- Email: [email protected]
The Tower of London: the impact of instructions, cueing, and learning on planning abilities
Cognitive Brain Research 17 (2003) 675–683 www.elsevier.com / locate / cogbrainres
Research report
The Tower of London: the impact of instructions, cueing, and learning on planning abilities J.M. Unterrainer a , *, B. Rahm a , R. Leonhart b , C.C. Ruff c , U. Halsband a a b
Neuropsychology, Institute of Psychology, University of Freiburg, Engelbergerstraße 41, D-79085 Freiburg, Germany Department of Rehabilitation Psychology, Institute of Psychology, University of Freiburg, D-79085 Freiburg, Germany c Institute of Cognitive Neuroscience, University College London, 17 Queen Square, London WC1 N 3 AR, UK Accepted 17 June 2003
Abstract The Tower of London (ToL) is a well-known test of planning ability, and commonly used for the purpose of neuropsychological assessment and cognitive research. Its widespread application has led to numerous versions differing in a number of respects. The present study addressed the question whether differences in instruction, cueing, and learning processes systematically influence ToL performance across five difficulty levels (three to seven moves). A total of 81 normal adults were examined in a mixed design with the between-subject factor instruction (online versus mental preplanning) and the within-subject factors cueing (cue versus non-cue test version) and learning processes ( first block and second block). We also assessed general intelligence for further analyses of differences between instruction groups. In general, there was a significant main effect across the difficulty levels indicating that the rate of incorrect solutions increased with problem difficulty. The participants who were instructed to make full mental plans before beginning to execute movements (preplanning) solved significantly more problems than people who started immediately with task-related movements (online). As for the cueing conditions, participants with the minimum number of moves predetermined (cue) could solve more trials than people who were only instructed to solve the problems in as few moves as possible (non-cue). Participants generally increased performance in the second part of the test session. However, an interaction of presentation order of the cueing condition with learning indicated that people who started the tasks with the non-cue version showed significantly better performance in the following cue condition, while participants who started with the cue condition stayed at the same performance level for both versions. These findings suggest that instruction, cueing conditions, and learning processes are important determinants of ToL performance, and they stress the necessity of standardized application in research and clinical practice. 2003 Elsevier B.V. All rights reserved. Theme: Neural basis of behaviour Topic: Cognition Keywords: Tower of London; Planning; Instruction; Cueing; Learning
1. Introduction The Tower of London (ToL) was originally developed by Shallice [30] in order to assess executive planning abilities. The ToL is a modification of the Tower of Hanoi that enables researchers and clinicians to test different levels of planning difficulties with validated psycho*Corresponding author. Tel.: 149-761-203-2464; fax: 149-761-2039438. E-mail address: [email protected] (J.M. Unterrainer). 0926-6410 / 03 / $ – see front matter 2003 Elsevier B.V. All rights reserved. doi:10.1016 / S0926-6410(03)00191-5
metrical standards. The original ToL test material included two identical tower structures, one for the patient and one for the examiner. Each structure consisted of three wooden sticks of different length, mounted on a block base. Three beads (red, green, and blue) were placed on the sticks in a prescribed start position. For each problem, the three beads had to be moved from the starting configuration to a target position in the minimum number of moves possible. In the original version, this minimum number of moves ranged from two to five moves per problem. Since 1982, the ToL has been employed in numerous studies of executive functioning. In clinical populations,
676
J.M. Unterrainer et al. / Cognitive Brain Research 17 (2003) 675–683
for instance, performance of the ToL was examined in patients with frontal lobe lesions and dysfunctions [5,6,15,20,30,31], schizophrenia [17,25,32], obsessive compulsive disorders [35], Huntington’s disease [12,37], and Parkinson disease [7,21]. However, the ToL has also been employed in behavioral and neuroimaging studies [2,3,7,8,11–13,16,19,23,26,28,29,36,38]. For these purposes, a broad range of different ToL versions were developed that differed from the original version (for review see Ref. [4]). In addition to the original three-rod design, Kafer and Hunter [11] used a modified version with four beads and four rods. Phillips and co-workers [26,27] and Ward and Allport [36] argued that the threedisc ToL, although useful for special populations, is too simple for the investigation of healthy subjects’ planning ability. They increased the number of discs to four and five and equalized the rods’ length to enable longer move sequences (see also Ref. [9]). Other differences arose in the exact instructions given to the subjects. While participants were often instructed to make full mental plans before beginning to execute movements [9,18,22,24], no explicit instructions were given in other studies [33]. It can clearly be expected that these differences in instructions have a considerable influence on performance, since preplanning the full set of moves before executing them appears a more promising strategy than solving the problems ‘online’, in a step-by-step fashion. The possible consequences of such different instruction modes were directly examined by Phillips et al. [27]. They compared three different types of instructions: (a) no specific planning instruction given, just to solve the problems in as few moves as possible; (b) asked to construct mentally a full plan of a minimum move path before executing the solution (c) informed of the minimum moves in which to solve the task, and then asked to construct mentally a full plan before execution. Interestingly, Phillips et al. [27] did not observe differences in the number of trials solved in minimum moves between the three instruction conditions, although the preplanning time was significantly longer in the conditions b and c as compared to a. It may be concluded from these results that instructions do only influence the time spent preplanning the solution, but not performance on the ToL. However, other studies have underlined that preplanning may be an important determinant of coming up with the correct solution. For example, Ward and Allport [36] and Unterrainer et al. [34] found that better performance was substantially correlated with longer preplanning time. In addition, Hodgson et al. [10] provided evidence that strategic patterns of eye movements during the planning phase are good indicators of whether a participant is an efficient or inefficient problem solver. Apart from the effects of preplanning, Phillips et al. [27] also failed to find differences between the non-cue (minimum number of moves not presented) and cue condition (minimum number of moves told). This is at odds with
introspection, since obviously the cue condition with its hint for effective planning should be easier to solve. One of the aims of the present study was to re-examine the effects of instruction and cueing on ToL performance. Since Phillips et al. [27] used a hierarchical study design, they were restricted to comparisons between the three groups. In order to overcome this limitation, we developed an experimental design for two instruction groups which also allowed within-subject comparisons between the two cueing conditions. In addition to effects of instructions and cueing, we examined whether previous experience with the planning operations required by the ToL also has an influence on task performance. For example, it appears possible that participants learn to solve planning tasks in the course of solving ToL problems and therefore increase their performance in the second part of a test session. Such effects could also interact with test instructions and cueing, e.g. participants may mostly benefit from cues about the minimum number of moves at the beginning of a test session, while they are able to solve the task efficiently without such cues at later stages. We thus set up an experimental design that allowed us to examine the effects of instruction, cueing, learning, and their interactions. Four main questions were addressed:
1. Does the instruction to make a full mental plan before beginning to execute movements result in more correctly solved ToL trials compared to the instruction of making the trials ‘online’? 2. Do people show better ToL performance when the number of moves to solve a problem is presented compared to people who are instructed to solve a problem in the minimum number of moves? 3. Are there any learning effects when people have to solve ToL problems? 4. Do different instruction and cueing conditions have an impact on the learning of planning abilities?
2. Methods
2.1. Participants A total number of 81 unpaid psychology students from the University of Freiburg participated in this study (63 females, 18 males). The mean age of the participants was 23.8 (S.D.55.9), ranging from 19 to 39 years. The experiment lasted for approximately 1 h.
2.2. Materials Participants were administered a computerized version of the ToL and the MWT-B, a test measuring crystallized intelligence.
