Intentional Action Processing across the Transition to Crawling: Does the Experience of Self-Locomotion Impact Infants’ Understanding of Intentional Actions?

Abstract

Motor developmental milestones in infancy, such as the transition to self-locomotion, have cascading implications for infants’ social and cognitive development. The current studies aimed to add to this literature by exploring whether and how crawling experience impacts a key social-cognitive milestone achieved in infancy: the development of intentional action understanding. Study 1 used a cross-sectional, age-held-constant design to examine whether locomotor (n = 36) and prelocomotor (n = 36) infants differ in their ability to process a failed intentional reaching action. Study 2 (n = 124) further probed this question by assessing how variability in locomotor infants’ experience maps onto variability in their failed intentional action understanding. Both studies also assessed infants’ tendency to engage in triadic interactions to shed light on whether self-locomotion impacts intentional action understanding directly or indirectly via changes in infants’ interactions with social partners. Altogether, results showed no evidence for the role of self-locomotion in the development of intentional action understanding. Locomotor and prelocomotor infants did not differ in their failed action understanding or levels of triadic engagement (Study 1) and individual differences in days of crawling experience, propensity to crawl during play, and maximum crawling speed failed to predict infants’ intentional action understanding or triadic engagement (Study 2). Explanations for these null findings and alternative influences on the development of intentional action understanding are considered.

Motor developmental milestones in infancy have cascading implications for infants’ social and cognitive development (Adolph & Robinson, 2013; Campos et al., 2000; Gibson, 1988; Thelen, 2005). A growing body of evidence suggests that the acquisition of new motor skills alters how infants attend to and engage with the world around them, thus serving as a catalyst for development across domains (for reviews, see Bertenthal, Campos, & Barrett, 1984; Campos et al., 2000; Iverson, 2010; Libertus & Hauf, 2017). For example, research suggests that crawling infants spend more time in interactive games and distal communicative interactions with their mothers than do non-crawling infants (Campos et al., 2000). Crawling also corresponds with changes in infants’ joint-attention behaviors, with crawling infants showing greater skill in following the gaze and gestures of others (Campos, Kermoian, Witherington, Chen, & Dong, 1997). The transition from crawling to walking is likewise associated with changes in infants’ engagement with the social environment, including increases in receptive and productive language (Oudgenoeg-Paz, Volman, & Leseman, 2012; Walle & Campos, 2014) and joint attention (Clearfield, 2011; Clearfield, Osborne, & Mullen, 2008; Karasik, Tamis-LeMonda, & Adolph, 2011, 2014; Walle, 2016). Even the acquisition of reaching early in infancy has been shown to have social-cognitive implications (Gerson & Woodward, 2013; Libertus & Needham, 2011; Skerry, Carey, & Spelke, 2013; Sommerville, Woodward, & Needham, 2005).

The goal of the current studies is to add to this growing literature on the role of motor transitions in infants’ developing social cognition. Specifically, we explore whether and how crawling experience impacts a key social-cognitive milestone achieved early in infancy: the development of intentional action understanding.

1.1. The Development of Intentional Action Understanding

The ability to reason about the intentional nature of human action is an important developmental milestone argued to lay the foundation for critical later social-cognitive achievements (e.g., Aschersleben, Hofer, & Jovanovic, 2008; Krogh-Jespersen, Liberman, & Woodward, 2015; Wellman, Lopez-Duran, LaBounty, & Hamilton, 2008). Intentional action understanding emerges early in infancy and expands across the first year of life (Woodward, Sommerville, Gerson, Henderson, & Buresh, 2009). For example, infants are sensitive to the intentional structure of human behavior and readily interpret others’ actions as goal-directed by 6 months of age (Woodward, 1998). Also during the first year, infants can use their understanding of human behavior to generate predictions about intentional actions as they unfold over time (e.g., Brandone, Horwitz, Aslin, & Wellman, 2014; Cannon & Woodward, 2012; Falck-Ytter, Gredebäck, & von Hofsten, 2006; Kochukhova & Gredebäck, 2010). These emerging abilities are important as success in our social world depends on the capacity to understand, predict, and learn from the actions of others (Woodward et al., 2009).

One key transition in the development of intentional action understanding is the acquisition of an understanding of failed actions. Evidence suggests that, although successful action understanding is in place relatively early in infancy (e.g., Woodward, 1998), the ability to reason about the intentional nature of actions that fail to achieve their goals does not appear until later in the first year. For example, both habituation (Brandone & Wellman, 2009) and anticipatory-looking studies (Brandone et al., 2014) have shown that infants as young as 8 months understand the goal-directedness of an action in which an agent reaches over a barrier and successfully retrieves a ball—but it is not until 10 months that infants show comparable understanding of a matching action in which the agent reaches over the barrier but fails to retrieve the ball (see Behne, Carpenter, Call, & Tomasello, 2005; Daum, Prinz, Aschersleben, 2008; Hamlin, Newman, & Wynn, 2009, for further evidence of infants’ understanding of unsuccessful or incomplete actions).

This transition from successful to failed action understanding is important because failed actions have been proposed to provide an especially stringent assessment of intention understanding (Meltzoff, 1995). Failed actions serve as one example of the many complex intentional actions infants experience in their daily lives: in addition to basic successful actions, infants regularly observe actions that are unsuccessful, interrupted, and incomplete. Mature action understanding requires the ability to process these more complex events (Brandone et al., 2014). In addition, failed actions are unique in that the goal of the action (e.g., retrieving the ball) is distinct from the action’s observed outcome (e.g., returning empty-handed). This stands in contrast to successful actions in which the actor’s goal is evident in the outcome of the action itself. Thus, to properly process a failed action, one must consider more than just its surface appearance (Brandone & Wellman, 2009; Meltzoff, 1995). For these reasons, failed action understanding has been proposed to be a significant conceptual achievement within the development of intentional action understanding, and research has begun to examine the mechanisms underlying its development.

