2005 Child Dev

Child Development, January/February 2005, Volume 76, Number 1, Pages 122 – 136

Are A-Not-B Errors Caused by a Belief About Object Location? Ted Ruffman

Lance Slade

University of Sussex and University of Otago

University of Sussex and University of Roehampton

Juan Carlos Sandino and Amanda Fletcher University of Sussex

Eight- to 12-month-olds might make A-not-B errors, knowing the object is in B but searching at A because of ancillary (attention, inhibitory, or motor memory) deficits, or they might genuinely believe the object is in A (conceptual deficit). This study examined how diligently infants searched for a hidden object they never found. An object was placed in A twice, and then in B. In a different task the object was placed beside A twice, and then in B. Infants made more A-not-B errors in the former task, and perseverating infants searched diligently in A rather than in B. Infants seemed to believe the object was in A, suggesting that both a conceptual deficit and ancillary deficits account for A-not-B errors.

For several decades researchers have tried to explain why 8- to 12-month-old infants make A-not-B errors. Having seen the object hidden in one location (A) and searched successfully for it at that location over several trials, infants then watch the object hidden at a second location (B). If delayed for a few seconds, infants often err by searching at A again for the object. Wellman, Cross, and Bartsch (1986) carried out a meta-analysis of prior research and concluded that A-not-B errors increase when infants are younger, with longer delays, and with fewer hiding locations. Markovitch and Zelazo (1999) carried out a more recent meta-analysis and found similar effects for age and delay, but also that errors increase with more A trials and with a decrease in the distance between the hiding locations. Despite the consistencies in empirical results over a large body of research, there remains no widespread agreement as to how to interpret such findings. We consider some of the more prominent ideas. Piaget Piaget (1954) used A-not-B tasks to demonstrate that infants around ages 8 to 10 months had an incomplete understanding of object permanence. He Many thanks to Jessica Redman and Linda Murdoch for helping code the data, to Jenny Arnott for her help in collecting the data, and to the Economic and Social Research Council of Britain for financial support (Grant R000238995). Correspondence concerning this article should be addressed to Ted Ruffman, Department of Psychology, University of Sussex, Brighton, East Sussex, BN1 9QG, UK. Electronic mail may be sent to [email protected].

thought that errors on these tasks provided evidence of an egocentric bias such that infants associated the appearance of the object with their searching at a particular location. Such infants, Piaget argued, did not have an understanding that the object existed independently of their own reaches. Other concept deficit accounts are possible. For instance, it is possible that infants search incorrectly because of a combination of genuine confusion (though not necessarily egocentrism) and ancillary deficits (e.g., errors due to inhibitory or attention deficits). We consider these ideas in greater detail.

Memory 1 Inhibition Diamond (e.g., Diamond, 1985; Diamond, Cruttenden, & Neiderman, 1994) argued that A-not-B errors are caused by a combination of flagging memory for the object at B and an inability to inhibit previous successful searches at A. Consistent with a memory deficit, Diamond (1985) showed that as infants grow older it is necessary to restrain them for longer periods to cause A-not-B errors. Munakata (1998) posited that competition between an active (object at B) and latent (object at A) memory trace results in A-not-B errors. Harris (1989) also argued that memory might play a role, in particular, proactive interference between a long-term memory of the object at A and a short-term memory of the object at B. Likewise, Schacter and Moscovitch (1984) argued that proactive interference causes A-not-B errors in amnesic adults r 2005 by the Society for Research in Child Development, Inc. All rights reserved. 0009-3920/2005/7601-0009

A-Not-B Errors and Belief

and found that these participants showed no insight into their mistakes. Indeed, the participants expressed surprise when they found nothing hidden at A and frequently suggested the experimenter was ‘‘playing some kind of a trick.’’ They also denied that the object had been hidden anywhere else. Such findings make it possible that infants might, too, genuinely believe the object is at A when they search there. If so, diligent searching at A and little searching at B would be expected. Arguments that infants suffer from memory deficits essentially amount to a form of conceptual deficit such that errors represent genuine confusion. In contrast, Diamond (1985) suggested that inhibitory deficits might be sufficiently strong that infants would search at A even when they know the object is at B. For instance, she pointed out that infants sometimes make A-not-B errors when the toy is placed in a transparent container or is left uncovered, although errors are less frequent (Bremner & Knowles, 1984; Butterworth, 1975). In addition, infants sometimes look to B (as if they know the object is there) yet search at A. As such, Diamond (1991) argued that ‘‘infants appear to be showing with their eyes that they know where the toy is hidden, even though they reach back to A anyway’’ (p. 85). Inhibition is typically construed in terms of the infant’s ability to resist previous motor responses, that is, ‘‘the ability to resist the conditioned tendency to reach back to A’’ (Diamond, 1985, p. 880). Like Diamond (1985; Diamond et al., 1984), Markovitch and Zelazo (1999) also argued that A-not-B errors are caused by an inhibitory failure. Consistent with this idea, A-not-B errors have been tied to the dorsolateral prefrontal cortex by means of its role in inhibition. For instance, Diamond and Goldman-Rakic (1983, 1989) demonstrated that primates with a lesioned dorsolateral prefrontal cortex make A-not-B errors whereas previously they had not. Bell and Fox (1992, 1997) showed that success on A-not-B tasks is associated with greater frontal EEG power. Motor Memories Thelen, Scho¨ner, Scheier, and Smith (2001) offered a dynamical systems account of A-not-B errors (see also Smith, Thelen, Titzer, & McLin, 1999). They argued that several factors interact to create the error, one of which is a motor memory, that is, a tendency to repeat previous actions. This notion bears some relation to inhibitory deficits. For instance, Smith et al. (1999) found that A-not-B errors occurred even when there was no object at A. On A trials, if the experimenter merely waved the lid at A, infants would reach to that location. After several A trials,

123

the experimenter then waved the lid at B. Despite there not having ever been an object at A, Smith et al. found that infants often made A-not-B errors by returning to the A well. They argued that A-not-B errors are caused in part by a motor memory and that errors are not caused by beliefs about object location at all. Yet, Smith et al.’s claims have been disputed by Munakata (1997), who compared performance in two conditions. In both conditions the experimenter just tapped an A lid, causing search there. In one condition a toy was then hidden in B, whereas in the other the experimenter tapped the B lid. A-not-B errors were more frequent when no object was present at B. This means that infants’ search is influenced by the presence of toys and is consistent with the idea that infants’ belief about object location affects search (although see the following). Attention Thelen et al. (2001) argued that the decrease in Anot-B errors when an object was present in the B location (Munakata, 1997) might have been caused not by a belief that the object was in B but by the presence of the object at B, which heightened the salience of that location. In other words, differences in the extent to which the B location captured attention in the two conditions might have accounted for the lower amount of A searching on object trials. In fact, many theories have posited that attention plays a role in causing A-not-B errors. Harris (1989) reinterpreted Diamond’s (1985) data demonstrating higher error rates with longer delays by positing that longer delays make it more likely that the infant’s attention will wander from B to A, causing search there. Munakata (1998) argued that the infant must maintain an active representation of the object at B to allow search there. Maintaining an active representation is akin to maintaining attention on B. Likewise, Bell and Fox (1992) posited that the infant must ignore distraction over a delay. Munakata also posited that searches at A on A trials make that location more salient in future trials. In reaching to the A well, the infant likely encodes itFthe toy, the well, that sideFas an important location so that A is more likely to capture attention in the future. Horobin and Acredolo (1986) showed that A-not-B errors were more likely when two wells were close together than when they were far apart, a factor confirmed in Markovitch and Zelazo’s (1999) meta-analysis. Harris (1989) argued that whereas the inhibitory requirements should remain roughly the same, wells in close proximity increase the likelihood that the infant’s attention will wander from the B to the A

