J.  S. Metcalfe*

Department of Kinesiology University of Maryland

College Park, MD 20742– 2611 E-mail: [email protected]


K.   McDowell*

Department of Kinesiology and Program in Neuroscience and Cognitive Science University of Maryland

College Park, MD 20742– 2611 E-mail: [email protected]


T-Y. Chang L-C. Chen

Department of Kinesiology University of Maryland

College Park, MD 20742– 2611 E-mail: [email protected]

E-mail: [email protected]


J. J. Jeka

  1. E. Clark

Department of Kinesiology and Program in Neuroscience and Cognitive Science University of Maryland

College Park, MD 20742– 2611 E-mail: [email protected]

E-mail: [email protected]

Development of Somatosensory- Motor Integration: An Event- Related Analysis of Infant Posture in the First Year of Independent Walking

ABSTRACT: The ability to integrate sensation with action is considered an important factor underlying the development of upright stance and locomotion. While many have studied sensory influences on posture, the nature of these influences and how they change with development have yet to be thoroughly characterized in infancy. Six infants were examined from 1 month prior to walk onset until 9 months of independent walking experience while standing quietly and touching either a static or a dynamic surface. Five adults were examined performing an analogous task. An event-related, time-frequency analysis was used to assess the relationship between postural sway and the motion of the somatosensory stimulus. Phase consistency between sway and stimulus was observed for both adults and infants, and with walking experience the infants increased their phase consistency rather than changing aspects of response amplitude. It is concluded that walking experience provides opportunities for an active tuning of sensorimotor relations for adequate estimation of body position in space and thus facilitates refined   control

over temporal aspects of postural sway. © 2004 Wiley Periodicals, Inc. Dev Psychobiol 46: 19– 35, 2005.

Keywords: posture; motor; development; somatosensory; sensorimotor; integra- tion; event-related; infancy; longitudinal; human


By the end of the first year of life, infants accomplish the challenging task of independent stance and locomotion. It has been proposed that the ability to integrate sensation with action may underlie this development (Barela, Jeka, & Clark, 1999; Bertenthal & Clifton, 1998; Bertenthal, Rose, & Bai, 1997). A number of investigators have addressed the issue of sensorimotor integration in the

*Co-first author.

Received 6 May 2003; Accepted 29 August 2004

Correspondence to: J. S. Metcalfe Contract grant sponsor: NSF Contract grant number: 9905315 Contract grant sponsor: NIH

Contract grant number: IF31 MH12963-01 Published online in Wiley InterScience

(www.interscience.wiley.com). DOI 10.1002/dev.20037

© 2004 Wiley Periodicals, Inc.

context of both adult and infant postural control. For adults, the integration of sensation with action has been robustly shown using somatosensation, vision, and multi- modal combinations of stimuli (Dijkstra, Scho¨ ner, & Gielen, 1994; Jeka, Oie, Scho¨ ner, Dijkstra, & Henson, 1998; Oie, Kiemel, & Jeka, 2002); however, neither the nature of this sensorimotor integration in infants nor how it changes with development have been well charac- terized. The majority of studies addressing sensorimotor integration in infant posture have focused on across-trial amplitude and/or average phase responses to visual cues (Barela, Godoi, Freitas, & Polastri, 2000; Bertenthal et al., 1997; Bertenthal, Boker, & Xu, 2000; Butterworth & Hicks, 1977; Delorme, Frigon, & Lagace´, 1989; Lee & Aronson, 1974). Other sensory modalities as well as other aspects of the postural response, such as within-trial amplitude and phase consistency, have been   relatively

neglected. To validate that postural development is dependent on adaptive sensorimotor integration, it is necessary to characterize how all relevant modalities are integrated into postural control, the response within each modality, and how these relations change with develop- ment. The purpose of the present study was to begin addressing these issues by examining the emergence of this sensorimotor relationship during the earliest period of upright stance.

Traditionally, the moving room paradigm (Lee & Aronson, 1974; Lee & Lishman, 1975) has been a primary means of examining the linkage between posture and sensation. The general method of the moving room in- volves the use of dynamic manipulations of the surround- ing environment to observe sensory-induced postural adjustments. For example, recent applications of this paradigm have shown that standing adults entrain their sway with small oscillations of a visual field (Dijkstra   et al., 1994) or a fingertip contact surface (Jeka et al., 1998). Characteristics of the adult’s response to low- frequency   (<1   Hz)   sensory   manipulations include

(a) increased sway amplitude at the frequency of the

sensory stimulus and (b) a consistent phase relationship between body sway and the sensory cue, which could be interpreted in the framework of Oie et al. (2002) as neces- sary to estimate body position relative to the environment. Stimulus frequencies in the range of rv0.2 to rv0.4 Hz typically lead to the largest amplitude and the most consistent phase responses in adults. This paradigm has been used with some success to examine sensorimotor integration in the development of posture in infancy; however, much remains to be learned about the specific nature of these responses and how they progress towards those that are so robustly observed in adults.

A few studies have examined the influence of dynamic sensory information on the postural sway of infants in bipedal standing (Delorme et al., 1989; Foster, Sveistrup, & Woollacott, 1996; Lee & Aronson, 1974; Stoffregen, Schmuckler, & Gibson, 1987). Of these studies, only one attempted to measure infants’ ability to continuously relate their body sway with an oscillating stimulus (Delorme et al., 1989). The data from this study suggested that anterior– posterior motion of the surrounding room influenced the amplitude of the infants’ anterior– posterior sway. That is, the amplitude of sway accounted for by the frequency of the stimulus was increased as compared with the surrounding frequency components; however, because frequency spectra were not reported for an appropriate control condition (static visual surround), it is difficult to attribute these results to an influence of the visual stimulus alone as opposed to a natural tendency for infants to show increased sway in the range of the stimulus frequency. Further, because the phase relationship between the stimulus and the response was not reported for either

condition, this study provided no insight into the ability of infants to consistently maintain a particular phase relation- ship over multiple cycles of sway behavior. Likewise, studies employing discrete movements of the visual surround have suggested developmental changes in the amplitude of sway responses such that in new walkers, initial responses are poorly scaled to the stimulus ampli- tude and often exceed biomechanical sway limits, thus resulting in staggers and falls (Foster et al., 1996; Lee & Aronson, 1974; Stoffregen et al., 1987); however, the magnitude of the perturbations in these discrete tasks is more representative of a transition between two stationary environments as opposed to a dynamic relationship within a continuously changing environment. As such, it is unclear what inferences may be made from these studies regarding the continuous integration of sensation and postural control.

Several studies have focused on how visual informa- tion is continuously integrated with body sway during a sitting task in infants (Barela et al., 2000; Bertenthal et al., 1997; Bertenthal et al., 2000). On the whole, these studies have provided evidence for relatively small sensory in- fluences on sway amplitude, and only one (Barela et al., 2000) provided insight into the within-trial consistency of the phase relationship between sway and sensory stimuli. In Barela et al.’s (2000) study of prewalking infants (6– 9 months old), the observation of large within-trial phase variability (>60 degrees) suggested that the phase

relationship between the infant’s sway and visual stimulus

was highly variable, if not completely random.

Taken together, studies of both sitting and standing

(a) support influences of visual stimuli on the amplitude of postural sway in infants and (b) hint that the temporal (phase) relationship between sway and a continuously oscillating stimulus is inconsistent prior to the onset of independent walking. These data suggest a hypothesis that the development of the amplitude and phase components of the sway response occurs differentially in ontogenetic time, a suggestion that has been made previously in studies ranging from goal-directed reaching (Konczak, Borutta, Topka, & Dichgans, 1995), visually evoked potentials (Sokol, Zemon, & Moskowitz, 1992), vestibulo- ocular reflexes (Wiener-Vacher, Toupet, & Narcy, 1996), and interlimb coordination in walking (Clark, Whitall, & Phillips, 1988). The suggestion of differential devel- opment of amplitude and phase relations in posture is further supported by the observation of greater temporal variability in the EMG responses of young infants to a physical perturbation of the base of support than in infants with independent walking experience (Sveistrup & Woollacott, 1996). Amplitude increases without consis- tent phase responses suggest an extraction of relevant information from the sensory cue, but a poor estimation of body position relative to the stimulus. As such, knowledge

of developmental changes in both amplitude and phase consistency are essential for a complete understanding of how infant responses progress towards those seen in adults. Further, whether these findings hold across different sensory modalities remains unknown, but is important to assess the validity of the claim that the ability to integrate sensation with action, in all modalities, is a general factor underlying postural development.

