Infant Behavior & Development 33 (2010) 159–167

Contents lists available at ScienceDirect

Infant Behavior and Development

Memory development throughout the second year: Overalldevelopmental pattern, individual differences, and developmentaltrajectories

Thorsten Kollinga,∗, Claudia Goertza, Frahsek Stefanieb, Monika Knopfa

a Department of Psychology, Goethe-University Frankfurt, Georg Voigt Strasse 8, 60054 Frankfurt am Main, Germanyb University of Würzburg, Universitätskinderklinik Würzburg, Josef-Schneider-Straße, 97080 Würzburg, Germany

a r t i c l e i n f o

Article history:Received 7 January 2009Received in revised form 29 June 2009Accepted 29 December 2009

Keywords:Memory developmentDeferred imitationLanguageSelfInfancy

a b s t r a c t

The present three-wave longitudinal study analyzed the development of declarative mem-ory in N = 92 infants (12-, 18- and 24-month-olds) using a deferred imitation task. Asexpected, overall memory performance improved throughout the second year. Previousresearch is also replicated insofar as stability of inter-individual differences was low tomoderate within this age range. In addition, cluster analyses identified two developmen-tal groups showing different growth and different stability patterns. Multivariate analysesrevealed specificities in language and self-development in these two developmental groupshaving different developmental trajectories.

© 2010 Elsevier Inc. All rights reserved.

1. Introduction

For learning and memory in infancy, imitation has been identified as a central mechanism. While in newborns immediateimitations are reported (Meltzoff & Moore, 1977), the imitative acts of infants become more and more decoupled from thepresence of a model, i.e. deferred imitations are found. In a classical study Piaget (1962) described deferred imitation throughbehavioural observations of his daughter Jacqueline in a real-life situation. In more recent years, an experimental task forassessing this ability has been proposed (e.g., Meltzoff, 1985). In this deferred imitation task, young infants observe shortseries of simple, object-based actions demonstrated by a model (demonstration phase). After a delay of minutes, hours oreven days the infants are given the props and their target behaviour is coded (imitation phase).

Experimental and clinical work provided evidence for the claim that deferred imitation can be classified as tappingdeclarative memory (Mandler, 2004; McDonough, Mandler, McKee, & Squire, 1995). First, the cross-modal character of thedeferred imitation task renders it improbable that infants’ imitations are due to priming processes, which are sensitiveto modality changes. Second, the possibility that infants learn these actions through incremental learning processes can beexcluded, since infants acquire instrumental actions via observation and not through motor learning. Next, deferred imitationtasks use new, unknown actions as memory material. These actions are not available in the infant’s knowledge before. Finally,it has been demonstrated that deferred imitation is not found in amnesics, whose declarative memory system is impaired.

∗ Corresponding author. Tel.: +49 69 79825008.E-mail address: T.Kolling@psych.uni-frankfurt.de (T. Kolling).

0163-6383/$ – see front matter © 2010 Elsevier Inc. All rights reserved.doi:10.1016/j.infbeh.2009.12.007

160 T. Kolling et al. / Infant Behavior & Development 33 (2010) 159–167

The cross-sectional developmental studies conducted so far using this type of memory task showed that 6-month-oldinfants are already capable of deferred imitation after short retention intervals. In the course of development, infants learnfaster as they need less exposure to the target actions and they retain actions in memory for an increasing amount of time(e.g., Barr, Dowden, & Hayne, 1996).

Generally consistent with cross-sectional deferred imitation studies, the rare longitudinal studies indicate that declarativememory improves with age (Goertz, Kolling, Frahsek, Stanisch, & Knopf, 2008; Heimann & Meltzoff, 1996; Kolling, Goertz,Frahsek, & Knopf, 2009; Nielsen & Dissanayake, 2004). In addition, it was demonstrated that correlations between fourindividual memory tests assessed in a longitudinal design are low to moderate in the second year (Nielsen & Dissanayake,2004). In other words, inter-individual variability of intra-individual change is high. The reasons for this huge amount of inter-individual variability of intra-individual change are still unclear. To shed further light on this issue the present multivariate,longitudinal study tracks two main goals besides the replication of previous longitudinal research on memory developmentthroughout the second year both with respect to overall developmental pattern as well as with respect to the developmentof inter-individual differences. The first goal of this longitudinal study is to identify developmental groups varying in theirdevelopmental trajectories. A second goal is to explain how these groups differ in terms of developmental determinants.

