responses in Experiment 3, especially on the manual task.

More importantly, the main effect of compatibility, F(1,66) = 6.49, p < .05, MSe = 6,131.80, was qualified by the two-way interaction between experiment and compatibility, F(1,66) = 5.87, p < .05, whereas the interaction between task and compatibility, and the three-way interaction of experiment, task, and compatibility did not reach significance (both F’s < 1), implying that the compatibility effects on both tasks were reduced in Experiment 3. The PI ANOVA that yielded the same significances as the RT analysis corroborates these findings.

The comparison of Experiment 3 and Experiment 2 shows a similar picture as the comparison of Experiment 1 and 3. The RT ANOVA reveals that Experiments 2 and 3 did not differ regarding overall RT level (main effect of experiment, F(1,66) < 1, MSe = 108,525.54), whereas all other factors were highly significant: The main effects of task, F(1,66) = 613.63, p < .001, MSe = 21,786.41, compatibility, F(1,66) = 21.46, p < .001, MSe = 2,527.39, as well as the interactions between task and experiment, F(1,66) = 5.32, p < .05, task and compatibility, F(1,66) = 9.72, p < .01, MSe = 455.07, and between compatibility and experiment, F(1,66) = 19.72, p < .001, that were qualified by the three-way interaction of experiment, task, and compatibility, F(1,66) = 18.96, p < .001, implying that the differences in compatibility effects were larger for manual responses. However, when each task was analyzed sepa[page 90↓]rately, the interaction between compatibility and experiment reached significance for both tasks, F(1,66) = 10.45, p < .01, MSe = 1,626.83 for verbal responses, and F(1,66) = 23.04, p < .001, MSe = 4,338.09 for manual responses, indicating that both the verbal and the manual compatibility effects were reduced in Experiment 3. The ANOVA on PI yielded similar results as the RT analysis, most notably a highly significant compatibility x experiment interaction, F(1,66) = 16.86, < .001, MSe = 7.3, that was not qualified by a three-way interaction, however, F(1,66) = 2.54, p > .11, MSe = 5.81.

Additional Analyses

One might still argue that the fact that responses (especially manual responses) were much slower in Experiment 3, resulting in larger IRIs (on average, mean median IRI was 469 ms and 458 ms on compatible and incompatible trials, respectively; in comparison, the corresponding IRIs in Experiment 2 were 372 ms vs. 399 ms, and 354 vs. 359 in Experiment 1), led to reduced compatibility effects. Therefore, I again determined the IRI quintiles and the median RTs for each participant, quintile, task, and compatibility condition. Figure 7 shows the means of the median RTs. As becomes clear from inspection of Figure 7, little happens across IRIs except for overall slowing of manual responses. This is reflected in the results from separate ANOVAs on each task.

In the analysis of verbal responses, no effect reached significance, neither the main effects of quintile, F(4,132) = 1.09, MSe = 5,864.69, and compatibility, F(1,33) = 1.6, p > .2, MSe = 1,730.23, nor the interaction between quintile and compatibility, F(4,132) = 1, MSe = 2,466.16. In the analysis of manual responses, the effect of quintile was significant, F(4,132) = 191.19, p < .001, MSe = 9,011.28; the main effect of compatibility, F(1,33), = 1.35, p > .25, MSe = 3,812.92, and the interaction between quintile and compatibility, F(4,132) = 1.32, p > .27, MSe = 3,181.16, were not.


[page 91↓]

Figure 7: Mean median verbal and manual RTs as a function of IRI and compatibility between verbal and manual responses in Experiment 3.

Mean median IRI quintiles were 289, 398, 469, 554, and 699 ms for compatible trials, and 294, 388, 458, 537, and 687 ms for incompatible trials.

Another objection that could be raised might be that the lack of effects may be due to trials on which the stimulus-to-(left/right) key mapping changed as a consequence of changing the color-to-key assignment on successive trials, thus requiring a re-binding of stimulus codes and left/right codes. To address this possibility, I determined median RTs and PIs according to task, compatibility, and change vs. no-change of color-to-key assignment on consecutive trials, and compared the compatibility effects for change and no-change trials. The compatibility effects did not differ across trial types. The Δs for the verbal task were 4 ms /0.1% invalid and 7 ms / -0.1% for change and no-change trials, respectively. The corresponding Δs for the manual task were -4 ms / -2.1% and 1 ms / -0.4%. Not surprisingly, trial type did not interact with compatibility (neither in the analysis of RTs, nor in the error analysis) for either task, indicating that the null-effects were not restricted to change trials. Moreover, a similar analysis for the Experiment 2 data revealed that the color compatibility effects were not affected by the change of color-to-key assignment either.