Infant postural sway is considerably morevariable than adult sway; rendering the assessment of within-trial amplitude and phase components of stimulus– response relationships a nontrivial task. To address this problem, we have adapted an event-related analysis that is based on methods used in the electroencephalographic (EEG) literature (Pfurtscheller, 1977; Sutton, Braren, Zubin, & John, 1965). This analysis relies on averaging to reveal componentsofasignalthataretimelockedtoanevent(e.g., a sensory drive), but would otherwise remain hidden by a low signal-to-noise ratio. Given the variability of infant sway, the averaging process involved in an event-related analysis is well suited for assessing phase variability and interactionsbetweenamplitudeandphasewhenexamining infant postural behavior. Further, time-frequency analysis has been introduced as a way to assess nonstationary time series across multiple cycles of observation (Kayhan, El-Jaroudi, & Chaparro, 1994) and has been suggested as an important way to examine human upright posture (Loughlin, Redfern,& Furman, 1996; Schumann, Redfern, Furman, El-Jaroudi, & Chaparro, 1995). This type of method is unique in that it can be used to study how specific components of the response unfold dynamically in time. In particular, this technique can be combined with an event- related analysis to characterize postural responses at any specific frequency, such as that of the driving stimulus. By using the event-related method alone or in combination with the time-frequency analysis, we can ask particular questions regarding the nature of the developing sensor- imotor relationship. For example, do infants who increase their response amplitude to both discrete and continuous stimuli prior to independent walking stabilize the temporal relationship between themselves and the environment with increasing upright experience? If this were true, it would support the hypothesis that the development of the ability to estimate body position relative to the environment, rather than the more general ability to integrate sensation with action, underlies the development of independent upright stance and locomotion.

In this article, our first aim is to provide convincing evidence of infants’ abilities to integrate haptic cues with postural sway. To accomplish this, we compare infant and adult postural responses to somatosensory cues to confirm the efficacy of our selected event-related, time-frequency analysis technique. That is, the validity of the method could be questioned if it lacked the power to detect dif-

ferences between the consistent responses of adults as compared with those of infants. The results of this comparison, if positive, would enable the pursuit of our primary purpose: to examine the longitudinal pattern of change in postural responses to somatosensory cues as infants gain experience in the upright. Relative to this primary purpose, our hypothesis is that from walk onset onwards, development of the stimulus– sway relationship will be characterized by increasing temporal consis- tency as opposed to changes in stimulus-related sway amplitude.



Six infants (3 females, 3 males), who were part of a larger longitudinal study, were included in this analysis. All infants were healthy, full-term, and without developmen- tal delay as assessed by the Bayley Scales of Infant Development (2nd ed.; Bayley, 1993). Infants entered the study when they were able to sit independently (mean age ¼ 6.14 þ 0.86 months) and were tested monthly until they reached 9 months of independent walking experience (mean age at walk onset ¼ 10.97 þ 1.22 months). For the purpose of this investigation, infants were assessed only at ages when they could maintain upright stance while using single-hand support, specifically from 1 month prior to walk onset onwards. Each infant’s parent or guardian provided written informed consent prior to inclusion in the longitudinal protocol, and a small payment was given to the parent or guardian for each laboratory visit.


To provide a control group for comparison, 5 healthy adults (2 females, 3 males) also were included in the analysis. The 5 adults (mean age ¼ 29.8 þ 8.2 years) were unpaid volunteers who provided written informed consent. The Institutional Review Board at the University of Maryland approved all experimental procedures for this study.


All data were remotely acquired with a Windows NT workstation (Intergraph TDZ-2000) using a National Instruments A/D board (BNC-2090) and custom LabView software (National Instruments, Inc., Austin, TX). All signals were sampled at 50.33 Hz, in real time, and synchronized to a manual trigger at trial onset. Figure 1 illustrates the experimental setup for infants, wherein each participant stood on a pedestal mounted on a force platform in parallel stance with eyes open and with the hand touching a dynamic (oscillatory) or static surface; adults stood on a pedestal in an analogous position.

Touch Apparatus. For the infants, an instrumented touch bar mounted on a support frame was positioned to the right

FIGURE 1 Illustration of the experimental setup. The ex- perimenter maintained the infant’s attention with a variety of books or toys. Omitted are the infant’s parent or guardian who sat close enough to prevent the infant falling, the second experimenter who monitored the infant’s hand contact with the touch apparatus, and the Logitech tracking system (see text).

of each participant and aligned with the top of his or her iliac crest. The touch bar was composed of a 4.4 cm diameter convex surface, formed by the top half of a

45.7 cm long PVC tube. The purpose of this convex surface was to be ‘‘touchable’’ without being ‘‘graspable’’ by the infants. The contact surface was attached atop two support columns, each instrumented with force transdu- cers (Interface MB-10) for resolving applied hand-contact forces. Vertical touch forces were recorded with negative values indicating downward application of force. The entire touch bar was mounted on a precision linear posi- tioning table (Daedal 105002BT) and was driven by a DC brushless motor (Compumotor SM231AE) controlled by a torque servo drive (Compumotor OEM675T). For the adults, the contact surface was a circular metal plate 5 cm in diameter mounted on a tripod and positioned to the right and forward of each participant at approximately hip level. Both the touch apparatus and the servo control system for the adults were identical to those described in previous reports (Jeka et al., 1998; Oie et al., 2002).

For all participants, the servomotor was experimen- tally controlled by specifying movement amplitude as well as peak velocity and acceleration. Touch surface position was measured using a precision optical encoder

attached to the end of the servomotor. The encoder produced 1,000 pulses per revolution, and a custom circuit monitored the motor’s direction and counted the number of encoder pulses to enable D/A conversion at a resolution of 0.004 mm. To account for the inertial properties of the different contact surface sizes, both servo controllers were tuned such that resulting motion profiles were equivalent.

Postural Sway. Center of pressure in the medial– lateral (CPML) and anterior– posterior (CPAP) directions were calculated from ground reaction forces measured by a force platform (Kistler 9261A). Three-dimensional shoulder girdle and approximate center of mass displace- ments also were sampled using a Logitech 6-dimensional position-tracking system (VR Depot; Boony Doon, CA), but are not reported in this analysis.

Videotaping. All infant testing sessions were displayed on a remote monitor and videotaped with a standard sVHS recorder (Panasonic AG-7350) for on-line observation of trials during acquisition as well as later behavioral coding. The videotape records were synchronized with the analog data using an event synchronization unit (PEAK Perfor- mance Technologies) and time stamped with an SMPTE code generator (Horita RM-50 II).

Design and Procedure

Data for this article are from a larger longitudinal study designed to examine quiet stance as well as the use of somatosensory information in the development of posture. Here, we describe in detail the procedures relevant to our questions, and only a summary of the full experimental protocol is provided to illustrate the context within which these data were obtained. The remaining data and procedures are to be presented elsewhere (cf. Metcalfe et al., 2004).

Infants. Upon entering the laboratory each month, the infant was provided a few minutes to acclimate to the testing environment. During this time, an experimenter questioned the parent or guardian about the infant’s general health and developmental progress during the previous month. Following the acclimation period, the infant was taken to a small testing room (2.1 x 5.5 m) that was enclosed by heavy black curtains and was introduced to a small pedestal (10 cm deep x 20 cm long x 11 cm tall) placed to the left of the infant touch bar and affixed to the force platform. The infant’s shoes were removed, and once placed on the pedestal, the two Logitech trackers were affixed and the position of the touch apparatus was adjusted such that the infant’s arm was abducted approximately 45 degrees and the hand was aligned with the top of the iliac crest.

Figure 1 provides a simplified illustration of the postural task for an infant in the touch condition. To facilitate participation, an experimenter was positioned in front of the infant to best maintain the infant’s visual attention on one of a variety of toys or books. The parent or guardian was always present and helped position the infant for each trial as well as prevent any possible falls. To ensure that the child performed the appropriate touch condition, a second experimenter was positioned to the infant’s right side and monitored hand contact with the touch bar.