A proper analysis of inter-individual differences in intra-individual change and the identification of developmental groupsnecessitates the use of person-centred analysis approaches, as they focus on the individual subject and its development (vonEye & Bogat, 2006). Exploratory and confirmatory person-centred analysis approaches, i.e. longitudinal cluster analysis andlongitudinal configural frequency analysis are applicable in this methodological context. These approaches allocate thesubjects under study into different developmental groups under the assumption that different subgroups exist. In a nextstep of analysis, external validity of groupings is established by explaining intra-individual change with significantly relatedvariables. Finally, externally validated groups are interpreted on the base of substantive theory (von Eye & Bogat, 2006). Toour knowledge, only one deferred imitation study used this analysis approach so far. In a person-centred analysis of twowaves of longitudinal data (12- and 18-month-old infants), Kolling et al. (2009) identified subgroups revealing differentialgrowth (high, low and moderate) and stability (higher stabilities for subgroups than for the total group) of declarativememory performance.

To understand and explain different developmental courses of declarative memory throughout the second year, theassessment of language, self, social interaction abilities (joint attention, turn-taking skills) and cognitive development ingeneral seem to be fruitful, since there is at least some recent theoretical work as well as empirical research linking thedevelopment of deferred imitation and the development of cognitive-, language- and self-aspects to one another. In severalstudies, joint development between declarative memory and language has been found. In one study enriching deferredimitation items with verbal cues, Herbert and Hayne (2000) showed that children at the age of 24 months, who outper-formed younger children, used language cues efficiently, whereas 18-month-olds were not able to do so. This analysis hasbeen realized for the overall age group and cross-sectionally, however, not for specific groups having different develop-mental trajectories. In a longitudinal study Heimann et al. (2006) found that visual recognition memory (at 6 months),deferred imitation (at 9 months) and turn-taking skills (at 14 months) predicted language skills at the age of 14 months,with deferred imitation accounting for the highest variance in the regression model. Furthermore, Strid, Tjus, Smith, Meltzoff,and Heimann (2006) reported that deferred imitation and joint attention were predictive for cognitive abilities at 4 years ofage demonstrating the important role of social interaction variables.

While the aforementioned theoretical considerations and findings assume that the development of declarative mem-ory are directly related to the improvement of abilities like language or social interaction, other theoretical considerationsassume qualitative changes in deferred imitation during the second year. Declarative memory performance and the devel-opment of self are believed to be interrelated in a complex manner both from a theoretical (Knopf, Mack, & Kressley-Mba,2005) and an empirical (Prudhomme, 2005) perspective in the second year. According to Tulving (2002) the emergence of a(conceptual) self, which develops around the middle of the second year, is a prerequisite for episodic (auto-noetic) remem-bering. With the development of the self, encoding of deferred imitation actions becomes more episodic-like in nature. Inaccordance with this line of thinking, Perner’s (1991) theory of the representational mind postulates significant changesin the representational system at the age of 18 months. Perner claims that around the age of 18 months a new quality ofrepresentation, namely secondary representations, emerges, being the basis of a new type of representational activity. Bylinking deferred imitation memory performance and self development assessed via mirror self-recognition to each other,Prudhomme (2005) demonstrated empirically that 20-month-old children who passed the mirror test (rouge test) wereless affected by a change of colours of the relevant target props given within the deferred imitation test than those whodid not pass the mirror test. This finding is seen as an evidence for the fact that a more advanced development of the selfallows a more flexible memory performance in a deferred imitation test. All in all, the research reviewed points out that thedevelopment of different cognitive and non-cognitive characteristics importantly determines declarative memory develop-ment in the second year of life. However, much more evidence is needed about the role these different concepts play for theoccurrence and explanation of inter-individual differences of intra-individual change of declarative memory throughout thesecond year.

By putting together the evidence reviewed above we hypothesize that while there is an overall improvement of memorythroughout the second year, inter-individual differences in intra-individual change of declarative memory performance arehigh. If the use of person-centred approaches (cluster analyses) leads to the extraction of developmental cluster groups dif-fering in growth, group differences will need to be described by important developmental determinants (external validity

T. Kolling et al. / Infant Behavior & Development 33 (2010) 159–167 161

of the groups). Following the aforementioned line of reasoning it seems to be reasonable to expect a large amount of discon-tinuity in deferred imitation assessed longitudinally, especially around the age of 18 months, where the representationalsystem improves as well as the self-related knowledge emerges gradually. Moreover, several developmental domains wereassessed in addition to memory development in order to be able to explain group differences to some amount.

2. Methods

2.1. Participants

The initial sample of the longitudinal study consisted of N = 92 infants (N = 48 boys) who were recruited via radioannouncement and advertisements in childcare centres and local paediatricians and by word of mouth. The criteria foradmission into the study were no known physical, sensory or mental handicaps, normal length of gestation (over 37 weeks)and normal birth weight (2500–4500 g). The infants in the study were on average 11.9 months of age (M = 362 days, SD = 8.7)at the first testing, 18.1 months (M = 551 days, SD = 7.9) at the second testing, and 24.0 months (M = 731, SD = 10.6) at the thirdtesting. Mean birth weight was M = 3393 g (SD = 507). Parents reported an APGAR index of M = 9.78/9.94 with a minimumscore of 7. Four subjects did not continue participating in the study because of relocation. An additional n = 10 infants wereexcluded because of fussiness, excessive crying and refusing to play with the objects at one of the three testings. Final samplesize therefore was n = 75 infants.

