The Development of Mother-Infant Coordination Across the First Year of Life

Abstract

Early mother-infant coordinated interactions play a critical role in infant development. The present study describes the development of the dyadic coordination of vocalization and gaze behavior between mothers and infants over the first year of life. In addition to describing developmental trajectories of behavior, the study contributes to our understanding of dyadic coordination by investigating how different measures of coordination relate to one another and how individual behaviors relate to dyadic coordination within the vocal and gaze domains.

Thirty dyads were recorded playing together with a standard set of toys when infants were 3, 6, 9, and 12 months of age, and mother and infant vocalization and gaze behaviors were coded from these videos on a moment-to-moment basis. Coordination was analyzed using both cross-recurrence quantitative analysis and event-based measures of analysis. Hierarchical linear modeling was used to examine developmental trajectories of vocalization and gaze coordination. Results indicated that dyadic coordination of the timing of vocalization and gaze behaviors is a very early emerging phenomenon that is supported by both mothers and infants. Vocal coordination did not show developmental change over time, whereas mothers and infants became increasingly coordinated in their gaze behaviors as infants got older. Taken together, findings from this study highlight the importance of considering bidirectional and developmental influences on parent and infant behavior.

Parent-infant social interactions play a crucial role in early development. In the first months of life, parents organize behavior toward their infants in contingent ways that facilitate the development of coordinated interactions; and by the time infants are 2–3 months old, these interactions begin to show a coordinated, bidirectional structure. Social interactions are inherently multi-modal and multifaceted in nature; and dyadic coordination of communicative behaviors (e.g., vocalizations, gaze) establishes structure within an otherwise complex experience. This structure creates predictability, allowing members of the dyad to anticipate their partner’s behavior and respond to it. As infants develop and refine the ability to perceive regularities and recognize relationships between their own behavior and their partner’s behavior, coordinated communicative interactions provide opportunities for learning and development (Feldman, 2007; Jaffe et al., 2001). For example, when infants experience prompt, contingent responses to their communicative behaviors, they learn about the efficacy of those behaviors within the interaction (Goldstein, Schwade, & Bornstein, 2009). Numerous studies have shown that features of communicative coordination, such as maternal vocal contingency and shared attention, predict infant development in a number of areas (e.g., Carpenter, Nagell, Tomasello, Butterworth, & Moore, 1998; Jaffe et al., 2001; Tamis-LeMonda, Bornstein, & Baumwell, 2001).

In fact, vocalization and gaze coordination are two of the most widely studied aspects of early parent-infant interaction. Research suggests that infants begin coordinating their vocal and gaze behaviors with those of social partners from very early in life (Beebe, Alson, Jaffe, Feldstein, & Crown, 1988; Van Egeren, Barratt, & Roach, 2001). Despite the empirical and theoretical importance of coordination in early parent-infant interactions, the literature on this topic has a number of limitations. First, longitudinal studies are limited and offer mixed results, making claims about development difficult to interpret. Second, researchers have used a variety of methods for measuring coordination across studies, making it difficult to compare results. Furthermore, past research has often focused on either maternal or infant behavior in the study of vocal and gaze coordination, thereby overlooking the transactional nature of behavior within interactions. The present study aims to address these gaps in the literature through a longitudinal examination of mother and infant vocal and gaze coordination during toy play interactions across the first year.

Defining Coordination

For the purposes of this paper, “coordination” refers to the relative timing of two individuals’ behaviors in relation to one another during an interaction. Previous research has used a variety of terms to refer to this process (e.g. synchrony, contingency, co-regulation) and a variety of methods for measuring and analyzing it (e.g., Feldman, 2007; Van Egeren et al., 2001; Yu & Smith, 2013). These methods generally focus on three overarching types of coordinated relationships between events (Feldman, 2007): 1) Concurrent relationships, or matching states of behavior (e.g., simultaneous speech; mutual gaze); 2) Sequential relationships between the behavior of one partner and the next behavior of the other partner (e.g. contingent vocal responses, gaze following), including latencies to respond or follow; and 3) The ongoing lagged associations of two individuals’ streams of behavior (i.e. the overall degree to which one individual’s behavior predicts or matches their partner’s lagged behavior). These measures of coordination all capture the structured timing of behaviors between two individuals in an interaction, but do so in different ways. Prior studies have typically measured only one (or sometimes two) types of coordination in a single study, making it difficult to compare results across studies, and leaving open the question of how these different measures of coordination relate to one another. Examining relations among different measures will help answer important theoretical and methodological questions about parent-infant coordination. For example, do various measures of coordination (e.g. simultaneous speech, contingent responsiveness, ongoing lagged associations between mother and infant vocalizations) relate strongly to one another, indexing the extent to which coordination is a property of the dyad, or do different measures provide unique information?

Coordination of Vocalizations and Gaze in Infancy

Dyadic coordination of vocalizations.

Research suggests that vocalizations are the most common communicative signal and response in parent-infant interactions from as early as 4 months of age (Van Egeren et al., 2001), and like adults, infants and their mothers adjust the timing of their vocalizations to coordinate with those of their social partners (Jaffe et al., 2001; Stevenson, Ver Hoeve, Roach, & Leavitt, 1986). Research on the development of vocal coordination across the first year is somewhat mixed. For example, a review of cross-sectional studies suggests that while simultaneous speech (i.e. mother and infant vocalizing concurrently) is quite common when infants are 4 months old (Jaffe et al., 2001), by 9 months of age caregivers and infants display relatively little simultaneous speech in vocal interactions, indicating an emerging tendency to inhibit vocalizations when a partner is speaking (Jasnow & Feldstein, 1986). A recent longitudinal study by Hilbrink, Gattis, & Levinson (2015) examined 12 mother-infant dyads from 3 to 18 months and reported that infants reduced the amount of simultaneous speech (i.e. proportion of infant vocalizations that began during a mother vocalization) over this time period, while maternal simultaneous speech did not change with infant age.

Another frequently examined measure of vocal coordination is “response time”, a measure of sequential relationships between maternal and infant speech that focuses on the speed and consistency with which mothers respond to their infants’ vocalizations, and vice versa. Vocal responses that occur within a short time frame (typically within 2 seconds, see Van Egeren et al., 2001) of a partner’s vocalization are often called “contingent responses;” and research indicates that both parents and infants respond contingently to their partner’s vocalizations with vocalizations of their own significantly more often than would be expected by chance (Van Egeren et al., 2001). Research on longitudinal change in vocal response timing is limited, but suggests that mothers do not change with regard to the number or timing of responses to infant vocalizations across the first year (Hilbrink et al., 2015; Hsu & Fogel, 2003). The developmental picture for infant responses to maternal vocalizations is less clear. A review of cross-sectional work suggests that infants either reduce or show no change in the duration of latency to respond between 4 and 9 months (Beebe et al., 1988; Elias, Hayes, & Broerse, 1986; Gratier et al., 2015; Jasnow & Feldstein, 1986), but the longitudinal study by Hilbrink et al. (2015) reported that infant latency to respond increased (i.e. became longer) between 5 and 9 months of age.

Research examining the lagged sequential structure of vocal turn-taking, using cross-recurrence quantitative analysis and/or time series analysis, indicates that infant vocal behavior is significantly correlated with lagged parent vocal behavior and parent vocal behavior is significantly correlated with lagged infant vocal behavior, controlling for chance recurrence (Jaffe et al., 2001; Warlaumont et al., 2010). Importantly, the relationship between parent and infant vocal behavior bidirectional. Caregivers’ responsiveness to their infant depends at least in part on the quality and structure of infants’ behavior, and infants’ communicative behaviors develop in the context of these coordinated social interactions. For example, parents are less likely to respond contingently to infant vocalizations that contain only vowel sounds than to vocalizations containing a consonant (Gros-Louis, West, Goldstein, & King, 2006). Furthermore, mothers and infants tend to be similar in the speed and frequency with which they respond to one another, indicating co-regulated interactional processes from early in life (Jasnow & Feldstein, 1986; Van Egeren et al., 2001). These reciprocal processes highlight the importance of examining aspects of individual infant and maternal vocal behavior in relation to dyadic coordination.