During the testing session, the infant completed five conditions including independent stance (without touch), touching a static surface, and three dynamic conditions of touching an oscillating surface (frequencies ¼ 0.1, 0.3, and 0.5 Hz; amplitudes ¼ 1.6, 0.59, and 0.36 cm, respectively). Three trials were collected in each condi- tion. All trials lasted 60 s, with the exception of the 0.1 Hz trials, which were 90 s. The 15 trials were presented in randomized order. An exception to the random presenta- tion was that an independent stance (‘‘no touch’’) trial never occurred within the first 5 trials. This is based on previous experience with this paradigm, which has shown that infants tend not to participate in touch conditions when independent stance trials are presented first. Follow- ing completion of all experimental conditions, the infant’s height and weight were recorded for future reference.

Adults. Upon entering the laboratory, the experimental procedures were explained to the participant and the informed consent form was provided for reading. After signing the consent form, the adult removed his or her shoes and was taken to a small testing room (2.1 x 5.5 m), which was enclosed by heavy black curtains. Within the testing room, the two Logitech trackers were affixed, and the participant stood on a block (19 cm deep x 40.5 cm long x 29.5 cm tall) that was centered on the force platform and placed to the left and behind the adult touch plate. The purpose of this block was to create a pedestal analogous to that used for the infants, but scaled to the adult’s larger body size. Therefore, the positioning of the adults’ feet on the pedestal as well as the location of the touch plate approximated the posture of the infants. Similar to Figure 1, the postural task analyzed in this study required the participant to stand quietly on the pedestal while touching the contact surface and main- taining visual attention on an object positioned at eye height. Following completion of all experimental condi- tions, the participant’s height and weight were recorded for future reference.

During the testing session, the adult participant completed four conditions including independent stance (without touch), touching a static surface, touching an oscillating surface similar to the infants (frequency ¼

0.3 Hz; amplitude ¼ 0.59 cm), and touching an oscillating surface in which the amplitude of oscillation at 30 s was reduced in half (to an amplitude of 0.3 cm) and then subsequently stopped oscillating for the last 30 s of the trial (60– 90 s). Two trials were collected in each condition. All trials lasted for 30 s, with the exception of the decreasing-amplitude trials, which were 90 s. The eight trials were presented in a randomized order.

For the purposes of this report, our analyses focused only on the conditions in which the infants and adults

(a) touched a static surface and (b) touched a dynamic surface (0.3 Hz oscillation). To examine the postural relationship with the stimulus, we specifically analyzed the response to the 0.3 Hz frequency because (a) it is in the range of frequencies in which adult sway responses typically show the largest amplitude and most consistent phase, and (b) the period of oscillation at 0.3 Hz is short enough (3.33 s) that the infants in this study could perform the task over multiple cycles of stimulus oscillation within a given data segment (minimum ¼ 15 s, rvfive cycles; discussed later). The data from the remaining conditions are beyond the scope of this investigation and will be the subject of a future report on frequency-response charac- teristics of infant sensorimotor integration.

Data Reduction

Behavioral Coding. Because infants rarely stand quietly for the entire duration of a trial, data analysis was based on individual segments of quiet stance within the completed trials. As such, following data acquisition, all infant trials were independently examined for valid segments of quiet posture by two trained coders.1 Criteria for valid data segments were: (a) minimum length of 22 s (15 s plus edges; discussed later); (b) continuously touching, but not grasping, the contact surface; (c) standing independently without assistance of the experimenter or caregiver; (d) no dancing or bouncing movements; and (e) no falling or stepping movements. Small head/trunk movements (i.e., turning) and upper-limb movements (i.e., pointing) that did not disrupt the maintenance of stance were considered as valid postural data and were not excluded; however, any movements resulting in a disruption of the task, such as a complete turn and lean toward the parent/guardian, were completely excluded from further analyses. Coders were instructed to record start and end times of segments to the nearest second, and these times were assessed by   a

1Coders   underwent   an   extended   training   protocol involving

(a) explanation and definition of the behavior of interest (i.e., quiet posture), (b) explanation of our behavioral coding scheme, (c) test coding of a subset of trials that had already been successfully and reliably coded by the lead experimenters, and (d) assessment of the trainee’s agreement with the established procedure. A coder was not allowed to assess data to be analyzed until their coding was in agreement with segments selected by the established procedure.

Table 1. Summary of the Average Amount of Data Contributed at Each Walk Age

Walk Age

(months)       Walk Age (days)      n       Amount of Data (s)

-1                      -20.40               4 66.25
(6.07) (41.06)
0 4.33 5 33.33
(3.08) (18.22)
1 33.50 5 59.33
(2.43) (22.53)
2 66.43 6 55.28
(6.08) (30.97)
3 96.40 5 58.67
(3.51) (29.66)
4 123.83 6 98.06
(6.62) (68.27)
5 155.00 5 73.67
(3.46) (37.87)
6 185.83 6 60.83
(2.14) (32.74)
7 212.83 5 59.44
(1.33) (29.96)
8 247.75 3 43.89
0.96 (22.99)
9 274.50 6 49.44
(7.37) (24.26)

Note. n ¼ number of infants contributing data at each walk age; SDs are reported in parenthesis under each mean.

third experimenter. Only those data segments that were in complete agreement (overlapping times) were taken as reliable data.2 Once the segments of quiet stance were determined, they were extracted from the raw files using an interactive data-extraction program. Table 1 presents a summary of the average and SD of the amount of data contributed after behavioral coding at each level of walk age. Adult data were not coded, as these participants were able to complete the task in the specified duration without actions that invalidated trial segments.

Signal Processing. All data extraction, reduction, and signal processing were performed using custom soft- warewritten in MATLAB (Mathworks, Inc., Natick, MA). To address specific aspects of the data, several analyses were conducted. Analyses were performed using two filtering parameters. For the first analysis, hereafter referred to as total sway, a recursive lowpass filter (second-order Butterworth; f3db ¼ 5 Hz) was applied to the raw data following removal of the mean. To speci- fically examine the influence of the 0.3 Hz drive on sway,

2Interrater reliability was set at 100% by definition. It was never the case that disagreement between coders occurred across trials. If a segment of quiet stance was found within a given trial, it was noted by both coders, and only the portions of time where both coders were in agreement were used for further analysis.

the second analysis, hereafter referred to as 0.3 Hz sway, used a time-frequency methodology. This time-frequency method used a recursive bandpass filter in a narrow range around the driving frequency of the stimulus (0.3 Hz) and is similar to techniques used in studies of EEG (Pfurtscheller, 1977). Specifically, the raw data, with the mean removed, were bandpass filtered ( f3db ¼ 0.2, 0.4 Hz) using a second-order Butterworth filter (see Figure 2b; raw data in Figure 2a). Of the number of time-frequency analysis techniques available, each suffers some limita- tion (Kayhan et al., 1994; Schumann et al., 1995). Our results will validate that the method chosen here was well suited to examine the questions posed in this study. Based on the characteristics of both filters, 3.5 s were uniformly removed from either end of all filtered data to account for edge effects.3

Sway Amplitude. The first measures generated from this analysis were amplitude variables, which were computed irrespective of stimulus motion. Amplitude (AMP) was the root mean square (RMS) of the filtered data and was calculated for total and 0.3 Hz sway. For total sway, AMP is equivalent to mean sway amplitude (the SD of the total time-series). For 0.3 Hz sway, AMP is analogous to the spectral amplitude of the 0.3 Hz sway component (e.g., the amplitude spectrum obtained using the Fourier transform) and thus is related to the amount of sway amplitude accounted for at 0.3 Hz (see Figure 2b, white sidebar). Amplitude variability (AMPVAR; Figure 2b, black side- bars) was determined as the SD of the AMP that was computed for both total and 0.3 Hz sway. Because AMP and AMPVAR are computed on the sway irrespective of the behavior of the stimulus, they yield only indirect measures of the influence of the dynamic stimulus on postural sway.

Event-Related Measures. To examine the relationship of sway and stimulus directly, an event-related technique was used to assess average phase, phase variability, and the potential interactions between amplitude and phase. A summary of all dependent measures from this event- related, time-frequency analysis is presented in Table 2.