2.2. Testing environment

All infants were tested individually in the laboratory. Testing took place in a small room that was unfurnished except forthe experimental apparatus. During the session, the infant was seated on his or her parent’s lap at a small rectangular tablejust opposite the experimenter. Behind and to the left of the experimenter was a video camera that was focused on the child’shead and torso and most of the tabletop. A second camera behind the infant on the right recorded the experimenter. Therecording apparatus was housed in an adjacent viewing room to reduce auditory distractions. The testing was electronicallytimed by a computer that mixed elapsed time in 0.10 s increments directly onto the videotaped records.

2.3. Material and apparatus

2.3.1. Frankfurt Imitation Tests for 12- to 24-month-olds (FIT 12, FIT 18, FIT 24)To obtain sound tests for deferred imitation, items for both 12-month-olds and 18-month-olds were pilot tested in

several independent studies. Furthermore, control groups (12- and 18-month-olds) were assessed to obtain mean baselineperformance and mean test performance (Goertz, Knopf, Kolling, Frahsek, & Kressley, 2006; Goertz, Kolling, Frahsek, &Knopf, 2008; Goertz, Kolling, Frahsek, Stanisch, & Knopf, 2008; Kolling et al., 2009). The deferred imitation items finallyadopted were chosen among the potential items in the pre-tests using the criteria that (1) they yield good inter-scorerreliability, (2) they involve uniformly distributed item difficulties with a total mean item difficulty of about 50%, (3) theycomprise formerly unknown actions that infants in the different age groups are able to perform in terms of motor skills,and (4) they are age-adapted. The deferred imitation tests finally applied consisted of these items adjusted for 12-, 18- and24-month-olds, respectively.

All the objects used were commercially available toys and were specifically adapted for the respective age-groups. Table 1gives an overview of the objects used, the target actions demonstrated and the operational definitions used for coding.1

The Frankfurt Imitation Test for 12-month-olds (FIT 12) (Goertz et al., 2006; Kolling et al., 2009) consists of five differentobject-based actions. Two of these are two-step actions and three consist of one-step actions; most of the objects and targetactions were used in previous research. The maximum score of the FIT 12 is 7. The scorers of FIT 12 reached an inter-rater reliability of r = 89.0–92.7% and a Cohen’s kappa of � = .78–.83. For all tests, discrepancies in scoring were resolved bydiscussion to 100% agreement. A test–retest study (1 week retest interval) demonstrated a rather high reliability (r = .52;r = .63 after correction for outliers) of the FIT 12.

The Frankfurt Imitation Test for 18-month-olds (FIT 18) (Goertz, Kolling, Frahsek, & Knopf, 2008; Kolling et al., 2009)consists of six object-based actions (1 three-step action, 4 two-step actions, 1 one-step action). The items of the FIT 18 wereconstructed such as the number of multi-staged items was increased (one 1-stepped, four 2-stepped, one 3-stepped) andone item included one goal-irrelevant step. The maximum score of the FIT 18 is 12. The scorers of the FIT 18 reached aninter-rater reliability of r = 95.7–96.4% and a Cohen’s kappa of � = .91–.92.2

1 More detailed descriptions of target actions, operational definitions and item statistics for FIT 12, FIT 18 and FIT 24 can be obtained on request from thecorresponding author.

2 Four scorers rated the video tapes. Scorer 1 and scorer 2 reached an inter-rater reliability of r = 96.1% (Cohen’s kappa � = .92), Scorer 2 and Scorer 3 aninter-rater reliability of r = 96.1% (Cohen’s kappa � = .92) and scorer 3 and scorer 4 an inter-rater reliability of r = 95.8% (Cohen’s kappa � = .91).

162 T. Kolling et al. / Infant Behavior & Development 33 (2010) 159–167

Table 1Overview and item examples of deferred imitation tests.

The Frankfurt Imitation Test for 24-month-olds (FIT 24) (Goertz, Kolling, Frahsek, & Knopf, 2008) was administeredwhich consisted of 8 object-based actions (1 six-step action, 1 five-step action, and 6 three-step actions). The maximumscore of the FIT 24 is 28. The scorers of the FIT 24 reached an inter-rater reliability of r = 93.9–96.3% and a Cohen’s kappa of� = .88–.93.3

3 Three scorers rated the video tapes. Scorer 1 and scorer 2 reached an inter-rater reliability of r = 96.3% (Cohen’s kappa � = .93), Scorer 2 and Scorer 3 aninter-rater reliability of r = 93.9% (Cohen’s kappa � = .87).

T. Kolling et al. / Infant Behavior & Development 33 (2010) 159–167 163

Table 2Overview and item examples of German developmental test.