1.2. Links between Self-Locomotion and Intentional Action Understanding

Efforts to explore how failed action understanding develops have revealed a possible role for infants’ motor development – in particular, the capacity for self-locomotion. To shed light on the mechanisms underlying changes in failed action understanding, Brandone (2015) tested a sample of 8-11-month-olds on their ability to visually anticipate the outcome of a failed reaching action. Infants watched a video in which an agent repeatedly reached over a barrierfor a ball. With the ball just out of reach, the agent hovered briefly with his hand above the ball and then brought his empty hand back to his torso. Infants’ anticipatory looks to the goal of the actor’s reach (i.e., the ball) were measured. Alongside this experimental assessment, parents also reported infants’ progress on various other developmental milestones, including achievements related to locomotor development (e.g., crawling, pulling to stand, walking with support). Results showed that infants’ ability to predict the goal of a failed reaching action was significantly related to parent-report assessments of infants’ self-locomotive development: infants who showed more advanced self-locomotion also showed superior performance on the failed action measure. Importantly, this relation was preserved when controlling for infant age. Overall, this study reveals the possibility of an important link between infants’ intentional action processing and their motor abilities (for related evidence of differences in infants’ reasoning about an agent’s movements based on self-locomotion experience, see Brand, Escobar, Baranès, & Albu, 2015; Stapel, Hunnius, Meyer, & Bekkering, 2016).

There are at least two possible explanations for this relation. First, in light of the proposal that motor developmental milestones impact infants’ development in a variety of domains (e.g., Adolph & Robinson, 2013; Campos et al., 2000; Gibson, 1988; Thelen, 2005), one possibility is that motor experience directly influences infants’ understanding of others’ actions. The emerging capacity to move independently through the environment may serve as a highly salient example of the intentional nature of infants’ own actions. Upon the onset of self-locomotion, infants can move independently to distant objects or people and can be derailed in their attempts to do so. This experience may provide an especially powerful type of first-hand agentive experience that leads to broad insights about both infants’ own and others’ actions (Brandone, 2015).

Indeed, a large body of work has supported the role of first-hand agentive experience in shaping infants’ understanding of others’ actions (Gallese, Rochat, Cossu, & Sinigaglia, 2009; Gredebäck & Falck-Ytter, 2015; Woodward et al., 2009). Infants’ appreciation of the goal-directed nature of basic reaching actions has been shown to coincide with infants’ ability to produce coordinated reaching actions themselves (Kanakogi & Itakura, 2011; Woodward, 1998). A similar correspondence exists between infants’ tendency to make anticipatory gaze shifts while watching an actor placing objects into a container and their ability to competently produce this action (Cannon, Woodward, Gredebäck, von Hofsten, & Turek, 2012; Falck-Ytter, Gredebäck, & von Hofsten, 2006). Compelling experimental findings also support the view that motor experience causally impacts infants’ understanding of others’ actions: when an experimental intervention (i.e., Velcro mittens) permitted 3-month-olds the opportunity to produce coordinated reaching actions for the first time, they subsequently interpreted an experimenter’s reaching actions as goal-directed (Gerson & Woodward, 2013; Sommerville et al., 2005).

The observed link between self-locomotion and failed action understanding (Brandone, 2015) can be interpreted in light of this evidence that insights into others’ actions are derived from first-person motor experience. However, there is an important distinction between these two sets of findings. While the examples described above demonstrate a specific mapping between infants’ production of a particular motor behavior (e.g., reaching for an object, putting an object in a container) and their understanding of that same behavior in others, the link between self-locomotion and failed action understanding (Brandone, 2015) does not share this specificity. Thus, if the experience of independent locomotion is linked to changes in infants’ failed action understanding, it would suggest that broader aspects of experience as an intentional agent also influence infants’ views of others’ intentional actions.

In addition to the potential direct link between self-locomotion and intentional action understanding, a second possibility is that self-locomotion impacts intentional action understanding indirectly, by way of the changes it promotes in infants’ social environment (Brandone, 2015). According to Campos and colleagues (2000), the acquisition of crawling—the most common early form of self-locomotion—has cascading effects on infants’ interactions with social partners. Compared to non-crawling infants, crawlers have been reported to spend a greater amount of time in interactive games involving reciprocity (e.g., peek-a-boo) and in distal communicative interactions with their mothers (Campos et al., 2000). Crawlers also show greater skill than do non-crawlers in following the gaze and point gestures of others (Campos et al., 1997). The underlying argument here is that crawling affords new opportunities for infants’ interactions with the world and that this motor transition evokes mutually influential changes in both caregivers and infants (for a review, see Campos et al., 2000).