For the dynamic condition, sections of sway data were time locked to the repeating cycles of stimulus motion, treating the beginning of each cycle as the event. For the static condition, time-locked measures were computed relative to a simulated stimulus. The purpose of using this simulated stimulus was to provide a basis of comparison

3Because filtering involves fitting polynomials to time series, some distortion always appears as transients at the extremes of the data series due to a small number of samples (e.g., ‘‘edge effects’’. This also has been discussed as the ‘‘end point problem’’ (Phillips & Roberts, 1983). The extent of the edge effects for these filters was estimated by comparing pre- and postfiltered idealized waveforms (sine waves) of different frequencies). The maximum transient length observed was 3.5 s, which was chosen as a uniform amount of time to be removed from all postfiltered data.

FIGURE 2 Exemplar of the methodology applied in the dynamic condition for a healthy adult. (A) Raw time series of postural sway (heavy line) and the stimulus (thin line). (B) Data from A bandpass filtered with a passband from 0.2 to 0.4 Hz, thus yielding the 0.3 Hz sway. Note the attenuation of the signals at each end; these edge effects were removed from all filtered data (see Signal Processing). The solid white bar at the end of the time series indicates the computed AMP, with the black bars on the top and bottom indicating AMPVAR. (C) Sway amplitude and

(D) sway-stimulus cross-correlations plotted across phase   for

each stimulus cycle. (E) Sway amplitude from C averaged across time-locked cycles; the RMS of this averaged signal yields AMPTL. (F) The cross-correlation functions from D averaged across time-locked cycles; horizontal dashed line indicates CCTL, and the vertical dashed line indicates LAGTL. These same steps also were applied to the total sway (lowpass filtered at a 5 Hz cutoff) (see Method for details).

between the static and dynamic conditions. The simulated stimulus was created as a zero-phase, 0.3 Hz sinusoid that was scaled to the amplitude of the actual 0.3 Hz stimulus from the dynamic condition. Because this stimulus did not exist for the static touch condition (i.e., the contact surface was actually stationary), measures computed on it served as an indication of how they should behave when the stimulus and response are independent of one another (i.e., creating the null hypothesis; any measure computed relative to the simulated stimulus in the static touch condition should represent an arbitrary relationship). Using these stimuli (real and simulated), several time- locked measures of amplitude and phase were then com-

puted for both conditions as described in the following paragraphs.

Time-locked amplitude (AMPTL) was computed by windowing the filtered data to a length equal to one cycle of the stimulus (sampling rate/drive frequency; 50.33/0.3, rv168 samples) with no overlap (i.e., cycles were locked at 360-degree intervals throughout the data segment). The time-locked sections of data (Figure 2c) were then averaged across individual cycles of the stimulus (Figure 2e). AMPTL was then determined by computing the RMS of the averaged signal across time-locked cycles, reflecting an interaction between the amplitude and the phase of the sway response.

Time-locked cross-correlations (CCTL) were used to index the phase as well as cycle-by-cycle phase variability between the signal and the sway response. Similar to AMPTL, CCTL between the filtered sway and the stimulus signals (signals shown in Figure 2a and 2b; correlation functions in Figure 2d) were computed for each cycle. The correlation functions were calculated using Hanning windows with a 50% overlap (i.e., cycles were locked at 180-degree intervals throughout the data segment). The correlation functions were then bias corrected and averaged across cycles. CCTL was defined as the absolute maximum of the averaged correlation function (Figure 2f, horizontal dashed line), and time-locked lag (LAGTL) was the time at which the maximum positive correlation was found (Figure 2f, vertical dashed line). In both dynamic and static conditions, because of the narrow bandpass filter at 0.3 Hz, individual cycle correlations with the 0.3 Hz stimulus will always reach a maximum of

1.0. Thus, the 0.3 Hz sway CCTL reflects only the variability of the phase of the postural signal; that is, if the phase is stable from cycle to cycle, CCTL will approach a value of 1 whereas if the phase varies from cycle to cycle, CCTL will be reduced. For total sway, CCTL reflects both the variability of the phase of the postural signal as well as the magnitude of the correlation between the sway and the stimulus. In both cases, LAGTL is analogous to the average phase.

Infant Touch Forces. Mean vertical touch force (TFV) was used as a global index of the extent to which infants were using the touch apparatus for mechanical support. Reduction of the raw touch-force signal included removal of analog spikes (i.e., data points exceeding 4 within-trial SDs from the mean were reduced to the perimeter of that range) followed by lowpass filtering with a recursive second-order Butterworth filter ( f3db ¼ 5 Hz). Absolute TFV was then calculated as the mean touch force during the data segment minus a baseline that was determined by the transducer output when the infant’s hand was not on the touch apparatus in the same trial. TFV was calculated for 58% of the trials with valid postural data

Table 2.     Summary of Event-Related/Time-Frequency Dependent Variables


Frequency Component     Variable                          Computation                                          Represents

Total Sway (lowpass filtered; 5 Hz cutoff)

0.3 Hz Sway (bandpass filtered; 0.2, 0.4 Hz cutoff)

AMP            RMS of lowpass filtered time-series

irrespective of stimulus oscillation.

AMPVAR             Standard deviation of AMP computed on

the lowpass filtered time-series.

AMPTL                RMS of the filtered time-series after

time-locking to repeated cycles of stimulus oscillation (3600 intervals).

CCTL                      Maximum of the averaged cross-correlation functions

across time-locked ½ stimulus cycles (1800 intervals).

LAGTL                  Value of the time-lag associated with the

maximum CCTL.

AMP            RMS of bandpass filtered time-series

irrespective of stimulus oscillation.

AMPVAR             Standard deviation of AMP computed on

the bandpass filtered time-series.

AMPTL                RMS of the 0.3 Hz time-series after

time-locking to repeated cycles of stimulus oscillation (3600 intervals).

CCTL                      Maximum of the averaged 0.3 Hz

cross-correlation functions across time-locked ½ stimulus cycles (1800 intervals).

LAGTL                  Value of the time-lag associated with the maximum 0.3 Hz CCTL

The average displacement from upright equilibrium; analogous to mean sway amplitude.

The variance of the mean displacement from upright equilibrium; measures within-trial consistency of sway amplitude.

The cycle-by-cycle consistency of both

amplitude and timing (phase) of the overall sway response to the stimulus.

The cycle-by-cycle consistency of the

relationship between the overall sway response and the stimulus; sensitive to both magnitude and time lag of the correlation.

The average time-difference between stimulus motion and the overall sway response across repeated cycles of stimulus oscillation.

The average amplitude of the sway occurring at

0.3 Hz; analogous to the spectral amplitude determined using a Fourier transform.

The variance of the amplitude of sway occurring at 0.3 Hz; measures within-trial consistency of the sway occurring at the driving frequency.

The cycle-by-cycle consistency of both amplitude and timing (phase) of the 0.3 Hz sway component.

The cycle-by-cycle consistency of the temporal relationship between the 0.3 Hz sway component and the stimulus; similar to coherence (shared power) between stimulus.

The averaged time-difference between stimulus motion and the 0.3 Hz sway component; analogous to the average relative phase between the stimulus and 0.3 Hz sway.


because some of the infants never removed their hands from the touch bar in a given trial and thus had no valid baseline. As continuous hand contact was a criterion for valid segments in the touch conditions, baseline data were never included in the touch data. The values of TFV that were tested statistically were evenly distributed across infants, conditions, and walk ages (infants:   w2 ¼ 0.35,

sway through a comparison of the postural responses of infants and adults. The 2 x 2 ANOVAs contained one between-subjects factor, Age (infant, adult), and one within-subjects factor, Condition (static, dynamic), and were performed to examine AMP, AMPVAR, AMPTL, and CCTL for total and 0.3 Hz sway in both medial– lateral and anterior– posterior sway directions. Because the

2                                            2

conditions: w1 ¼ 0.001, walk ages: w9 ¼ 0.57, ps > 0.9).

Statistical Analysis

The analysis procedure for this experiment was performed in two steps. In the first step, univariate repeated measures ANOVAs were used to provide convincing evidence of infants’ abilities to integrate haptic cues with postural

simulated stimulus used in the static condition was at an

arbitrary phase relative to the sway, LAGTL had no meaning for the static condition. As such, LAGTL from the dynamic condition was tested using one-way ANOVAs with Age (infant, adult) as a between-subjects factor.

While the first step of the analysis addressed the issues

of detecting a postural response to the stimulus and discriminating differential effects based on age   group,

the second step was designed to examine longitudinal changes in the sway response of infants. For this purpose, linear mixed-model regression (PROC MIXED; SAS v. 8.02; The SAS Institute, Inc., Cary, NC) was used to examine all dependent variables for the influence of Condition (static, dynamic) and Walk Age. Walk Age was used to normalize all data to the individual infant’s developmental level and was computed as both months and days elapsed from walk onset. Walk onset was defined as the age at which the infant first took three independent steps.