J.M. Unterrainer et al. / Cognitive Brain Research 17 (2003) 675–683
2.2.1. ToL A computerized version of the ToL displayed the goal and start configurations simultaneously in the upper and the lower half of the screen, respectively. Participants were instructed to transform the start state (lower field) into the goal state by means of a computer mouse, following three basic rules: (1) only one ball may be moved at a time, (2) a ball may only be moved if no other ball is on top of it, and (3) three balls may be placed on the tallest peg, two balls on the middle peg, and one ball on the shortest peg. Note that the program did not allow rule-incongruent moves. With respect to the different cueing conditions, one version was designed to ask for ‘as few moves as possible’ (non-cue) whereas the second version provided the minimal number of moves to solve the problem on each screen (cue). Forty ToL problems had to be solved (20 in the first and 20 and in the second block, respectively), ranging from three to seven moves, and presented in randomized order. There was no time limit set to solve a problem. Participants did not get any feedback on success. The following measures were recorded: • Number of correctly solved trials defined as the number of problems solved in minimum number of moves • Preplanning time (defined as the time between the presentation of each problem and the first touch of a ball) • Movement execution time (defined as the time between the first touch of a ball and the final solution of the problem) For additional analyses, the total number of moves across all problems and the total amount of excess moves (defined as the number of moves exceeding the minimum number of moves) were assessed.
2.2.2. General intelligence The MWT-B (Mehrfachwahl-Wortschatz-Test) [14] was administered to compare the intelligence level of the two instruction groups. This test measures crystallized verbal intelligence; it consists of 37 rows providing five words in each line, only one of which is an existing German word. These words have to be identified by the participant. The number of recognized words was recorded. 2.3. Procedure Each participant was randomly assigned to one of the two instruction groups: preplanning (n539) or online (n5 42). For the preplanning condition, the instruction was to make a full mental plan before initiating the execution of the moves (preplanning condition). The online group, in contrast, was instructed to start solving the problem as soon as possible. In order to examine learning-related
677
differences, we defined the start condition completed by a participant as the ‘first block’ and the continue condition as ‘second block’. In order to balance the design, both instruction groups were divided into participants who started with the cue condition (n520 for preplanning and n521 for online) and people who began with the non-cue condition (n519 for preplanning and n521 for online).
3. Results The following results present three ANOVAs for preplanning time, movement execution time, and performance (number of correctly solved trials), respectively. Note that due to our experimental design, the main effect of cueing condition is represented by the interaction between learning (first and second block) and the presentation order of the cueing conditions (starting with the cue versus the non-cue condition).
3.1. Preplanning times To answer the question whether the different instruction groups, cueing, and learning conditions yielded different preplanning times, a 2323235 ANOVA on the ToL preplanning time across the five difficulty levels was computed. There was a significant main effect between the online and preplanning group, F(1,77)548.07, P,0.001. The planning time of the preplanning group was twice as long as time taken by the online group (mean: 14 932 vs. 7211 ms). There was also a significant main effect in the cueing condition [F(1,77)539.54, P,0.001], indicating longer preplanning times in the cue compared to the non-cue condition (13 084 vs. 9058 ms). No main effect was found for the learning conditions, F(1,77)5.46. An additional interaction between instruction, cueing, and learning condition [F(1,77)58.41, P50.005] indicated that participants who started with the cue condition had significantly longer preplanning times than people who began with the non-cue condition. This effect was primarily present in the preplanning group (see Fig. 1). A significant effect of the five difficulty levels on preplanning time was found [F(4,308)568.99, P,0.001]. Preplanning time increased with difficulty level, and post hoc analyses with Bonferroni correction showed that there were significant differences across all levels except for levels 5 and 6. There was also a significant interaction between instruction and difficulty level [F(4,308)517.94, P,0.001], indicating that the more difficult the problems, the more time participants spent for mental planning in the preplanning compared to the online condition (see Fig. 2).
3.2. Movement execution times A 2323235 ANOVA on the ToL movement execution time across the five difficulty levels was computed. There
678
J.M. Unterrainer et al. / Cognitive Brain Research 17 (2003) 675–683
Fig. 1. Preplanning times in milliseconds of different instruction groups and cueing conditions with respect to block 1 and 2.
were no significant main effects of both instruction [F(1,77)5.29] and cueing [F(1,77)5.12], but there was a significant effect of learning [F(1,77)595.51, P,0.001]. People showed shorter movement execution times in the second (mean: 12 641 ms) than in the first block (mean: 16 792 ms). A significant interaction between learning and instruction was also found [F(1,77)56.81, P,0.01]: as shown in Fig. 3 participants in the preplanning group had longer movement execution time in the start condition than
people in the online group, whereas in the continue condition the inverse pattern was present. A significant effect of difficulty level could be found [F(4,308)5246.6, P,0.001]; movement execution time increased with more difficult problems. Multiple comparisons (Bonferroni) revealed significant differences between all difficulty levels. In addition, a significant interaction between difficulty level and learning could be observed [F(4,308)514.101, P,0.001], showing that the more difficult the problems, the less time participants spent
Fig. 2. Preplanning time in milliseconds of the instruction groups across five difficulty levels.
Fig. 3. Movement execution times in milliseconds of different instruction groups with respect to block 1 and 2.
J.M. Unterrainer et al. / Cognitive Brain Research 17 (2003) 675–683
Fig. 4. Movement execution times across the five difficulty levels for block 1 and 2.
for movement execution in the second compared to the first block (see Fig. 4).
3.3. Number of correctly solved trials The effects of instruction (online vs. preplanning), cueing (cue vs. non-cue), and learning (first / second block) on the number of correctly solved trials for the five difficulty levels of the ToL trials were again examined
679
with a repeated-measure ANOVA (2323235). A significant main effect was found for instruction, F(1,77)511.53, P50.001. Post-hoc tests showed that the preplanning group attained more correctly solved trials than the online group (preplan: mean526.03 versus online: mean523.11). Additional analyses revealed that the participants in the preplanning group also made fewer number of moves than those in the online condition (mean: 243.6, S.D.: 20.9 vs. mean: 265.8, S.D.: 31.7 t(79)523.98, P,0.001). There was also a significant main effect for the cueing conditions, F(1,77)519.18, P,0.001, indicating that cued problems were easier to solve than non-cued problems (mean 12.92 versus mean 11.65). The analysis of the learning conditions revealed that people solved significantly more problems correctly in the second block (mean 12.95) compared to the first block (mean 11.57), F(1,77)5 23.04, P,0.001. In order to evaluate differences between participants starting with the cued condition and continuing with the non-cued in the second block and vice versa, additional t-tests were computed. They showed that people had a significant increase in performance when they began with the non-cued condition and had to go on with the cued condition (mean: 10.7 vs. mean: 13.38), t(40)52 5.77, P,0.001. In contrast, no significant difference in performance in the first block and in the second block was found when people started with the cue condition and proceeded with the non-cue condition (mean: 12.42 vs. mean: 12.54, see Fig. 5), t(40)520.36, P5ns. Overall, there was a significant effect of difficulty level F(4,308)5295.76, P,0.001. A linear relationship was found between difficulty level and number of correctly solved trials. Post-hoc test of significance (Tukey’s H.S.D. test with Bonferroni correction) showed that significant
Fig. 5. Correctly solved trials of cueing conditions for block 1 and 2.
680
J.M. Unterrainer et al. / Cognitive Brain Research 17 (2003) 675–683
Fig. 6. Number of correctly solved trials across the five difficulty levels for block 1 and 2.
differences existed between all difficulty levels. In addition, an interaction between learning condition and difficulty level [F(4,308)56,523, P,0.001] indicated that participants attained better performance in the second than in the first block in the higher difficulty levels (see Fig. 6). Additional correlational analyses showed that there was a significant relationship between preplanning time and correctly solved trials (r5.47; P,0.001) and a negative correlation between correctly solved trials and movement execution time (r520.33; P50.002). After excluding the total number of excess moves from movement execution time, this correlation decreased but remained significant (r520.28; P50.012).
3.4. Intelligence level of the different groups A Student’s t-test comparing general intelligence between the two instruction group showed no significant differences (online: mean531.16; S.D.53.2; preplanning: mean: 31.28, S.D.53.0), t(79)520.18, P5ns.