Finally, as in the previous experiments, I ran distribution analyses to check whether the (null-) effects varied as a function of response speed (see Figure 8 for means of medians for each RT quintile, task, and condition).


[page 92↓]

Figure 8: Mean median verbal and manual RT quintiles as a function of compatibility between verbal (primary) and manual (secondary) responses in Experiment 3.

The omnibus ANOVA including RT quintile, task, and compatibility as factors only yielded significant main effects of task (F(1,33) = 317.4, p < .001, MSe = 125,967.81) and RT quintile (F(4,132) = 401.03, p < .001, MSe = 13,460.72). It neither showed a main effect of compatibility (F(1,33) < 1, MSe = 2,006.64), nor did compatibility interact with task (F(1,33) = 1.63, p > .21, MSe = 2,123.18). Most important, compatibility did not interact with quintile, (F’s < 1 for both, the two-way interaction of compatibility and quintile and the three-way interaction involving task). Because there seemed to be a hint of an effect at the slowest verbal quintiles, I tested whether the interaction was significant when each task was considered in isolation. These analyses showed that the interaction of compatibility and quintile did not reach significance for either task alone, F(4,132) = 1.26, p > .29, MSe = 653.55; and F(4,132) < 1, MSe = 932.56, for verbal and manual responses, respectively. The distribution analysis thus implies that the null effects in the main analysis were not due to a reversal of positive effects at particular RT bins.


[page 93↓]

Practice

As in the previous experiments, RTs and PIs for all blocks (including the practice block) were calculated as a function of task, compatibility, and block-cluster in order to assess how the effects develop with practice (for group means, see Table 10).

Table 10: Mean Median Reaction Times (RT, in ms) and % Invalid (PI) for Verbal (Primary) and Manual (Secondary) Responses as a Function of Response-Response Compatibility and Block Cluster in Experiment 3.

  

Block Cluster

  

1

2

3

4

 

overall

Response

R1-R2

RT

PI

RT

PI

RT

PI

RT

PI

 

RT

PI

             

Verbal

compatible

698

8.1

661

4.0

662

2.4

632

2.1

 

663

4.2

incompatible

720

5.3

663

4.8

656

3.9

639

1.6

 

670

3.9

 

Δ

22

-2.8

2

0.8

-6

1.5

7

-0.5

 

7

-0.3

             

Manual

compatible

1207

11.3

1165

6.4

1147

6.1

1100

5.6

 

1155

7.3

incompatible

1218

11.8

1165

4.7

1130

4.2

1083

5.0

 

1149

6.4

 

Δ

11

0.5

0

-1.7

-17

-1.9

-17

-0.6

 

-6

-0.9

Note: Rows labeled Δ indicate effect sizes of the compatibility effects, whereby negative values signal faster responses or fewer errors in the incompatible condition. The two rightmost columns contain the means across blocks.

RT. From Table 10, it seems as if there were compatibility effects in RTs, especially for verbal responses, on the first block-cluster. However, this impression did not receive support from the RT analysis. Although the omnibus ANOVA revealed a significant interaction of task and compatibility (F(1,31) = 4.29, p < .05, MSe = 1,122.94), in addition to significant main effects of task (F(1,31) = 339, p < .001, MSe = 89,079.42), block-cluster (F(3,93) = 10.76, p < .001, MSe = 19,268.46), and an almost significant interaction between task and block-cluster (F(3,93) = 3.06, p < .06, MSe = 5,337.42), both the two-way interaction of block-cluster and compatibility (F(3,93) < 1, MSe = 5,067.91), and the three-way interaction of block-cluster, compatibility, and task (F(3,93) < 1, MSe = 1,183.62) were far from significance. Moreover, in separate ANOVAs for each task, compatibility did not reach significance, neither for verbal responses, F(1,31) = 1,53, p > .22, MSe = 1,830.39, nor for manual responses, F(1,31) < 1, MSe = 3,502.65. Moreover, compatibility did not interact with block-cluster when each task was considered separately (both F’s ≤ 1), indicating that the task x compatibility interaction in the omnibus analysis was due to nonsignificant opposite effects in (overall) verbal and manual RTs.