124

Ruffman, Slade, Sandino, and Fletcher

location, increasing search errors. Keenan (2002) found that infants who were distressed during the B trial were more likely to make A-not-B errors. He suggested that negative affect might disrupt attention and impair performance. Ruffman and Langman (2002) argued that errors on multiwell tasks, as well as other two-location A-not-B tasks typically construed as caused by inhibitory deficits, can best be understood as due to deficits in attention. These accounts gain support from the fact that the dorsolateral prefrontal cortex, the brain region linked to A-not-B errors, has also been linked to attention (e.g., Stuss, Binns, Murphy, & Alexander, 2002). One way of understanding this link is hinted at by Diamond (1990; Diamond et al., 1994) who has sometimes pointed out that inhibitory deficits might be better conceptualized as attention deficits in that the infant must maintain attention on B. She noted that when infants do sustain attention on B throughout the delay they typically reach correctly. Present Research In summary, some researchers believe that infants might search at A even when they know the object is at B, in part, because they cannot inhibit previous successful searches at A, because they have motor memories for A, or because A captures infants’ attention. Alternatively, it is possible that A-not-B errors are accompanied by a genuine belief that the object is there, much as seems to be the case for amnesic adults. These different possibilities were examined in Experiment 1. One characteristic of previous A-not-B tasks is that when infants make an A-not-B error, if they then look in B they will find the object there. In this situation, search will usually be terminated because infants just explore the object. In the present studies we examined what infants would do if they returned to the B well but then did not find the object there either. In Experiment 1 we compared infants’ searching in two conditions. In the AAB condition, infants searched successfully for the object twice in the A well. The object was then placed in the B well but hidden in a false bottom (in a pocket below the movable floor of the well) so that no object would be found on this third (B) hiding trial. Performance on this condition was compared with that on the AAA condition, in which infants searched successfully for the object twice in A, and then the object was hidden again in A but in a false bottom so that it could not be found at either location. The question was whether infants who initially perseverated at A in the AAB task would search for

as long (or longer) at the B well as they did at the A well. It was reasonable to expect infants to return to a location or search diligently after not initially finding the object there because they often reached initially with their hands. It would be easy to not find an object when reaching and therefore further searching would be required. Furthermore, even adults often return to search for lost objects in places they had previously looked. We reasoned that if infants who make A-not-B errors are uncertain (e.g., because of competition between short- and long-term memory traces or active and latent representations), more diligent searching at A might be seen in the AAA task than in the AAB task. This is because there should be little uncertainty in the AAA task. Short- and long-term memory traces would be consistent with one another, as would active and latent representations (object at A). In contrast, in the AAB task memory traces would be in conflict with one another, which should lead to more searching at B than in the AAA task even when infants initially searched at A. Furthermore, we reasoned that if A-not-B errors are caused by attention deficits, inhibitory deficits, or motor memories, either of two results might be expected. First, infants might know that the object is in B (Diamond, 1991) and attention, inhibitory, and motor memory deficits might be relatively mild. If so, after initially searching at A, infants might return to B and search diligently there. Alternatively, the ancillary deficits might be sufficiently large that infants would search diligently at A in both tasks. Both tasks bias the infant toward A because they each involve two previous searches at A. Moreover, once infants had initially searched in A on the B trial, this would further enhance the salience of the A location or increase the inhibitory bias, making subsequent diligent A searching more likely. Finally, besides attention, inhibitory, or motor memory deficits, another reason for searching diligently in A in an A-not-B task is that infants genuinely believe the object is in A (i.e., possess a conceptual deficit). Our task is different from previous tasks, even those employing three hiding locations and allowing infants a second search after an initial failure (Reznick, Fueser, & Bosquet, 1998). After initially erring at one well and having this well removed, Reznick et al. (1998) found that 9-month-olds usually searched at the correct hiding well from among the remaining two wells. Having found the object at the correct well, subsequent searching tends to cease (with some exceptions as Reznick et al., 1998, noted). What is unknown, then, is whether infants would have

A-Not-B Errors and Belief

returned to the incorrect well where they initially searched and search diligently there, or whether they would have searched more diligently at the correct well. Our task sheds light on this question because the infant never finds the object.

Experiment 1 Method Participants. We tested 58 infants between 8 and 12 months of age. Most infants were White and middle class. Testing was abandoned for 10 infants because of fussiness and for 5 because of a failure to reach on the third trial. This left 43 infants (25 girls and 18 boys) with a mean age of 9.62 months (SD 5 1.13 months). Mothers responded to advertisements in local doctors’ offices and nurseries and were given small gifts for their participation. Materials. The hiding box was 44 cm wide, 28 cm long, 18 cm high, and colored pink. The two hiding wells were 11  11 cm across, 10 cm deep (16 cm including the false bottom), with 13 cm between them. We did not use lids for the hiding wells in this experiment or in Experiments 2 and 3 because infants frequently become distracted by the lid and fail to look further in a hiding well. For the hiding object we used a red and yellow cloth sunflower that rattled, a small purple doll’s purse with sequins, keys, or a small plastic musical toy (depending on infants’ preferences). In the warm-up phase we used a hiding box with a single opening and a caterpillar or rattle as the toy. Procedure. Infants were tested in their own homes and all infants were given both the AAA and AAB tasks. They sat on the floor in front of their mothers and faced the experimenter. Infants were first given some warm-up trials. The toy was placed on the floor in front of and in full view of the infant who was encouraged to reach for it. Once having grasped the toy, the experimenter retrieved it from the infant after allowing a brief play and placed it half in and half out of the single hiding well. Once the infant had retrieved the toy, the experimenter pulled the hiding well backward, retrieved the toy after letting the infant play briefly with it, and then placed it right inside the single hiding well so that it was completely hidden. When the infant had retrieved the toy on this trial, the experimenter removed the single hiding well and the toy used in the warm-up, and presented the two-well hiding box and a new toy. In all phases of the experiment when hiding the toy, the experimenter placed it in the hiding box and then

125

pushed it forward to a point 30 cm from the infant. The box was pushed toward the infant immediately after hiding in the warm-up phase and on initial A trials, and after a 5-s delay on the third (test) trial of the AAA and AAB tasks. The infant was then encouraged to search. No form of distraction was employed during the delay. On each A and B hiding trial, the experimenter drew the box toward himself, captured the infant’s attention by calling his or her name and, shaking the toy, placed the object in a well and pushed the box toward the infant. The wells were deep enough so that the infant could not tell where the object was hidden without either leaning forward or reaching inside. Each trial lasted for 30 s or until the infant retrieved the object or became distressed. If the infant did not search correctly on the first two A trials, the experimenter secretly retrieved the object and repeated the hiding procedure. On both tasks infants needed to retrieve the object twice successfully at A before moving onto the third (test) trial. A successful search occurred only when infants used their hand to retrieve the object from A. In the third trial, the object was hidden in a false bottom at either A or B so that it was not visible at either location when the infant searched. These trials lasted for 30 s or until the infant became distressed. Infants received the two conditions in a counterbalanced order. For half of the infants the A well was on their left and for the other half it was on their right. Infants were tested in their own homes. Scoring. On the third (test) trial of the AAA and AAB conditions, we categorized infants’ initial search at a location as correct or incorrect. A search on this trial entailed either reaching into a hiding well or bending over and looking inside a particular hiding well. Subsequent search at a particular location (in all experiments) entailed either searching in the location (e.g., with their hands) or immediately around the location (e.g., by leaning forward and looking with their eyes). All searching was coded from videotapes of the testing procedure. The videotapes were played frame by frame and examined for the initial search location and the total amount of time searching at each location. Because coding was frame by frame, estimates of search time were accurate to within .04 s. Two coders rated looking time, and interrater reliability was calculated over 25% of trials by computing the amount of time each observer said a child searched at the A (or B) location over the 30-s coding period. For each location, we then subtracted Coder 1’s time at each location from Coder 2’s to calculate the mean difference score. Reliability was high, with a mean difference score of

126

Ruffman, Slade, Sandino, and Fletcher

0.37 s over the 30-s coding period. In cases of dispute the primary coder’s looking time was used.