Together, these findings support the idea that mothers and infants coordinate the timing of their vocalizations in a way that bolsters predictability and the perception of interpersonal contingencies. However, the limited number of longitudinal studies examining both parent and infant behavior has made it difficult to fully examine the nature and mechanisms of development in this area.

Dyadic coordination of gaze.

Over the first months of life infants rapidly develop the ability to orient their attention to and disengage from stimuli in their environment and begin to coordinate their looking behavior with social partners (Feldman, 2007). At around 3 months of age, infants engage in concurrent mutual gaze with their mothers (i.e. simultaneous attention to one another’s faces) approximately 30–50 percent of the time in face-to-face interactions (Harel, Gordon, Geva, & Feldman, 2011; Tronick, Als, & Brazelton, 1980). Longitudinal studies of gaze coordination reveal mixed results, possibly stemming from subtle methodological differences. Several studies report decreasing concurrent mutual gaze to face and increasing shared attention to objects over the course of the first year (Bakeman & Adamson, 1984; Kaye & Fogel, 1980). However, more recent research is not consistent with these findings. For example, de Barbaro, Johnson, Forster, & Deák (2016) found high levels of shared attention to objects at 4 months, with a reduction over the course of the first year, when infants were seated in an upright position (i.e. placed in a high chair) with objects in view. Other cross-sectional research reports that shared attention to objects may be quite high at young ages (Deák, Krasno, Triesch, Lewis, & Sepeta, 2014; Rossmanith, Costall, Reichelt, López, & Reddy, 2014). More longitudinal research examining mutual gaze to faces and shared attention to objects is needed in order to clarify how these forms of coordinated attention change over time.

In addition to understanding changes in the amount of shared attention mother-infant dyads achieve across time, it is important to examine how dyads achieve moments of gaze coordination. While there has been much research examining how infants follow an adult’s gaze to distal targets in the context of structured laboratory experiments (e.g., Corkum & Moore, 1998; Morales, Mundy, Delgado, Yale, Neal, et al., 2000), there has been relatively little research on how infants establish shared gaze with parents in the context of unstructured social interactions, a significantly more complex context than the structured tasks typically used in gaze following research.

An exception is a cross-sectional study by Deák et al. (2014), who examined gaze-following in unstructured toy play interactions between parents and 3- to 11-month-old infants. Infants of all ages spent only a small proportion of time looking at parents faces, and gaze following (i.e. looking to parent’s face and then following the parent’s gaze to a new location) was not a primary method by which infants achieved shared attention. Instead, concurrent shared attention most often occurred when infants followed their parents’ actions (e.g. holding or gesturing toward objects). Yu and Smith (2013) used lag-based cross-recurrence models to analyze coordination between parents’ and 12-month-old infants’ gaze and found that infants and mothers led and followed one another to shared attention equally and within a 5 second lag of their partner’s gaze. However, consistent with Deak et al., (2014) infants rarely looked to their mothers’ faces in these interactions, again providing evidence that eye gaze following is not a primary method through which infants establish shared attention with parents. Thus, while research has shown that infants have the ability to follow a partner’s gaze early in the first year (Corkum & Moore, 1998; Morales, Mundy, Delgado, Yale, Neal, et al., 2000), there appears to be a meaningful distinction between what infants can do and what they actually do in their everyday environments.

As with the literature on vocal coordination, gaze coordination is inherently a bidirectional process, with parent and infant behavior both playing a role in the construction of coordinated interaction. For example, Mendive, Bornstein, and Sebastián (2013) reported that moments of coordinated joint engagement were most likely to occur when mothers followed infants’ engagement with an object and less likely to occur following mothers’ redirection of infant attention. While research provides some evidence that following infants’ attention supports moments of shared attention, it has not considered how infant gaze behavior might influence parent gaze behavior. For example, infants who shift gaze more or less frequently may have parents who are more likely to redirect their looking, and may also be less likely to have lasting instances of shared attention. In other words, parent behavior could simply be a reflection of infant behavior. Just as in the literature on vocal coordination, research simultaneously examining parent and infant gaze behavior will improve understanding of the bidirectional nature of these processes.

The Present Study

The present study takes a microanalytic approach to examining parent and infant vocalization and gaze behavior during naturalistic play interactions longitudinally across the first year of life. The primary contribution of this research is the longitudinal examination of dyadic coordination in multiple modalities, allowing us not only the opportunity to examine change in coordination over time, but also to understand how mother and infant behaviors relate to one another and to coordination. Accordingly, we describe longitudinal change in coordination of vocalizations and gaze across the first year of life and address two primary questions.

How do different measures of coordination relate to one another?

Within domain (i.e. vocal, gaze), we examine several aspects of coordination and use multiple methods for quantifying coordination (explained in detail in the Methods section below). This multi-measure approach allowed us to test two hypotheses. First, we hypothesized that within the vocal and gaze domains, different measures of coordination would relate to one another, forming a coherent “coordination” scale. Second, we hypothesized that due to the bidirectional nature of dyadic coordination, measures of mother leading/infant following (i.e. mother produces behavior first, infant follows¹) would be related to measures of infant leading/mother following (i.e. infant produces behavior first, mother follows). In other words, dyads with more “responsive” mothers would also have more “responsive” infants (Jaffe et al., 2001; Van Egeren et al., 2001).

How do individual mother and infant vocal and gaze behaviors relate to the coordination of vocalizations and gaze?

Previous research suggests that coordination of vocalizations and gaze is supported by the individual behaviors of infants and mothers (Gros-Louis et al., 2006; Van Egeren et al., 2001). We tested two hypotheses regarding how individual mother and infant vocal and gaze behaviors relate to the coordination of vocalizations and gaze respectively. First, we predicted that dyads with infants who vocalize more and with more advanced speech-like vocalizations would be more coordinated with their mothers (Gros-Louis et al., 2006). Second, we hypothesized that the frequency and duration with which mothers and infants shifted their gaze would be related to gaze coordination, although we did not have a prediction about the direction of this relationship. On the one hand, longer gaze shifts may make it easier for a partner to coordinate his/her gaze. On the other hand, shorter and more frequent gaze shifts may be indicative of flexibility, which may allow for high levels of coordination.

Methods

Participants

Participants for this study were recruited from a larger longitudinal study of reaching, posture, object exploration, and language in infants. All infant participants in the larger study were full-term, from uncomplicated pregnancies and deliveries, free from known genetic syndromes, sensory impairments, and non-febrile seizures, and from English-speaking homes. Beginning in December 2013 (when the present study received IRB approval), all families entering the larger study were asked if they would like to participate in a study of parent-infant interaction. This continued until the larger study had completed recruitment of a grand total of 46 participants, a sample size chosen given that study’s larger aims. Out of a total of 35 participants enrolled after December 2013 in the larger study, 30 (85.7%) agreed to participate. This convenience sample of 30 infants is consistent with (or larger than) similar research examining dynamics in parent-infant interactions (Beebe et al., 1988, n = 15; de Barbaro, Johnson, Forster, & Deák, 2016, n = 26; Deák et al., 2014, n = 35; Jasnow & Feldstein, 1986, n = 29; Yu & Smith, 2013, n = 17), and is further strengthened by the large amounts of data available for each infant given the use of micro-analytic coding techniques and a longitudinal design.