Developmental domains Developments assessed

Body motor development Trunk control, head control, walking, crawlingHand motor development Grasping and releasing, manipulation of propsCognitive development Object concept, ego vs. allocentricity, planning, causal understanding, numerosityBody self-awareness Infant shows/knows own/other person’s body partsReceptive language Word understanding, sentence understandingExpressive language 1,2 syllables (e.g., baba, dada, gaga), uses 1,2 word utterances, vocabulary spurtSocial development (peers) Parallel playSocial development (adults) Shared/joint attention, social smilingSocial development (autonomy) Infant/child wants to do things on his/her own, walks away from caretakerEmotional development Development of primary and secondary emotions, attachment, gender-awareness

2.3.2. Data scoringAll experimental sessions were videotaped by two cameras, one taping the infant and the second the experimenter.

Two naive and independent observers scored both target action completion using operational definitions and the infant’sbehaviour (e.g., attention). They were blind to the hypotheses of the study and how exactly the experimenter demonstratedthe target actions. Target actions were scored as “Yes” or “No” responses.

2.3.3. Developmental Test for 6-month to 6-year-olds (ET 6-6)The German Entwicklungstest für 6-Monate bis 6 Jahre [Developmental Test for 6-month to 6-year-olds] aims to (1)

assess normal development, (2) assess strengths and weaknesses of the individual infant/child, (3) diagnose developmentaldeficits early on, and (4) formulate developmental prognostics about individual development. The assessment procedureof the total test takes about 60 min. The test is comprised of 10 developmental scales (body motor development, handmotor development, cognitive development, body self-awareness, receptive language development, expressive languagedevelopment, social development – interaction with peers; social development – interaction with adults, social development– autonomy; emotional development), which are assessed with both questionnaire and examiner-assessed tasks, i.e., theexaminer scored the performance of the infant in specific tasks. The test is a standardized, reliable measure with populationnorms. The developmental quotients (M = 100, SD = 15) were calculated using the population norms. This test resembles theBayley Scales of Infant Development not only with respect to the mode of administration (questionnaire and examiner-assessed tasks) but also with respect to the developmental milestones assessed and the structure of the developmentalscales and subscales.

Table 2 depicts the different dimensions (with exemplary sample items) of the German developmental test.All subjects were within normal range (mean DQ’s ranged from 97.9 to 113.6 at the first testing, from 87.3. to 111.0 at

the second testing, and from 92.1 to 104.1 at the third testing) in all developmental dimensions tested.

2.4. Procedure

In all tests, actions were presented to the infants successively three times (12 months: four times) by a model. Followinga delay of 30 min, infants were given the props successively in the same order as in the demonstration phase for a given timeinterval. Infants’ playing behavior was videotaped. At all three waves, the 30-min-delay was used to interview the parent(APGAR, birth weight, pregnancy, etc.), to obtain informed consent, and to fill out the questionnaire of the developmentaltest.

3. Results

3.1. Data analysis

The data are analyzed through three analysis steps. In a first step, variable-centred analyses of both mean memoryperformance and stability correlations are calculated for the total sample. Further, distinct, developmental subgroups are

Fig. 1. Stability correlations of deferred imitation performance (n = 78).

164 T. Kolling et al. / Infant Behavior & Development 33 (2010) 159–167

separated with cluster analytic procedures by decomposing the data set into subgroups (person-centred). Finally, differencesbetween these developmental groups are reported (multivariate analysis) to establish external validity of developmentalsubgroups.

3.2. Variable-centred analysis – Total group

Fig. 1 depicts the growth of the mean memory performance as well as inter-individual stability correlations.Mean memory performance increased from Mt1 = 4.2 (SDt1 = 1.5, min = 1, max = 7, opt = 7) over Mt2 = 6.9 (SDt2 = 1.9, min = 3,

max = 11, opt = 12) to Mt3 = 17.0 (SDt3 = 3.8, min = 6, max = 25, opt = 28). An analysis of variance for repeated measures showedthat the linear development trend is significant, F = 963, df = 1, p < .05. No gender differences were found for the developmentaltrend (F = 0.02, df = 1, ns). This trend indicates that, as expected and replicating prior findings, infants are able to retain moretarget actions in declarative memory with increasing age. The stability correlations show that inter-individual differencesof intra-individual change are higher between the first and second than between second and third testing. However, bothstability correlations are rather moderate replicating earlier longitudinal findings. This demonstrates, as expected, that inter-individual differences in intra-individual change of declarative memory performance are high. Therefore, in a next statisticalstep, we used a person-centred analysis to extract developmental cluster groups.