These changes in infants’ social interactions resulting from the onset of self-locomotion have the potential to subsequently shape infants’ understanding of intentional actions. Recent evidence suggests that experience in triadic interactions in which infants share attention and action with a social partner around an object of mutual interest supports infants’ intentional action understanding: infants who showed more triadic engagement with a researcher in a structured play task also showed superior intentional action understanding both in concurrent assessments and when observed three months later (Brandone, Stout, & Moty, 2019; see Barresi & Moore, 1996 for a related theoretical argument). Additional studies have also supported links between infants’ intentional action knowledge and their triadic engagement tendencies or skills (Brune & Woodward, 2007; Dunphy-Lelii, La Bounty, Lane, & Wellman, 2014). If the onset of crawling results in changes in infants’ experience in triadic interactions (Campos et al., 2000), it may also indirectly influence infants’ intentional action understanding. Initial support for this possibility comes from an exploratory mediation model tested in Brandone (2015): analyses showed that the link between self-locomotion and failed action understanding was partially mediated by parent reports of infants’ tendency to initiate joint attention (e.g., pointing to direct attention, showing objects). Together these findings suggest that self-locomotion may serve as a catalyst for the development of intentional action understanding by facilitating the kinds of rich social interactions necessary to build an understanding of intentional actions.

In sum, there is intriguing preliminary support in the literature for links between self-locomotion and failed action understanding (Brandone, 2015). However, the existing data are limited by the parent-report nature of the self-locomotion assessment. Further research is needed to confirm this relation using observed measures of self-locomotion. Additionally, questions remain concerning the underlying nature of the association between self-locomotion and failed action understanding, including whether self-locomotion impacts intentional action understanding directly or indirectly via changes in infants’ social interactions. The goal of the current studies was to disentangle these possibilities through a systematic exploration of the relation between self-locomotion and intentional action understanding in infancy.

1.3. The Current Studies

Here we present two studies investigating whether and how the experience of self-locomotion impacts the development of intentional action understanding in the first year of life. We focus specifically on self-locomotion via crawling as this represents the most common type of early successful self-locomotion. Failed action processing was used as our measure of intentional action understanding because it provides a rigorous test of infants’ ability to interpret a complex action in the absence of outcome information (see Brandone et al., 2014) and because prior work has established its association with locomotor experience (Brandone, 2015).

In Study 1, we explored the basic question of whether age-matched locomotor and prelocomotor infants differ in intentional action understanding. This cross-sectional, age-held-constant design allowed us to test whether locomotor experience matters for intentional action understanding while carefully controlling for associated maturation differences based on age. In Study 2, we moved beyond a categorical definition of locomotor status by exploiting the natural variability that exists in locomotor infants’ experience. Using a large sample of infants varying in locomotor experience, we examined whether individual differences in self-locomotive abilities map onto individual differences in intentional action processing. Finally, to shed light on whether self-locomotion impacts intentional action understanding directly or indirectly via changes in infants’ rich interactions with social partners, in both studies we assessed infants’ spontaneous tendency to engage in triadic interactions.

If the experience of self-locomotion contributes to infants’ understanding of intentional actions, locomotor infants should show better intentional action understanding than age-matched prelocomotor infants in Study 1, and greater locomotor experience should predict better intentional action understanding in Study 2. Additionally, if self-locomotion serves as a catalyst for the development of intentional action understanding by facilitating rich interactions with social partners, triadic engagement should mediate relations between infants’ locomotor experience and their intentional action understanding in both studies. Alternatively, if locomotor experience impacts intentional action understanding directly rather than indirectly, there should be no evidence for mediation.

2. Study 1

The primary goal of Study 1 was to test whether age-matched locomotor and prelocomotor infants differ in their ability to process a failed intentional action. Locomotor status was determined based on parental report of infants’ ability to crawl a distance of 8 to 10 feet without stopping and was confirmed through an observed locomotion task. This relatively strict locomotor criterion was applied here because (a) it features prominently in prior work as an effective marker of locomotor proficiency and experience (e.g., Adolph, Vereijken, & Shrout, 2003; Adolph, Berger, & Leo, 2011), and (b) the central explanations for the influence of self-locomotion on intentional action understanding tested here assume more than minimal levels of locomotor experience. Following Brandone (2015), locomotor infants were predicted to outperform prelocomotor infants in their intentional action processing. To further explore the nature of self-locomotion’s effects on infants’ social-cognitive abilities, we also examined whether infants’ tendency to participate in triadic engagement with a social partner differed based on locomotor status and whether triadic engagement mediated the effects of self-locomotion on intentional action understanding.

2.1. Participants

A total of 72 7.5- to 9-month-old infants (M = 8.46 months, SD = .40; 19 males, 53 females¹; 68% White, 17% Hispanic or Latino, 3% Asian, 1% Black or African American, 10% selected two or more races/ethnicities, 1% did not respond). This age range is justified given normative data (Bayley, 1969; Frankenburg et al., 1992) and empirical studies (e.g., Adolph et al., 2011; Davis, Moon, Sachs, & Ottolini, 1998) showing that the approximate age of crawling onset is between 8 and 8.5 months. Targeting this age range also ensured variability in infants’ locomotor experience, and enabled us to recruit both prelocomotor and locomotor infants. A power analysis using the effect size from Brandone (2015), setting alpha = .05 and power = .80, determined a sample of this size is sufficient to detect group differences in infants’ failed action understanding. Eight additional infants were excluded from the final sample due to insufficient eye-tracking data (n = 7; see Data Processing section) and equipment malfunction during eye tracking (n = 1). Infants came from families that were predominantly middle class (11% reported household incomes less than $35,000, 10% $35-52,000, 18% $52-74,000, 28% $74-121,000, 28% greater than $121,000, 5% declined to respond) and college-educated (3% of mothers had no more than a high school degree, 13% had some college, 42% had a college degree, 38% had a graduate degree, 5% declined to respond). Participants were recruited from birth records and a database of families interested in developmental research in a midsize city in the Northeastern United States. Families were compensated $20 for participation and informed consent was obtained from all parents prior to data collection. A subset of data from these participants was reported in Brandone et al. (2019). Fourteen participants completed the same focal tasks previously at approximately 6.5 months of age as part of a longitudinal investigation of intentional action understanding.²