The larger study included infants with and without an older sibling with autism spectrum disorder (ASD), thus the present study included 17 (12 male) infants with no first- or second-degree relatives with ASD and at least one neurotypically developing older sibling, and 13 infants (9 male) with an older sibling diagnosed with ASD. Although an initial aim of this study was to examine differences between dyads of families with and without an older child with ASD, there were virtually no significant differences between these groups, and they were therefore combined for the present study in order to examine development in a heterogenous group of infants. All subjects provided written informed consent as approved by the University of Pittsburgh Institutional Review Board (PRO13090529; ‘Parent-Infant interactions and the development of infants at risk for ASD’).

Table 1 displays demographic information for infant participants and their mothers. Infants were primarily Caucasian and non-Hispanic (83%) and most mothers were college educated (90%). The sample included 2 African-American, 1 Hispanic, and 3 mixed-race infants.

Table 1.

Demographic Information

	(N = 30)
Gender Male (%)	21	(70%)
Racial or ethnic minority (%)	5	(16.7%)
Mean age for mothers (SD)	33.37	(3.89)
Mean age for fathers (SD)	34.63	(6.07)
Maternal Education
Graduate of Professional School (%)	7	(41.2%)
Some College of College Degree (%)	10	(58.8%)
High School (%)	0	(0%)
Paternal Education
Graduate of Professional School (%)	15	(50%)
Some College of College Degree (%)	13	(43.3%)
High School (%)	2	(6.7%)

Note. SD = Standard Deviation

Procedure

Infants and mothers were observed in their homes playing with a standard set of toys (a rattle, stacking rings, a spherical puzzle, and a book) for 10 minutes at 3, 6, 9, and 12 months of age. Mothers were simply instructed to play face-to-face with their infants as they normally would. A boppy pillow was given to mothers to provide additional postural support to infants if desired, but mothers were free to position their infant however they wished throughout the interaction. Interactions were video recorded by two hand-held cameras, one focused on the infant and one focused on the mother, in order to capture the behavior of both members of the dyads. To enhance the quality of the audio component of the videos, infants wore a small wireless microphone clipped to a cloth vest worn over their clothing.

Data were available for 20 infant-mother dyads at 3 months (4 enrolled after 3 months, 1 cried inconsolably after 2 minutes, 5 missed visits), 25 dyads at 6 months (5 missed visits), 27 dyads at 9 months (3 missed visits), and 24 dyads at 12 months (2 began study before 12 month visit was added, 5 missed visits).

Coding

A 5-minute segment from each 10-minute observation was coded. The segment from 2:00 to 7:00 was chosen unless an interruption (e.g. parent asks experimenter a question, older sibling enters the room) occurred during this time period. If an interruption occurred, the uninterrupted segment closest to the middle of the observation was chosen. The segment closest to the middle of the 10-minute interaction was chosen as it allowed for a short warm up period but minimized fatigue. Additionally, obtaining an uninterrupted segment of observation was important due to our interest in analyzing parent-infant coordination.

Videotapes were coded by independent observers trained to criterion (achievement of at least 80% reliability on three consecutive clips). All behaviors were coded using a time-locked annotation program (ELAN; Brugman & Russel, 2004) to allow for detailed analyses of the relative timing of mother and infant communicative behaviors. For coding of gaze, video of mother and infant was synchronized and watched simultaneously side-by-side in order to enhance the ability of coders to accurately assess the object of each subject’s gaze. However, for coding of both gaze and vocalization, mother and infant behavior were coded at separate times.

Gaze.

Mother and infant gaze was coded when it was directed at either their partner’s face or a toy, and the location of gaze (i.e. to partner vs. to object) was identified. For each instance of gaze directed toward a toy, coders noted which specific object was being looked at. Periods during which mother and infant gaze was not coded were identified as either “undirected” (if mother or infant was looking at something other than their partner’s face or one of the toys), or “unclear” (if the angle or quality of the video made this moment impossible to code). “Unclear” codes were rare (M_mothers = 3.5 seconds, SD = 5.14; M_infants = 2.3 seconds, SD = 3.55). One 6-month mother gaze video could not be coded due to poor quality of the video; therefore mother gaze and mother-infant gaze coordination could not be calculated for this dyad (however, vocalization data and infant gaze data are included in analyses).

Vocalizations.

Vocalizations were coded in two passes completed by separate coders. The first pass involved identification of vocalizations. All voluntary sounds (e.g. raspberries, babbles, words, etc.) made by the infant or mother were coded (involuntary sounds such as sneezes, coughs, and hiccups were excluded). The second involved categorizing each infant vocalization as a word, linguistic vocalization (containing prespeech sounds), non-linguistic vocalization (e.g. growls, raspberries, squeals), or affective vocalization (e.g. fussing, laughing). Vocalizations that included both linguistic and non-linguistic or affective sounds were categorized as non-linguistic or affective respectively. Due to the low frequency of words, they were combined with linguistic vocalizations for purposes of analyses. Finally, linguistic vocalizations were coded for the presence of a consonant (e.g. [aaba], [gaa]).

Reliability.

Approximately 18–24% of videos were double coded for each behavior, balancing across age and infant risk status (N = 21/96 infant gaze; N = 18/95 mother gaze; N = 19/96 infant vocalization identification; N = 21/29 mother vocalization identification; N = 23/96 infant vocalization type). For gaze and vocalization identification, reliability was calculated on frame-by-frame coding based on the nature of the data reduction and analysis used in this project (see below). Average Cohen’s kappa values were 0.75 for infant gaze, 0.73 for mother gaze, 0.89 for infant vocalizations, and 0.91 for mother vocalizations. Average kappa values for categorizing infant vocalizations were .70 for infant vocalization type (linguistic, non-linguistic, affective) and .83 for presence of a consonant.

Data Reduction and Analysis

Coordination Variables.

Table 2 contains definitions and visual depictions for each of the cross-recurrence and event-based measures of vocal and gaze coordination.

Table 2.

Descriptions of Measures of Coordination

Domain	Variable	Definition
Vocal Coordination
Cross-Recurrence Measures	Total Simultaneous Recurrence	Recurrence rate at lag 0; the total proportion of time mothers and infants spent vocalizing at the same moment
	Maximum recurrence within 5 second lag	Maximal recurrence rate to the right/left of 0, the maximum lagged match for Partner A’s vocalizations following Partner B’s vocalizations.
Event-based Measures	Frequency Contingency Response	Frequency Person B’s Vocalization follows offset of Person A’s within 2 seconds.
	Mean Latency to Respond	Mean latency between offset of Person A’s vocalization and onset of Person B’s vocalization
	Frequency Simultaneous Speech	Frequency Person B’s vocalization begins during Person A’s Vocalization
Gaze Coordination
Cross-Recurrence Measures	Total Simultaneous Recurrence	Recurrence rate at lag 0; the total proportion of time mothers and infants spent in simultaneous attention
	Mean recurrence within 5 second lag	Mean recurrence rate to the right/left of 0; the mean of all lagged matches for person A looking where person B was looking previously at varying degrees of lag up to 5 seconds.
Event-based Measures	Frequency follow to simultaneous attention	Frequency Person B looks to location Person A is looking (while Person A is still looking there).
	Mean latency follow to simultaneous attention	Mean latency between when Person A looks to a new location and Person B looks to that same location (while Person A is still looking there).

Cross-recurrence measures of coordination.

Cross-recurrence measures of coordination were used to examine concurrent and lagged temporal relationships (up to 5 seconds of lag) between mother and infant vocalizations and between mother and infant gaze (e.g. Warlaumont et al., 2010; Yu & Smith, 2013). In this type of analysis, two categorical temporal data streams are aligned, and their temporal coordination is measured with varying degrees of lag (see Supplemental Materials for details).

Vocalizations.