The mixed-model analysis was chosen because it separately controls fixed (i.e., Condition) and random (i.e., Infant) sources of variation as well as provides tools to account for variance heterogeneity and correlated measures. For this analysis, class-level fixed-effect variables were specified as Condition (dummy variable; 0 ¼ static, 1 ¼ dynamic) and Walk Age in months of data acquisition (integer intervals). Random effects were speci- fied as due to Infant as well as the Infant x Condition and Infant x Walk Age interactions. Residuals were blocked within infant and stimulus condition, and the variance- covariance matrix was structured with a first-order autoregressive function. This structure was selected to account for correlations between subsequent intervals of Walk Age with the assumption that ages that are closer together (e.g., 1 and 2 months) display a higher correlation than those that are separated by larger time intervals (e.g., 1 and 9 months). With the aforementioned parame- terization, the mixed model was applied to a linear regression across Condition, Walk Age in days, and the Condition x Walk Age interaction. Values for LAGTL were again examined only in the dynamic condition and, thus, simply regressed across Walk Age.

For both steps in the analysis, hypothesis tests were conducted on weighted averages within each individual and condition, using the amount of data obtained (e.g., number of cycles) as the weighting factor. Further, correlation coefficients were normalized using a Fisher’s Z transformation, and the amplitude variables (AMP, AMPVAR, AMPTL) were normalized using the natural logarithm prior to hypothesis tests. For clarity, variables are reported in their untransformed metric in all plots and descriptive statistics. All effects were tested at a significance level of a ¼ 0.05. Finally, with the exception of age-related variables, which are shown as means ± SDs, all effects are represented with means and SEs.


Exemplar sway responses for both adults and infants are illustrated in Figure 3 for the dynamic condition   (0.3 Hz stimulus). Two-dimensional plots of lowpass

FIGURE 3 Exemplars of adult and infant center of pressure sway responses in the dynamic condition (0.3 Hz stimulus). Two-dimensional stabilograms of the lowpass filtered sway response are illustrated for (A) an adult and (B) an infant with 6 months of walking experience. Likewise, the 0.3 Hz sway component is shown for (C) the adult and (D) the infant. In adults (E), a close relationship is observed between the lowpass filtered time series of medial– lateral sway (heavy line) and the stimulus (thin line). In infants (F), the relationship between lowpass filtered time series of medial– lateral sway (heavy line) and the stimulus (thin line) is difficult to observe.

filtered (Figure 3a and 3b) and bandpass filtered (Figure 3c and 3d) sway trajectories reveal important aspects of the postural response. First, the total sway of the adult (Figure 3a) showed dramatically reduced amplitude compared to that of an infant performing an analogous task (Figure 3b). Second, when the 0.3 Hz component was extracted for the adult sway (Figure 3c), the amplitude of the 0.3 Hz sway appeared similar to the amplitude of the total-sway response. In contrast, the infant’s 0.3 Hz component was proportionally a smaller component of the total sway (Figure 3d). Further, in all trajectories, one can see that the response to the medial– lateral stimulus was not confined only to the medial– lateral direction. In both total sway and 0.3 Hz sway, the postural trajectories appeared oriented in a resultant direction, intermediate to the anterior– posterior and medial– lateral axes. Finally, the exemplars of the medial– lateral sway component (Figure 3e and 3f) highlight the importance of the analysis

technique employed in this study. While the adults’ sway demonstrated a relatively strong influence of the stimulus (Figure 3e), which could be analyzed with standard techniques (Dijkstra et al., 1994), the infants’ response to the dynamic stimulus was much less apparent and masked by large variability (Figure 3f). Data in these figures illustrate why alternative analysis techniques discussed in the Method section earlier were considered appropriate for understanding infant postural control.

Infants Versus Adult Controls

The first step of the analysis attempted to provide convincing evidence of infants’ abilities to integrate haptic cues with postural sway as well as to verify that the methods used herein had the power to detect stimulus influences on the sway of the infants. For this analysis, all infants (n ¼ 6) at 6 months’ postwalking (M ¼ 185.83 ±

2.14 days of walking experience) were compared with the adult controls using univariate 2 (Age: infant, adult) x 2 (Condition: static, dynamic) repeated measures ANOVAs for all variables except LAGTL, which was examined only for age effects in the dynamic condition. All significant results for this comparison have been summarized in Table 3.

Total Sway. To examine the differential effects of the dynamic stimulus on the total-sway response of the infants and adult controls, 2 x 2 ANOVAs were applied to all measures computed on the lowpass filtered data. In this analysis, effects of condition were revealed for CCTL, which was computed in relation to the stimulus in the dynamic condition but versus an identical, but simulated, stimulus in the static condition. As can be seen in Figure 4a, CCTL was significantly increased in the dynamic stimulus condition for both medial– lateral, F(1, 9) ¼ 24.59, p < 0.01, and anterior– posterior, F

(1, 9) ¼ 7.29, p < 0.05, sway. The only amplitude measure that revealed a main effect for condition was AMPTL in the medial– lateral direction, F(1, 9) ¼ 5.52, p < 0.05. For the static condition, AMPTL was reduced (0.14 ±

0.02 cm; M ± SE) as compared with that observed in the dynamic condition (0.21 ± 0.04 cm).

Age main effects were observed in the measures of AMP, AMPVAR, and CCTL. For adults, CCTL was greater than that for infants in both the medial– lateral, F(1, 9) ¼ 10.59, p < 0.01, and anterior– posterior, F(1, 9) ¼ 6.27, p < 0.05, directions (Figure 4b). Conversely, Figure 5a demonstrates that AMP computed in both medial– lateral and anterior– posterior directions was significantly greater for infants than adults, medial– lateral: F(1, 9) ¼ 15.28,

p < 0.01; anterior– posterior: F(1, 9) ¼ 21.67, p < 0.01. Similarly, AMPVAR was significant in both medial– lateral, F(1, 9) ¼ 19.84, p < 0.01, and anterior– posterior, F(1, 9) ¼ 24.35, p < 0.01, directions wherein infants showed higher variability than adults (Figure 5b). No effects were observed for age in either AMPTL or LAGTL.

0.3 Hz Sway. The pattern of results in the 0.3 Hz sway was similar to, but had a further interaction than, that observed in total sway. For these narrow lowpass filtered data, a significant Age x Condition interaction was revealed for CCTL in the medial– lateral direction, F(1,

  • ¼ 09, p < 0.05. Figure 6 illustrates this interaction wherein adults showed a much larger increase in medial– lateral CCTL from static to dynamic conditions than did the infants. Follow-up comparisons using Dunn’s method performed within age indicated that while the increase in

adults was significant, t(9) ¼ 5.27, p < 0.01, from static to dynamic conditions, the same was not true for the infants, t(9) ¼ 1.29, p > 0.05. Because of the orthogonal nature of this Age x Condition interaction, main effects for med- ial– lateral CCTL, which were found for both condition and age, will not be discussed further.

Table 3.   Summary of Infant Versus Adult Results

Medial– Lateral Sway                                                                     Anterior-Posterior Sway

Age Condition Age x Condition Age Condition Age x Condition
AMP ** **
Total AMPVAR ** **
Sway AMPTL *
CCTL ** ** * *
AMP ** *
0.3 Hz AMPVAR ** *
Sway AMPTL * **
CCTL ** ** * * **

Note. LAGTL results are not summarized as a different statistical design was performed on this measure and no significant results were observed.

* p < .05.

** p < .01.

FIGURE 4   Time-locked cross-correlations (CCTL) computed on total sway. Main effects for

  • condition and (B) age for medial– lateral (white bars) and anterior– posterior sway (dark bars). Bars represent M ± SE.