Locomotor status was initially determined by parental report of whether or not infants met the criterion for self-locomotion: traveling 8-10 feet across the room without stopping (Adolph et al., 2011). Locomotor status was also confirmed in the lab (see task below). Half of the participants met the criterion for self-locomotion (n = 36; M = 8.47 months; SD = .39; Range = 7.63 to 9.27 months; 11 males, 25 females). Based on parental report, locomotor infants had been crawling to criterion for an average of one month (M = 33.77 days; Range = 3 to 117; SD = 23.88) and locomoting in some capacity (although less than 8-10 feet) for roughly two months (M = 56.97 days; Range = 15 to 129; SD = 28.57). Twenty-seven of the infants crawled on their hands and knees and 9 moved on their bellies. None of the infants had begun walking.

The other half of participants failed to meet the locomotion criterion (n = 36; M = 8.43 months; SD = .42; Range = 7.53 to 9.27 months; 8 males, 28 females). Twenty-three of these infants had begun self-locomoting but were not yet able to travel 8-10 feet without stopping. These infants had been self-locomoting in some capacity for an average of 20.56 days (Range = 0 to 61; SD = 19.15). Of these infants, 5 moved on their hands and knees, 14 on their bellies, and 4 in an upright position (e.g., bum shuffling). Importantly, none of the prelocomotor infants were able to travel 8-10 feet across the room without resting. Locomotor and prelocomotor infants did not differ significantly in age, gender composition, or socioeconomic status (all ps > .43).

2.2. Measures and Procedure

During a laboratory visit, infants completed an eye-tracking measure of intentional action understanding, a behavioral assessment of triadic engagement, as well as several other tasks that are beyond the scope of this paper. A self-locomotion task was given to all infants reported to have reached the locomotor criterion in order to confirm the parental report. (Detailed descriptions of all tasks and coding procedures are available at https://osf.io/c92z6/ and video examples of all procedures are also available at https://nyu.databrary.org/volume/1152.) All infants completed the intentional action understanding task first, followed by the triadic engagement measure; the self-locomotion task occurred later in the experimental session.

2.2.1. Intentional action understanding

Video stimuli were presented on a 24-inch Tobii T60XL corneal reflection eye-tracking system with a sampling rate of 60 Hz and accuracy between .5 and 1 visual degree (Tobii Technology, Sweden). Infants sat in a high chair or on a parent’s lap approximately 60 cm from the screen. The eye tracker was calibrated for each participant using a 5-point procedure implemented through Tobii Studio software (Tobii Technology, Sweden). Video presentation was controlled by PsyScope X (available at http://psy.ck.sissa.it/) and the TobiiPlus interface (Papp & Fillippin, 2014).

The video presented to infants depicted a failed reaching event identical to that used in prior work (Brandone et al., 2014; Brandone & Wellman, 2009; see Figure 1). The video shows a man seated at a table in front of a barrier and a ball. The actor first peers over the barrier at the ball and then reaches in an arcing path over the barrier, narrowly missing the ball. After hovering with his hand briefly suspended above the ball, the actor brings his empty hand back to his torso and the video freezes. The full reaching event was 7200 ms in duration, followed by an 800 ms still frame. Half of infants watched a version of the video where the actor was seated on the left side of the screen and reached for a ball on the right side; the other half watched the inverse video. Participants were presented with 10 repetitions of this event in between which infants saw an attention-getting stimulus. On the first trial only, the initial frame of the video was presented for 2000 ms before the video began in order to orient infants to the scene.

Figure 1.

Time course depiction of the failed intentional reaching action. The highlighted region indicates the window of opportunity for a look to the AOI—beginning when the actor’s reaching motion was launched (1000 ms) and ending when the actor’s hand had crossed back to the near side of the barrier (6000ms). Looks to the AOI occurring before 3500 ms are considered anticipatory.

Data processing

Eye-tracking data processing was carried out in R (Version 3.3.2; R Core Team, 2017). Trials on which a participant watched the full screen for less than 50% of the reaching action were dropped. Across participants, 14.7% of trials were dropped for this reason. Participants who contributed usable data on fewer than 50% of trials were excluded from the sample as this pattern suggests inattentiveness or poor eye-tracking data quality (n = 7; 6 locomotor and 1 prelocomotor infant), leaving a final sample of 72 infants.

A circular area of interest (AOI) was defined manually around the ball. The AOI extended approximately 1° beyond the outer limit of the ball. This buffer was selected based on previous work (Brandone et al., 2014), standards in the field (see Gredebäck, Johnson, & von Hofsten, 2009), and estimates of inaccuracies in the Tobii system (0.5 to 1.0°).

Analyses examined two assessments of infants’ intentional action processing: the latency of gaze shifts to the ball AOI and whether or not gaze shifts were considered anticipatory. These measures provide complementary but unique assessments of infants’ action processing. The latency measure provides a pure evaluation of how long it takes for infants to look at the AOI, but excludes both trials on which infants do not look at the AOI (39.4% of trials) and infants who never look at the AOI (n = 18; 10 locomotor and 8 prelocomotor). The anticipatory-looking measure includes all infants, but relies on an anticipatory-looking criterion that treats looking behavior categorically. Generally, an anticipatory look is defined as a gaze shift to the goal of an action before the action is completed (e.g., Falck-Ytter et al., 2006); however, because failed actions are never technically completed, as in Brandone et al. (2014), we defined anticipatory looks using a criterion that involves a distance between the actor’s hand and the ball. We selected a distance of roughly 2 visual degrees because this reflects the distance between the ball and the actor’s hand at the full extension of the failed reach.