Cross-recurrence profiles for vocalizations indicate the degree to which mothers’ and infants’ vocalizations occurred simultaneously and at varying degrees of lag. For cross-recurrence measures examining lagged coordination (maximum recurrence within 5 second lag), we removed instances of simultaneous speech and fixed the duration of all vocalizations to 1 second (while maintaining the exact lag times between vocalizations). This process has the effect of controlling for the duration of vocalizations so that the lagged recurrence measures better reflect the timing between offsets and onsets of infant and adult vocalizations (see Abney, Warlaumont, Oller, Wallot & Kello, 2017). All vocal coordination measures (including event-based coordination measures described below) were calculated using only non-affective vocalizations. Cross-recurrence measures of vocal coordination included: 1) Total simultaneous recurrence: a measure of the total amount of time mothers and infants spent vocalizing at the same moment; and 2) Max recurrence of infant leading/mother following and mother leading/infant following: A measure of the maximum lagged match for mother vocalizations occurring after infant vocalizations and infant vocalizations occurring after mother vocalizations.

Gaze.

Cross-recurrence profiles for gaze indicate the degree to which mother and infants attended to the same place at the same time and on varying degrees of lag. These measures were calculated separately for gaze to objects and gaze to faces. Cross-recurrence measures of gaze coordination included: 1) Total simultaneous recurrence: a measure of the total amount of time infants spend in simultaneous attention; and 2) Mean recurrence within 5-second lag for infant leading/mother following and mother leading/infant following: measure of the degree to which the location mother is looking matched the location where infant was looking previously (and vice versa) at varying degrees of lag up to 5 seconds.

As can be seen in the example recurrence plots for vocalizations and gaze depicted in Table 2 (and in the actual mean recurrence plots depicted in Figures 1, 2, and 3), recurrence shows a very different structure for vocalizations and gaze. For vocalizations, simultaneous recurrence is low (i.e. mothers and infants do not often vocalize at the same time) and lagged recurrence (mother leading/infant following and infant leading/mother following) increases and peaks sharply. For this reason, looking at the maximum recurrence provides the best picture of how coordinated mothers and infants are at a lag. For gaze coordination, on the other hand, recurrence peaks at lag 0 (i.e. mothers and infants are most recurrent simultaneously) and slowly declines over increasing degrees of lag. Thus, looking at the mean (rather than the max) of recurrence within a 5 second lag provides a better overall picture of lagged recurrence for gaze.

Figure 1.

Adjusted mean recurrence plots for lagged vocal coordination at 3, 6, 9, and 12 months. Shaded area represents 95% Confidence Interval. Points to the left of zero indicate infant leading/mother following and points to the right of zero indicate mother leading/infant following. Zero on the vertical axis represents chance recurrence.

Figure 2.

Adjusted mean recurrence plots for mutual gaze coordination at 3, 6, 9, and 12 months. Shaded area represents 95% Confidence Interval. Points to the left of zero indicate infant leading/mother following and points to the right of zero indicate mother leading/infant following. Zero on the vertical axis represents chance recurrence.

Figure 3.

Adjusted mean recurrence plots for object gaze coordination at 3, 6, 9, and 12 months. Shaded area represents 95% Confidence Interval. Points to the left of zero indicate infant leading/mother following and points to the right of zero indicate mother leading/infant following. Zero on the vertical axis represents chance recurrence.

Event-based measures of coordination.

While cross-recurrence measures provide a broad view of the match between mother and infant both simultaneously and at varying lags, event-based measures provide a clearer sense of the specific leader/follower dynamics in simultaneous and lagged coordination. These variables were calculated separately for simultaneous attention to objects and for simultaneous attention to faces.

Vocalizations.

We examined the following event-based vocal coordination variables: 1) Frequency Contingent Responses: instances of mother and infant vocalizations followed by a partner’s non-overlapping vocalization within 2 seconds of the offset of the original vocalization. 2) Average Latency to Respond: the duration of all pauses between the offset of one individual’s vocalizations and the onset of their partner’s vocalization were calculated and averaged. 3) Frequency Simultaneous Speech: instances of mother and infant vocalizations that are “interrupted” by a partner’s vocalization were counted. This variable is attributed to the individual who speaks second (for example, frequency mother simultaneous speech refers to the number of times the mother begins speaking during an infant’s vocalization).

Gaze.

With regards to gaze, we examined the Frequency and Mean Latency of Follows to Simultaneous Attention: instances when one individual shifted his/her gaze to a location and his/her partner subsequently looked to that same location (thus beginning a moment of simultaneous attention) were counted, and the duration of the lags between the initial person’s gaze shift and the beginning of simultaneous attention were calculated and averaged.

Adjusting for chance occurrence.

To determine whether measures of coordination exceeded chance, randomized baselines were created by randomly shuffling individual mother and infant event profiles 500 times (i.e. randomizing the timing of events while maintaining durations), calculating each dyadic variable from the randomly shuffled profiles of each mother-infant dyad, and taking the average of these randomly generated variables. Randomized baselines were then subtracted from the true coordination variables in order to adjust for chance. This method controls for the baseline frequency and duration of a behavior’s occurrence, thus controlling for “opportunity” for coordination. If this adjusted coordination is significantly different from zero, we can conclude coordination has occurred at levels greater than would be expected by chance (see Supplemental Materials for further details). All variables reported below are “adjusted” for chance in this way. Thus, variables should be interpreted as the degree to which coordination is occurring above and beyond what would be expected by chance (represented by zero).

Data Analytic Plan.

Hierarchical linear modeling (HLM; Raudenbush & Bryk, 2002) was used to create and compare growth trajectories for measures of coordination. For each variable, a multi-step process was used to determine the best and most parsimonious model for the data. This process began with fitting a fully unconditional random intercept model (without predictors at Level 1 or 2). In order to determine the most appropriate model of individual change, we examined change in model fit from the means only model to a linear model, and subsequently to a quadratic model. Chi-square tests of deviance were calculated to determine whether the linear or quadratic model lead to a significant reduction in deviance compared to the previous model (i.e., was a better fit for the data). Higher order growth models were retained only if they significantly reduced the deviance (i.e. improved the fit) of the model and the growth term was significantly greater than zero. A random effect was included on the growth term if the variance was significantly greater than zero. HLM can accommodate missing data, thus all available data can be used without the need for listwise deletion (Huttenlocher, Haight, Bryk, Seltzer, & Lyons, 1991; Willett, Singer, & Martin, 1998). Data were analyzed using Version 7.03 of HLM for Windows (Raudenbush, Bryk, Cheong, Congdon, & Du Toit, 2011).

Assumptions of normality and homoscedasticity were assessed through examination of residuals. All variables met the assumption of homoscedasticity. In the case of modest violations of normality assumptions for some variables, robust standard errors (which enable computation of sensible confidence intervals and tests even when residuals are not normally distributed; Raudenbush and Bryk 2002) are reported (indicated with an ‘RSE’ subscript in tables below).

To analyze how different measures of coordination within each domain (i.e., vocal coordination, mutual gaze, object gaze) related to one another, we first ran repeated measures correlations (Bakdash & Marusich, 2017) to assess the overall intra-individual association between measures. Repeated measures correlation is a statistical technique for examining associations between variables assessed on two or more occasions without violating independence assumptions. Our goal with these analyses was to provide an overall description of how different types of measures related to one another and how measures of mother leading/infant following related to measures of infant leading/mother following. Repeated measures correlations were run using the “rmcorr” package in R (Bakdash & Marusich, 2017). Second, in order to assess how closely related the various coordination variables were as a group, Cronbach’s alphas were calculated to assess internal consistency at each age. Cronbach’s alphas above .7 are generally considered to represent adequate internal consistency (Nunnally, 1978).

Results

The overarching goal of this study was to describe the development of the dyadic coordination of vocalization and gaze between mothers and infants across the first year of life. Analyses relevant to coordination of vocalizations and gaze are presented in turn, with coordination of gaze to faces (mutual gaze) and toys examined separately. For each coordination modality (vocalizations, mutual gaze, gaze to toys), analyses: a) describe development of concurrent and sequential coordination; b) examine relations among measures of coordination; and c) examine relations between coordination and individual mother and infant behaviors.