In the anterior– posterior direction, condition main effects were observed for the time-locked measures CCTL and AMPTL. A significant condition effect for CCTL, which reflects phase consistency for 0.3 Hz sway (see Table 2), was observed in the anterior– posterior direc- tion, F(1, 9) ¼ 14.07, p < 0.01; dynamic ¼ 0.48 ± 0.05,

static ¼ 0.20 ± 0.04. For AMPTL, main effects for condi-

tion also were observed in both the medial– lateral, F (1, 9) ¼ 8.21, p < 0.05, and anterior– posterior directions, F(1, 9) ¼ 11.95, p < 0.01. Across age group, AMPTL was greater in the dynamic (medial– lateral ¼ 0.16 ± 0.03 cm; anterior– posterior ¼ 0.12 ± 0.02 cm) than in the static (medial–lateral ¼ 0.07 ± 0.01; anterior– posterior ¼ 0.05 ±

0.01 cm) condition.

FIGURE 5 Age main effects for (A) total sway AMP, (B) total sway AMP variability, (C) 0.3 Hz sway AMP, and (D) 0.3 Hz sway AMP variability. In all plots, medial– lateral is plotted with white bars and anterior– posterior with dark bars. Bars represent M ± SE.

FIGURE 6 Medial– lateral time-locked correlations for infants (white bars) and adults (dark bars) plotted across condition (static, dynamic). Bars represent M ± SE.

Age main effects were again seen within the 0.3 Hz sway for AMP, AMPVAR, and CCTL. Measures of both AMP (Figure 5c) and AMPVAR (Figure 5d) demonstrated larger values for infants than adults. For 0.3 Hz AMP, infants showed larger sway amplitude than adults, medial– lateral: F(1, 9) ¼ 10.42, p < 0.01; anterior–

posterior: F(1, 9) ¼ 9.23, p < 0.05, with a corresponding increase in AMPVAR, medial– lateral: F(1, 9) ¼ 15.85,

p < 0.01; anterior– posterior: F(1, 9) ¼ 9.02, p < 0.05. For CCTL, which showed a significant Age x Condition interaction in the medial– lateral direction, infants had consistently lower values than adults in the anterior– posterior direction, F(1, 9) ¼ 5.54, p < 0.05; infants ¼ 0.27 ± 0.04; adults ¼ 0.40 ± 0.04. No age effects were observed for AMPTL or LAGTL.

Infants Longitudinally

Longitudinal changes in infants’ postural responses were assessed using linear mixed-model regression of all dependent measures across Condition and Walk Age. As with the first analysis step, all measures were ex- amined for both total and 0.3 Hz sway.

Touch Forces. As a check for the possibility of infants changing their reliance upon the contact surface for mechanical support, the absolute level of vertical touch forces (TFV) were assessed with a linear mixed-model regression that included Walk Age (in days) and Condition as factors. In this analysis, no evidence was found for differential application of touch force in either stimulus condition or across the 10-month duration of this ex- periment. Irrespective of walk age and condition, these infants applied 3.84 ± 0.53 N of vertical touch force on the contact surface, which is consistent with previous reports

at these walk ages (Barela et al., 1999; Metcalfe et al., 2004). As no changes were observed in this variable, it is unlikely to have had a strong influence on our pattern of results and thus will not be discussed further.

Total Sway. For total sway, significant effects were ob- served only in the medial– lateral direction. Specifically, Walk Age x Condition interactions were revealed for both CCTL, F(1, 19.7) ¼ 7.35, p < 0.05, and AMP, F(1,

19.7) ¼ 4.96, p < 0.05. Figure 7a, in comparison with

Figure 7b, illustrates that for CCTL this interaction was due to a significant increase across Walk Age in the dynamic stimulus condition, slope ¼ 0.0003 ± 0.0001; t(46.8) ¼ 2.28, p < 0.05, that was not observed in the static con- dition, which had a mean coefficient of 0.12 ± 0.01 across the range of walk ages observed. While the Walk Age x

Condition interaction also was significant for medial– lateral AMP, neither of the within-condition slopes alone contributed to this effect. Plotted in Figure 8a and 8b, the two conditions were differentiated such that across Walk Age, medial– lateral AMP tended to decrease in the dynamic condition while it increased in the static condition; however, on a qualitative level, this interaction does not appear to have any practical significance.

FIGURE 7 Medial– lateral time-locked correlations, CCTL, for all infants in (A) dynamic and (B) static conditions. Separate markers , , , , , ) indicate data corresponding to each individual infant. The solid lines are the mixed-model fits reflecting within- and between-infant sources of variation.

FIGURE 8 Medial– lateral sway amplitude, AMP, for all infants in (A) dynamic and (B) static conditions. Separate markers , , , , , ) indicate data corresponding to each individual infant. The solid lines are the mixed-model fits reflecting within- and between-infant sources of variation.

0.3 Hz Sway. Unlike the findings for total sway, main effects rather than interactions were found for the narrow bandpass filtered 0.3 Hz sway. The only main effect for walk age was observed for AMP in the anterior– posterior direction, F(1, 56.8) ¼ 4.16, p < 0.05. Acrossthe 10 months

of walking experience, AMP reflected an average increase

in anterior– posterior sway of 0.003 ± 0.001 cm/day, t(56.8) ¼ 2.04, p < 0.05, for all infants. In the medial– lateral direction, a main effect for condition was observed for time-locked amplitude, F(1, 27.5) ¼ 5.19, p < 0.05, such that AMPTL reflected an increased 0.3 Hz sway in the dynamic (0.15 ± 0.02 cm) as compared with the static condition (0.11 ± 0.02 cm), indicating a consistent within-trial effect of the stimulus on sway amplitude that persisted across all levels of walk age.


The changing integration of somatosensory information with the postural system was examined in infants as they stood quietly while touching a contact surface that was either stationary or gently oscillating in the medial–

lateral direction. The results indicated that infants not only used cues from the contact surface for postural control but that this integration improved with upright locomotor experience. The comparison of infants with adult controls who performed analogous tasks revealed marked differ- ences between the two groups as well as evidence for an influence of the dynamic stimulus on the posture of the infants. Most important, these data demonstrated the strongest longitudinal change in the phase consistency, as opposed to the amplitude, of the sway response that was associated with increased walking experience. This supports the general hypothesis that dynamic experience in the upright affords infants opportunities that facilitate the improvement of the ability to estimate body position relative to the environment.

Infants Versus Adult Controls

Comparisons of infants at 6 months of walking age with adult controls demonstrated that infants integrate postural sway with dynamic somatosensory stimuli. By separately examining within-trial phase and amplitude components of the sway response, evidence indicated that changes in infant sensorimotor integration were primarily due to increasingly consistent phase. These findings were similar to and in the same direction as data from the adult controls who performed an analogous task. Together, the infant and adult data lend credence to the hypotheses that infants integrate dynamic somatosensory stimuli with their sway and that developmental changes are due to increased phase stability by 6 months of walking experience.

Analysis of the sway response in the medial– lateral direction revealed differences between infants and adults in several measures including AMP, AMPVAR, and CCTL. The finding of increased AMP for both total and 0.3 Hz sway, which are analogous to standard measures of sway amplitude (total sway) and spectral amplitude (0.3 Hz sway), is consistent with previous research with toddlers (2– 3 years); that is, infants demonstrated larger sway amplitude than adults (Newell, Slobounov, Slobounova, & Molenaar, 1997; Riach & Hayes, 1987). Also consistent with the results of Riach and Hayes (1987), the AMPVAR results suggest that the infant’s large sway amplitude was accompanied by a large within-trial amplitude variance. While AMP and AMPVAR are sensitive only to amount of sway, the time-locked measures offer somewhat different insights. For example, because CCTL is sensitive to within-trial phase stability, it provides a measure of how consistently the individual swayed with the temporal properties of the stimulus. In this experiment, CCTL revealed a condition effect for total sway, indicating that both infants and adults entrained their medial– lateral postural sway with the stimulus; however, for the 0.3 Hz sway, an Age x Condition interaction revealed that only

adults achieved phase consistency in the 0.3 Hz com- ponent of medial– lateral sway. The discrepancy between these two findings suggests that the infant’s response to the stimulus was nonlinear; that is, the infant’s sway response was distributed across the frequency spectrum rather than simply reflecting the 0.3 Hz stimulus in   the

0.3 Hz response component. Taken together, the medial– lateral results support sensorimotor integration through consistent phasing of the response to the stimulus rather than stimulus-related amplitude increases.