For focal analyses examining differences in intentional action understanding based on locomotor experience, we created two composite variables: (1) infants’ proportion of anticipatory looks on Trials 2-5, and (2) their mean latency of looking to the AOI across Trials 2-5. The composite variables included the first half of trials only because initial analyses showed that the likelihood of producing unusable data increased across trials and infants contributed significantly more unusable data in the second half of trials (see the Supplemental Materials for a full reporting of these analyses). Focusing analyses on the first half of trials thus minimized the effects of dropped trials and trials without looks to the ball. We also chose to exclude Trial 1 from our composite variables for three reasons. First, the timing of Trial 1 differs from that of subsequent trials because, on this trial only, the initial frame of the video was presented for 2000 ms before the reaching action began. Second, Trial 1 is also conceptually distinct because it assesses how infants viewed the reaching event before they witnessed the failed outcome. Finally, initial analyses showed that infants shifted their gaze to the AOI later on Trial 1 than on all subsequent trials³ (see the Supplemental Materials).

2.2.2. Triadic engagement

Triadic engagement was assessed through a structured 2-minute play session involving the infant and an experimenter. We examined interactions with an experimenter rather than a familiar caregiver in order to provided standardized input and eliminate variability in caregiver interactive style as a factor in infants’ triadic engagement (Gaffan, Martins, Healy, & Murray, 2010). Infants were seated in a highchair across from the experimenter at a table on which a set of toys was arranged. At 30-second increments, the experimenter engaged the infant with the toys. During the remainder of the session, she sat neutrally and only responded if infants engaged by looking at her face. Following infants’ face looks, the experimenter responded by making eye contact, emoting positively, and providing brief verbal comments (e.g., “What do you have?”). The experimenter also followed infants’ gaze if they shifted back to the toy, and re-established eye contact if they looked back at her face.

Coding

Video recordings of the interactions were coded using Datavyu (Datavyu Team, 2014) by individual members of a team of trained researchers. In an initial coding pass, a coder identified the onset and offset of each look to the researcher’s face. In a subsequent pass, a coder examined each face look to identify bouts of triadic engagement. Following Carpenter et al. (1998), we defined triadic engagement as gaze alternations between an object, the experimenter’s face, and the same object. Bouts of triadic engagement could include one or more of these object-face-object sequences so long as infants’ attention alternated between the researcher and the same object. A bout was terminated if infants’ attention shifted to a new object. For the full coding scheme, see https://osf.io/c92z6/. For each infant, we calculated the total number of bouts of triadic engagement during the 2-minute task.

Reliability

To establish coding reliability, 30% of videos were independently coded by an additional member of the coding team. Coders agreed that a face look occurred 95% of the time and the total number of face looks reported for each participant was strongly correlated across coders, r > .99. Coders agreed that a bout of triadic engagement occurred 81% of the time and the total number of bouts reported for each participant was also strongly correlated across coders, r = .88. As a final verification of the coding accuracy, all bouts of triadic engagement identified by the initial coder in 100% of videos were reviewed by an expert coder to confirm that they met the coding criteria. Of these bouts, 89% were confirmed. When disagreements occurred, the expert coder’s coding was retained. Although this final verification process deviates from conventional reliability procedures, it is appropriate here due to the expertise of the expert coder and the fact that all coders, including the expert, were blind to infants’ performance on the eye-tracking task.

2.2.3. Self-locomotion

Locomotor status was initially determined via parental report. In a structured interview, parents were asked whether they had observed their child independently travel a distance of 8-10 feet without stopping to rest (Adolph et al., 2011). For infants whose parents indicated yes, a self-locomotion task was administered at the end of the laboratory visit to verify their locomotor status. In this task, infants were given 3-4 opportunities to crawl across a 12-foot mat to their caregivers (see Karasik et al., 2011). Three infants whose parents initially reported they met the locomotor criterion were ultimately assigned to the prelocomotor group because their self-locomotion abilities were not verified in the observed self-locomotion task: these infants were unable to travel the required 8-foot distance without stopping. All other infants in the locomotor group provided clear evidence that they met the stringent criterion for self-locomotion⁴.

2.3. Analytic Plan

Primary analyses examined potential differences between age-matched locomotor and prelocomotor infants. Analysis of variance (ANOVA) was used to examine the effects of locomotor status (locomotor vs. prelocomotor) on failed action understanding (proportion of anticipatory looks and mean latency of looking to the AOI) and triadic engagement. Because prior research has demonstrated gender differences in triadic engagement with girls showing more bouts of triadic interaction than boys (Brandone et al., 2019), gender and the interaction between gender and locomotor status were included in the analysis of triadic engagement. (See the Supplemental Materials for an additional exploratory analysis revealing a significant main effect of gender on infants’ intentional action understanding.)

Bayes factors (BF) were also used to assess the strength of evidence for the alternative hypothesis (H1: locomotor infants outperform prelocomotor infants) relative to the null (H0: locomotor infants do not outperform prelocomotor infants). By convention, a BF of greater than 3 indicates positive evidence for H1 over H0, and a BF of less than .33 indicates positive evidence for H0 over H1. BFs between .33 and 3 indicate data insensitivity (Dienes, 2014; Jeffreys, 1961; Kass & Raftery, 1995). Following Dienes (2014), the predictions of H1 were modelled as a half-normal distribution centered on zero with a standard deviation equal to the rough expected effect size derived from prior research (Brandone, 2015). The half-normal distribution is appropriate given our directional predictions that locomotor infants will show better intentional action understanding and more triadic engagement than prelocomotor infants.