Results Task order (AAA first vs. AAB first) and gender did not affect the probability of the infant perseverating at A on the third (B) trial of the AAB task, so that these variables were dropped from subsequent analyses. We compared errors in the AAA and AAB tasks. There were 17 infants who erred on the third (B) trial of the AAB task (40%) by searching at A, and 3 who erred on the third (A) trial of the AAA task (7%) by searching at B. Comparing the AAA and AAB tasks directly, there were 16 infants who erred on the third trial of the AAB task but not the third trial of the AAA task, and only 2 who obtained the opposite error pattern. Thus, infants were significantly more likely to err on the AAB task, McNemar’s w2(1, n 5 18) 5 10.89, po.01. Given that errors were so infrequent in the AAA task, we eliminated from subsequent analyses the 3 infants who failed to search at A on the third (A) trial, leaving 40 infants in total. All remaining infants searched correctly on their initial search on the third trial of the AAA task. Figure 1 presents the total amount of time over the 30-s coding period that these infants searched in A and B on the third trial of the AAA and AAB tasks. Our interest was in whether infants who perseverated by initially searching at A in the AAB task searched diligently there. Thus, we analyzed perseverators and nonperseverators separately, and compared search in the AAB with search in the AAA task. Of the 40 infants, 24 searched correctly at B and 16 perseverated at A. In all experiments we used search times in the analyses rather than number of searches because infants might search for longer at a given location even when the number of searches at each

12 10

AAB: Correct Initial Search (in B) AAB: Incorrect Initial Search (in A)

8 6 4 2 0

AAA: Search Time in A

AAA: Search Time in B

AAB: Search Time in A

AAB: Search Time in B

Figure 1. Mean amount of time (seconds) searching at each location (and standard errors) in Experiment 1.

location was equal. Nevertheless, in each experiment we carried out identical analyses for number of searches and the pattern of results was the same. Because the data were positively skewed, with standard deviations in some cases proportional to the means, we used a logarithmic transformation. Using the transformed data, the tendency to look diligently in A or B was examined with two 2 (task: AAA vs. AAB)  2 (location: time spent searching in A vs. time spent searching in B) ANOVAs. The dependent variable was the amount of time searching on the third trial of the AAA or AAB tasks. Diligent A searching by perseverators would be indicated by two results. First, for perseverators there would not be a two-way interaction between time searching at a particular location (A vs. B) and task (AAA vs. AAB), but for nonperseverators (infants who initially searched at B in the AAB task), there would be a Location  Task interaction. Second, perseverators would be expected to search significantly longer at A than at B in the AAB task. Consistent with diligent A searching for the perseverators, there was a main effect for location, F(1, 15) 5 11.12, po.01 (with longer searching in A), but no effect for task and no interaction, F(1, 35) 5 0.10, p4.76. As planned, we then examined searching in each task separately. Perseverators searched more at A than at B in the AAA condition, t(15) 5 2.89, po.05, and in the AAB condition, t(15) 5 2.95, po.05. For the nonperseverators, the main effect for location was not significant but the interaction was, F(1, 23) 5 10.55, po.01. Nonperseverators searched more at A than at B in the AAA condition, t(23) 5 3.10, po.01, and more at B than at A in the AAB condition, t(23) 5 – 1.89, po.05, one-tailed. Discussion We were particularly interested in infants who perseverated at A on the B trial of the AAB task. There were 16 such infants, all of whom searched correctly at A on the third (A) trial of the AAA task. Our interest was in whether these infants would search more diligently at A in the AAA task than in the AAB task. This might be expected because it has been hypothesized that search errors occur even when infants know the object is at B (Diamond, 1991), meaning infants should return to search in B in the AAB task. Furthermore, the change in the object’s location in the AAB but not in the AAA task might have created a conflict between two memory representations, making confusion and uncertainty regarding object location more likely. If so, infants might return to B to search and spend more time

A-Not-B Errors and Belief

searching at B in the AAB task than in the AAA task. However, this did not happen. Infants who perseverated at A in the AAB task tended to search diligently at that location. Their tendency to search diligently at A was similar to their searching in the AAA task, as evidenced by the lack of an interaction between searching at A versus B in the AAA task relative to the AAB task. This was not due to a general lack of statistical sensitivity because: (a) there was a significant interaction in the group of nonperseverators (who searched longer at A in the AAA task and at B in the AAB task), and (b) the perseverators searched significantly longer at A than B in the AAB task. The question, then, is why perseverators searched so diligently at A in the AAB task. It is possible that infants’ ancillary deficits are sufficiently strong that they cannot inhibit their previous successful searches there, because their motor memory for reaching to A is too strong or because the A location is so salient by virtue of having searched there previously that infants’ attention is drawn to A for an extended period. These ancillary deficit views can be contrasted with a concept deficit explanation; infants’ belief that the object is at A might be sufficiently strong that it overrides all understanding that the object is at B. That is, perseverating infants might be so convinced that the object is in A that they search diligently at that location. Experiment 2 was designed to adjudicate between these two views. Experiment 2 We gave infants two tasks. One was the AAB-inside task, identical to the AAB task of Experiment 1. The other was the AAB-on-top task, identical to the AABinside task except that on A trials the object was not placed inside the A well; instead, it was placed beside the A well, on top of the box, immediately adjacent to and to the outside of the A well opening but in full view of the infant. Thus, in the A trials of the AAB-inside task the infant formed a belief that the object was inside the A well when it was not visible. This was not the case in the A trials of the AAB-ontop task. The infant never formed a belief that the object was inside the A well when it was not visible. Indeed, the object was always visible but was just placed beside the A well. Thus, on the B trials, genuine confusion regarding the object’s location was more likely in the AAB-inside task. If search errors are caused by a belief about object location, infants should be more likely to err and should search more diligently at A in the AAB-inside task than in the AAB-on-top task.

127

If, in contrast, search errors are caused by attention deficits, inhibitory deficits, or motor memories, infants should search equally often and with equal diligence at A in the two tasks. This is because in both tasks the infant reaches toward the A location twice on A trials. This reach should focus attention on that location, should create a motor memory of reaching to the A side, and should create a bias (to be inhibited) to that location. Furthermore, Experiment 2 has an advantage over previous ‘‘waving lids’’ tasks (Munakata, 1997; Smith et al., 1999). Recall that A-not-B errors are reduced when an object is hidden at B on the B trial in contrast to when the B lid is simply waved. Although one could argue that errors are reduced because of the infant’s belief that the object is in B, Thelen et al. (2001) pointed out that object trials enhance the salience of the B location and this, rather than a belief about object location, might reduce errors. In contrast, in Experiment 2 both tasks involve placing an object around the A location and therefore should make this region of roughly equal salience. Likewise, both tasks involve placing an object in B on the B trial. Despite these similarities, the A trials of the AAB-inside task are much more likely to create a lingering belief that the object could be in A. This belief would plausibly create more diligent searching at A when the object is hidden in B on the B trial. A final difference in Experiment 2 was that we coded not just where infants searched but also where they glanced. Glancing involved the infants’ looking toward one location but not bending forward to discover what was inside and not simultaneously reaching with their hand. Whereas searches provided visual or tactile information as to whether an object was present at a location, glances provided no such information. For instance, an infant might initially glance at B but then search at A either by bending forward and looking in A or reaching to A with his or her hand. Diamond (1988) reported anecdotes of infants looking at the correct location but searching incorrectly in A-not-B tasks. Hofstadter and Reznick (1996) obtained experimental evidence in favor of this idea (see also Ahmed & Ruffman, 1998). Our glancing measure might indicate earlier developing knowledge of object location than the searching measure. Method Participants. There were 61 participants tested originally. Most infants were White and middle class. Thirty-one were assigned to the AAB-inside condition and 30 to the AAB-on-top condition. Subse-