Sensorimotor integration was more clearly observed in infant as well as adult anterior– posterior sway. For 0.3 Hz sway, both AMPTL and CCTL revealed main effects for condition, providing evidence for infant and adult sen- sorimotor integration. For total sway, a CCTL main effect for condition also supports consistent phasing between anterior– posterior sway and the stimulus. Further, the age effects observed for amplitude measures were similar, but in general smaller, than the same effects in the medial– lateral direction. The finding of condition effects in the anterior– posterior direction across the age groups is con- sistent with previous data for this type of postural task. A medial– lateral stimulus is typically associated with a postural response in the same direction; however, we have observed that this effect is dependent in part on the position of the touch apparatus relative to the individual’s base of support. That is, in cases where the stimulus presented to the hand has required an arm position forward and to the side of an individual standing in parallel stance, effects could be observed in anterior– posterior sway (Jeka, Ribeiro, Oie, & Lackner, 1998). As such, we intentionally placed the touch apparatus for adults in a location that was analogous to the infants, and found anterior– posterior effects in both age groups. Addit- ionally, theuse of parallel stancemayhaveledtodecreased anterior– posterior stability. The finding of similar statis- tical effects but decreases in overall amplitude and phase stability of the anterior– posterior rather than medial– lateral sway suggests that the anterior– posterior direction was a smaller and less stable component of the overall sway. Thus, the effects of the stimulus appeared stronger in the anterior– posterior direction due to this lower baseline stability. Taking all of these factors into account, the clear effects in the anterior– posterior sway for both adults and infants further validate the conclusion that the infants demonstrated an ability to integrate the somatosensory stimulus with their postural control system.

Previous research on visual manipulations of the environment suggests that prior to walking onset, infants’ responses to dynamic visual stimuli while in a seated posture are characterized by an inconsistent within-trial phase relationship (Barela et al., 2000). We observed that both the adults and 6-month walkers demonstrated consistencies in the phase relationship with the stimulus.

These data point to clear signatures of integration between the sensory cue and the postural control system. While the current data alone cannot address whether amplitude and phase relationships develop differentially, taken in the context of the previous research (Barela et al., 2000; Clark et al., 1988; Konczak et al., 1995; Sokol et al., 1992; Wiener-Vacher et al., 1996), these data contribute to understanding sensorimotor integration by showing improved phase, as opposed to amplitude, consistency in infants by 6 months of walking experience.

Infants Longitudinally

The comparison with adults performing an analogous postural task established that by 6 months of walking experience, these infants were capable of adopting a temporally consistent relationship with the somatosensory stimulus, particularly in the anterior– posterior direction. In the second step of this analysis, the goal was to probe developmental changes in this sensorimotor integration as a function of upright postural experience from the onset of independent walking. Thus, all infants were assessed using the same event-related measures in both the static and the dynamic conditions from 1 month prior to walk onset until 9 months of walking experience.

The longitudinal analysis revealed that the infants’ postural responses showed increasingly stable phase in the medial– lateral direction with walking experience. Specifically, because time-locked correlations (CCTL) are influenced by phase consistency and because this measure was sensitive to longitudinal change, it was concluded that changes in this sensorimotor integration are due to increasingly stable phase responses to the dynamic stimulus. While for medial– lateral AMP, a Walk Age x Condition interaction reached statistical significance, qualitative assessment suggested that this effect may not have any practical significance (Figure 8). By contrast, Figure 7 displays an obvious effect such that prior to walk onset, the estimated regressions did not differ across condition (compare CCTL at walk age ¼- 30 in Figure 7a & 7b), indicating a lack of phase consistency that im- proved with increased walking experience.

This constellation of observations, taken in combina- tion with the information, albeit sparse, hinting that the temporal aspects of stimulus-induced postural responses are highly variable prior to walk onset at both the level of behavior (Barela et al., 2000) and muscular activation (Sveistrup & Woollacott, 1996), suggests that the phase stability may develop concurrent with walking experi- ence. What underlies the relationship between locomotor experience and this changing phase consistency, however, remains to be fully understood. One suggestion may be derived from the fact that similar to the medial– lateral results from the adult versus infant comparison, the

longitudinal data indicated nonlinearities in the develop- ment of sensorimotor integration. That is, effects asso- ciated with walking experience were found in total rather than 0.3 Hz sway. What this indicates is a lack of pre- cision in adopting the appropriate temporal coordination between postural sway and the information specified by the stimulus.

A precisely defined temporal relationship between body sway and sensory stimuli is necessary for veridical estimation of body position relative to the environment. While this estimation of postural state is considered integral in control-theory-based models of posture for adults (Kiemel, Oie, & Jeka, 2002; Kuo, 1995; Lestienne & Gurfinkel, 1988; van der Kooij, Jacobs, Koopman, & van der Helm, 2001), its role has had, until recently, minimal influence on explanations of postural develop- ment. In part, this is due to a lack of investigations identi- fying how the specific properties of the stimulus may interact with different sensory modalities to influence the nature of infant’s postural responses. For example, prior assessments of the frequency-response characteristics of adult posture have led to the suggestion of modality- specific use of position versus velocity cues in the sen- sorimotor control of posture (cf. Dijkstra, 2000). Recent theoretic modeling (Kiemel et al., 2002) has further indicated the importance of velocity information for the formation of stable estimates of postural state. The finding of improvements in phase consistency in the current study indicates a changing ability to stabilize the temporal aspects of postural coordination with external environ- mental information (see also Metcalfe et al., 2004), and thus is consistent with conclusions regarding the inter- action between upright walking experience and the ability to utilize somatosensory cues for prospective estimation of body position relative to the environment (Barela et al., 1999). These arguments further reinforce the conclusion that infants in this period of ontogeny exploit the opportu- nities provided by upright walking experience to actively tune their sensorimotor relations for adaptive stance control (Metcalfe & Clark, 2000). Future investigations, however, will aim to more thoroughly characterize the nature (e.g., frequency response) of this type of sensor- imotor integration across development as well as how the role of state estimation changes in concert with other aspects of stance control.

Event-Related Time-Frequency Analysis

The method used in the present study was adapted from event-related and time-frequency analyses used in the EEG literature. These types of analyses have proven useful to uncover cortical signatures of a variety of motor and cognitive-motor tasks including simple finger and toe movements (Pfurtscheller, Neuper, Andrew, & Edlinger,

1997), choice reaction-time tasks (McDowell, Jeka, Scho¨ ner, & Hatfield, 2002), and elite marksmanship (Kerick et al., 2001). The strength of these techniques is that they discriminate specific components of a signal embedded within noisy data. While this has not been a major problem for postural research with adults, due to relatively high-gain sway responses to dynamic stimuli, this is a more relevant issue when attempting to under- stand the infant’s dramatically more variable postural sway. In the present study, the number and the consistency of the effects suggest that event-related time-frequency analysis was a powerful tool for examining dynamic aspects of infant posture across development. The method was further validated by the fact that the same general pattern of results was observed in the measures based on total sway and 0.3 Hz sway, providing a strong indica- tion that the observed sensorimotor relationship was not merely an artifact of the narrow bandpass filter at the stimulus frequency.

As this study was an initial foray into applying such techniques to the problem of infant posture, future re- search may be guided by the results of this exploratory effort. For example, in the present study, time-locked correlation appeared to be the measure that was most sensitive to age and condition effects on sensorimotor integration. In part, the correlations were advantageous because they were computed on approximately double the number of cycles as the time-locked measures of amplitude. In studies such as this where participants entrain to periodic stimulus with a specific frequency, this is a relatively powerful measure that will be useful in future characterizations of infant posture across a range of stimulus properties.


In this study, we observed convincing evidence that infants adopted a consistent relationship between their postural sway and a somatosensory stimulus following onset of independent walking. Further, the temporal stability of this relationship increased concurrently with increasing walking experience. This observation is con- sistent with the hypothesis that the ability to integrate a sensory cue into the postural response and the ability to estimate relative body position develops differentially in time. Marked differences between the postural responses of adults and infants with 6 months of walking experi- ence also were observed. In particular, changes in overall postural stability as well as task and stance-specific stability influenced the manner in which infants used the stimulus to control upright standing and stabilize phase relations. Future investigations of this developing sen- sorimotor relationship, across the life span as well as at

multiple levels of analysis, would benefit from further assessment of the differential development of amplitude and timing components of postural sway across a range of dynamic sensory environments. Insights from such investigations will provide the information necessary to understand how infants tune their sensorimotor relations for the development of stable estimates of body position in space, thus leading to increased capacity for adaptive stance control.