3.3. Person-centred analysis

Cluster analysis. First, the deferred imitation data were separately z-standardized (per subject and time point) to eliminatemean and variance differences of the deferred imitation tests. For the analysis of individual consistency, several scaling pro-cedures were considered for analysis, i.e. (1) absolute difference scores, |D| (Ghiselli, 1956, 1960), (2) individual consistencyscores (Asendorpf, 1990), and (3) relative difference scores (Zedeck, 1971). Then cluster analyses with Ward’s method andthe Euclidian distance (both squared and un-squared) measures with these scores were computed. This method providesresults with specific properties, i.e., (1) inclusion of non-overlapping clusters, (2) distance rather than correlation measure,and (3) preservation of unequal cluster sizes. The decision for an optimal number of clusters was guided by the followingcriteria: (1) the accepted solution has to be meaningful, (2) reasonably equal sample sizes per cluster should must result,and (3) there has to be adequate validity of the cluster solution.

Following these criteria the best cluster solution was established with Ward’s method (Euclidean distance) using relativedifference scores. As deferred imitation data are very little understood with respect to inter-individual differences in intra-individual change, only two cluster solutions were taken into account. The more clusters a cluster solution takes into account,the more the fuzziness of cluster interpretation increases. This hierarchical cluster analysis with the D scores as variablesresulted in (1) the optimal cluster solution with respect to equal cluster group sizes (nc1 = 45, nc2 = 33), (2) no floor or ceilingeffects and (3) the cluster solution showed adequate validity with respect to the developmental correlates assessed. In anext step of the analysis, stability correlations and means of deferred imitation performance were calculated for the twocluster groups, which are reported below.

3.4. Cluster group analyses

The first developmental group (n = 45) improves from Mt1 = 3.4 (SDt1 = 1.4) over Mt2 = 7.7 (SDt2 = 1.9) to Mt3 = 17.0(SDt3 = 3.5). The second developmental group (n = 33) improves from Mt1 = 5.1 (SDt1 = 1.0) over Mt2 = 5.8 (SDt2 = 1.3) toMt3 = 17.0 (SDt3 = 4.3). An analysis of variance for repeated measures revealed a significant linear trend, F = 958, p < .05, and asignificant interaction, F = 4.0, p < .05. No gender differences were found for the developmental trend (F = 0.03, df = 1, ns). Therepeated-measurement ANOVA demonstrates that both developmental cluster groups increase their deferred imitation per-formance with age, but that the developmental groups differ with respect to their growth pattern. The first developmentalgroup starts with a lower value (intercept) than the second developmental group on the first testing (t = 5.8, p < .01). The firstdevelopmental group then increases approx. 4 action steps (second testing) whereas the second developmental group muchslower (<1 action step) than the first developmental group. Now the first developmental group outperforms the seconddevelopmental group (t = 5.0, p < .01). On the third testing, the differences between the two developmental groups equal outas the first developmental group increases 9.3 action steps and the second developmental group increases 11.2 action steps.

Stability. Fig. 2 depicts the stability correlations for both developmental groups. The first developmental group (n = 45)shows high, significant correlations between t1 and t2, r (45) = .66, p < 0.01, and between t2 and t3, r (45) = .54, p < 0.01.The second developmental group (n = 33) shows a high, significant correlation between t1 and t2, r (33) = .53, p < 0.01, and amoderate, not significant, correlation between t2 and t3 r (33) = .29, ns. The comparison of the total sample stability correlationand developmental group stability correlations indicates that longitudinal stabilities for the two developmental groups arehigher than the overall stability found for the total sample. We therefore assume that it is interesting to analyze these twodevelopmental trajectories in more detail. In a final analysis step, external validity of the cluster grouping was establishedin order to get more insight in the developmental processes that underlie the specific developmental growth patterns.

Multivariate analysis. First and second developmental group were compared with a multivariate analysis of variance(MANOVA) with respect to the developmental scales of the ET6-6. Table 3 depicts the descriptive and inferential statisticsand effect sizes between developmental groups for the developmental factors of the ET6-6.

T. Kolling et al. / Infant Behavior & Development 33 (2010) 159–167 165

Fig. 2. Stability correlations of deferred imitation performance of developmental group 1 (n = 45) and developmental group 2 (n = 33).

Table 3Developmental differences for developmental groups.

Developmental subscales t1 (12 months) t2 (18 months) t3 (24 months)