4. Discussion The present findings show that differences in instructions, cueing, and learning have a considerable impact on ToL performance. As we employed the original Tower of London configuration of Shallice [30] for all conditions, we could also show that the ToL in its original version provides a broad spectrum of problems applicable for healthy normals without gaining ceiling effects. As assumed in our first hypothesis, people who were
instructed to make full mental plans before beginning with movement execution solved more trials correctly than participants who were not explicitly told to preplan the problems. The significant increase in the number of correctly solved trials in the preplanning group was rather small (mean of three problems). It is important to stress that we did not find differences in general intelligence between the instruction groups. The better results of the preplanning group thus cannot be explained by a higher overall intelligence of the participants. The ‘preplanning’ group spent the double amount of time on planning compared to the ‘online’ group; this confirms that the participants complied with the demanded instructions. Preplanning time also increased with difficulty level, and it was higher for correctly solved trials. All these findings underline the assumption that longer, or more elaborate, preplanning is accompanied by better ToL performance. A recent study by Phillips et al. [27] also showed that participants who were instructed to preplan spent more time on mental planning before initiating movement than participants who were just instructed to solve the problems. In addition, the biggest differences in preplanning time between the groups were also present on the most difficult trials. However, in contrast to the present results, Phillips et al. [27] failed to find an effect of preplanning and cueing on the accuracy of solving ToL trials. In their discussion, they asked if preplanning is generally ‘wasted time’, since participants in all conditions took a similar numbers of moves to the solution. The results of our study clearly contradict this assumption. Participants in the preplanning group made significantly fewer moves to solution than the online group; they therefore seemed to have used the preplanning time quite effectively. This is in agreement with results by Ward and Allport [36], who found that individuals who spent more time on planning made fewer errors on the ToL task. What, then, happens during the preplanning phase? Based on our data and the data by Ward and Allport [36], one may argue that people mentally visualize the moves leading to the solution of the problem before executing them. This view is, for example, supported by Hodgson [10], who demonstrated that problem solvers show timerelated eye-movement patterns in different preplanning phases during the ToL-task. Efficient and poor performers even displayed different gaze direction patterns, again underlining that imagining, or mentally ‘testing’ possible moves, is related to how well a problem is solved. However, an important question remains whether problem solvers work out the full sequence of moves during planning, encode it in memory, and merely execute it afterwards, or whether they simply keep in mind some identified ‘key manoeuvres’ that benefit the subsequent problem solution during movement execution. The first of these hypotheses would clearly predict that movement execution times are shorter for conditions with preplanning than for conditions where participants plan the movements
J.M. Unterrainer et al. / Cognitive Brain Research 17 (2003) 675–683
online, as participants merely have to ‘replay’ a generated sequence maintained in working memory. This effect should be most marked for sequences involving a large number of moves. Interestingly, the results of the present study contradict these assumptions, since the movement execution time did not differ between the preplanning and the online group (preplanning: 14 952 ms vs. online: 14 485 ms), t(80)50.6, P5ns. It thus appears that for the preplanning group, movement execution did not simply provide a ‘pure’ replay of the complete mental plan, but rather is used for online planning of the next step. Taken together with the rather small increase of performance when preplanning, these results support the second notion described above, stating that the main operations performed during preplanning are related to the identification of critical ‘key manoeuvres’ that benefit subsequent online planning during movement execution. This means that preplanning is effective inasmuch as it facilitates subsequent planning operations, but it appears rather timeconsuming, and it does not provide a problem solver with a fully worked-out set of moves to problem solution. With respect to the cueing manipulation, participants clearly showed better performance if they were cued with the minimum number of moves necessary to solve the problem. Overall, there was a main effect on the number of correctly solved trials and on preplanning time between the cue and non-cue condition. We had expected such a pattern of results, since the predetermined number of moves should help participants to find the correct solution. For instance, the hint how many moves will be necessary may structure and reduce the problem space, therefore increasing performance. It is noteworthy in this context that our participants spent more time on mental planning in the cue as compared to the non-cue condition. This effect was predominantly found in the preplanning group; they spent about 4 s longer on preplanning in the cue condition than in the non-cue condition (Fig. 1). Obviously, the instructions to plan ahead and the hint on how many moves the problem can be solved lead to a speed–accuracy trade-off. Under these circumstances, participants try as long as possible to elaborate the optimal solution, which enhances but slows performance. In contrast, if participants are not cued and instructed to plan ahead, they tend to produce faster but more often incorrect solutions. Crucially, people are also not aware of these failures, since they do not get any feedback on the correctness of their responses. Irrespective of these putative mechanisms, the present findings clearly support the assumption that mental preplanning and cueing increase performance on the ToL trials. This is not in line with the results by Phillips et al. [27], who failed to find any significant effects of these manipulations on performance. However, the graph given in Phillips et al. [27] also states that their ‘inform group’ (mental preplanning plus minimum number of moves told) solved more problems than the online group, and the lack
681
of significance of this difference can easily be attributed to the considerably lower statistical power of the study (three groups were compared on the performance of 20 ToL problems only). Additionally, Phillips et al. [27] used the five disc ToL version, which further limits the comparability of the two studies. As for learning effects, participants on average solved one problem more in the second than in the first block. These learning effects were more pronounced for more difficult problems (as shown in Fig. 6). This clearly demonstrates that people learnt to develop planning strategies in the course of the two blocks, as already demonstrated with the Tower of Hanoi, a similar test of executive planning [1]. Interestingly, learning effects were not observed for the preplanning times. There was no main effect between the first and second block, and no decrease in preplanning time between the second block over the two learning periods. However, there was a significant overall reduction of time spent for the problems between the first and the second block if the preplanning and the movement execution time was combined (first block mean: 27 499 ms, second block mean: 23 778), t(80)53.8, P,0.001. This effect remained significant even after controlling for unnecessary wrong moves, although the level of significance decreased. Thus learning effects were present for movement execution, but not for preplanning time. One possible explanation for this finding is that participants may have used movement execution time to further elaborate their non-finished solutions. Since preplanning time did not differ between the first and second block, the effect of learning on execution time could thus have been reflected by less corrective or trial-and-error processes in movement execution. This interpretation is supported by the finding that longer movement execution times were observed in the first block for the preplanning as compared to the online group, while this pattern changed in the opposite direction in the second block (see Fig. 3). As a fourth major question of this paper, we asked whether differences in instructions and cueing have an impact on the learning of solving ToL problems. Interestingly, in our study learning was not influenced by planning instructions (preplan vs. online). With respect to cueing, we found significant differences in the first and second block only if people started with the non-cue condition, while those participants that started with the cue-condition could not significantly increase performance level in the following non-cue condition. Additional analyses showed that there was a significant difference between non-cue in the first compared to non-cue in the second block [mean: 10.7 versus mean: 12.54; t(39)53.31, P,0.001]. This rise in performance in the second block can thus clearly be interpreted as a learning effect. Therefore one could argue that after beginning with the easier cue version, performance in the more difficult non-cue version was better in the second block since participants had the chance to become familiar with an easier version of the problem
682
J.M. Unterrainer et al. / Cognitive Brain Research 17 (2003) 675–683
solving beforehand. In contrast, starting the experiment with the more difficult non-cue version was expressed in the low start performance, and resulted in a highly significant increase for the easier cue version in the second block. It would be an interesting question for further studies to examine learning effects separately for cued and non-cued versions of the ToL. In addition, an analysis of the performance on single trials of different complexity levels over the full testing period could provide new insights into learning processes in problem solving.
[9]
[10]
[11]
[12]
5. Conclusions The results of the present paper clearly demonstrate that different instructions, cueing conditions, and learning effects have a strong impact on the ToL performance. Our findings therefore help to explain divergences in the results of the numerous publications on the Tower of London, and they imply that due to differences in the versions of the test employed, comparisons between the results of different studies are hardly possible (see also Ref. [4]). It follows that one standardized version of the ToL should be applied in research and clinical practice, or that at least all necessary parameters should be reported. In addition, our study showed that the original ToL version by Shallice [30] offers a variety of problems, which appear suitable for the application in research with both special patient groups and healthy volunteers. The employment of this original version would thus clearly facilitate the comparability of different samples.