128

Ruffman, Slade, Sandino, and Fletcher

quently, 7 infants were dropped because of fussiness, 1 was dropped because the mother cued the infant’s search, and 9 were dropped because they never searched on the B trial. Three other babies in the AABon-top condition were dropped because they accidentally knocked the toy into the A well on an A trial. This left 41 infants: 26 in the AAB-inside condition (10 girls and 16 boys, M 5 10.23 months, SD 5 1.28, range 5 8.21 to 12.60 months) and 15 in the AAB-ontop condition (8 girls and 7 boys, M 5 10.26 months, SD 5 1.21, range 5 8.81 to 12.54 months). Recruitment and rewards were as in Experiment 1. Materials. The hiding boxes were identical to those used in Experiment 1. In the experimental phase we used the small plastic musical toy employed in the experimental phase of Experiment 1. In the warm-up phase we used the caterpillar or rattle used in Experiment 1. Procedure. Infants were tested in a university laboratory, but the procedure for the AAB-inside task was otherwise identical to that of Experiment 1. On A trials of the AAB-on-top task, the object was placed immediately adjacent to and to the outside of the A well on top of the box. On A trials of both tasks, two successful searches (retrieval of the object) were required before moving onto the B trial. On B trials of both tasks, the delay between hiding and search was 5 s. During the delay the experimenter waved his hand in front of and at the center of the hiding boxes and counted to maintain infants’ attention at the box midline. Hofstadter and Reznick (1996) coded infants’ first glance immediately after lowering an opaque screen, revealing a table with wells in which a toy had been hidden. In contrast, as in Experiment 1, we used a hiding box and pushed this forward after the hiding event. Unlike Hofstadter and Reznick, we did not completely obscure the hiding box after a hiding event. Infants’ attention was drawn to the box midline as far as possible by the experimenter but their visual contact with the box and the hiding well was not always broken. Thus, in our task first glance was determined not necessarily by an expectation of where the toy was but sometimes simply by where the infant happened to be looking when the box was pushed forward. For this reason, we coded glancing time at A and B over the 30-s coding period. This involved the amount of time infants spent looking toward one location but not bending forward to discover what was inside and not simultaneously reaching with their hands. Interrater reliability was calculated over 15% of trials, as in Experiment 1. Searching reliability was high, with a mean difference score of 0.34 s over the 30-s coding period.

Glance reliability was calculated in the same way and was slightly lower but still acceptable with a mean difference score of 0.72 s over the 30-s coding period.

Results Searching. First, we examined infants’ tendency to perseverate on their initial search on the B trial of each task by incorrectly searching at A. In the AABinside task, 17 of 26 infants (65%) searched initially at A, and in the AAB-on-top task this figure was 5 of 15 (33%). Infants were significantly more likely to perseverate in the AAB-inside task, Fisher’s exact test: po.05, one-tailed. Figures 2a and 2b present the total amount of time infants searched and glanced at A and B over the 30-s coding period. Because the data were positively skewed, with standard deviations proportional to the means, we used a logarithmic transformation. Using these transformed data, the tendency to search diligently in A or B was examined with two 2 (location: time spent searching in A vs. B)  2 (initial

8

B Trial: Correct Initial Search (in B)

6

B Trial: Incorrect Initial Search (in A)

4 2 0 AAB-Inside: Search Time in A

AAB-Inside: Search Time in B

AAB-On Top: Search Time in A

AAB-On Top: Search Time in B

AAB-On Top: Glance Time at A

AAB-On Top: Glance Time at B

8 6 4 2 0 AAB-Inside: Glance Time at A

AAB-Inside: Glance Time at B

B Trial: Correct Initial Search (in B)

B Trial: Incorrect Initial Search (in A)

Figures 2a and 2b. Mean amount of time (seconds) searching and glancing at each location (and standard errors) in Experiment 2. Looking times for the AAB-inside condition are based on whether the infant initially did or did not perseverate in that condition. Similarly, looking times for the AAB-on top task are based on whether the infant initially did or did not perseverate in that condition.

A-Not-B Errors and Belief

AAB response: perseverate vs. do not perseverate) ANOVAs. The dependent variable in one analysis was search time in the AAB-inside task, and for the other it was search time in the AAB-on-top task. In each case we were interested in whether there was a Location  AAB Response interaction and whether infants who perseverated by initially searching at A in each task searched more at A than at B overall. If perseverators in the AAB-inside task searched at A because of a belief that the object was in A, we would expect them to search diligently there as compared with perseverators in the AABon-top task whose searches at A might simply have indicated wandering attention or an inhibitory deficit. In the AAB-inside task there was only one significant effect: the Location  AAB response interaction, F(1, 13) 5 12.81, po.01. As in Experiment 1, the 10 children who perseverated searched significantly longer at A than at B, t(9) 5 3.22, po.05. The 5 children who initially searched correctly in B, searched longer in B overall but not significantly so (possibly because of the low participant numbers), t(4) 5 – 2.05, ns. In the AAB-on-top task there was also only one significant effect in the initial ANOVA: again, the Location  AAB Response interaction, F(1, 24) 5 18.13, po.001. The 9 children who perseverated searched significantly longer at A than at B, t(8) 5 2.00, po.05, one-tailed. The 17 children who initially searched correctly in B searched longer in B overall, t(16) 5 – 4.54, po.001. Glancing. Figure 2b reveals that the pattern for searching and glancing is roughly similar except that infants glanced more than they searched, and they tended not to be as polarized when glancing at A versus B as compared with searching. We carried out identical 2  2 ANOVAs for glancing time as for searching time. In the AAB-inside task there was only one significant effect: the Location  AAB Response interaction, F(1, 13) 5 4.04, po.05, one-tailed. As earlier, the 10 children who perseverated glanced significantly longer at A than at B, t(9) 5 2.19, po.05, one-tailed. The 5 children who initially searched correctly in B glanced longer at B overall but not significantly so, t(4) 5 – 0.90, ns. In the AAB-on-top task, there was also only one significant effect in the initial ANOVA: again, the Location  AAB Response interaction, F(1, 24) 5 8.39, po.01. The 9 children who perseverated glanced longer at A than at B but not significantly so, t(8) 5 1.15, ns. The 17 children who initially searched correctly in B glanced longer at B overall, t(16) 5 – 3.44, po.01. Finally, we compared searching with glancing. Infants were categorized as passing the glance measure if they glanced more at the correct location