This research was supported by National Science Foundation Grant 9905315 (J. E. Clark) and National Institutes of Health Grant 1 F31 MH12963-01 (K. McDowell). Portions of this research were presented at the annual meeting of the Society for Neuroscience in Orlando, FL, in 2002.

The authors thank Larry Douglass for guidance on the statistical analysis and all of our research assistants who helped collect and code the data included in this article. We also thank three anonymous reviewers who provided valuable comments in shaping our presentation and analysis. Finally, we express special gratitude to the parents and infants who gave willingly of their time and effort throughout the duration of this longitudinal study.


Barela, J. A., Godoi, D., Freitas, P. B., & Polastri, P. F. (2000). Visual information and body sway coupling in infants during sitting acquisition. Infant Behavior and Development, 23, 285– 297.

Barela, J. A., Jeka, J. J., & Clark, J. E. (1999). The use of somatosensory information during the acquisition of inde- pendent upright stance. Infant Behavior and Development, 22, 87– 102.

Bayley, N. (1993). Bayley scales of infant development (2nd ed.) San Antonio, TX: Psychological Corporation.

Bertenthal, B. I., Boker, S. M., & Xu, M. (2000). Analysis of the

perception– action cycle for visually induced postural sway in 9-month-old sitting infants. Infant Behavior and Devel- opment, 23, 299– 315.

Bertenthal, B. I., & Clifton, R. K. (1998). Perception and action. In D. Kuhn & R. S. Seigler (Eds.), Handbook of child psychology, Vol. 2: Cognition, perception, and language (5th ed., pp. 51– 102). New York: Wiley.

Bertenthal, B. I., Rose, J. L., & Bai, D. L. (1997). Perception– action coupling in the development of visual control of posture. Journal of Experimental Psychology: Human Perception and Performance, 23, 1631– 1643.

Butterworth, G., & Hicks, L. (1977). Visual proprioception and postural stability in infancy. A developmental study. Percep- tion, 6, 255– 262.

Clark, J. E., Whitall, J., & Phillips, S. J. (1988). Human interlimb coordination: The first 6 months of independent walking. Developmental Psychobiology, 21, 445– 456.

Delorme, A., Frigon, J., & Lagace´, C. (1989). Infants’ reactions to visual movement of the environment. Perception, 18, 667– 673.

Dijkstra, T. M. H. (2000). A gentle introduction to the dynamic set-point model of human postural control during perturbed stance. Human Movement Science, 19, 567– 595.

Dijkstra, T. M. H., Scho¨ ner, G., & Gielen, C. C. A. M. (1994). Temporal stability of the action– perception cycle for postural control in a moving visual environment. Experimental Brain Research, 97, 477– 486.

Foster, E. C., Sveistrup, H., & Woollacott, M. H. (1996). Transitions in visual proprioception: A cross-sectional developmental study of the effect of visual flow on postural control. Journal of Motor Behavior, 28, 101– 112.

Jeka, J. J., Oie, K. S., Scho¨ ner, G. S., Dijkstra, T. M. H., & Henson, E. (1998). Position and velocity coupling of postural sway to somatosensory drive. Journal of Neurophysiology, 79, 1661– 1674.

Jeka, J. J., Ribeiro, P., Oie, K., & Lackner, J. R. (1998). The structure of somatosensory information for human postural control. Motor Control, 2, 13– 33.

Kayhan, A. S., El-Jaroudi, A., & Chaparro, L. F. (1994). Evolutionary periodogram for nonstationary signals. IEEE Transactions on Signal Processing, 42, 1527– 1536.

Kerick, S. E., McDowell, K., Hung, T. M., Santa Maria, D. L., Spalding, T. W., & Hatfield, B. D. (2001). The role of the left temporal region under the cognitive motor demands of shooting in skilled marksmen. Biological Psychology, 58, 263– 277.

Kiemel, T., Oie, K. S., & Jeka, J. J. (2002). Multisensory fusion and the stochastic structure of postural sway. Biological Cybernetics, 87, 262– 277.

Konczak, J., Borutta, M., Topka, H., & Dichgans, J. (1995). The development of goal-directed reaching in infants: Hand trajectory formation and joint torque control. Experimental Brain Research, 106, 156– 168.

Kuo, A. D. (1995). An optimal control model for analyzing human postural balance. IEEE Transactions on Biomedical Engineering, 42, 87– 101.

Lee, D. N., & Aronson, E. (1974). Visual proprioceptive control of standing in human infants. Perception & Psychophysics, 15, 529– 532.

Lee, D. N., & Lishman, J. R. (1975). Visual proprioceptive control of stance. Journal of Human Movement Studies, 1, 87– 95.

Lestienne, F. G., & Gurfinkel, V. S. (1988). Posture as an organizational structure based on a dual process: A formal basis to interpret changes of posture in weightlessness. Progress in Brain Research, 76, 307– 313.

Loughlin, P. J., Redfern, M. S., & Furman, J. M. (1996). Time- varying characteristics of visually induced postural sway. IEEE Transactions on Rehabilitation Engineering, 4, 416–424. McDowell, K., Jeka, J. J., Scho¨ ner, G., & Hatfield, B. D. (2002). Behavioral and electrocortical evidence of an interaction between probability and task metrics in movement prepara-

tion. Experimental Brain Research, 144, 303– 313. Metcalfe, J. S., Chen, L-C., Chang, T-Y., McDowell, K.,   Jeka,

  1. J., & Clark, J. E. (2004). The temporal organization of

posture changes during the first year of independent walking. Experimental Brain Research [Online]. DOI: 10.1007/ s00221-004-2082-z.

Metcalfe, J. S., & Clark, J. E. (2000). Somatosensory infor- mation affords the exploration of posture in newly walking infants and toddlers. Infant Behavior and Development, 23, 391– 405.

Newell, K. M., Slobounov, S. M., Slobounova, E. S., & Molenaar,

  1. C. (1997). Stochastic processes in postural center-of- pressure profiles. Experimental Brain Research, 113, 158–164. Oie, K. S., Kiemel, T., & Jeka, J. J. (2002). Multisensory fusion: Simultaneous re-weighting of vision and touch for the control of human posture. Cognitive Brain Research, 14, 164– 176. Pfurtscheller, G. (1977). Graphical display and statistical evaluation of event-related desynchronization (ERD). Electro- encephalography and Clinical Neurophysiology, 43, 757–760. Pfurtscheller, G., Neuper, C., Andrew, C., & Edlinger, G. (1997). Foot and hand area mu rhythms. Journal of Psychophy-

siology, 26, 121– 135.

Phillips, S. J., & Roberts, E. M. (1983). Spline solution to terminal zero acceleration problems in biomechanical data. Medicine and Science in Sports and Exercise, 15, 382– 387. Riach, C. L., & Hayes, K. C. (1987). Maturation of postural sway in young children. Developmental Medicine and Child

Neurology, 29, 650– 658.

Schumann, T., Redfern, M. S., Furman, J. M., El-Jaroudi, A., & Chaparro, L. F. (1995). Time-frequency analysis of postural sway. Journal of Biomechanics, 28, 603– 607.

Sokol, S., Zemon, V., & Moskowitz, A. (1992). Development of lateral interactions in the infant visual system. Visual Neuroscience, 8, 3– 8.

Stoffregen, T. A., Schmuckler, M. A., & Gibson, E. J. (1987). Use of central and peripheral optical flow in stance and locomotion in young walkers. Perception, 16, 113– 119.

Sutton, S., Braren, M., Zubin, J., & John, E. R. (1965, November 26). Evoked-potential correlates of stimulus uncertainty. Science, 150, 1187– 1188.

Sveistrup, H., & Woollacott, M. H. (1996). Longitudinal development of the automatic postural response in infants. Journal of Motor Behavior, 28, 58– 70.

van der Kooij, H., Jacobs, R., Koopman, B., & van der Helm, F. (2001). An adaptive model of sensory integration in a dynamic environment applied to human stance control. Biological Cybernetics, 84, 103– 115.

Wiener-Vacher, S. R., Toupet, F., & Narcy, P. (1996). Canal and otolith vestibulo-ocular reflexes to vertical and off vertical axis rotations in children learning to walk. Acta Otolar- yngolica, 116, 657– 665.

Copyright of Developmental Psychobiology is the property of John Wiley & Sons Inc. and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder’s express written permission. However, users may print, download, or email articles for individual use.