Vocal Coordination

Table 3 displays the results of the final HLM models for measures of vocal coordination (for descriptive statistics of vocal coordination measures at each age, see Supplemental Material, Table 2). Figure 1 displays the chance-adjusted mean recurrence plots for mother and infant lagged vocal coordination at 3, 6, 9, and 12 months.

Table 3.

HLM Models for Vocal Coordination Measures

		β	SE	t (df)	p-value
Simultaneous Recurrence r	Intercept, β00	–0.003	0.001	–4.27(29)	<0.001
Caregiver Frequency Simultaneous Speech e	Intercept, β00	–0.76	0.30 RSE	–2.5 (29)	0.016
Infant Frequency Simultaneous Speech e	Intercept, β00	–1.58	0.31 RSE	–5.12(29)	<0.001
Infant Leading, Mother Following
Maximum Recurrence r	Intercept, β00	0.01	0.001 RSE	15.41 (29)	<0.001
Frequency Contingent Response e	Intercept, β00	1.75	0.25	6.98 (29)	<0.001
Latency to Respond e	Intercept, β00	1.75	0.19 RSE	9.43 (29)	<0.001
Mother Leading, Infant Following
Maximum Recurrence r	Intercept, β00	0.01	0.001 RSE	10.22 (29)	<0.001
Frequency Contingent Response e	Intercept, β00	1.23	0.26	4.81 (29)	<0.001
Latency to Respond e	Intercept, β00	2.44	0.23 RSE	10.48 (29)	<0.001

Note. SE = Standard Error;

^rsubscript indicates cross-recurrence measures,

^esubscript indicates event-based measures.

RSE = Robust Standard Errors.

Simultaneous speech.

Three measures of simultaneous speech were analyzed: total simultaneous recurrence of mother and infant vocalizations (recurrence at lag zero), frequency infant simultaneous speech (i.e. infant vocalizes during mother vocalization), and frequency mother simultaneous speech (i.e. mother vocalizes during infant vocalization). HLM models for these three variables revealed no significant growth terms, and thus unconditional means models were used. The unconditional means model reports the grand mean across ages for the variable in question. The significance of the fixed effect on the intercept (β₀₀) in these models indicates whether coordination is significantly different from zero (i.e. significantly different than would be expected by chance). As can be seen in Table 3, rates were significantly lower than would be expected by chance for all three measures of simultaneous speech, indicating that both mothers and infants inhibited speech when their partners vocalized.

Infant leading/mother following.

With regard to mother vocalizations following infant vocalizations, we analyzed the maximum recurrence rate (maximum recurrence rate for mother vocalizations following infant vocalizations), frequency of contingent vocal responses, and average duration of latency to respond (see Table 3). For all three measures of infant leading/mother following, HLM models revealed no significant change with infant age. Maximum recurrence for infants leading/mothers following was significantly greater than would be expected by chance and mothers responded to their infants’ vocalizations contingently (i.e. within 2 seconds) more often than would be expected by chance. On average, mothers’ latency to respond to an infant vocalization was about 1.75 seconds.

Mother leading/infant following.

Analysis of measures of infant vocal behavior following mother vocal behavior also revealed no change across age. Maximum recurrence for infants following mothers was significantly different from zero and similar to the recurrence rate for mothers following infants. Infants also responded to their mothers contingently more than would be expected by chance and at a similar rate as mothers. Average duration of infants’ latency to respond was slightly longer than that for mothers, averaging around 2.44 seconds.

Relations among vocal coordination variables.

Table 4 displays the repeated measures correlations among measures of vocal coordination. Consistent with our hypothesis that various measures of vocal coordination would be related to one another, the three measures of infant leading/mother following all related significantly to one another, as did the three measures of mother leading/infant following. Measures of simultaneous speech were also variably associated with these measures of lagged coordination. In addition, measures of infant leading/mother following were related to respective measures of mother leading/infant following (although this was marginal in the case of frequency of contingent response, p = .068), providing support for our hypothesis that mothers who had more and faster vocal responses would have infants with more and faster vocal responses. Cronbach’s alphas for standardized vocal coordination variables (with scores reversed for measures of simultaneous speech and latency to respond) were close to or exceeded .7 at every age (3 months α = .73; 6 months: α = .78; 9 months α = .84; 12 months: α = .69), indicating that these measures of vocal coordination generally index the same underlying construct.

Table 4.

Repeated Measures Correlations between Vocal Coordination Variables

Variable	1	2	3	4	5	6	7	8	9
1. Simultaneous Recurrence r	-	.50***	.63***	−0.23	−.23	0.09	−0.07	−.31*	.25*
2. Caregiver Adjusted Frequency Simultaneous Speech e		-	0.14	−0.04	−.46***	0.21	−.240*	−0.06	0.22
3. Infant Adjusted Freq. Simultaneous Speech e			-	−.21	−0.18	0.08	−0.05	−.41**	.23
4. Max Recurrence Infant Leading, Mother Following r				-	.42***	−.33**	.41**	.51***	−.24**
5. Freq Mother Contingent Responses e					-	−0.39**	0.12	0.22	−0.19
6. Mother Latency to Respond e						-	−0.32**	−0.25*	0.29*
7. Max Recurrence Mother Leading, Infant Following r							-	0.43***	−0.49***
8. Freq Infant Contingent Response e								-	−0.47***
9. Infant Latency to Respond e									-

Notes.

^*<.05

^**<.01

^***<.001;

^rsubscript indicates cross-recurrence measures,

^esubscript indicates event-based measures.

Relations between vocal coordination composite and individual behaviors.

Vocal coordination composite variables were created by averaging together standardized scores for all vocal coordination variables. We examined bivariate correlations between the vocal coordination composite and measures of individual mother and infant vocal behavior (see Supplemental Material, Table 1 for descriptive statistics) at each age. At every age, higher vocal coordination composite scores were related to higher frequency of infant vocalizations (3 months: r = .57, p = .009 ; 6 months: r = .39, p = .049; 9 months: r = .46, p = .016; 12 months: r = .49, p = .018 ) and higher frequency of mother vocalizations (3 months: r = .63, p = .003; 6 months: r = .67, p <.001; 9 months: r = .79, p <.001 ; 12 months: r = .65, p = .001; note that this relationship occurs despite the fact that all coordination measures control for chance recurrence). Additionally, at 6 and 9 months, infants with higher proportions of consonants also had higher vocal coordination composite scores (6 months: r = .66, p < .001; 9 months: r = .66, p < .001 ).

Gaze Coordination

Figures 2 and 3 display the mean recurrence plots for mutual gaze coordination and coordination of gaze to objects respectively at 3, 6, 9, and 12 months. Table 5 displays the final HLM models for measures of mutual gaze coordination (i.e., infant and mother looking at one another’s faces) and object gaze coordination (descriptive statistic for these measures at each age are presented in Supplemental Material, Table 3).

Table 5.