[13]
[14] [15]
[16]
[17]
[18]
[19]
References [1] J. Ahonniska, T. Ahonen, T. Aro, A. Tolvanen, H. Lyytinen, Practice effects on visuomotor and problem-solving tests by children, Percept. Motor Skills 92 (2001) 479–494. [2] N.C. Andreasen, K. Rezai, R. Alliger, V.W. Swayze, M. Flaum, P. Kirchner, G. Cohen, D.S. O’Leary, Hypofrontality in neurolepticnaive patients and in patients with chronic schizophrenia. Assessment with xenon 133 single-photon emission computed tomography and the Tower of London, Arch. Gen. Psychiatry 49 (1992) 943– 958. [3] S.C. Baker, R.D. Rogers, A.M. Owen, C.D. Frith, R.J. Dolan, R.S. Frackowiak, T.W. Robbins, Neural systems engaged by planning: a PET study of the Tower of London task, Neuropsychologia 34 (1996) 515–526. [4] K.W. Berg, D. Byrd, The tower of london spatial problem-solving task: enhancing clinical and research implementation, J. Clin. Exp. Neuropsychol. 24 (2002) 586–604. [5] D. Carlin, J. Bonerba, M. Phipps, G. Alexander, M. Shapiro, J. Grafman, Planning impairments in frontal lobe dementia and frontal lobe lesions, Neuropsychologia 38 (2000) 655–665. [6] J. Cockburn, Performance on the Tower of London test after severe head injury, J. Int. Neuropsychol. Soc. 1 (1995) 537–544. [7] A. Dagher, A.M. Owen, H. Boecker, D.J. Brooks, Mapping the network for planning: a correlation PET activation study with the TOL task, Brain 122 (1999) 1973–1987. [8] A. Dagher, A.M. Owen, H. Boecker, D.J. Brooks, The role of the
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
striatum and hippocampus in planning: a PET activation study in Parkinson’s disease, Brain 124 (2001) 1020–1032. K.J. Gilhooly, V. Wynn, L.H. Phillips, R.H. Logie, S. Della Sala, Visuo-spatial and verbal working memory in the five-disc Tower of London task: An individual approach, Think. Reason. 8 (2002) 165–178. T.L. Hodgson, A. Bajwa, A.M. Owen, C. Kennard, The strategic control of gaze direction in the Tower-of-London task, J. Cogn. Neurosci. 12 (2000) 894–907. K.L. Kafer, M. Hunter, On testing the face validity of planning / problem-solving tasks in a normal population, J. Int. Neuropsychol. Soc. 3 (1997) 108–119. K.W. Lange, B.J. Sahakian, N.P. Quinn, C.D. Marsden, T.W. Robbins, Comparison of executive and visuospatial memory function in Huntington’s disease and dementia of Alzheimer type matched for degree of dementia, J. Neurol. Neurosurg. Psychiatry 58 (1995) 598–606. R.H. Lazeron, S.A. Rombouts, W.C. Machielsen, P. Scheltens, M.P. Witter, H.B. Uylings, F. Barkhof, Visualizing brain activation during planning: the tower of London test adapted for functional MR imaging, Am. J. Neuroradiol. 21 (2000) 1407–1414. S. Lehrl, Mehrfachwahl-Wortschatz-Intelligenztest (MWT-B). 4. ¨ ¨ uberarbeitete Auflage, Hogrefe Test, Gottingen, 1999. H.S. Levin, D. Mendelsohn, M. Lilly, J.M. Fletcher, K.A. Culhane, S.B. Chapman, H. Harward, L. Kusnerik, D. Bruce, H.M. Eisenberg, Tower of London Performance in relation to magnetic resonance imaging following closed head injury in children, Neuropsychology 8 (1994) 171–179. R.G. Morris, S. Ahmed, G.M. Syed, B.K. Toone, Neural correlates of planning ability: frontal lobe activation during the Tower of London test, Neuropsychologia 31 (1993) 1367–1378. R.G. Morris, T. Rushe, P.W. Woodruffe, R.M. Murray, Problem solving in schizophrenia: a specific deficit in planning ability, Schizophr. Res. 14 (1995) 235–246. R.G. Morris, E.C. Miotto, J.D. Feigenbaum, P. Bullock, C.E. Polkey, The effect of goal–subgoal conflict on planning ability after frontaland temporal-lobe lesions in humans, Neuropsychologia 35 (1997) 1147–1157. R. Morice, A. Delahunty, Frontal / executive impairments in schizophrenia, Schizophr. Bull. 22 (1996) 125–137. A.M. Owen, J.J. Downes, B.J. Sahakian, C.E. Polkey, T.W. Robbins, Planning and spatial working memory following frontal lobe lesions in man, Neuropsychologia 28 (1990) 1021–1034. A.M. Owen, M. James, P.N. Leigh, B.A. Summers, C.D. Marsden, N.P. Quinn, K.W. Lange, T.W. Robbins, Fronto-striatal cognitive deficits at different stages of Parkinson’s disease, Brain 115 (1992) 1727–1751. A.M. Owen, B.J. Sahakian, J. Semple, C.E. Polkey, T.W. Robbins, Dopamine-dependent fronto-striatal planning deficits in early Parkinson’s disease, Neuropsychology 9 (1995) 1–24. A.M. Owen, J. Doyon, M. Petrides, A. Evans, Planning and spatial working memory: a PET study in humans, Eur. J. Neurosci. 8 (1996) 353–364. A.M. Owen, Cognitive planning in humans: neuropsychological, neuroanatomical and neuropharmacological perspectives, Prog. Neurobiol. 53 (1997) 431–450. C. Pantelis, T.R. Barnes, H.E. Nelson, S. Tanner, L. Weatherley, A.M. Owen, T.W. Robbins, Frontal-striatal cognitive deficits in patients with chronic schizophrenia, Brain 120 (1997) 1823–1843. L.H. Phillips, V. Wynn, K.J. Gilhooly, S. Della Sala, R.H. Logie, The role of memory in the Tower of London task, Memory 7 (1999) 209–231. L.H. Phillips, V.E. Wynn, S. McPherson, K.J. Gilhooly, Mental planning and the Tower of London task, Q. J. Exp. Psychol. 54 (2001) 579–597. K. Rezai, N.C. Andreasen, R. Alliger, G. Cohen, V. Swayze, D.S. O’Leary, The neuropsychology of the prefrontal cortex, Arch. Neurol. 50 (1993) 636–642.
J.M. Unterrainer et al. / Cognitive Brain Research 17 (2003) 675–683 [29] J.B. Rowe, A.M. Owen, I.S. Johnsrude, R.E. Passingham, Imaging the mental components of a planning task, Neuropsychologia 39 (2001) 315–327. [30] T. Shallice, Specific impairments of planning, Philos. Trans. R. Soc. Lond. B 298 (1982) 199–209. [31] T. Shallice, From Neuropsychology To Mental Structure, Cambridge University Press, Cambridge, 1988. [32] W.G. Staal, R. Hijman, H.E. Hulshoff Pol, R.S. Kahn, Neuropsychological dysfunctions in siblings discordant for schizophrenia, Psychiatr. Res. 95 (2000) 227–235. [33] C.M. Temple, R.A. Carney, S. Mullarkey, Frontal lobe function and executive skills in children with Turner’s syndrome, Dev. Neuropsychol. 12 (1996) 343–363. ¨ [34] J. Unterrainer, K. Quiske, K. Hoppe-Seyler, C. Meier, C. Muller, B. Rahm, R. Leonhart, U. Halsband, Planning abilities and the Tower
[35]
[36] [37]
[38]
683
of London: Is this task measuring a discrete cognitive function? J. Clin. Exp. Neuropsychol. (2003) submitted for publication. D.M. Veale, B.J. Sahakian, A.M. Owen, I.M. Marks, Specific cognitive deficits in tests sensitive to frontal lobe dysfunction in obsessive–compulsive disorder, Psychol. Med. 26 (1996) 1261– 1269. G. Ward, A. Allport, Planning and problem-solving using the 5-disc Tower of London Task, Q. J. Exp. Psychol. 50 (1997) 49–78. L.H. Watkins, R.D. Rogers, A.D. Lawrence, B.J. Sahakian, A.E. Rosser, T.W. Robbins, Impaired planning but intact decision making in early Huntington’s disease: implications for specific fronto-striatal pathology, Neuropsychologia 38 (2000) 1112–1125. M.C. Welsh, T. Satterlee-Cartmell, M. Stine, Towers of Hanoi and London: contribution of working memory and inhibition to performance, Brain Cogn. 41 (1999) 231–242.