Cluster 1 Cluster 2 d Cluster 1 Cluster 2 d Cluster 1 Cluster 2 d

Body motor development 108.9 (14.4) 110.3 (14.2) .1 101.1(13.7) 107.3 (17.7) .4 99.2 (16.5) 98.9 (13.7) .0Hand motor development 112.8 (15.0) 117.5 (12.3) .3 104.2 (22.2) 107.3(19.4) .2 99.4 (18.8) 102.9(15.5) .2Cognitive development 107.6 (13.6) 110.2(10.7) .2 111.5 (15.3) 111.4(12.1) .0 102.8 (15.5) 102.0(15.5) .1Body self-awareness 99.3 (14.7) 100.7(15.3) .1 98.7 (16.8) 107.7 (15.7) .6 105.5 (10.2) 107.6 (5.3) .3Receptive language 100.4 (18.3) 101.0(20.5) .0 108.9 (22.6) 119.6 (16.1) .6 104.8 (15.7) 108.8 (9.8) .3Expressive language 105.3 (16.2) 109.8(16.5) .3 109.8 (16.4) 113.8(14.6) .3 103.0 (17.2) 107.7(11.3) .3Social development – peers 104.5 (13.1) 107.2(10.8) .2 96.1 (24.5) 96.0 (25.6) .0 96.0 (15.7) 93.2 (21.2) .2Social development – adults 98.1 (18.4) 101.8(15.3) .2 87.0 (23.9) 91.3 (21.1) .2 93.8 (17.2) 96.0 (19.3) .1Social development – auton. 103.3 (18.1) 110.2(15.9) .4 101.5 (16.1) 101.2(16.1) .0 95.5 (17.3) 94.6 (15.5) .1Emotional development 97.7 (16.3) 98.5 (15.9) .1 100.8 (20.7) 101.4(24.1) .0 97.1 (21.1) 98.2 (22.6) .1

Note. Standard deviations shown in parentheses. Bold text indicates significant differences.

The second developmental group has significantly higher scores in body self-awareness and receptive language devel-opment at the second testing (18 months) than the first developmental group. As the statistical analyses are constrained bythe fact that thirty comparisons were calculated in the multivariate analyses, we also grouped the predictor variables intofunctional domains (6 functional domains) by calculating an average of the different subscales. This analysis demonstratedthat language and body self-awareness still differ significantly in the two clusters. These analyses do not completely ruleout chance differences. But, however, as it is reasonable to assume theoretically that language and self-awareness relate todeferred imitation, the effect found in the data is more likely to be a true, theoretical effect than a mere statistical one.

4. Discussion

The present multivariate, longitudinal deferred imitation study extends previous research by focusing on inter-individualdifferences of intra-individual change and its determinants in infancy. As expected and replicating cross-sectional andlongitudinal results (Heimann & Meltzoff, 1996; Heimann et al., 2006; Nielsen & Dissanayake, 2004; Strid et al., 2006),declarative memory performance increased throughout the second year indicating that, with development, young childrenare able to retain more target actions in memory. As the tests used in the present study consist of more target actions(different steps and structures) than the tests used in previous studies, it is shown here that the improvement of memoryperformance is more pronounced than formerly reported.

Critics may focus on the fact that the total number of target actions differs at the three time points. This argument, however,omits reasoning from a test-theoretical perspective (structural and differential continuity, the relationship between itemdifficulty and differential validity, the longitudinal shrinkage of variance, floor and ceiling effects) as a test adaptation isimportant in each longitudinal study, especially in periods of rapid developmental growth. Discussion may probably remainover how well the present tests are able to provide structural continuity, which concerns the degree of constancy in the

166 T. Kolling et al. / Infant Behavior & Development 33 (2010) 159–167

operational definition of a trait over time or developmental measurement equivalence (Bates & Novosad, 2005). However, theauthors are confident because of empirical results and theoretical reasoning that the tests used are reasonable for assessingindividual differences from a longitudinal perspective. A comparison of the present study with recent longitudinal studies(Heimann & Meltzoff, 1996; Heimann et al., 2006; Nielsen & Dissanayake, 2004; Strid et al., 2006) shows that these studieschanged the target objects across testings as well. A further empirical argument is provided by the fact that it is relativelyimprobable that younger infants produce more target actions even if they are confronted with a longer test. Kressley-Mba etal. (2005) demonstrated in a sample with younger infants (6-month-olds) that a longer, not age-adequate series of presentedactions leads to a lower memory performance than an age-adequate shorter action series.

Stability correlations of deferred imitation for the overall group were lower between 12 and 18 months (r = .17) thanbetween 18 and 24 months (r = .39*). The amount of inter-individual differences in declarative memory performance isalmost identical to those reported by Nielsen and Dissanayake (2004) providing a cross-validation and demonstrating that,as predicted, inter-individual differences of intra-individual change (developmental dynamics) are high throughout thesecond year of life.

The core analysis of the present study (person-centred cluster analysis) revealed that two developmental groups of infantsshowing different developmental trajectories are found. The analysis of variance indicated a significant time effect and asignificant interaction effect demonstrating that both developmental groups improve with respect to declarative memoryperformance. Furthermore, the significant interaction effect shows that both developmental groups develop differently. Withrespect to stability correlations the first developmental group shows high, significant correlations between all three testings(rt1t2 = .66∗, rt2t3 = .54∗) indicating a high amount of continuity in development of the individuals. The second develop-mental group shows a high, significant correlation between t1 and t2 (rt1t2 = .53∗) but no significant correlation between thesecond and third testing. This finding, that the second developmental group has a lower continuity of memory developmentthan the first one in the second half of the second year, indicates a larger developmental variability within this group.