[page 94↓]

PI. The PI effects tended to show a somewhat different pattern than the RT effects across block-clusters, at least where verbal PIs are concerned. Moreover, the values in Table 10 indicate that the pattern of the PI effects differ across tasks. More specifically, whereas the verbal PI effects tended to show an inverse quadratic trend across block-clusters, the trend for the manual PI effect was in the opposite direction. This observation was supported by the omnibus PI ANOVA that revealed a significant three-way interaction between task, block-cluster, and compatibility, F(3,93) = 5.66, p < .01, MSe = 12.61, in addition to significant, but theoretically less interesting, main effects of task, F(1,31) = 15.12, p < .001, MSe = 69.64, block-cluster, F(3,93) = 28.53, p < .001, MSe = 28.58, and a nearly significant interaction between task and block-cluster (F(3,93) = 2.8, p < .07, MSe = 30.99; no other effects reached significance). Separate analyses for each task showed that the interaction between compatibility and block-cluster was significant for verbal responses, F(3,93) = 5, p < .01, MSe = 11.45. Contrasts that tested the verbal PI compatibility effects for each block (without adjusting degrees of freedom) showed that the (reversed) compatibility effect only reached significance in the first and third block-cluster, but not in the second and fourth block-cluster. In contrast, the interaction between compatibility and block-cluster was not significant for manual responses (F(3,93) = 1.6, p > .2, MSe = 12.25), whereas the main effect of compatibility almost was, F(1,31) = 2.93,< .1, MSe = 18.78, corroborating the main results that showed a marginally significant reversed compatibility effect for manual PIs.

Because of the obvious discrepancies between RTs and PIs, and because there already was some evidence for a trade-off of effects (i.e., a slightly negative correlation between the RT effect and the PI effect on the verbal task) in the analysis without block-cluster (see Main Results, above), I again ran doubly multivariate MANOVAs that simultaneously considered PI and RT as dependent variables, this time including block-cluster as a factor. The centroids on which the MANOVA results were based closely followed the pattern of PIs because PI was weighted more strongly than RT on the discriminant function that primarily discriminated between conditions13. As a consequence, the omnibus MANOVA including task, block-cluster, and compatibility as factors led to similar outcomes as the PI analyses. Specifically, the interaction of task, block-cluster, and compatibility was marginally significant, F(6,26) = 2.35, p < .07. Moreover, whereas the analysis of verbal responses again revealed a [page 95↓]significant interaction of block and compatibility, F(6,26) = 3.6, p < .01, the interaction did not reach significance for manual responses, F(6,26) < 1.

4.3.3 Discussion

In the RT ANOVAs neither the 4 ms backward compatibility effect on the verbal task, nor the –2 ms forward compatibility effect on the manual task reached significance, and the effects were significantly smaller than in Experiments 1 and 2. The MANOVAs that conjointly considered RT and PI suggest that the lack of effects in Experiment 3 cannot be attributed to a tradeoff between (effects in) RTs and PI. In addition, the significant experiment x compatibility interactions in the experimental comparisons, combined with the facts that (a) the N’s were rather large, and that (b) the statistical error terms (i.e., the MSe’s) for the compatibility effects in Experiment 3 did not exceed those of the previous experiments, indicate that the null effects in Experiment 3 were not due to power problems. Finally, the lack of effects was not restricted to trials on which the color-to-key assignment, and hence the stimulus-to-location mapping, changed, but also held for no-change trials, indicating that re-binding of location codes to different stimulus attributes cannot be the main cause for the outcome.

Rather, the results seem to suggest that R‑R compatibility effects are extremely reduced when instructed response dimensions do not overlap, even when the two tasks share a common implicit response dimension (i.e., location). This result corroborates the findings by Logan and colleagues (e.g., Logan & Schulkind, 2000) who also demonstrated a lack of inter-task effects when categorization tasks changed from Task 1 to Task 2.

Interestingly, regarding the forward compatibility effect (i.e., the effect in manual responses) there even was a (marginally significant) tendency for a reversal of the compatibility effect in PIs that also manifested itself in the mean median IRIs (remember that mean median IRI was 469 ms and 458 ms on compatible and incompatible trials, respectively). This observation suggests that manual responding was slightly more difficult on compatible than on incompatible trials, and may imply that some sort of reset mechanism inhibits response “repetition” on compatible trials. I will return to this issue in the General Discussion section (see Section 4.4).