129

than at the incorrect location, and failing if they glanced equally or glanced more at the incorrect location. Search was coded in an analogous fashion. In the AAB-inside task, there were 12 infants who searched and glanced correctly, 7 who searched and glanced incorrectly, 4 who glanced correctly but searched incorrectly, and 3 who searched correctly but glanced incorrectly. Glancing was not better than searching, McNemar’s w2(n 5 7) 5 0.14, ns. In the AAB-on-top task, there were 3 infants who searched and glanced correctly, 8 who searched and glanced incorrectly, 2 who glanced correctly but searched incorrectly, and 2 who searched correctly but glanced incorrectly. Again, glancing was not better than searching, McNemar’s w2(n 5 4) 5 0.00, ns. Novelty. Infants often look longer at novel stimuli. One might argue that hiding an object in B would be more novel in the on-top task because unlike the inside task, the fingers of the hand had not been inside a hiding well on A trials of the on-top task (although see the following discussion for counterarguments). If correct, infants might look at B more in the on-top task than in the inside task, and this increased attention to B might result in more (correct) looking at B in the on-top task. We examined this idea with a 2 (condition: AAB-inside vs. AABon-top)  2 (location: A vs. B) ANOVA, collapsing over whether infants perseverated. Total glancing time at A versus B (logarithmically transformed) was the dependent variable (see Figure 2b for infant glancing). If novelty determines looking time, the interaction should be significant such that there is more diligent glancing at B in the AAB-on-top task (because fingers go inside the A well on A trials of the inside task but not in the on-top task). However, the interaction was not significant, F(1, 39) 5 2.07, ns, indicating no difference in looking at B in the on-top and inside tasks. Discussion Schacter and Moscovitch (1984) found that amnesic adults who made A-not-B errors genuinely believed that the object was in A. Infants’ initial searching in Experiment 2 demonstrates that they often appear to share this characteristic. If initial search errors were caused by a belief that the object was in A, more errors would have been expected in the AAB-inside task than in the AAB-on-top task. This was precisely what we found. The crucial difference between the AAB-on-top task and the AABinside task was that the object was placed beside A in full view in the AAB-on-top task rather than being hidden inside A. The motor memory and inhibitory

130

Ruffman, Slade, Sandino, and Fletcher

bias to A, caused by reaching to A on two previous A trials, should have been about the same in the two tasks. Likewise, the attentional pull to A should have been about the same in the two tasks given that infants had reached and focused attention on this location on two previous trials. The different error rates in the AAB-on-top and AAB-inside tasks are consistent with the idea that a belief about object location affects error rates in Anot-B tasks. In each task the object is not visible on the third (B) trial. However, whereas in the AABinside task the object had also not been visible on A trials, in the AAB-on-top task the object had always been visible beside A on A trials. Thus, in the AABinside task, when the infant cannot see the object on the B trial, there is considerable potential for thinking it could be in A where it had previously been hidden. In contrast, in the AAB-on-top task, the object had never been in A; therefore, there is less potential for thinking it could be there now. In sum, the different error rates in the AAB-inside and AAB-on-top tasks are consistent with the idea that infants have a belief about object location rather than an inhibitory bias, motor memories, or attention deficits. Furthermore, unlike previous waving lids tasks (Munakata, 1997; Smith et al., 1999), the Experiment 2 results cannot be explained along the same lines as in Munakata’s (1997) study. Once again, Munakata showed that A-not-B errors are reduced when an object is hidden at B on the B trial in contrast to when the B lid is simply waved. Thelen et al. (2001) pointed out that object trials enhance the salience of the B location, and this rather than a belief about object location (object in B) might reduce errors. In contrast, in Experiment 2 both tasks involve placing an object around the A location and therefore should make this region of roughly equal salience. Likewise, both tasks involve placing an object in B on the B trial. The plausible explanation for infants’ greater number of errors on B trials of the AAB-inside task is that this task created the belief that the object could be in A. Another finding concerns the amount of time infants spent searching in the AAB-inside task of Experiment 2. This was consistent with Experiment 1 in that infants who initially perseverated in A searched diligently there, whereas infants who initially searched correctly in B searched diligently there. An additional aim of Experiment 2 was to compare infants’ glancing with their searching. Glancing tended to mirror searching in that perseverating infants glanced more in A than in B. A final finding was that infants were not more likely to glance correctly than to search correctly. This result is different from that obtained by Hofstadter and Reznick (1996).

There are several reasons why this might be. First, our methodology was different so that we did not code first glance as did Hofstadter and Reznick. Second, the advantage for looking over searching was only slight in Hofstadter and Reznick. For instance, in Experiment 1 the average percentage correct was 61% in the reach condition and 72% in the gaze condition. Third, in Hofstadter and Reznick, infants either searched or looked but they did not do both, whereas in our task infants did both. Searching in a particular location tends to lead infants to look in that location as well; therefore, arguably, dissociations should not be so prevalent. Finally, we examined whether more diligent searching in B in the on-top task might have been because the B hiding event was more novel. In the B hiding events the experimenter’s fingers went slightly inside the B well. Infants had seen fingers do this on A trials of the inside task but not of the on-top task. However, infants were not more likely to look to B in the on-top task compared with the inside task, providing no evidence that their attention was drawn more to B where they then searched in the ontop task.

Experiment 3 In Experiment 2, infants were less likely to make Anot-B errors when, on A trials, the toy had been placed in full view on top of the A well rather than hidden inside A. Because this task is novel we collected more data in Experiment 3 to examine how reliable our Experiment 2 findings were. In Experiment 3, however, we used longer delays. Whereas Experiments 1 and 2 used 5-s delays, Experiment 3 used both 5-s and 30-s delays. We expected that longer delays might lead to higher rates of A-not-B errors, either because infants’ memory for the object’s location would fade (Diamond, 1985) or because longer delays would make it more likely that infants’ attention would wander from the B location to A (Harris, 1989). Higher error rates would thus be expected unless the shorter delay was sufficient to cause errors by itself. Regardless, a primary interest was in how diligent searching and glancing would be at a particular location. It was possible that with longer delays the infant’s memory trace would fade altogether so that searching and glancing would become somewhat random. Alternatively, if infants searched diligently in a particular location for an object even at a longer delay, this would provide more evidence for the idea that searching is related to a belief about object location.

A-Not-B Errors and Belief

131

Table 1 Tasks of Experiment 3

Task AAB-inside-5 AAB-inside-30 AAB-on-top-30 AAA-inside-30

Hiding location on two A trials

Hiding location on third trial

Delay (seconds)

Mean age (months)

Age SD (months)

Inside A Inside A On top of A Inside A

Inside B Inside B Inside B Inside A

5 30 30 30

10.07 10.38 10.94 10.51

1.24 0.96 1.11 1.27

We used four conditions summarized in Table 1. The AAB-inside-5 task was identical to the AAB-inside task of Experiment 2. The AAB-inside-30 task was identical to the AAB-inside-5 task except that a delay of 30 and instead of 5 s was used between hiding on the third (B) trial and search. The AAB-ontop-30 task was similar to the AAB-on-top task of Experiment 2 except that a 30-s delay was used. Finally, the AAA-inside-30 task was similar to the AAA task of Experiment 1 except that a 30-s delay was used. We were interested in whether similar rates of Anot-B errors and committed A searching would occur on the third trial of the AAB-inside-5 and AABinside-30 tasks. If memory faded with the longer delay, less diligent searching might be expected with the 30-s delay. We were also interested in whether we would obtain converging results for the results from Experiment 2 in that more A-not-B errors would be committed in the AAB-inside tasks than in the AABon-top task. Finally, we were interested in whether we would obtain converging results for the results from Experiment 1, finding that infants search as diligently at A on the third trial of the AAB-inside-30 task as they do on the third trial of the AAA-inside30 task. The longer delays would potentially weaken memory traces, leading to more confusion in the AAB task as weak memory traces for the object at each location could be more easily confused. In contrast, the single memory trace in the AAA task (object at A ) could not be so easily confused with a trace of the object at B because the object had never been at B. Method Participants. There were 61 participants tested originally. Infants were mainly White and middle class. Thirteen babies were dropped because of fussiness or failure to search, leaving 12 infants per condition. Infants ranged between 8.37 and 12.90 months. Table 1 lists the mean age of infants and gender distribution in each condition. There were no