To explain these two developmental trajectories and the reason for the developmental variability in the second develop-mental group, the findings of multivariate analyses were taken into account. They indicated a significant difference betweenthe two developmental groups for the factors body self-awareness as well as receptive language at the second testing (18months). Previous research also demonstrated that language and deferred imitation are related to each other (Heimann etal., 2006; Herbert & Hayne, 2000). The findings of the present study both add more evidence to this relation and demonstratethat language seems to be an important factor for a differentiation and explanation of different memory trajectories in thisage range. In addition, these findings point to the necessity of differentiating between receptive and expressive languagemore thoroughly in future studies.

The significant difference in body self-awareness (representation and knowledge of own body and body of others) indi-cates that the self may be an important variable for the developmental dynamics found in the 18-month period. For explainingthis difference, we would like to refer to episodic memory theory and conclude – against the background of recent episodicmemory (Tulving, 2002) and representational theories of development (Perner, 1991) – that around 18 months deferredimitation performance becomes more episodic-like in nature. It is assumed here that in the second developmental group,which shows a significantly better body self-awareness than the first developmental group, deferred imitation memory mayhave characteristics of episodic memory to a larger amount than that is the case in the first developmental group. We, thus,are inclined to name the first developmental group “late episodic memory developers” and the second developmental group,“early episodic memory developers”. The early episodic memory developers show higher deferred imitation in comparison tothe late episodic memory developers at the age of 12 months, an age where it is theoretically feasible to assume that memory isstill semantic. Quite possibly, the early episodic memory developers are advanced in their memory development with respectto encoding and retrieving actions than the late episodic memory developers. At the second testing (18 months) the relationsbetween the groups with respect to memory performance reverse. Now it is the late episodic memory development clusterwhich shows higher memory performance than the early episodic memory development cluster. One might speculate thatthe beginnings of episodic memory are associated with a memory overload of the infants and therefore the overall memoryperformance level decreases temporarily by the time episodic memory shows up for the first time. At the age of 24 months,the third wave in this longitudinal study, differences between the two developmental groups equal out. This finding maypoint to the fact that the role of self and language development for deferred imitation is most important around 18 monthsof age and does not play an important role in later ages, when these concepts are acquired by most of the children to a similarand fairly good amount.

We do not imply by our proposition of early episodic memory developers and late episodic memory developers that semanticand episodic memories are completely separate processes. It is assumed, however, that with an increasing age of the infantsand somewhat earlier in the second than in the first developmental group, deferred imitation has an episodic character. Thismeans that at the end of the second year action-related memory assessed via the deferred imitation task includes details likespace (the location of the demonstration) time (when did the demonstration happen) as well as own experience (mentallytravel in time from present to past experience). However, this can only be speculated by the data set obtained in this study.

Overall the equal effect sizes of self and receptive language show that both developmental aspects are important toexplain inter-individual differences of intra-individual change of deferred imitation performance. These findings are new intwo ways: firstly, no longitudinal study before assessed the self as a correlate of deferred imitation under the assumptionthat the relation between self and memory development point to changes in the way actions are encoded in the deferredimitation task. Secondly, as the present study differentiated between receptive and expressive language, it was shown that

T. Kolling et al. / Infant Behavior & Development 33 (2010) 159–167 167

the receptive part of language relates more strongly to deferred imitation than expressive language. As (deferred) imitationalso is a receptive, i.e., observing, learning strategy stronger relations between receptive language and deferred imitationthan between expressive language and deferred imitation are feasible.

Future deferred imitation research should take into account in addition to language and measures of social interaction(joint attention, care taking) the self-concept as important determinants of the developmental course in multi-domain (mul-tivariate) experimental and longitudinal studies. With larger longitudinal infant studies and proper and powerful statisticalmethods (path modelling, growth curve modeling, person-centred analysis) infant researcher will be able to disentangle therelations between declarative memory, self, language and social interaction measures in the infancy period. By achievingthis goal, not only basic research will benefit in terms of theory building, but also test developers and diagnosticians will beable to understand infant development more properly.

Acknowledgements

This study was supported by a grant from the German Research Foundation (Deutsche Forschungsgemeinschaft) toMonika Knopf (KN 275/3-1). We thank our research assistants for their help during data collection and analysis. We greatlyappreciate the cooperation of all the families who participated in the study.