Research report
The Tower of London: the impact of instructions, cueing, and learning on planning abilities J.M. Unterrainer a , *, B. Rahm a , R. Leonhart b , C.C. Ruff c , U. Halsband a a b
Neuropsychology, Institute of Psychology, University of Freiburg, Engelbergerstraße 41, D-79085 Freiburg, Germany Department of Rehabilitation Psychology, Institute of Psychology, University of Freiburg, D-79085 Freiburg, Germany c Institute of Cognitive Neuroscience, University College London, 17 Queen Square, London WC1 N 3 AR, UK Accepted 17 June 2003
Abstract The Tower of London (ToL) is a well-known test of planning ability, and commonly used for the purpose of neuropsychological assessment and cognitive research. Its widespread application has led to numerous versions differing in a number of respects. The present study addressed the question whether differences in instruction, cueing, and learning processes systematically influence ToL performance across five difficulty levels (three to seven moves). A total of 81 normal adults were examined in a mixed design with the between-subject factor instruction (online versus mental preplanning) and the within-subject factors cueing (cue versus non-cue test version) and learning processes ( first block and second block). We also assessed general intelligence for further analyses of differences between instruction groups. In general, there was a significant main effect across the difficulty levels indicating that the rate of incorrect solutions increased with problem difficulty. The participants who were instructed to make full mental plans before beginning to execute movements (preplanning) solved significantly more problems than people who started immediately with task-related movements (online). As for the cueing conditions, participants with the minimum number of moves predetermined (cue) could solve more trials than people who were only instructed to solve the problems in as few moves as possible (non-cue). Participants generally increased performance in the second part of the test session. However, an interaction of presentation order of the cueing condition with learning indicated that people who started the tasks with the non-cue version showed significantly better performance in the following cue condition, while participants who started with the cue condition stayed at the same performance level for both versions. These findings suggest that instruction, cueing conditions, and learning processes are important determinants of ToL performance, and they stress the necessity of standardized application in research and clinical practice. 2003 Elsevier B.V. All rights reserved. Theme: Neural basis of behaviour Topic: Cognition Keywords: Tower of London; Planning; Instruction; Cueing; Learning
1. Introduction The Tower of London (ToL) was originally developed by Shallice [30] in order to assess executive planning abilities. The ToL is a modification of the Tower of Hanoi that enables researchers and clinicians to test different levels of planning difficulties with validated psycho*Corresponding author. Tel.: 149-761-203-2464; fax: 149-761-2039438. E-mail address: [email protected] (J.M. Unterrainer). 0926-6410 / 03 / $ – see front matter 2003 Elsevier B.V. All rights reserved. doi:10.1016 / S0926-6410(03)00191-5
metrical standards. The original ToL test material included two identical tower structures, one for the patient and one for the examiner. Each structure consisted of three wooden sticks of different length, mounted on a block base. Three beads (red, green, and blue) were placed on the sticks in a prescribed start position. For each problem, the three beads had to be moved from the starting configuration to a target position in the minimum number of moves possible. In the original version, this minimum number of moves ranged from two to five moves per problem. Since 1982, the ToL has been employed in numerous studies of executive functioning. In clinical populations,
676
J.M. Unterrainer et al. / Cognitive Brain Research 17 (2003) 675–683
for instance, performance of the ToL was examined in patients with frontal lobe lesions and dysfunctions [5,6,15,20,30,31], schizophrenia [17,25,32], obsessive compulsive disorders [35], Huntington’s disease [12,37], and Parkinson disease [7,21]. However, the ToL has also been employed in behavioral and neuroimaging studies [2,3,7,8,11–13,16,19,23,26,28,29,36,38]. For these purposes, a broad range of different ToL versions were developed that differed from the original version (for review see Ref. [4]). In addition to the original three-rod design, Kafer and Hunter [11] used a modified version with four beads and four rods. Phillips and co-workers [26,27] and Ward and Allport [36] argued that the threedisc ToL, although useful for special populations, is too simple for the investigation of healthy subjects’ planning ability. They increased the number of discs to four and five and equalized the rods’ length to enable longer move sequences (see also Ref. [9]). Other differences arose in the exact instructions given to the subjects. While participants were often instructed to make full mental plans before beginning to execute movements [9,18,22,24], no explicit instructions were given in other studies [33]. It can clearly be expected that these differences in instructions have a considerable influence on performance, since preplanning the full set of moves before executing them appears a more promising strategy than solving the problems ‘online’, in a step-by-step fashion. The possible consequences of such different instruction modes were directly examined by Phillips et al. [27]. They compared three different types of instructions: (a) no specific planning instruction given, just to solve the problems in as few moves as possible; (b) asked to construct mentally a full plan of a minimum move path before executing the solution (c) informed of the minimum moves in which to solve the task, and then asked to construct mentally a full plan before execution. Interestingly, Phillips et al. [27] did not observe differences in the number of trials solved in minimum moves between the three instruction conditions, although the preplanning time was significantly longer in the conditions b and c as compared to a. It may be concluded from these results that instructions do only influence the time spent preplanning the solution, but not performance on the ToL. However, other studies have underlined that preplanning may be an important determinant of coming up with the correct solution. For example, Ward and Allport [36] and Unterrainer et al. [34] found that better performance was substantially correlated with longer preplanning time. In addition, Hodgson et al. [10] provided evidence that strategic patterns of eye movements during the planning phase are good indicators of whether a participant is an efficient or inefficient problem solver. Apart from the effects of preplanning, Phillips et al. [27] also failed to find differences between the non-cue (minimum number of moves not presented) and cue condition (minimum number of moves told). This is at odds with
introspection, since obviously the cue condition with its hint for effective planning should be easier to solve. One of the aims of the present study was to re-examine the effects of instruction and cueing on ToL performance. Since Phillips et al. [27] used a hierarchical study design, they were restricted to comparisons between the three groups. In order to overcome this limitation, we developed an experimental design for two instruction groups which also allowed within-subject comparisons between the two cueing conditions. In addition to effects of instructions and cueing, we examined whether previous experience with the planning operations required by the ToL also has an influence on task performance. For example, it appears possible that participants learn to solve planning tasks in the course of solving ToL problems and therefore increase their performance in the second part of a test session. Such effects could also interact with test instructions and cueing, e.g. participants may mostly benefit from cues about the minimum number of moves at the beginning of a test session, while they are able to solve the task efficiently without such cues at later stages. We thus set up an experimental design that allowed us to examine the effects of instruction, cueing, learning, and their interactions. Four main questions were addressed:
1. Does the instruction to make a full mental plan before beginning to execute movements result in more correctly solved ToL trials compared to the instruction of making the trials ‘online’? 2. Do people show better ToL performance when the number of moves to solve a problem is presented compared to people who are instructed to solve a problem in the minimum number of moves? 3. Are there any learning effects when people have to solve ToL problems? 4. Do different instruction and cueing conditions have an impact on the learning of planning abilities?
2. Methods
2.1. Participants A total number of 81 unpaid psychology students from the University of Freiburg participated in this study (63 females, 18 males). The mean age of the participants was 23.8 (S.D.55.9), ranging from 19 to 39 years. The experiment lasted for approximately 1 h.
2.2. Materials Participants were administered a computerized version of the ToL and the MWT-B, a test measuring crystallized intelligence.
J.M. Unterrainer et al. / Cognitive Brain Research 17 (2003) 675–683
2.2.1. ToL A computerized version of the ToL displayed the goal and start configurations simultaneously in the upper and the lower half of the screen, respectively. Participants were instructed to transform the start state (lower field) into the goal state by means of a computer mouse, following three basic rules: (1) only one ball may be moved at a time, (2) a ball may only be moved if no other ball is on top of it, and (3) three balls may be placed on the tallest peg, two balls on the middle peg, and one ball on the shortest peg. Note that the program did not allow rule-incongruent moves. With respect to the different cueing conditions, one version was designed to ask for ‘as few moves as possible’ (non-cue) whereas the second version provided the minimal number of moves to solve the problem on each screen (cue). Forty ToL problems had to be solved (20 in the first and 20 and in the second block, respectively), ranging from three to seven moves, and presented in randomized order. There was no time limit set to solve a problem. Participants did not get any feedback on success. The following measures were recorded: • Number of correctly solved trials defined as the number of problems solved in minimum number of moves • Preplanning time (defined as the time between the presentation of each problem and the first touch of a ball) • Movement execution time (defined as the time between the first touch of a ball and the final solution of the problem) For additional analyses, the total number of moves across all problems and the total amount of excess moves (defined as the number of moves exceeding the minimum number of moves) were assessed.