Unlike Experiments 1 and 2, the compatibility effect in PIs on the verbal task seemed to inconsistently change with practice, namely from negative to positive back to negative. However, this pattern did not show up in RTs and was inconsistent across tasks. At present, it [page 96↓]is unclear, which factors (e.g., strategies) may have led to this pattern of results, and whether it is systematic or not.

Nevertheless, whatever the exact reasons for the practice results, the IRI quintile analyses suggest that the lack of RT effects (or their reduction) was not due to the fact that verbal and manual responses in the main analysis were, on average, scheduled further apart in Experiment 3 than in the other two experiments, thus resulting in reduced overlap in Task 1 and Task 2 processing. First, the interaction between IRI quintile and compatibility was not significant for either task. Second, although there seemed to be a hint of a compatibility effect at the first IRI quintile (the mean median IRI of which was 291 ms), in the other two experiments the compatibility effects were significant at much higher IRIs (e.g., both the verbal and the manual compatibility effects reached significance at the highest IRI quintile, the mean medians were 586 ms and 596 ms in Experiments 1 and 2, respectively). Similarly, the distribution analysis indicates that the size or the direction of the null-effects did not significantly differ as a function of response speed.

In sum, instructing spatially organized responses in terms of color reduces the spatial (backward and forward) compatibility effects observed under spatial response instructions, suggesting that manual responses were arbitrarily coded as instructed, thereby providing strong support for the intentional weighing hypothesis.

4.4 General Discussion Experiments 1-3

Experiments 1-3 used a dual task approach requiring consistent viz. inconsistent responses on two tasks to investigate which of three different views on how response labels used in task instructions influence response coding is correct. According to the spatial coding hypothesis, responses are spatially coded, regardless of instructions. Therefore, the spatial coding hypothesis predicted cross-task compatibility effects on Experiment 1 and 3 where responses overlapped in terms of ‘location’. In contrast, no congruency effects were expected with respect to arbitrary instructed response dimensions such as color (Experiment 2). The direct coding hypothesis, on the other hand, assumes that arbitrary response codes are included into response representations when so instructed. Hence, cross-task compatibility should extend to arbitrary response dimensions (Experiment 2). Whereas the weak version of the direct coding hypothesis assumes that non-instructed (i.e. spatial) response codes are weighed as strongly as explicitly mentioned response dimensions, the strong version proposed that response codes can be intentionally weighed. Accordingly, inter-task consistency effects [page 97↓]resulting from implicit (non-instructed) overlap on the spatial dimension (Experiment 3) were expected by the weak direct coding hypothesis, but not by its strong version.

In Experiment 1, both a verbal and a concurrently performed manual task required left and right responses to univalent, non-integrated stimuli that were arbitrarily mapped to responses. Responses on both tasks were faster and less error prone when the two responses were left or right as opposed to one being right and one being left, thus replicating Hommel (1998, Exp. 1) using a slightly modified paradigm. Experiment 2 showed that such (forward and backward) R‑R compatibility effects are not restricted to the spatial response dimension, but also occur with abstract response dimensions that cannot be assumed to have been part of the manual response representations prior to instruction (i.e., color). This result extends findings on response-effect compatibility showing that arbitrary attributes become integrated into response representations after training. Both forward and backward compatibility effects were extremely reduced when different response dimensions were instructed for the two tasks in Experiment 3 (i.e., when the verbal task required “left” and “right” responses, whereas the left and right keys on the manual task were instructed as blue vs. green). The latter finding corroborates the results obtained by Logan and colleagues and suggests that inter-task consistency effects depend on overlap of instructed response dimensions.

Implications for Response Coding. The compatibility effects in Experiments 1 and 2 suggest that the same relatively abstract conceptual response codes (abstract in the sense that they do not necessarily contain information necessary for motor responding) were used for response selection on the verbal and the manual task. Furthermore, they suggest that the same responses can be controlled differently, depending on response instructions, indicating a high degree of flexibility in response coding. In line with the prediction of the direct coding hypothesis the color dimension was used for response selection on the manual task when response instructions primed the color dimension for both tasks in Experiment 2.