References

Asendorpf, J. (1990). The measurement of individual consistency. Methodika, 4, 1–22.Barr, R., Dowden, A., & Hayne, H. (1996). Developmental changes in deferred imitation by 6- to 24-month-old infants. Infant Behavior & Development, 19,

159–170.Bates, J. E., & Novosad, C. (2005). Measurement of individual difference constructs in child development, or taking aim at moving targets. In D. M. Teti (Ed.),

Blackwell handbook of research methods in developmental science (pp. 319–339). Oxford: Blackwell Publishing.Ghiselli, E. E. (1956). Differentiation of individuals in terms of their predictability. Journal of Applied Psychology, 40, 374–377.Ghiselli, E. E. (1960). The prediction of predictability. Educational and Psychological Measurement, 20, 3–8.Goertz, C., Knopf, M., Kolling, T., Frahsek, S., & Kressley, R. (2006). Der Frankfurter Imitations-Test für 12 Monate alte Kinder (FIT12): Erprobung einer

Aufgabenserie zur Erfassung des deklarativen Gedächtnisses Einjähriger [The Frankfurt Imitation Test for 12-month-old infants: Testing a series ofitems for measuring declarative memory in 1-year-olds]. Zeitschrift für Entwicklungspsychologie und Pädagogische Psychologie, 38, 88–96.

Goertz, C., Kolling, T., Frahsek, S., & Knopf, M. (2008). Die Frankfurter Imitationstests für 18 Monate alte Kinder (FIT 18) und für 24 Monate alte Kinder (FIT24): Entwicklung altersangepasster Instrumente für die Verzögerte Imitation [The Frankfurt Imitation Test for 18- and 24-month-old infants]. Zeitschriftfür Entwicklungspsychologie und Pädagogische Psychologie, 40, 152–160.

Goertz, C., Kolling, T., Frahsek, S., Stanisch, A., & Knopf, M. (2008). Assessing declarative memory in 12-month-old infants: A test–retest reliability study ofthe deferred imitation task. European Journal of Developmental Psychology, 5, 492–506.

Heimann, M., & Meltzoff, A. N. (1996). Deferred imitation in 9- and 14-month-old infants: A longitudinal study of a Swedish sample. British Journal ofDevelopmental Psychology, 14, 55–64.

Heimann, M., Strid, K., Smith, L., Tjus, T., Ulvund, S. E., & Meltzoff, A. N. (2006). Exploring the relation between memory, gestural communication, and theemergence of language in infancy: A longitudinal study. Infant and Child Development, 15, 233–249.

Herbert, J., & Hayne, H. (2000). Memory retrieval by 18–30-month-olds: Age-related changes in representational flexibility. Developmental Psychology, 36,473–484.

Knopf, M., Mack, W., & Kressley-Mba, R. (2005). Wissen und Erinnern [Knowing and remembering: On episodic memory in infants and preverbal children].Psychologische Rundschau, 56, 113–122.

Kolling, T., Goertz, C., Frahsek, S., & Knopf, M. (2009). Stability of deferred imitation in 12- to 18-month-old-infants: A closer look into developmentaldynamics. European Journal of Developmental Psychology, 6, 615–640.

Kressley-Mba, R. A., Lurg, S., & Knopf, M. (2005). Testing for deferred imitation of 2- and 3-step action sequences with 6-month-olds. Infant Behavior &Development, 28, 82–86.

Mandler, J. M. (2004). A synopsis of The foundations of mind: Origins of conceptual thought. Developmental Science, 7, 499–505.McDonough, L., Mandler, J., McKee, R., & Squire, L. (1995). The deferred imitation task as a nonverbal measure of declarative memory. Proceedings of the

National Academy of Science, 92, 7580–7584.Meltzoff, A. N. (1985). Immediate and deferred imitation in fourteen- and twenty-four-month-old infants. Child Development, 56, 62–72.Meltzoff, A. N., & Moore, M. (1977). Imitation of facial and manual gestures by human neonates. Science, 198, 75–78.Nielsen, M., & Dissanayake, C. (2004). Pretend play, mirror self-recognition and imitation: A longitudinal investigation through the second year. Infant

Behavior & Development, 27, 342–365.Perner, J. (1991). Understanding the representational mind. Cambridge: MIT Press.Piaget, J. (1962). Play, dreams and imitation in childhood. New York: Norton.Prudhomme, N. (2005). Early declarative memory and self-concept. Infant Behavior & Development, 28, 132–144.Strid, K., Tjus, T., Smith, L., Meltzoff, A. N., & Heimann, M. (2006). Infant recall memory and communication predicts later cognitive development. Infant

Behavior & Development, 29, 545–553.Tulving, E. (2002). Episodic memory and common sense: How far apart? In A. Baddeley, J. P. Aggleton, & M. A. Conway (Eds.), Episodic memory: New directions

in research (pp. 269–287). NY, USA: Oxford University Press.von Eye, A., & Bogat, G. (2006). Person-oriented and variable-oriented research: Concepts, results, and development. Merrill-Palmer Quarterly, 52, 390–420.Zedeck, S. (1971). Identification of moderator variables by discriminant analysis in a multipredictable group validation model. Journal of Applied Psychology,

55, 364–371.