The final set of analyses explored whether self-locomotion impacts intentional action understanding indirectly via changes in infants’ rich interactions with social partners. The indirect effect of locomotor status on intentional action understanding through triadic engagement was estimated using a bootstrapping approach implemented with the PROCESS model Version 3 (Hayes, 2017). Separate models were built predicting infants’ proportion of anticipatory looks and their mean latency of looking to the AOI.

2.4. Results

2.4.1. Does failed intentional action understanding differ by locomotor status?

Neither ANOVA yielded significant effects of locomotor status (locomotor vs. prelocomotor) on intentional action understanding: Locomotor and prelocomotor infants did not differ in their proportion of anticipatory looks, F(1, 70) = .213, p = .646, η_p² = .003 (locomotor: M = .21, SD = .30; prelocomotor: M = .24, SD = .30; see Figure 2A), nor in their mean latency of looking to the AOI, F(1, 52) = .032, p = .860, η_p² = .001 (locomotor: M = −105 ms relative to the anticipatory look cutoff, SD = 803; prelocomotor: M = −62, SD = 950; see Figure 2B).

Figure 2.

(A) Proportion of anticipatory looks, (B) mean latency of looking to the AOI relative to the anticipatory look cut-off (in ms), and (C) bouts of triadic engagement by locomotor status. Box plots reflect the median and interquartile range (IQR). Dots reflect individual data points. Data points outside of the whiskers are greater or less than 1.5 times the IQR.

Bayesian analyses revealed BFs of 0.21 (95% CI: 0.01, 0.44) for the proportion of anticipatory looks, and 0.38 (95% Cl: 0.01, 0.61) for the mean latency of looking to the AOI. These values suggest that the observed data are 4.8 and 2.6 times more likely to occur under H0 (locomotor infants do not outperform prelocomotor infants) than under H1 (locomotor infants outperform prelocomotor infants). Thus, data can be interpreted as anecdotal to tentative positive evidence in favor of the null hypothesis—that locomotor infants do not outperform prelocomotor infants in intentional action understanding.

2.4.2. Does triadic engagement differ by locomotor status?

The ANOVA examining the effects of locomotor status, gender, and their interaction on infants’ tendency to engage in triadic interaction with the experimenter yielded no significant results (all ps > .31). Critically, locomotor (M = 2.59, SD = 1.54) and prelocomotor infants (M = 2.61, SD = 1.95) did not differ in the number of bouts of triadic engagement produced, F(1, 69) = .102, p = .751, η_p² = .001 (see Figure 2C).

Bayesian analyses showed BF = 0.28 (95% CI: 0.01, 0.51), meaning that the observed data are 3.6 times more likely to occur under H0 (locomotor infants do not outperform prelocomotor infants) than H1 (locomotor infants outperform prelocomotor infants). This value suggests tentative positive evidence in favor of the null hypothesis—that locomotor infants do not outperform prelocomotor infants in triadic engagement.

2.4.3. Does triadic engagement mediate the relation between locomotor status and intentional action understanding?

Bootstrapping analyses using PROCESS (Version 3) provided no support for an indirect effect of locomotor experience through triadic engagement on either measure of intentional action understanding: proportion of anticipatory looks: β = −.004, SE = .08 (95% CI = −.15, .18); mean latency of looking to the AOI: β = −.028, SE = .11 (95% CI = −.31, .13). Although triadic engagement was significantly associated with both infants’ proportion of anticipatory looks, r(72) = .324, p = .006, and their mean latency of looking to the AOI, r(55) = −.350, p = .010, there was no evidence that triadic engagement mediated a link between locomotor status and intentional action understanding.

2.5. Discussion

Overall, the results of Study 1 provide no evidence for the hypothesis that crawling experience matters for infants’ emerging social-cognitive abilities. The current data showed that age-matched locomotor and prelocomotor infants did not differ significantly in their intentional action processing or in their tendency to participate in bouts of triadic engagement. Thus, contrary to prior research (Brandone, 2015), the current findings fail to find support for either direct or indirect effects of the transition to self-locomotion on infants’ intentional action understanding, and Bayesian analyses provide tentative positive evidence in favor of the null hypothesis.

Although the null effects observed here argue against the transformative power of self-locomotion for intentional action understanding, there are at least two reasons not to discount this hypothesis entirely based on the findings of Study 1 alone. First, Study 1 used a categorical definition of locomotor status that grouped infants based on a relatively strict locomotor criterion (i.e., the ability to crawl 8-10 feet without stopping). Although this kind of categorical approach has been used successfully in prior work examining differences between locomotor and prelocomotor infants (e.g., Campos et al., 2000), it may have glossed over meaningful variation in infants’ experience of self-locomotion that matters for their development of intentional action understanding. To further test our hypothesis, additional work exploiting the natural variability that exists in locomotor infants’ experience is needed. Second, although the findings of Study 1 fail to provide positive evidence for the hypothesis that locomotor infants outperform prelocomotor infants in intentional action understanding, according to the Bayesian analyses presented, the current data also fail to provide strong positive evidence for the null hypothesis that locomotor infants do not outperform prelocomotor infants in intentional action understanding. Thus, further exploration with a larger sample that offers greater power to detect the hypothesized effects is warranted.