Gender distribution 6 6 4 4

girls, girls, girls, girls,

6 6 8 8

boys boys boys boys

differences in the ages of infants in each condition, F(3, 44) 5 1.17, ns. Recruitment and rewards were as in Experiments 1 and 2. Materials. The hiding boxes were identical to those used in Experiments 1 and 2. In the experimental phase we used the small plastic musical toy employed in Experiments 1 and 2, and in the warmup phase we used a similar musical figure. Procedure. Infants were tested in a university laboratory using the same procedures as in Experiments 1 and 2 except for the length of the delays, which are described in Table 1. During the delay the experimenter waved his hand in front and at the center of the hiding boxes and counted to maintain infants’ attention at the box midline. Interrater reliability was calculated over 16% of search trials in the same way as in Experiments 1 and 2. Reliability was high, with a mean difference score of 0.55 s over the 30-s coding period. Glance reliability was calculated the same way and was slightly lower but still acceptable, with a mean difference score of 0.84 s over the 30-s coding period. Results Searching. First, we examined infants’ initial search on the third trial of each task. Eleven of 12 infants initially searched incorrectly (made A-not-B errors) in the AAB-inside-30 task and in the AABinside-5 task. In contrast, 11 of 12 infants searched correctly (at A) in the AAA-inside-30 task. In the AAB-on-top-30 task, 6 of 12 infants were correct. Thus, the number of incorrect infants was identical in the AAB-inside tasks over the 5- and 30-s delays, and infants were significantly more likely to err in each of these tasks than in the AAA-inside-30 task, Fisher’s exact test: po.001 (replicating Experiment 1), and the AAB-on-top-30 task, Fisher’s exact test: po.05, one-tailed (replicating Experiment 2). In addition, infants made significantly more errors in the AAB-on-top task than in the AAA-inside task, Fisher’s exact test: po.05, one-tailed.

132

Ruffman, Slade, Sandino, and Fletcher

Next, we examined search times (see Figure 3a). There were too few infants who passed the AABinside-5 and AAB-inside-30 conditions, and who failed the AAA-inside-30 condition to include the initial response on the third trial (correct vs. incorrect) as a factor in the analysis or to include these infants in the analyses of search and glance time. Furthermore, for the AAB-on-top-30 task, we included only infants who incorrectly searched at A on their first committed search. This was to examine whether infants who erred by initially searching at A in the AAB-on-top-30 task tended to search diligently at A relative to infants who searched at A in the other conditions. Thus, in the initial analyses of search time, infants’ initial search was at A in all 10 8 6 4 2 0 AAB-Inside-30

AAB-Inside-5 AAB-On Top-30 AAA-Inside-30

Initial Search Correct: Search Time in B

Initial Search Correct: Search Time in A

Initial Search Incorrect: Search Time in B

Initial Search Incorrect: Search Time in A

10

8

6

4

2

0 AAB-Inside-30

AAB-Inside-5 AAB-On Top-30 AAA-Inside-30

Initial Search Correct: Glance Time at B

Initial Search Correct: Glance Time at A

Initial Search Incorrect: Glance Time at B

Initial Search Incorrect: Glance Time at A

Figures 3a and 3b. Mean amount of time (seconds) searching and glancing at each location (and standard errors) in Experiment 3. Looking times for each condition are based on whether the infant initially did or did not perseverate in that condition.

cases. Our interest was in how diligently they then searched at A relative to B. The data were log transformed and analyzed with a 4 (task)  2 (location: time spent looking in A vs. time spent looking in B) ANOVA. There was a main effect for location, F(1, 35) 5 47.47, po.001 (with more looking in A than in B), and a Task  Location interaction, F(3, 35) 5 3.85, po.05. The interaction was explored with 2 (location)  2 (task: all pairwise comparisons of tasks) ANOVAs. There were six such comparisons and, hence, six ANOVAs. Our main interest was in the interaction between task and location. This interaction was significant only when comparing the AAB-inside-30 task with the AAAinside-30 task, F(1, 20) 5 14.44, po.01 (such that searching at A relative to B tended to be longer for infants in the AAA-inside-30 task). As a conservative measure, we examined whether the main effect for location was due to greater A looking in all tasks. Infants whose initial search was in A searched longer in A than in B in the AAB-inside-5 task, t(10) 5 3.16, po.05; the AAB-on-top-30 task, t(5) 5 2.75, po.05; and the AAA-inside-30 task, t(10) 5 6.32, po.001. Infants also searched longer in A than in B in the AAB-inside-30 task, but the effect failed to reach significance, t(10) 5 1.50, ns. Infants whose initial search in the AAB-on-top-30 task was in B searched longer in B than in A but not significantly so, t(5) 5 – 1.81, ns. Glancing. The glancing data were log transformed and, as with searching, analyzed with a 4 (task)  2 (location: time spent glancing at A vs. time spent glancing at B) ANOVA. As with searching, there was a main effect for location, F(1, 35) 5 25.15, po.001 (with more glancing at A than B), and a Task  Location interaction, F(3, 35) 5 3.78, po.05. The interaction was explored with 2 (location)  2 (task: all pairwise comparisons of tasks) ANOVAs. As earlier, there were six such comparisons. As with searching, the Task  Location interaction was significant when comparing the AAB-inside-30 task with the AAAinside-30 task, F(1, 20) 5 8.70, po.01 (such that glancing at A relative to B tended to be longer for infants in the AAA-inside-30 task). Additionally, the interaction was significant when comparing the AAB-inside-30 task with the AAB-inside-5 task, F(1, 20) 5 7.58, po.05 (such that glancing at A relative to B tended to be longer for infants in the AAB-inside-5 task). We then considered whether infants were more likely to be correct on the glance measure than on the search measure. As in Experiment 2, infants were never better on the glance measure in any of the four tasks (all ps4.50).

A-Not-B Errors and Belief

Novelty. As in Experiment 2, we examined whether more searching at B in the AAB-on-top-30 task relative to the AAB-inside-30 task might be because of increased gaze at B in the on-top task because of the novelty of the object going in a well for the first time, and increased attention to B, leading to more searching there. We used a 2 (condition: AAB inside-30 vs. AAB-on-top-30)  2 (location: A vs. B) ANOVA, with total glancing time (logarithmically transformed) as the dependent variable. Again, we analyzed all infants together, collapsing over whether they perseverated or did not perseverate (see Figure 3b). The novelty account predicts more glancing at B in the on-top task. However, the interaction did not approach significance, F(1, 22) 5 0.49, ns. Discussion First, we considered infants’ initial searches. Almost all infants (11 of 12) made A-not-B errors in the AAB-inside-5 and AAB-inside-30 tasks. With this group of infants, 5 s was sufficient to create A-not-B errors, and there was no increase in errors with the longer delay. The number of infants erring in the two AAB-inside tasks was significantly greater than the number erring in the AAA-inside-30 task and the AAB-on-top-30 task. Thus, as in Experiment 2, infants made fewer errors in the AAB-on-top task relative to the AAB-inside task. Because the demands on motor memory, inhibition, and attention were about the same in the AAB-inside and AAB-on-top tasks, the different error rates are consistent with the idea that a belief about object location affects error rates in A-not-B tasks. There is much more potential for thinking the toy could be in A where it had previously been hidden in the AAB-inside tasks than in the AAB-on-top task (in which the object had never been in A). We then examined total search times at A and B. When comparing searching and glancing time in the AAB-inside and AAA-inside tasks, only infants who initially searched in A for the toy were included. Our interest was in how diligently infants would search in A in the different conditions. As in Experiments 1 and 2, infants who perseverated in A in the standard A-not-B task (the AAB-inside-5 task in Experiment 3) searched significantly longer in A than in B overall. Again, then, there was no indication that infants knew the object was in B when they searched in A. Another finding of particular interest was infants’ searching in the AAA-inside-30 task relative to the AAB-inside-30 task because in each task infants had searched at A twice previously and encountered a