If one adopts the view that responses are represented in a distributed fashion (Allport, 1993; Hommel et al. 2001; Keele, Cohen, & Ivry, 1990) such that every response is coded in terms of its features (e.g., as being blue, left, manual, …), then the compatibility effects in Experiment 1 and 2 can be explained in a similar way as SRC effects, namely as a result of a conceptual match (vs. mismatch) of overlapping response features that are part of the response representations of both tasks (cf. Lien & Proctor, 2002, for a similar view). According to this view, response representations that share features across tasks tend to be activated in [page 98↓]parallel by the stimulus representations that are assigned and temporarily linked to them, leading to facilitation if the same code is activated by both stimuli, but to response competition if not (see Figure 9, panel A and B for an illustration of how this might lead to response competition in the incompatible conditions of Experiments 1 and 2). Note that this view, in accordance with most coding accounts of compatibility, implies that response selection primarily occurs at the level of conceptual response codes that (automatically) activate their ‘corresponding’ motor programs (e.g., the left and right hand motor codes Mm_l and Mm_r in Figure 9, Panel A and B), rather than at the level of motor programs.

Under this assumption, the lack (or the reduction) of the spatial backward compatibility effect in Experiment 3 suggests that arbitrary response codes were not only part of the manual response representation under color instructions, but that color coding actually dominated spatial coding. That is, in line with the predictions of the strong version of the direct coding hypothesis, Experiment 3 showed that R‑R compatibility effects are only observed when instructed response dimensions overlap, indicating that (a) manual responses were primarily coded and accessed in terms of color even when the verbal task required spatial coding, and (b) location codes did not play an important role in selecting manual responses when responses were instructed in terms of color. Figure 9 (Panel C) depicts how the lack of effects in Experiment 3 might be explained. More specifically, the stimuli for the two tasks in Experiment 3 were assigned to different response dimensions. Hence, responses for the two tasks were accessed by different codes, leading to less overlap regarding activation and use of location codes.

I do not argue, however, that location codes were completely substituted by color codes and omitted from manual response representation and selection in Experiment 3 (and Experiment 2). Rather, I believe that they were still part of the manual response representation – though less strongly weighed –, and were integrated with color codes (see Figure 9, Panel B and C) to allow manual responding, much as anatomical codes contribute to the Simon effect under crossed-hand conditions (see Chapter 3.1.2). This is so because if no spatial codes were included in the manual response representation at all, the response representation of the two tasks would not share codes. Accordingly, one would expect a clear-cut null-effect in both verbal and manual responses. However, in Experiment 3 there was a tendency for a reversed compatibility effect, that is, costs on compatible as compared to incompatible trials, in manual task errors and in mean median IRIs.

[page 99↓]
Figure 9: An illustration of the hypothetical activation flow in incompatible conditions in Experiment 1 (Panel A), Experiment 2 (Panel B) and Experiment 3 (Panel C).

Panel A: In Experiment 1, both tone stimuli (S1) and form stimuli (S2) activate highly weighted (conceptual) location codes, which in turn activate their corresponding motor programs (Mv and Mm for verbal and manual responses, respectively; subscripts _l and _r, stand for left and right, respectively). On incompatible trials, different location codes receive activation from the stimuli, leading to interference on the two tasks. Panel B: When color is the instructed response dimension for both tasks, S1 and S2 activate the color codes that are assigned to them, leading to slower responses on both tasks if different color codes are required for responding. Panel C: Response instructions prime the location dimension for the verbal task and the color dimension for the manual task, and S1 activates a location code, whereas S2 activates a color code. Therefore, S1 and S2 do not provide diverging evidence for codes on the common location dimension (see text for details). Note: Fv-2 and Fm-2 represent response features that are unique to verbal and manual response representations, respectively.

Response repetition costs have also been reported in the task switching literature (cf. Meiran, 2000), where enhanced switch costs were observed when the same response had to be repeated on switch trials. Recently Schuch and Koch (submitted) observed similar response repetition and even (forward) response compatibility costs in dual task performance using a PRP procedure more similar to my experiments. Most of these studies used bivalent stimuli (e.g., numbers) on two tasks (e.g., magnitude judgments on Task 1 and parity judgments on [page 100↓]Task 2), and interpret response repetition or response compatibility costs as indicating recoding of the responses in terms of what they signal (see Logan & Gordon, 2001, for a potentially similar interpretation of the lack of correspondence effects when categorizations differ across tasks). Accordingly, a left response that is recoded as, say, odd, on Task 1 hampers performance on a subsequent task when this task requires a left response to signal the property small, either because the competing (now relevant) category-response rule has been inhibited during Task 1 processing, and/or because re-binding of the left code with a different meaning during Task 2 processing requires prior unbinding of this code.