3. Study 2

The goal of Study 2 was to further probe relations between self-locomotion and intentional action understanding by assessing how variability in crawling experience maps onto variability in intentional action understanding. Moving beyond the binary definition of locomotor status applied in Study 1 (locomotor vs. prelocomotor), in Study 2 we capitalized on the natural variability in experience present in crawling infants. Specifically, we assessed locomotor experience in three distinct yet complementary ways. First, we derived a duration measure reflecting the number of days of crawling experience for each infant. Second, we observed infants in a free-play interaction with their caregiver during which we coded their spontaneous tendency to crawl during play. Third, we administered a standardized self-locomotion task in which we calculated infants’ maximum locomotion speed. Each of these assessments represents a conceptually distinct way of examining variability in locomotor experience (i.e., time since crawling onset, typical crawling behavior, peak locomotor speed) with potentially unique implications for infants’ intentional action understanding. Finally, to test the possibility that self-locomotion serves as a catalyst for the development of intentional action understanding by enhancing infants’ levels of triadic engagement, we also assessed infants’ tendency to participate in triadic interactions with an experimenter.

3.1. Participants

A total of 124 8- to 12-month-old infants (M = 9.72 months, SD = .92; 79 males, 45 females⁵; 73% White, 11% Hispanic or Latino, 2% Black or African American, 1% Asian, 10% selected two or more races/ethnicities, 4% did not respond) participated. None of the infants had participated in Study 1. To maximize variability in locomotor experience among infants who had all met the locomotion criterion established in Study 1, in this study we recruited a larger sample representing a wider and moderately older age range than in Study 1. Power analyses showed that a sample of this size is sufficient to detect a relatively small effect (f ² = .06) given alpha and power levels specified at .05 and .80. Eight additional infants were excluded from the final sample due to insufficient eye-tracking data (n = 5; see below), equipment malfunction during eye tracking (n = 1), fussiness that prevented completing the study (n = 1), or having transitioned to walking (n = 1). Infants came from families that were predominantly middle class (6% reported household incomes less than $35,000, 10% $35-52,000, 15% $52-74,000, 38% $74-121,000, 25% greater than $121,000, 6% declined to respond) and college-educated (3% of mothers had no more than a high school degree, 15% had some college, 37% had a college degree, 40% had a graduate degree, 5% declined to respond). Participants were recruited from birth records and a database of families interested in developmental research in a midsize city in the Northeastern United States. Families were compensated $20 for participation and informed consent was obtained from all parents prior to data collection. A subset of data from these participants was reported in Brandone et al. (2019). Ninety participants completed the same focal tasks at approximately 6.5 months of age—before they were proficient crawlers—as part of a longitudinal study of intentional action understanding.⁶

3.2. Measures and Procedure

During a laboratory visit, infants completed an eye-tracking measure of intentional action understanding, a behavioral assessment of triadic engagement, an 8-minute free play session with a caregiver, a self-locomotion task, and several tasks that are beyond the scope of this paper. (Detailed descriptions of all tasks and coding procedures are available at https://osf.io/c92z6/ and video examples of all procedures are also available at https://nyu.databrary.org/volume/1152.) All infants completed the intentional action understanding task first, followed by the free play session and the triadic engagement task; the self-locomotion task occurred later in the lab session.

3.2.1. Intentional action understanding

The intentional action understanding task and data processing procedures were identical to those used in Study 1. Across participants, 11.8% of trials were dropped for participants watching the full screen for less than 50% of the reaching action. Participants who contributed usable data on fewer than 50% of trials were excluded from analyses as this pattern suggests inattentiveness or poor eye-tracking data quality (n = 5), leaving a final sample of 124 infants. The latency of infants’ gaze shifts to the ball AOI and whether or not gaze shifts were considered anticipatory were examined. As in Study 1, composite measures of infants’ proportion of anticipatory looks and their mean latency of looking to the AOI across Trials 2-5 were calculated for use in central analyses. (See Supplemental Materials for analysis of the effects of trial on infants’ eye tracking data.)

3.2.2. Triadic engagement

Triadic engagement was assessed and coded following the procedure described in Study 1. To establish reliability, 30% of videos were independently coded by an additional coder. Coders agreed that a face look occurred 94% of the time and the total number of face looks reported for each participant was strongly correlated across coders, r = .98. Coders agreed that a bout of triadic engagement occurred 86% of the time and the total number of bouts reported for each participant was also strongly correlated across coders, r = .93. As a final confirmation of the accuracy of the coding, all bouts of triadic engagement identified by the initial coder in 100% of videos were verified by an expert coder. Of these bouts, 92% were confirmed. When disagreements occurred, the expert’s coding was retained.

3.2.3. Self-locomotion experience

Variability in infants’ self-locomotion experience was assessed in three ways: (1) parent-report of days of locomotor experience, (2) observations of self-locomotion during free play, and (3) a locomotor speed task.

Parent-report of days of locomotor experience

Using a structured interview procedure validated in previous research (Adolph et al., 2011), caregivers were asked to provide retrospective data on the onset dates of various motor milestones including prelocomotor (e.g., rocking on hands and knees, turning 180 degrees) and locomotor (e.g., belly, inchworm, hands and knees, or bear crawling) skills. To aid memory, parents were asked to consult calendars and personal records. As in Study 1, locomotor onset was defined as the first day infants independently traveled a distance of 8-10 feet without stopping to rest. Locomotor experience was then calculated as the number of days between locomotor onset and the test date. Based on parental report, infants had been crawling to criterion for an average of one month (M = 27.39 days; Range = 4 to 121; SD = 18.41) and locomoting in some capacity (although less than 8-10 feet) for nearly two months (M = 53.87 days; Range = 14 to 159; SD = 27.94). Ninety-two of the infants crawled on their hands and knees, 25 on their bellies, and 7 in an upright position (e.g., bum shuffling).