2.2.2. General intelligence The MWT-B (Mehrfachwahl-Wortschatz-Test) [14] was administered to compare the intelligence level of the two instruction groups. This test measures crystallized verbal intelligence; it consists of 37 rows providing five words in each line, only one of which is an existing German word. These words have to be identified by the participant. The number of recognized words was recorded. 2.3. Procedure Each participant was randomly assigned to one of the two instruction groups: preplanning (n539) or online (n5 42). For the preplanning condition, the instruction was to make a full mental plan before initiating the execution of the moves (preplanning condition). The online group, in contrast, was instructed to start solving the problem as soon as possible. In order to examine learning-related
677
differences, we defined the start condition completed by a participant as the ‘first block’ and the continue condition as ‘second block’. In order to balance the design, both instruction groups were divided into participants who started with the cue condition (n520 for preplanning and n521 for online) and people who began with the non-cue condition (n519 for preplanning and n521 for online).
3. Results The following results present three ANOVAs for preplanning time, movement execution time, and performance (number of correctly solved trials), respectively. Note that due to our experimental design, the main effect of cueing condition is represented by the interaction between learning (first and second block) and the presentation order of the cueing conditions (starting with the cue versus the non-cue condition).
3.1. Preplanning times To answer the question whether the different instruction groups, cueing, and learning conditions yielded different preplanning times, a 2323235 ANOVA on the ToL preplanning time across the five difficulty levels was computed. There was a significant main effect between the online and preplanning group, F(1,77)548.07, P,0.001. The planning time of the preplanning group was twice as long as time taken by the online group (mean: 14 932 vs. 7211 ms). There was also a significant main effect in the cueing condition [F(1,77)539.54, P,0.001], indicating longer preplanning times in the cue compared to the non-cue condition (13 084 vs. 9058 ms). No main effect was found for the learning conditions, F(1,77)5.46. An additional interaction between instruction, cueing, and learning condition [F(1,77)58.41, P50.005] indicated that participants who started with the cue condition had significantly longer preplanning times than people who began with the non-cue condition. This effect was primarily present in the preplanning group (see Fig. 1). A significant effect of the five difficulty levels on preplanning time was found [F(4,308)568.99, P,0.001]. Preplanning time increased with difficulty level, and post hoc analyses with Bonferroni correction showed that there were significant differences across all levels except for levels 5 and 6. There was also a significant interaction between instruction and difficulty level [F(4,308)517.94, P,0.001], indicating that the more difficult the problems, the more time participants spent for mental planning in the preplanning compared to the online condition (see Fig. 2).
3.2. Movement execution times A 2323235 ANOVA on the ToL movement execution time across the five difficulty levels was computed. There
678
J.M. Unterrainer et al. / Cognitive Brain Research 17 (2003) 675–683
Fig. 1. Preplanning times in milliseconds of different instruction groups and cueing conditions with respect to block 1 and 2.
were no significant main effects of both instruction [F(1,77)5.29] and cueing [F(1,77)5.12], but there was a significant effect of learning [F(1,77)595.51, P,0.001]. People showed shorter movement execution times in the second (mean: 12 641 ms) than in the first block (mean: 16 792 ms). A significant interaction between learning and instruction was also found [F(1,77)56.81, P,0.01]: as shown in Fig. 3 participants in the preplanning group had longer movement execution time in the start condition than
people in the online group, whereas in the continue condition the inverse pattern was present. A significant effect of difficulty level could be found [F(4,308)5246.6, P,0.001]; movement execution time increased with more difficult problems. Multiple comparisons (Bonferroni) revealed significant differences between all difficulty levels. In addition, a significant interaction between difficulty level and learning could be observed [F(4,308)514.101, P,0.001], showing that the more difficult the problems, the less time participants spent
Fig. 2. Preplanning time in milliseconds of the instruction groups across five difficulty levels.
Fig. 3. Movement execution times in milliseconds of different instruction groups with respect to block 1 and 2.
J.M. Unterrainer et al. / Cognitive Brain Research 17 (2003) 675–683
Fig. 4. Movement execution times across the five difficulty levels for block 1 and 2.
for movement execution in the second compared to the first block (see Fig. 4).
3.3. Number of correctly solved trials The effects of instruction (online vs. preplanning), cueing (cue vs. non-cue), and learning (first / second block) on the number of correctly solved trials for the five difficulty levels of the ToL trials were again examined
679
with a repeated-measure ANOVA (2323235). A significant main effect was found for instruction, F(1,77)511.53, P50.001. Post-hoc tests showed that the preplanning group attained more correctly solved trials than the online group (preplan: mean526.03 versus online: mean523.11). Additional analyses revealed that the participants in the preplanning group also made fewer number of moves than those in the online condition (mean: 243.6, S.D.: 20.9 vs. mean: 265.8, S.D.: 31.7 t(79)523.98, P,0.001). There was also a significant main effect for the cueing conditions, F(1,77)519.18, P,0.001, indicating that cued problems were easier to solve than non-cued problems (mean 12.92 versus mean 11.65). The analysis of the learning conditions revealed that people solved significantly more problems correctly in the second block (mean 12.95) compared to the first block (mean 11.57), F(1,77)5 23.04, P,0.001. In order to evaluate differences between participants starting with the cued condition and continuing with the non-cued in the second block and vice versa, additional t-tests were computed. They showed that people had a significant increase in performance when they began with the non-cued condition and had to go on with the cued condition (mean: 10.7 vs. mean: 13.38), t(40)52 5.77, P,0.001. In contrast, no significant difference in performance in the first block and in the second block was found when people started with the cue condition and proceeded with the non-cue condition (mean: 12.42 vs. mean: 12.54, see Fig. 5), t(40)520.36, P5ns. Overall, there was a significant effect of difficulty level F(4,308)5295.76, P,0.001. A linear relationship was found between difficulty level and number of correctly solved trials. Post-hoc test of significance (Tukey’s H.S.D. test with Bonferroni correction) showed that significant
Fig. 5. Correctly solved trials of cueing conditions for block 1 and 2.
680
J.M. Unterrainer et al. / Cognitive Brain Research 17 (2003) 675–683
Fig. 6. Number of correctly solved trials across the five difficulty levels for block 1 and 2.
differences existed between all difficulty levels. In addition, an interaction between learning condition and difficulty level [F(4,308)56,523, P,0.001] indicated that participants attained better performance in the second than in the first block in the higher difficulty levels (see Fig. 6). Additional correlational analyses showed that there was a significant relationship between preplanning time and correctly solved trials (r5.47; P,0.001) and a negative correlation between correctly solved trials and movement execution time (r520.33; P50.002). After excluding the total number of excess moves from movement execution time, this correlation decreased but remained significant (r520.28; P50.012).
3.4. Intelligence level of the different groups A Student’s t-test comparing general intelligence between the two instruction group showed no significant differences (online: mean531.16; S.D.53.2; preplanning: mean: 31.28, S.D.53.0), t(79)520.18, P5ns.