While the tendency for costs in manual responses in my Experiment 3 cannot be explained as resulting from changes in what they signal across tasks (otherwise there should have been forward compatibility costs in Experiments 1 and 2, too), unbinding and re-binding of (overlapping) location codes into a response representation that is considered “different” because of response instructions (e.g., integrating a left code into a green response after accessing a verbal “left” response via the left code) may have contributed to costs in my Experiment 3 as well. Alternatively, some sort of reset mechanism similar to that proposed by Milliken and colleagues (Milliken, Joordens, Merikle, & Seiffert, 1998) in the domain of negative priming might have adjusted the weights of location codes and thus hampered manual performance on compatible trials. Whatever the exact nature of the reset mechanism (see Schuch & Koch, submitted, for a comprehensive discussion of possible mechanisms underlying response repetition costs), the tendency for compatibility costs indicates that spatial codes were not completely excluded from manual response representations.

In sum, the color-based backward compatibility effect in Experiment 2 and the lack of a spatial backward compatibility effect in Experiment 3 indicate that manual responses were primarily coded in terms of color when so instructed. Location codes may still have been part of the manual action representations, but played a minor role in accessing manual responses, indicating that subjects have considerable control over how they code and access their responses. Together, Experiments 2 and 3 thus provide evidence in favor of the strong version of the direct coding hypothesis and against obligatory spatial response coding.

As already noted in the discussion section of Experiment 2, one argument that could be raised against this conclusion is that spatial coding did not have a “fair” chance in the current Experiments 2 and 3. More specifically, one may argue that unpredictably changing the color-to-key assignment from trial to trial might have reduced the likelihood of spatial response [page 101↓]coding. While, in principle, participants could have intentionally recoded instructed mappings (e.g., square < > blue) into location based coding (e.g., square < > left) when they were informed about the color-to-key mapping on the forthcoming trial, varying the color-to-key assignment strongly discouraged such a strategy. If the pattern of results in my Experiments 2 and 3 resulted from varied color-to-key assignment one would expect no or extremely reduced color-compatibility effects and substantially enhanced spatial compatibility effects on replications of Experiments 2 and 3 with constant color-to-key assignment, respectively. That is, when both the color and the location route are viable alternatives to response selection, but instruction suggests color coding. This is what I am currently testing in replication experiments in which the color plates presented in the beginning of each trial always signal the same assignment.

As already mentioned (see Section 4.2.3 above), preliminary data from the Experiment 2 replication experiment with constant key-assignment requiring color responses on both tasks indicate substantial color-based forward and backward compatibility effects, suggesting that manual responses are coded in terms of color even when spatial re-coding of S‑R mappings is easy. Moreover, while at present there are not yet any data available from the replication of Experiment 3, the experiments presented in Chapter 5 below indicate that varied assignment was not a key factor for the reduction of the effects in the current Experiment 3, either. In the experiments presented below the color-to-key assignment remained constant throughout a complete block of trials.

Implications for S‑R Translation. The backward compatibility effects (i.e., the compatibility effects on the verbal task) in the first two experiments suggest that S2-R2 translation overlapped in time with S1-R1 processing. The compatibility effects seem to be larger when verbal RTs are slow, indicating that the effect increases with increased overlap in processing of the two tasks. This finding appears to be consistent with other results (e.g., Lien & Proctor, 2000; Logan & Schulkind, 2000) that show larger compatibility effects at short as opposed to long SOAs. Moreover, the compatibility effects in Experiments 1 and 2 (as well as the lack thereof in Experiment 3) were largely independent of IRI, indicating that such effects do not depend on temporal response grouping, that is, on withholding R1 execution until R2 is selected. Rather, these results imply parallel activation or retrieval of R1 and R2 information in dual-task performance, leading to compatibility effects when task relevant response dimensions overlap, and to a lack of effects when they do not overlap.