HLM Models for Mutual Gaze Coordination and Object Gaze Coordination

		Mutual Gaze Coordination				Object Gaze Coordination
		β	SE	t (df)	p	β	SE	t (df)	p
Simultaneous Recurrence r	Intercept, β00	0.011	0.003 RSE	3.90 (29)	<0.001	0.13	0.01	8.92 (29)	<0.001
	Linear Growth, β10	0.002	0.001 RSE	2.52 (63)	0.014	0.01	0.003	4.54 (29)	<0.001
Infant Leading, Mother Following
Mean Recurrence within 5 seconds r	Intercept, β00	0.01	0.002 RSE	5.65 (29)	<0.001	0.11	0.01	8.06 (29)	<0.001
	Linear Growth, β10	-	-	-	-	0.01	0.002	3.33 (29)	0.002
Frequency Follows to JA e	Intercept, β00	0.789	0.363	2.17 (29)	0.038	3.61	0.82	4.43 (29)	<0.001
	Linear Growth, β10	−0.411	0.176	−2.33 (29)	0.027	2.18	0.5	4.36 (29)	<0.001
	Quadratic Growth, β20	0.035	0.017	2.08 (29)	0.047	−0.14	0.05	−2.67 (29)	0.012
Latency Follows to JA e	Intercept, β00	-	-	-	-	1.63	0.18	9.19 (29)	<0.001
	Linear Growth, β10	-	-	-	-	−0.07	0.02	−2.9 (29)	0.007
Mother Leading, Infant Following
Mean Recurrence within 5 seconds r	Intercept, β00	0.004	0.002 RSE	1.80 (29)	0.083	0.11	0.01	7.8 (29)	<0.001
	Linear Growth, β10	0.001	0.001 RSE	2.35 (63)	0.022	0.01	0.002	3.28 (29)	0.003
Frequency Follows to JA e	Intercept, β00	1.163	0.324	3.59 (29)	0.001	4.16	0.74	5.66 (29)	<0.001
	Linear Growth, β10	0.157	0.068	2.30 (29)	0.029	0.71	0.15	4.84 (29)	<0.001
Latency Follows to JA e	Intercept, β00	2.732	0.270 RSE	10.10 (29)	<0.001	0.97	0.05 RSE	19.46 (29)	<0.001
	Linear Growth, β10	−0.138	0.048 RSE	−2.91 (29)	0.007	-	-	-	-

Note. SE = Standard Error;

^rsubscript indicates cross-recurrence measures,

^esubscript indicates event-based measures.

RSE = Robust Standard Errors.

Mutual gaze coordination.

Simultaneous Gaze.

Recurrence of infants’ and mother’s gaze to partner’s face (i.e. proportion of time spent in mutual gaze) was significantly greater than chance at 3 months and increased over time, showing significant positive linear growth (see Table 5 and Figure 2). Importantly, this positive growth in amount of mutual gaze adjusting for chance recurrence occurred despite an overall decrease in the raw amount of time mothers and infants spent in mutual gaze (3 months: M = .11, SD = .14; 6 months: M = .04, SD = .04; 9 months: M = .04, SD = .04; 12 months: M = .05, SD =.08).

Infant leading/mother following.

Mean recurrence of infant leading/mother following within 5 seconds of mutual gaze (a measure of the degree to which the location mother looked matched the location where infant was looking previously at varying degrees of lag up to 5 seconds) was significantly greater than would be expected by chance and did not show significant change over time.

In contrast, the average adjusted frequency with which mother followed infant to mutual gaze (i.e. infant looks to mother’s face and then mother looks to infant’s face) showed a quadratic model of growth. At 3 months, mother following infant to mutual gaze was infrequent, but significantly more common than would be expected by chance, with significant negative instantaneous linear growth and an accelerating pattern over time. This indicates a U-shaped pattern whereby mothers were most likely to follow their infants’ gaze to face at 3 months, reduced sharply between 3 and 6 months, and then increased somewhat between 9 and 12 months. The frequency with which mothers followed infants’ gaze to face was not significantly different from chance at any age other than 3 months. This is likely because the frequency measure requires that mothers were not looking to their infant’s face at the moment the infant shifted gaze to mother’s face (while the recurrence measure does not make this distinction), and thus better captures how moments of mutual gaze begin. Given how much mothers looked at infants’ faces and how little infants looked to mothers’ faces overall (see Supplemental Materials, Table 1), it was rare for infants to look to their mothers’ faces at a time when mothers were not already looking at them. In fact, 24% of dyads at 3 months, 52% of dyads at 6 months, 33.3% of dyads at 9 months, and 45.5% of dyads at 12 months had zero instances of mother following infant to mutual gaze. Given how infrequently mothers followed infants to mutual gaze, we did not analyze latencies for this variable.

Mother leading/infant following.

Mean recurrence for mother leading/infant following mother was not significantly different from zero at 3 months but showed significant positive linear growth over time and was significantly greater than chance by 6 months (p <.001). Frequency of infant following mother to mutual gaze (i.e. mother looks at infant’s face and then infant looks to mother’s face to achieve mutual gaze) was much more common than mothers following infants. Infants followed their mother to mutual gaze more than would be expected by chance at 3 months and showed positive linear growth in following over time. Infants also became faster at following their mother’s gaze to mutual gaze over time. At 3 months, infants’ latency to follow was approximately 2.73 seconds on average, and they showed significant negative linear growth over time such that by 12 months infants were following their mothers’ gaze in 1.48 seconds on average.

Relations among mutual gaze coordination variables.

Table 6 displays repeated measures correlation coefficients for the relations among mutual gaze coordination variables. Overall, our hypothesis that the various measures of mutual gaze coordination would relate to one another was generally supported for measures of mother leading/infant following. However, contrary to expectation, the two measures of infant leading/mother following were not significantly related to one another. Furthermore, frequency of mother following infant’s gaze was negatively correlated with frequency of infant following mother’s gaze. These unexpected findings are likely indicative of the fact (discussed above) that mothers following infant gaze to mutual gaze is a rare event.

Table 6.

Repeated Measures Correlations between Mutual Gaze and Object Gaze Coordination Variables

Mutual Gaze Coordination Variables	1	2	3	4	5	6
1. Simultaneous Recurrence r	-	0.88***	−0.15	0.89***	0.44***	−0.23
2. Mean Recurrence Infant Leading, Mother Followingr		-	−0.11	0.95***	0.20	−0.08
3. Freq Mother Follows Infant e			-	−0.23	−0.33**	−0.05
4. Mean Recurrence Mother Leading, Infant Following r				-	0.24*	−0.07
5. Freq Infant Follows Mother e					-	−0.34**
6. Infant Latency to Follow e						-
Object Gaze Coordination Variables	1	2	3	4	5	6	7
1. Simultaneous Recurrence r	-	0.96***	0.35***	−0.32**	0.96***	0.48***	−0.07
2. Mean Recurrence Infant Leading, Mother Following r	-	0.16	−0.29*	0.98***	0.33**	−0.02
3. Freq Mother Follows Infant e			-	−0.37***	0.14	0.36***	−0.09
4. Mother Latency to Follow e				-	−0.26*	−0.19	0.05
5. Mean Recurrence Mother Leading, Infant Following r				-	0.36***	0.01
6. Freq Infant Follows Mother e						-	−0.12
7. Infant Latency to Follow e							-

Note.

^*<.05

^**<.01

^***<.001;

^rsubscript indicates cross-recurrence measures,

^esubscript indicates event-based measures

Given the non-significant and negative relations between frequency of mother following infant’s gaze to mutual gaze and other coordination variables, we excluded this variable from estimates of Cronbach’s alpha. With all other variables included, Cronbach’s alpha exceeded .7 at all ages (3 months: α = .71; 6 months: α = .76; 9 months α = .75; 12 months: α = .71), indicating good internal consistency among various measures of mutual gaze coordination.

Relations between mutual gaze coordination composite and individual behaviors.

A mutual gaze coordination composite variable was created for each age by standardizing and averaging all coordination variables (except frequency mother follows infant). We hypothesized that mutual gaze coordination would be related to the frequency and duration of mother and infant gaze shifts. Bivariate correlations between the mutual gaze coordination composite at each age and individual measures of gaze behavior (i.e. mother and infant frequency and mean duration of gaze shifts; see Supplemental Material, Table 1 for descriptive statistics) revealed that at 6 and 9 months, dyads with infants who shifted their gaze more frequently (6 months: r = .55, p = .004; 9 months: r = .54, p = .004) and had shorter gaze durations (although only marginal at 6 months; 6 months: r = −.34, p = .09; 9 months: r = −.52, p = .005) had higher mutual gaze coordination. Frequency and duration of mothers’ gaze shifts were not related to coordination.