133

30-s delay between hiding and search. Furthermore, in each condition, 11 infants initially searched at A for the object on the third trial. Despite these similarities, there was a difference in how committed infants were to the A location. Infants in the AAAinside-30 task tended to search more diligently in A than infants in the AAB-inside-30 task (the task x location interaction). Likewise, infants glanced more at A in the AAA-inside-30 task than in the AAB-inside-30 task. Also, recall that the Task  Location interaction had not been significant for perseverators in Experiment 1, although that experiment had also used AAA and AAB tasks. A plausible explanation for this difference in results stems from differences in methodology. Experiment 1 employed 5-s delays, whereas Experiment 3 employed 30-s delays. We hypothesized at the outset that 30-s delays should weaken memory traces, leading to more confusion in the AAB task, as weak memory traces for the object at each location could be more easily confused. In contrast, the single memory trace in the AAA taskFobject at AFcould not be so easily confused with a trace of the object at B because the object had not been at B. Thus, whereas with standard (5-s) delays committed A searching in the AAA and AAB tasks is indistinguishable, longer delays weaken the memory traces leading to subtle differences in committed searching. A further finding was that infants glanced significantly longer at the A location in the AAB-inside5 task than in the AAB-inside-30 task. This result is consistent with the idea that attention to A (e.g., an active memory trace) is stronger after a shorter delay because the object has more recently been in A. Finally, as in Experiment 2, there was no evidence for the novelty account. That is, infants looked equally at B in the on-top and inside tasks. Their attention to B was equal so that the increased searching at B in the on-top task could not have been due to greater gazing through novelty. General Discussion In Experiment 1, infants participated in two tasks. In each task they searched at A twice for the object. In the AAB task the object was then hidden at B, whereas in the AAA task the object was hidden again at A. Despite having just seen the object hidden at B in the AAB task, many infants made A-not-B errors, looking in A. We then examined only infants who perseverated at A in the AAB task, all of whom also looked in A in the AAA task. These infants looked diligently in A in the AAB task as well as in

134

Ruffman, Slade, Sandino, and Fletcher

the AAA task (see Experiment 3 also). The data provided no evidence for the idea that infants know the object is in B when they perseverate at A (Diamond, 1985; 1991). If infants knew the object was in B, it would be reasonable to expect them to return to B and search there more as compared with the AAA task. This result was replicated in the AAB-inside tasks of Experiments 2 and 3 as well, providing no evidence for the idea that an inhibitory bias causes A-not-B errors even when infants know the object is in B. Another idea is that confusion between memory traces in A-not-B tasks (object at A, object at B) could cause a conceptual deficit such that infants believe the object is in A when they search there (e.g., Diamond, 1985; Harris, 1989; Munakata, 1998). Infants’ relative ease on the AAB-on-top tasks relative to the AAB-inside tasks in Experiments 2 and 3 is consistent with this idea. After hiding the object in B, infants were less likely to make A-not-B errors when the object had only ever been beside A on A trials, compared with when it had been in A on A trials. When the object had been in A and it is then hidden in B, infants would be more likely to be uncertain as to its present location. Previously, when the object had not been visible, it had been in A; therefore, it was possibly in A again. In contrast, when the object had previously been beside A in full view, the infant was less likely to think it could currently be in A because it had never been there previously. Whereas this result is consistent with the idea that object search is based at least partially on infants’ belief about object location, it is inconsistent with several other explanations. The motor memory and inhibitory bias to reach to A should have been about the same in both the AAB-inside and AAB-on-top tasks. Likewise, the salience and attentional pull of the A location should have been about the same in the two tasks. Infants’ apparent confusion as to the object’s location is reminiscent of Schacter and Moscovitch’s (1984) findings for amnesic adults who frequently claimed the experimenter had played some kind of trick. At the same time, our results do not rule out the idea that motor memories, inhibition, or attention have some impact on infant searching. Many infants did still search at A on the B trial of the AAB-on-top tasks. Furthermore, in Experiment 3, infants made significantly more errors in the AAB-on-top-30 task than in the AAA-inside-30 task. Errors in the on-top task might have occurred if placing the object in B on the B trial led infants to think the object had also been in A on the A trial. This sort of confusion might have caused searching at A in the AAB-on-top task, or

alternatively, there might have been some influence from attention, inhibition, or motor memories. Thus, A-not-B errors might be due to ancillary deficits some of the time in addition to a belief about object location. A counterargument to our claims might be that they are based in some cases on failing to find a difference between groups (i.e., a lack of statistical sensitivity). For instance, when examining infants who perseverated on the AAB task of Experiment 1, we found that they searched diligently in the A location in the AAA and AAB tasks. Likewise, when infants perseverated on the AAB tasks of Experiment 2 and 3, they searched equally diligently at A in the AAB-inside and AAB-on-top tasks. Yet, such an argument fails to do justice to the pattern of findings. First, there was no general failure to achieve significance. For instance, in Experiments 1 and 2 we found significant Location  Task and Location  Initial Response interactions. Thus, infants searched differently in different tasks depending on whether they first perseverated. Second, in the standard Anot-B (inside) task in all experiments, infants who perseverated in A searched significantly more at A than at B. This is important in showing that our tests were sensitive measures of searching and in demonstrating perseverators’ tendency to search diligently at A. Third, in Experiment 3 infants searched significantly longer at A (relative to B) in the AAAinside-30 task compared with the AAB-inside-30 task, indicating that search times were sufficiently sensitive to uncover subtle differences between tasks. Thus, there was no indication that our tasks were insensitive measures of infant performance, or of a general failure to achieve statistical significance. Another argument is that the search demands on A trials of the on-top and inside tasks are different, and this accounts for search differences on B trials. On A trials, both tasks require infants to represent the object at the A location (see Perner, 1991, for characteristics of representations), although the representation is informed by current perceptual information only in the on-top task. Both tasks also require infants to initiate a reach and then reach for the object on the basis of their representation. In sum, the additional demand on A trials of the inside task is that infants must represent the object at A with no current perceptual information that it is there. Yet, all infants searched correctly at A twice for the toy, indicating that they were ultimately able to represent the object in A on the inside task. Furthermore, the additional task demand on A trials of the inside task should make infants generally less likely to access a

A-Not-B Errors and Belief

representation of the object at A, the opposite of what we found on B trials. Still another argument is that infants do not search at A on the B trial of the on-top task simply because they did not see the object there in full view. Yet, the object was not on A in view in the B trial of the inside task either, but in this task infants did search at A. Thus, neither an analysis of task demands on A trials nor the object’s absence on B trials explains the pattern of results. It is only the combination of events over both trials and tasks that can explain the different error rates. When an infant in the inside group sees that the object is not present on the B trial, he or she might think it is in the A well, where it had been before. When an infant in the on-top group sees that the object is not present on the B trial, there is less scope for thinking it is in the A well because it has not been in A before. Furthermore, it is not just that infants were more likely to reach for the object at A in the inside task but that they then searched diligently at A for the object. Diligent A searching cannot be the result of increased attention to A because the infant had reached to A on two previous A trials in both the inside and on-top tasks. For instance, compare infants’ searching in the AAA-inside-30 task with the AAB-inside-30 task in Experiment 3. In each task infants searched in A twice, and on the third trial there was a 30-s delay whereupon 11 infants searched at A. Despite these similarities, infants in the AAA-inside-30 task tended to search more diligently in A than infants in the AAB-inside-30 task (the Task  Location interaction). Because the object had been placed in A on A trials of both tasks, and the infant had searched there twice, task demands on the A trials should have been identical. Likewise, the object was not in plain view on the B trial of either inside task. There is no explanation for why infants searched more diligently in some conditions unless this is where they believed the object to be, a sensible explanation given that the object had never been in B in the AAA task. The difference in searching time in the AAA-inside-30 and AAB-inside-30 tasks helps rule out another explanation. One could argue that the amount of time spent searching at A is simply a result of the initial search there. Much as infants sometimes seem to get distracted by lids in A-not-B tasks such that they play with the lid and fail to search further for the object in the hiding location, one might argue that search time is caused by similar distraction. That is, having begun a search at a particular location, infants get distracted by their searching at that location and search diligently there. However, infants’