[page 102↓]

Evidence for parallel response activation is difficult to reconcile with strong response selection bottleneck interpretations of the PRP effect that assume that S2-R2 translation has to wait until S1-R1 translation is completed. I do not argue, however, that parallel S‑R translation eliminates the usual PRP effect. While I did not include an SOA manipulation in the present experiments, other studies did (e.g., Lien & Proctor, 2000; Logan & Schulkind, 2000). Their results show a typical PRP pattern (i.e., slowed R2 responses when SOA is short) over and above inter-task consistency effects. Hommel (1998; see also Lien & Proctor, 2002) therefore suggested to distinguish between response activation, on the one hand, and response identification or decision processes, on the other hand, much as the DO model of SRC does (e.g., Kornblum et al., 1990; also see Logan & Gordon, 2001). According to this view, response activation proceeds in parallel, whereas response identification is serial. Consistency effects have been attributed to parallel response activation processes, whereas the PRP effect has been explained by serial response identification.

Interestingly, the size of the consistency effects in my Experiments did not (consistently) change with practice. This finding is consistent with the results obtained by Hommel (1998), and presumably implies that, whatever changes in processing (such as short cuts, instance based processing, or strengthening of rules) may lead to more efficient translation with practice (i.e. faster overall RTs), these changes do not affect the codes used for response selection in tasks like the ones used here. Furthermore, – when present at all – substantial forward as well as backward effects were already observed in the first block cluster. Therefore it seems as if automatic translation develops relatively quickly even when S‑R mappings are arbitrary. This conclusion receives support from studies that demonstrated automatic response activation according to implemented mappings after relatively little practice even when (a) no responses were required on the secondary task on a complete block of trials (Azuma, Prinz, & Koch, in press), (b) the old mapping was no longer valid (i.e., when new mappings had been instructed for the secondary task; Hommel & Eglau, 2002; also see Wenke & Frensch, 2000, for analogous findings with a paradigm more similar to the one used here), and when (c) memory load was increased (Hommel & Eglau, 2002).

According to one interpretation of such findings (cf. footnote 6, Chapter 3.2), direct S‑R links are established after relatively little practice that lead to automatic response activation although stimuli and responses do not (conceptually) overlap (i.e., no SRC exists; Hommel & Eglau, 2002; Proctor & Lu, 1999). Alternatively, activation is transmitted automatically and [page 103↓]in parallel along the translation (conditionally automatic) route once it is implemented (Lien & Proctor, 2002; Tagliabue et al., 2000). Although both explanations are generally consistent with a response coding interpretation of my results, according to the latter an alternative interpretation is also possible.

More specifically, according to the ‘automatic translation’ account it could be argued that such effects as those observed in Experiment 2 (and the lack thereof in Experiment 3) are ‘located’ at some intermediate translation stage instead of reflecting type of response coding. Such a view would, for instance, attribute inter-task consistency effects to verbal codes (i.e., location or color names) that mediate responding on the manual task effects, and not necessarily to response coding per se. Although this possibility seems unlikely, given that it is only tenable if one assumes that incorrect location or color names are even retrieved when no longer valid (Wenke & Frensch, 2000) or needed (Azuma et al., in press), it cannot be ruled out with the experimental approach chosen in Experiments 1-3.

Therefore, Experiments 4 and 5 (Chapter 5) tried to extend the present findings to a task that is commonly accepted to be associated with automatic response activation, namely the Simon task.


Footnotes and Endnotes

7 Actually, the total sample (N = 47, see Section Participants, above) contains participants that were excluded according to these criteria and subjects that were tested in their place. Therefore, mappings and quasi-random sequences were approximately counterbalanced in the remaining (N = 34) sample. This also holds for Experiments 2 and 3.

8 Separate analyses on the data of excluded participants revealed a similar pattern of results as in the original analyses.

9 Excluding the participant that produces compatibility effect outlier in RTs from the PI analysis did not affect the error rates in any way (see Table 5). Therefore I conducted the PI analysis on the complete data set.

10 The outlier subject from the main analysis did not fulfill the pre-defined error criteria when the practice block was included and was therefore excluded from the practice analyses.

11 As in Experiment 1, these data describe the total sample, including the participants that were excluded and those that were tested in their place (see data analysis section for details).

12 Running a MANOVA instead of separate ANOVAs on RT and PI leads to the same conclusions. This also applies to the comparison between Experiments 3 and 2.

13 MANOVAs on the practice results of Experiments 1 and 2 that were run for reasons of comparability also revealed a relatively strong dependence on PI. However, in those experiments PI and RT effects go in the same direction. Therefore, those MANOVA outcomes do not contradict the RT analyses.



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