4. Discussion The present findings show that differences in instructions, cueing, and learning have a considerable impact on ToL performance. As we employed the original Tower of London configuration of Shallice [30] for all conditions, we could also show that the ToL in its original version provides a broad spectrum of problems applicable for healthy normals without gaining ceiling effects. As assumed in our first hypothesis, people who were
instructed to make full mental plans before beginning with movement execution solved more trials correctly than participants who were not explicitly told to preplan the problems. The significant increase in the number of correctly solved trials in the preplanning group was rather small (mean of three problems). It is important to stress that we did not find differences in general intelligence between the instruction groups. The better results of the preplanning group thus cannot be explained by a higher overall intelligence of the participants. The ‘preplanning’ group spent the double amount of time on planning compared to the ‘online’ group; this confirms that the participants complied with the demanded instructions. Preplanning time also increased with difficulty level, and it was higher for correctly solved trials. All these findings underline the assumption that longer, or more elaborate, preplanning is accompanied by better ToL performance. A recent study by Phillips et al. [27] also showed that participants who were instructed to preplan spent more time on mental planning before initiating movement than participants who were just instructed to solve the problems. In addition, the biggest differences in preplanning time between the groups were also present on the most difficult trials. However, in contrast to the present results, Phillips et al. [27] failed to find an effect of preplanning and cueing on the accuracy of solving ToL trials. In their discussion, they asked if preplanning is generally ‘wasted time’, since participants in all conditions took a similar numbers of moves to the solution. The results of our study clearly contradict this assumption. Participants in the preplanning group made significantly fewer moves to solution than the online group; they therefore seemed to have used the preplanning time quite effectively. This is in agreement with results by Ward and Allport [36], who found that individuals who spent more time on planning made fewer errors on the ToL task. What, then, happens during the preplanning phase? Based on our data and the data by Ward and Allport [36], one may argue that people mentally visualize the moves leading to the solution of the problem before executing them. This view is, for example, supported by Hodgson [10], who demonstrated that problem solvers show timerelated eye-movement patterns in different preplanning phases during the ToL-task. Efficient and poor performers even displayed different gaze direction patterns, again underlining that imagining, or mentally ‘testing’ possible moves, is related to how well a problem is solved. However, an important question remains whether problem solvers work out the full sequence of moves during planning, encode it in memory, and merely execute it afterwards, or whether they simply keep in mind some identified ‘key manoeuvres’ that benefit the subsequent problem solution during movement execution. The first of these hypotheses would clearly predict that movement execution times are shorter for conditions with preplanning than for conditions where participants plan the movements
J.M. Unterrainer et al. / Cognitive Brain Research 17 (2003) 675–683
online, as participants merely have to ‘replay’ a generated sequence maintained in working memory. This effect should be most marked for sequences involving a large number of moves. Interestingly, the results of the present study contradict these assumptions, since the movement execution time did not differ between the preplanning and the online group (preplanning: 14 952 ms vs. online: 14 485 ms), t(80)50.6, P5ns. It thus appears that for the preplanning group, movement execution did not simply provide a ‘pure’ replay of the complete mental plan, but rather is used for online planning of the next step. Taken together with the rather small increase of performance when preplanning, these results support the second notion described above, stating that the main operations performed during preplanning are related to the identification of critical ‘key manoeuvres’ that benefit subsequent online planning during movement execution. This means that preplanning is effective inasmuch as it facilitates subsequent planning operations, but it appears rather timeconsuming, and it does not provide a problem solver with a fully worked-out set of moves to problem solution. With respect to the cueing manipulation, participants clearly showed better performance if they were cued with the minimum number of moves necessary to solve the problem. Overall, there was a main effect on the number of correctly solved trials and on preplanning time between the cue and non-cue condition. We had expected such a pattern of results, since the predetermined number of moves should help participants to find the correct solution. For instance, the hint how many moves will be necessary may structure and reduce the problem space, therefore increasing performance. It is noteworthy in this context that our participants spent more time on mental planning in the cue as compared to the non-cue condition. This effect was predominantly found in the preplanning group; they spent about 4 s longer on preplanning in the cue condition than in the non-cue condition (Fig. 1). Obviously, the instructions to plan ahead and the hint on how many moves the problem can be solved lead to a speed–accuracy trade-off. Under these circumstances, participants try as long as possible to elaborate the optimal solution, which enhances but slows performance. In contrast, if participants are not cued and instructed to plan ahead, they tend to produce faster but more often incorrect solutions. Crucially, people are also not aware of these failures, since they do not get any feedback on the correctness of their responses. Irrespective of these putative mechanisms, the present findings clearly support the assumption that mental preplanning and cueing increase performance on the ToL trials. This is not in line with the results by Phillips et al. [27], who failed to find any significant effects of these manipulations on performance. However, the graph given in Phillips et al. [27] also states that their ‘inform group’ (mental preplanning plus minimum number of moves told) solved more problems than the online group, and the lack
681
of significance of this difference can easily be attributed to the considerably lower statistical power of the study (three groups were compared on the performance of 20 ToL problems only). Additionally, Phillips et al. [27] used the five disc ToL version, which further limits the comparability of the two studies. As for learning effects, participants on average solved one problem more in the second than in the first block. These learning effects were more pronounced for more difficult problems (as shown in Fig. 6). This clearly demonstrates that people learnt to develop planning strategies in the course of the two blocks, as already demonstrated with the Tower of Hanoi, a similar test of executive planning [1]. Interestingly, learning effects were not observed for the preplanning times. There was no main effect between the first and second block, and no decrease in preplanning time between the second block over the two learning periods. However, there was a significant overall reduction of time spent for the problems between the first and the second block if the preplanning and the movement execution time was combined (first block mean: 27 499 ms, second block mean: 23 778), t(80)53.8, P,0.001. This effect remained significant even after controlling for unnecessary wrong moves, although the level of significance decreased. Thus learning effects were present for movement execution, but not for preplanning time. One possible explanation for this finding is that participants may have used movement execution time to further elaborate their non-finished solutions. Since preplanning time did not differ between the first and second block, the effect of learning on execution time could thus have been reflected by less corrective or trial-and-error processes in movement execution. This interpretation is supported by the finding that longer movement execution times were observed in the first block for the preplanning as compared to the online group, while this pattern changed in the opposite direction in the second block (see Fig. 3). As a fourth major question of this paper, we asked whether differences in instructions and cueing have an impact on the learning of solving ToL problems. Interestingly, in our study learning was not influenced by planning instructions (preplan vs. online). With respect to cueing, we found significant differences in the first and second block only if people started with the non-cue condition, while those participants that started with the cue-condition could not significantly increase performance level in the following non-cue condition. Additional analyses showed that there was a significant difference between non-cue in the first compared to non-cue in the second block [mean: 10.7 versus mean: 12.54; t(39)53.31, P,0.001]. This rise in performance in the second block can thus clearly be interpreted as a learning effect. Therefore one could argue that after beginning with the easier cue version, performance in the more difficult non-cue version was better in the second block since participants had the chance to become familiar with an easier version of the problem
682
J.M. Unterrainer et al. / Cognitive Brain Research 17 (2003) 675–683
solving beforehand. In contrast, starting the experiment with the more difficult non-cue version was expressed in the low start performance, and resulted in a highly significant increase for the easier cue version in the second block. It would be an interesting question for further studies to examine learning effects separately for cued and non-cued versions of the ToL. In addition, an analysis of the performance on single trials of different complexity levels over the full testing period could provide new insights into learning processes in problem solving.
[9]
[10]
[11]
[12]
5. Conclusions The results of the present paper clearly demonstrate that different instructions, cueing conditions, and learning effects have a strong impact on the ToL performance. Our findings therefore help to explain divergences in the results of the numerous publications on the Tower of London, and they imply that due to differences in the versions of the test employed, comparisons between the results of different studies are hardly possible (see also Ref. [4]). It follows that one standardized version of the ToL should be applied in research and clinical practice, or that at least all necessary parameters should be reported. In addition, our study showed that the original ToL version by Shallice [30] offers a variety of problems, which appear suitable for the application in research with both special patient groups and healthy volunteers. The employment of this original version would thus clearly facilitate the comparability of different samples.