Object gaze coordination.

Simultaneous gaze.

Mothers and infants engaged in simultaneous attention to objects significantly more than would be expected by chance starting at 3 months of age, and recurrence increased over time. As can be seen in Table 5 and Figure 3, dyads spent approximately 13% of the time in simultaneous attention to objects (controlling for chance recurrence) at 3 months, and increased steadily over time. The proportion of time mothers and infants spent attending to the same toy at the same moment was significantly greater than chance at 3 months and showed significant linear growth over time (see Table 5).

Infant leading/mother following.

Mean recurrence of infant leading/mother following within 5 seconds of lag zero was significantly greater than chance at 3 months and showed significant linear growth over time. The frequency with which mothers followed their infants’ gaze to simultaneous object attention was significantly greater than chance at 3 months and showed a pattern of positive, but decelerating quadratic growth over time, such that mothers’ following increased more between 3 and 6 months than between 9 and 12 months. Mothers’ latency to follow their infants’ gaze also became shorter over time. Mothers followed their infants’ gaze in approximately 1.6 seconds on average at 3 months, and gaze following times decreased linearly by about .07 seconds per month, such that by 12 months, mothers were following their infants’ gaze in a little less than a second on average.

Mother leading/infant following.

Recurrence of mother leading/infant following within a 5 second lag showed a similar pattern. Mean recurrence was significantly greater than chance at 3 months and showed significant linear growth over time. Infants also followed their mothers’ gaze to simultaneous object gaze more than would be expected by chance starting at 3 months and showed significant positive linear growth in this variable. Finally, infants followed their mothers’ gaze in approximately 1 second on average and did not show significant change in latency to follow mother’s gaze over time.

Relations among object gaze coordination variables.

Table 6 displays the bivariate repeated measures correlations among measures of gaze coordination to objects. As can be seen in the table, various measures of object gaze coordination were generally related to one another. Furthermore, as hypothesized, mothers with higher recurrence and more follows had infants with higher recurrence and more follows. Notably, however, infant latency to follow mother was not related to any coordination variable.

Given the lack of correlation between infant latency to follow and all other variables, this variable was excluded from analyses of internal consistency. Examination of Cronbach’s alpha with remaining variables revealed adequate internal consistency at 3, 9, and 12 months, but somewhat lower internal consistency at 6 months: 3 months: α = .70; 6 months: α = .62; 9 months α = .68; 12 months: α = .72. At 6 months, examination of item statistics revealed that the further exclusion of frequency caregiver follows infant improved the Cronbach’s alpha to .794, suggesting that, at this age, it may not contribute to overall coordination.

Relations between Gaze Coordination Composite and Individual Behaviors.

Object gaze coordination composite scores for each month were created by standardizing and averaging coordination variables (except latency for infant to follow mother) at each age. Bivariate correlations between the object gaze coordination composite and measures of individual mother and infant gaze behavior (i.e. frequency and duration of gaze shifts; see Supplemental Material, Table 1 for descriptive statistics) revealed that at 3 months mothers with more (r = .68, p = .001) and shorter (r = −.51, p = .02) gaze shifts had higher gaze coordination with their infants. These relations were not apparent at later ages and object gaze coordination was not significantly related to infant frequency or duration of gaze shifts at any age.

Discussion

The overarching goal of this study was to describe the development of the dyadic coordination of vocalization and gaze behavior between mothers and infants across the first year of life. In addition, the study contributes to our understanding of dyadic coordination by investigating how different measures of coordination relate to one another and how individual behaviors relate to dyadic coordination within the vocal and gaze domains. Findings reveal that dyadic coordination is a very early emerging phenomenon that is supported by both mothers and infants and highlight the importance of considering bidirectional and developmental influences on parent and infant behavior.

The Development of Vocal Coordination in Mother-Infant Dyads

Dyadic coordination of vocalizations was apparent from the time infants were 3 months old and showed little developmental change over time. Mothers and infants adjusted the timing of their vocalizations to coordinate with one another by inhibiting vocalizations during partner’s speech, vocalizing more than would be expected by chance in the period immediately following a partner’s vocalizations, and responding promptly to partner’s vocalizations. Previous research has shown that mothers and infants coordinate their vocalizations with one another by the time infants are 4 months old (Beebe et al., 1988; Jaffe et al., 2001), and this study suggests an even earlier onset of coordination. Prior work on longitudinal change in vocal turn-taking has been largely cross-sectional or based on small sample sizes, and findings have been mixed (Gratier et al., 2015; Hilbrink et al., 2015; Jaffe et al., 2001; Jasnow & Feldstein, 1986). The present longitudinal study indicated that in the context of toy play interactions, the degree to which mothers and infants coordinated the timing of their vocalizations did not show significant change across the first year.

At first glance, the “lack” of developmental change may be surprising, but this result must be interpreted in the context of development in other areas. At 3 months interactions have a large face-to-face component and fewer stimuli competing for attention, thus both mothers and infants may devote more of their attention to their partners’ vocal behaviors rather than to other aspects of the environment. As interactions become more triadic, vocalization timing is likely driven by more than just the timing of a partner’s vocalizations, but also by the infant’s emerging skills and by other features of the interaction and environment. This possibility is supported by the high duration of gaze to partner’s face at 3 months, followed by a reduction in gaze to face and increase in gaze to objects as infants got older (see Supplemental Material, Table 1).

The Development of Gaze Coordination in Mother-Infant Dyads

Longitudinal examination of mother and infant gaze behavior in a naturalistic toy play interaction revealed that mothers and infants achieved greater than chance levels of simultaneous attention to both objects and faces beginning as early as 3 months of age, and that the degree to which dyads coordinated attention increased significantly across the first year.

Examination of the development of mutual gaze coordination revealed several new and intriguing findings. Although the raw amount of time mothers and infants spent in mutual gaze reduced dramatically across the first year, the degree to which those moments occurred above chance (i.e. given baseline levels of mother and infant face looking) increased over time. This is an important addition to our understanding of mutual gaze in a developmental context. Although mothers and infants reduce the amount of time they spend looking to each other’s faces across the first year, they become more likely (relative to chance) to time those looks with one another, suggesting that moments of mutual gaze remain a meaningful part of these social experiences.

While mothers and infants led and followed moments of simultaneous attention to objects to similar degrees, the picture was different for mutual gaze coordination. Mothers rarely followed infants into mutual gaze. This is likely due to the fact that mothers spend much more time looking to infants’ faces than infants spend looking to mothers’ faces (see Supplemental Material, Table 1). Mothers likely use gaze to infants’ faces for a variety of purposes, including assessing infants’ affective states and following infants’ gaze shifts to establish shared attention to objects and one another. Mothers’ relative height gives them opportunities to respond to changes in infants’ head and body movements and therefore anticipate shifts in infant gaze. Thus, nearly every time infants looked to their mothers’ face, mothers were already looking at them, making it unlikely that an infant would lead in the onset of a moment of mutual gaze. Consistent with findings from Yu and Smith (2013) and Deak et al. (2014), this study provides further evidence that in naturalistic interactions infants seldom use looks to their mother’s face as a method of gaze following. Instead, moments of coordinated gaze to face likely serve other purposes, such as checking for reactions or sharing affective responses (Mundy et al., 2007; Yale, Messinger, Cobo-Lewis, & Delgado, 2003).