135

longer A searching (and glancing) in the AAA-inside-30 task relative to the AAB-inside-30 task demonstrates that the search time in a location is not simply a function of where infants initially search because all infants included in the analyses had initially searched in A on the third trial and had searched at A on the two A trials. Instead, search time is consistent with how strongly infants believe an object is in a particular location. Belief would be stronger in the AAA task because the object had only ever been in A; therefore, there would be less confusion. A final argument against our claims is that novelty accounted for searching differences between the inside and on-top tasks. That is, perhaps the hand going inside the B well in the on-top task is more novel than in the inside task because the hand had never before been inside a well in the on-top task. This might draw infants’ gaze to B, where they then search in the on-top task. There are several difficulties with this argument. First, the difference in hand position in the two tasks was minimal. The hand never did go inside the well in the inside tasks, but rather, a 1- or 2-cm portion of the fingers. Thus, the difference in perceptual characteristics between the on-top and inside tasks was trivial. Far more salient was the fact that the hand had changed position from the A well side to the B well side, and this was common to both tasks. Second, all infants received the warm-up trials immediately before the A trials; therefore, they had seen the fingers inside a well previously. Third, novelty should have resulted in longer gazing at the B well in the on-top task relative to the inside task. Yet, this was not the case in either experiment. For all these reasons we conclude that greater B searching in the on-top task relative to the inside task is due a belief that the object might be in A in the inside task but not in the on-top task. In sum, we agree with most theorists (e.g., Diamond, 1985; Munakata, 1998; Thelen et al., 2001) that there may be more than one cause of A-not-B errors. The value of the research described here is in helping narrow down the options as to what the causes are. Although attention, inhibitory ability, or motor memories may cause A-not-B errors in some circumstances, errors are not purely a function of ancillary deficits. Our results are consistent with the idea that A-not-B errors are sometimes caused by a belief that the object is in A. Even when conditions were equated for their demands on infants’ attention, inhibitory ability, and motor memories, infants were more likely to make A-not-B errors when they had seen an object placed in A previously as opposed to being fully visible beside A. The in-A

136

Ruffman, Slade, Sandino, and Fletcher

condition would lead to more confusion as to object location on B trials when, again, the object was not visible. In addition, our results provided no evidence that A-not-B errors are accompanied by knowledge that the object is in B. References Ahmed, A., & Ruffman, T. (1998). Why do infants make A not B errors in a search task, yet show memory for the location of hidden objects in a non-search task? Developmental Psychology, 34, 441 – 453. Bell, M. A., & Fox, N. A. (1992). The relations between frontal brain electrical activity and cognitive development during infancy. Child Development, 63, 1142 – 1163. Bell, M. A., & Fox, N. A. (1997). Individual differences in object permanence performance at 8 months: Locomotor experience and brain electrical activity. Developmental Psychobiology, 31, 287 – 297. Bremner, J. G., & Knowles, L. S. (1984). Piagetian Stage IV errors with an object that is directly accessible both visually and manually. Perception, 13, 307 – 314. Butterworth, G. (1975). Object identity in infancy. Child Development, 46, 866 – 870. Diamond, A. (1985). Development of the ability to use recall to guide action, as indicated by infants’ performance on AB. Child Development, 56, 868 – 883. Diamond, A. (1988). Differences between adult and infant cognition: Is the crucial variable presence or absence of language? In L. Weiskrantz (Ed.), Thought without language (pp. 337 – 370). Oxford, England: Oxford University Press. Diamond, A. (1990). The development and neural bases of memory functions as indexed by the AB and delayed response tasks in human infants and infant monkeys. Annals of the New York Academy of Sciences, 608, 267 – 317. Diamond, A. (1991). Neuropsychological insights into the meaning of object concept development. In S. Carey & R. Gelman (Eds.), The epigenesis of mind (pp. 67 – 110). Hillsdale, NJ: Erlbaum. Diamond, A., Cruttenden, L., & Neiderman, D. (1994). AB with multiple wells: 1. Why are multiple wells sometimes easier than two wells? 2. Memory or memory1inhibition? Developmental Psychology, 30, 192 – 205. Diamond, A., & Goldman-Rakic (1983). Comparison of performance on a Piagetian object permanence task in human infants and rhesus monkeys: Evidence for involvement of prefrontal cortex. Society for Neuroscience Abstracts (Pt. I), 9, 641. Diamond, A., & Goldman-Rakic, P. S. (1989). Comparison of human infants and rhesus monkeys on Piaget’s AB

task. Evidence for dependence on dorsolateral prefrontal cortex. Experimental Brain Research, 74, 24 – 40. Harris, P. L. (1989). Object permanence in infancy. In A. Slater & G. Bremner (Eds.), Infant development (pp. 103 – 121). Hove, England: Erlbaum. Hofstadter, M., & Reznick, J. S. (1996). Response modality affects human infant delayed-response performance. Child Development, 67, 646 – 658. Horobin, K., & Acredolo, L. (1986). The role of attentiveness, mobility history, and separation of hiding sites on Stage IV search behavior. Journal of Experimental Child Psychology, 41, 114 – 127. Keenan, T. (2002). Negative affect predicts performance on an object permanence task. Developmental Science, 5, 65 – 71. Markovitch, S., & Zelazo, P. D. (1999). The A-not-B error: Results from a logistic meta-analysis. Child Development, 70, 1297 – 1313. Munakata, Y. (1997). Perseverative reaching in infancy: The roles of hidden toys and motor history in the AB task. Infant Behavior and Development, 20, 405 – 416. Munakata, Y. (1998). Infant perseveration and implications for object permanence theories: A PDP model of the AB task. Developmental Science, 1, 161 – 184. Perner, J. (1991). Understanding the representational mind. Cambridge, MA: MIT Press. Piaget, J. (1954). The construction of reality in the child. New York: Basic Books. Reznick, J. S., Fueser, J. J., & Bosquet, M. (1998). Self-corrected reaching in a three-location delayed-response search task. Psychological Science, 9, 66 – 70. Ruffman, T., & Langman, L. (2002). Infants’ reaching in a multi-well A not B task. Infant Behavior & Development, 25, 237 – 246. Schacter, D. L., & Moscovitch, M. (1984). Infants, amnesics, and dissociable memory systems. In M. Moscovitch (Ed.), Infant memory (pp. 173 – 216). New York: Plenum. Smith, L. B., Thelen, E., Titzer, R., & McLin, D. (1999). Knowing in the context of acting: The task dynamics of the A-not-B error. Psychological Review, 106, 235 – 260. Stuss, D. T., Binns, M. A., Murphy, K. J., & Alexander, M. P. (2002). Dissociations within the anterior attentional system: Effects of task complexity and irrelevant information on reaction time speed and accuracy. Neuropsychology, 16, 500 – 513. Thelen, E., Scho¨ner, G., Scheier, C., & Smith, L. B. (2001). The dynamics of embodiment: A field theory of infant perseverative reaching. Behavioral and Brain Sciences, 24, 1 – 86. Wellman, H. M., Cross, D., & Bartsch, K. (1986). Infant search and object permanence: A meta-analysis of the A-not-B error. Monographs of the Society for Research in Child Development, 51(3 Serial No. 214), 1 – 50.