[13]
[14] [15]
[16]
[17]
[18]
[19]
References [1] J. Ahonniska, T. Ahonen, T. Aro, A. Tolvanen, H. Lyytinen, Practice effects on visuomotor and problem-solving tests by children, Percept. Motor Skills 92 (2001) 479–494. [2] N.C. Andreasen, K. Rezai, R. Alliger, V.W. Swayze, M. Flaum, P. Kirchner, G. Cohen, D.S. O’Leary, Hypofrontality in neurolepticnaive patients and in patients with chronic schizophrenia. Assessment with xenon 133 single-photon emission computed tomography and the Tower of London, Arch. Gen. Psychiatry 49 (1992) 943– 958. [3] S.C. Baker, R.D. Rogers, A.M. Owen, C.D. Frith, R.J. Dolan, R.S. Frackowiak, T.W. Robbins, Neural systems engaged by planning: a PET study of the Tower of London task, Neuropsychologia 34 (1996) 515–526. [4] K.W. Berg, D. Byrd, The tower of london spatial problem-solving task: enhancing clinical and research implementation, J. Clin. Exp. Neuropsychol. 24 (2002) 586–604. [5] D. Carlin, J. Bonerba, M. Phipps, G. Alexander, M. Shapiro, J. Grafman, Planning impairments in frontal lobe dementia and frontal lobe lesions, Neuropsychologia 38 (2000) 655–665. [6] J. Cockburn, Performance on the Tower of London test after severe head injury, J. Int. Neuropsychol. Soc. 1 (1995) 537–544. [7] A. Dagher, A.M. Owen, H. Boecker, D.J. Brooks, Mapping the network for planning: a correlation PET activation study with the TOL task, Brain 122 (1999) 1973–1987. [8] A. Dagher, A.M. Owen, H. Boecker, D.J. Brooks, The role of the
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
striatum and hippocampus in planning: a PET activation study in Parkinson’s disease, Brain 124 (2001) 1020–1032. K.J. Gilhooly, V. Wynn, L.H. Phillips, R.H. Logie, S. Della Sala, Visuo-spatial and verbal working memory in the five-disc Tower of London task: An individual approach, Think. Reason. 8 (2002) 165–178. T.L. Hodgson, A. Bajwa, A.M. Owen, C. Kennard, The strategic control of gaze direction in the Tower-of-London task, J. Cogn. Neurosci. 12 (2000) 894–907. K.L. Kafer, M. Hunter, On testing the face validity of planning / problem-solving tasks in a normal population, J. Int. Neuropsychol. Soc. 3 (1997) 108–119. K.W. Lange, B.J. Sahakian, N.P. Quinn, C.D. Marsden, T.W. Robbins, Comparison of executive and visuospatial memory function in Huntington’s disease and dementia of Alzheimer type matched for degree of dementia, J. Neurol. Neurosurg. Psychiatry 58 (1995) 598–606. R.H. Lazeron, S.A. Rombouts, W.C. Machielsen, P. Scheltens, M.P. Witter, H.B. Uylings, F. Barkhof, Visualizing brain activation during planning: the tower of London test adapted for functional MR imaging, Am. J. Neuroradiol. 21 (2000) 1407–1414. S. Lehrl, Mehrfachwahl-Wortschatz-Intelligenztest (MWT-B). 4. ¨ ¨ uberarbeitete Auflage, Hogrefe Test, Gottingen, 1999. H.S. Levin, D. Mendelsohn, M. Lilly, J.M. Fletcher, K.A. Culhane, S.B. Chapman, H. Harward, L. Kusnerik, D. Bruce, H.M. Eisenberg, Tower of London Performance in relation to magnetic resonance imaging following closed head injury in children, Neuropsychology 8 (1994) 171–179. R.G. Morris, S. Ahmed, G.M. Syed, B.K. Toone, Neural correlates of planning ability: frontal lobe activation during the Tower of London test, Neuropsychologia 31 (1993) 1367–1378. R.G. Morris, T. Rushe, P.W. Woodruffe, R.M. Murray, Problem solving in schizophrenia: a specific deficit in planning ability, Schizophr. Res. 14 (1995) 235–246. R.G. Morris, E.C. Miotto, J.D. Feigenbaum, P. Bullock, C.E. Polkey, The effect of goal–subgoal conflict on planning ability after frontaland temporal-lobe lesions in humans, Neuropsychologia 35 (1997) 1147–1157. R. Morice, A. Delahunty, Frontal / executive impairments in schizophrenia, Schizophr. Bull. 22 (1996) 125–137. A.M. Owen, J.J. Downes, B.J. Sahakian, C.E. Polkey, T.W. Robbins, Planning and spatial working memory following frontal lobe lesions in man, Neuropsychologia 28 (1990) 1021–1034. A.M. Owen, M. James, P.N. Leigh, B.A. Summers, C.D. Marsden, N.P. Quinn, K.W. Lange, T.W. Robbins, Fronto-striatal cognitive deficits at different stages of Parkinson’s disease, Brain 115 (1992) 1727–1751. A.M. Owen, B.J. Sahakian, J. Semple, C.E. Polkey, T.W. Robbins, Dopamine-dependent fronto-striatal planning deficits in early Parkinson’s disease, Neuropsychology 9 (1995) 1–24. A.M. Owen, J. Doyon, M. Petrides, A. Evans, Planning and spatial working memory: a PET study in humans, Eur. J. Neurosci. 8 (1996) 353–364. A.M. Owen, Cognitive planning in humans: neuropsychological, neuroanatomical and neuropharmacological perspectives, Prog. Neurobiol. 53 (1997) 431–450. C. Pantelis, T.R. Barnes, H.E. Nelson, S. Tanner, L. Weatherley, A.M. Owen, T.W. Robbins, Frontal-striatal cognitive deficits in patients with chronic schizophrenia, Brain 120 (1997) 1823–1843. L.H. Phillips, V. Wynn, K.J. Gilhooly, S. Della Sala, R.H. Logie, The role of memory in the Tower of London task, Memory 7 (1999) 209–231. L.H. Phillips, V.E. Wynn, S. McPherson, K.J. Gilhooly, Mental planning and the Tower of London task, Q. J. Exp. Psychol. 54 (2001) 579–597. K. Rezai, N.C. Andreasen, R. Alliger, G. Cohen, V. Swayze, D.S. O’Leary, The neuropsychology of the prefrontal cortex, Arch. Neurol. 50 (1993) 636–642.
J.M. Unterrainer et al. / Cognitive Brain Research 17 (2003) 675–683 [29] J.B. Rowe, A.M. Owen, I.S. Johnsrude, R.E. Passingham, Imaging the mental components of a planning task, Neuropsychologia 39 (2001) 315–327. [30] T. Shallice, Specific impairments of planning, Philos. Trans. R. Soc. Lond. B 298 (1982) 199–209. [31] T. Shallice, From Neuropsychology To Mental Structure, Cambridge University Press, Cambridge, 1988. [32] W.G. Staal, R. Hijman, H.E. Hulshoff Pol, R.S. Kahn, Neuropsychological dysfunctions in siblings discordant for schizophrenia, Psychiatr. Res. 95 (2000) 227–235. [33] C.M. Temple, R.A. Carney, S. Mullarkey, Frontal lobe function and executive skills in children with Turner’s syndrome, Dev. Neuropsychol. 12 (1996) 343–363. ¨ [34] J. Unterrainer, K. Quiske, K. Hoppe-Seyler, C. Meier, C. Muller, B. Rahm, R. Leonhart, U. Halsband, Planning abilities and the Tower
[35]
[36] [37]
[38]
683
of London: Is this task measuring a discrete cognitive function? J. Clin. Exp. Neuropsychol. (2003) submitted for publication. D.M. Veale, B.J. Sahakian, A.M. Owen, I.M. Marks, Specific cognitive deficits in tests sensitive to frontal lobe dysfunction in obsessive–compulsive disorder, Psychol. Med. 26 (1996) 1261– 1269. G. Ward, A. Allport, Planning and problem-solving using the 5-disc Tower of London Task, Q. J. Exp. Psychol. 50 (1997) 49–78. L.H. Watkins, R.D. Rogers, A.D. Lawrence, B.J. Sahakian, A.E. Rosser, T.W. Robbins, Impaired planning but intact decision making in early Huntington’s disease: implications for specific fronto-striatal pathology, Neuropsychologia 38 (2000) 1112–1125. M.C. Welsh, T. Satterlee-Cartmell, M. Stine, Towers of Hanoi and London: contribution of working memory and inhibition to performance, Brain Cogn. 41 (1999) 231–242.