Relations among Coordination Variables

In addition to describing the development of vocal and gaze coordination in the first year of life, we used a multi-measure approach to defining and describing coordination, allowing us to examine how different measures of coordination related to one another. Consistent with our hypothesis and with previous research demonstrating that mothers with greater and faster responsiveness tend to have infants with greater and faster responsiveness (Jaffe et al., 2001; Van Egeren et al., 2001), we found that for vocal coordination and object gaze coordination, measures of infant leading/mother following were generally related to measures of mother leading/infant following. These relations suggest that understanding coordinated interactions requires consideration of both mother and infant behavior. The fact that, within dyads, mothers and infants appeared to match in the degree and speed with which they respond suggests that both members are attuned to their partner’s patterns of behavior and are also timing their own behavior in predictable ways. This is consistent with the view that coordinated interactions are indicative of the creation of structure and predictability in the dyad.

We further hypothesized that overall, different measures of coordination would relate to one another, forming a coherent coordination “scale” and suggesting that coordination is a feature of the dyad. This hypothesis was largely supported for measures of vocal coordination, and moderately supported for measures of mutual gaze and object gaze coordination. For mutual gaze coordination, measures of overall recurrence and measures of coordination for mother leading/infant following were generally inter-related. However, the frequency with which mothers followed infants to mutual gaze was not related to any recurrence measures, likely due to the infrequency with which infants looked to mother’s face when the mother was not already looking at the infant’s face. Results for object gaze coordination suggested overall internal consistency between measures, with the exception of infant latency to follow mother, which was not significantly related to any other variable. Although infants increased the speed with which they followed their mothers’ gaze to simultaneous attention over the course of the first year, this may be a less important component of overall coordination.

This was the first study to our knowledge to examine cross-recurrence and event-based measures of coordination simultaneously, allowing us the opportunity to comment on the relative contributions of these measures to our understanding of coordination. For both mutual gaze and object gaze coordination, correlations between recurrence at lag 0 and lagged recurrence measures were so high (range .88-.99) as to suggest that they may not be measuring distinct aspects of coordination. Unlike event-based measures of leading and following dynamics, lagged recurrence measures do not consider who looked at a location first, nor do they require the dyad to actually achieved simultaneous attention. Instead, lagged recurrence measures calculate how often one partner is looking to the location the other partner was looking X lags ago. Given the extremely high correlations between simultaneous recurrence and lagged recurrence as well as the lack of specificity with which lagged recurrence measures were calculated, we believe that the recurrence measures included here are less ideal than event-based measures for understanding leader and follower dynamics in gaze coordination. They are, however, an excellent tool for examining overall simultaneous attention while controlling for chance occurrence.

Lagged recurrence measures of vocal coordination related to event-based measures as expected. This is likely because vocalizations are discrete events rather than continuous events (like gaze). In addition, the method used to analyze lagged vocalization recurrence (removing simultaneous speech and fixing the durations of vocalizations) has the impact of guaranteeing that any lagged recurrence that occurs represents one partner’s vocalization following another partner’s vocalization (at varying degrees of lag). In fact, recurrence measures may have an advantage over event-based measures in describing coordination for vocalizations, as they do not require the researcher to decide what degree of lag qualifies as a contingent response and instead allow examination of lagged recurrence across all lags within a specific time period.

Relations between Coordination and Individual Behaviors

Several findings revealed ways in which individual mother and infant behaviors related to dyadic coordination of behavior, highlighting the bidirectional nature of coordination. For example, the establishment of vocal coordination was related to the frequency of both mother and infant vocalizations at all four ages, despite the fact that coordination variables controlled for chance occurrence of coordination. It is likely that mothers and infants who vocalize less have fewer opportunities to establish coordinated interactions with one another and may therefore be less predictable partners. While this relation was strong across the first year of life, the majority of relations changed with infant age, suggesting that the mechanisms through which mothers and infants achieve coordination may also change over time. For example, the relation between infant production of consonant sounds and vocal coordination was only apparent at 6 and 9 months, ages when consonant sounds are first developing and being refined (Oller, 2000). This finding replicates and extends previous research suggesting that quality of infant vocalizations has an impact on dyadic coordination (Gros-Louis et al., 2006). It further suggests that the aspects of infants’ vocal production that impact vocal coordination were not a constant feature of the interaction, but changed over time as infants developed and refined new skills.

Relations between object and mutual gaze coordination and individual gaze behaviors also changed with infant age. At 3 months, dyads with mothers who shifted their gaze more frequently and had shorter gaze durations had better coordination of gaze to objects. At this age, mothers and infants spent much of their time in face-to-face interactions, typically with only a single toy. Mothers’ gaze shifts likely represent shifts between infant’s face and a toy the infant is attending to. It may be that mothers who shift their gaze more frequently give their infants more opportunities to jointly attend to objects by shifting between face-to-face interaction and object interactions. As infants get older and develop the ability to sit upright and initiate interactions with objects, mothers’ gaze shifts likely take on less importance in establishing coordination.

Relations between individual gaze behavior and mutual gaze coordination showed a different pattern. At 6 and 9 months, dyads with infants with more gaze shifts and shorter gaze durations had better coordination of gaze to faces. The importance of infant behavior to mutual gaze coordination is consistent with the finding that mutual gaze was most often achieved with mothers leading and infants following. While mothers have the ability to quickly and frequently gaze at their infants’ faces, this is a bigger task for young infants, particularly around 6 to 9 months of age as they are newly developing the ability to sit independently (Fogel, Dedo, & McEwen, 1992). In fact, prior research suggests that the developmental shift to upright posture is associated with decreased looking to mother’s faces (Fogel et al., 1999), and infants placed in an upright posture (e.g. a highchair) during toy play interactions with their mothers appear to engage in increased shared attention to objects (de Barbaro et al., 2016; Deák et al., 2014). In an upright posture, infants must disengage their attention from an object and shift their entire head and body weight in order to look up at their mother’s faces. In the present study, mothers were free to place or hold their infants however they pleased (using only an optional boppy pillow). Future research examining how infants’ posture impacted the coordination of gaze in these interactions could provide valuable insight into the findings reported here.

Limitations and Conclusions

This research has a number of strengths, including micro-coding of infant and mother behavior, a longitudinal design, and a multi-measure approach to examining dyadic coordination. As mentioned above, the sample size of 30 infants is similar to or exceeds sample sizes in similar research (e.g. Beebe et al., 1988; de Barbaro et al., 2016; Deák et al., 2014; Jasnow & Feldstein, 1986; Yu & Smith, 2013), and is further strengthened by the use of micro-coding and a longitudinal design. That being said, the relatively small N (particularly given missing data at 3 and 12 months) may have limited power to detect some effects. In particular, an initial aim of this study was to examine differences between dyads of families with and without an older child with ASD, but we found no differences between these groups. These null results should be interpreted with caution given the small sample size, and further research is needed to know whether early differences in mother-infant coordination emerge in the younger siblings of children with ASD. Intensive moment-by-moment coding of behavior requires a large investment of time, placing limits on reasonable sample sizes. However, recent advances in more automated systems for micro-coding behavior, such as head mounted eye trackers and vocal recognition technology, may make this type of research more feasible with larger samples in the near future.

Second, although examining coordination within modalities is an important first step, coordination undoubtedly occurs across modalities, with mothers and infants shifting their gaze in response to partners’ vocalizations and vocalizing in response to partners’ gaze shifts. In future research we plan to examine how mother and infant vocal behavior and gaze behavior interact and impact one another across the first year of life.

Our findings emphasize the importance of considering mother-infant interactions as bidirectional and impacted by the vastly changing context and dynamics of interactions across infants’ first year of life. The bidirectional relations revealed here make clear that concepts like “maternal responsiveness” are not individual features of mothers, but emerge out of the dyadic interaction. Infant behavior is both impacted by and impacts how responsive mothers are. This notion of coordination emerging as a complex interaction between multiple aspects of mother and infant behavior is even more noteworthy when considered in the developmental context. Findings with regard to the early onset of dyadic coordination, the lack of change in vocal coordination over time, and the changing relations between individual behaviors and coordination of behaviors lend support to the notion that coordination emerges out of the structure and dynamics of the interaction.