Observation of self-locomotion during free play

Individual differences in self-locomotion were also observed in an 8-minute, digitally-recorded free play session in the laboratory. Infants were observed in play with a familiar caregiver to create the most natural play environment possible in order for infants to demonstrate their typical locomotor tendencies. Caregivers and infants were seated on the floor with a standard set of toys and encouraged to play as they normally would at home. The free-play task took place in an 11 by 7.5-foot room. A table and chair were positioned against one wall of the room and a small shelf was positioned against the opposite wall. Aside from this furniture, the space was open and infants were permitted to navigate freely. The goal of this task was to provide measures of infants’ spontaneous tendency to engage in locomotion during naturalistic play.

Coding. Video recordings of the play sessions were coded using Datavyu (Datavyu Team, 2014) by members of a team of trained researchers. See https://osf.io/c92z6/ for a detailed coding scheme. A member of the coding team first identified episodes of locomotion, defined here as clear, intentional progress in a specific direction of travel (typically forward). Episodes began at the onset of locomotion with the movement of the leg in a crawling motion; episodes ended when the infant moved to a sitting position or onto his or her back, or when the infant paused for longer than 1.5 seconds in a balanced, stable position (often to engage with a toy).

For each episode, the coder then identified the number of strides taken. To accommodate locomotion of a variety of different forms (e.g., belly, inchworm, hands and knees, bear crawling, bum shuffling; see Adolph et al., 1998), strides were defined as drives forward with the legs in the service of locomotion. Because we were interested in assessing variability in sustained locomotion, episodes with greater than 2 strides were retained and defined as bouts of locomotion; episodes with 1 or 2 strides only were discarded (see Karasik, Adolph, Tamis-LeMonda, & Zuckerman, 2012 for similar definitions and exclusion criteria). For each infant, we calculated the total number of strides produced across the free-play session.

Reliability. To establish coding reliability, 30% of videos were independently coded by an additional member of the coding team. Coders showed high levels of agreement on the number of strides per bout. Using a standard of plus or minus 1 stride, coders were in agreement on 92% of bouts. The total number of strides reported for each participant was also strongly correlated across coders, r = .98.

Locomotor speed task

As a final assessment of infants’ locomotor experience, a locomotor speed task was administered at the end of the lab visit. Infants were given between 3 and 6 trials during which they were encouraged to crawl across a 12-foot foam mat with demarcations indicating every foot (see Karasik et al., 2011). At the beginning of each trial, infants were placed in their preferred crawling posture by an experimenter at one end of the mat; caregivers were positioned at the opposite end of the mat with a desired object. Together with a member of the research team, caregivers encouraged infants to crawl as fast as they could across the mat by calling to them excitedly and motioning with the desired object. Infants were given multiple opportunities to cross the mat until caregivers indicated that performance was typical of their top speed. Infants were given breaks between trials to rest as needed.

Coding. Video recordings of the locomotion task were coded using Datavyu (Datavyu Team, 2014) by trained coders. See https://osf.io/c92z6/ for the full coding scheme. A coder evaluated each trial in which infants traveled 8 or more feet (as determined by the demarcations on the mat). For each of these trials, the coder identified the onset and offset of the locomotor pass, defined, respectively, as the frame in which the infant takes his or her first stride forward and the frame in which the infant crosses the 8-foot line. For trials in which infants traveled greater than 8 feet, the onset and offset of each 8-foot pass was identified. Using this coding, we calculated the total duration of each 8-foot locomotor pass (M = 10.9 seconds; Range = 4.2 to 31.1; SD = 5.8). For the fastest pass, we calculated infants’ locomotor speed by dividing the distance traveled (8 feet) by the amount of time it took. Twelve infants failed to produce an 8-foot pass without stopping for 5 seconds or more. These infants were excluded from subsequent analyses of locomotor speed.

Reliability. To establish coding reliability, 30% of videos were independently coded by an additional member of the coding team. Using a standard of plus or minus 1 second, coders were in agreement on the duration of 91% of crawling passes. Pass durations were strongly correlated across coders, r = .99.

3.3. Analytic Plan

Focal analyses examined the central question of whether variability in locomotor experience predicted infants’ intentional action understanding. A series of regression models were built predicting infants’ proportion of anticipatory looks and mean latency of looking to the AOI. Separate models were tested for each of the three measures of locomotor experience: days of locomoting to criterion, crawling during free play, maximum crawling speed. Each model also included age as a control variable based on prior evidence of changes in failed action understanding with age (Brandone, 2015). (See the Supplemental Materials for an exploratory examination revealing a main effect of gender on infants’ intentional action understanding.) A similar set of regression models were built predicting the number of bouts of triadic engagement observed during the interaction task from each of the three measures of locomotor experience defined above. Each model also included age and gender as control variables based on prior evidence of the role of these variables in triadic engagement (Brandone et al., 2019).

Bayes factors were also used to assess the strength of evidence for the alternative hypothesis (H1: locomotor experience is associated with better intentional action understanding / more triadic engagement) relative to the null (H0: locomotor experience is not associated with better intentional action understanding/more triadic engagement). These analyses used a default uniform prior for a one-sided correlation.

Finally, we explored whether triadic engagement mediates the effect of locomotor experience on intentional action understanding by estimating the indirect effect using a bootstrapping approach implemented with the PROCESS model Version 3 (Hayes, 2017). Separate models were built predicting infants’ proportion of anticipatory looks and their mean latency of looking to the AOI, and each measure of locomotor experience was examined in a separate model (6 total models).

3.4. Results

3.4.1. Descriptions of locomotor experience

Descriptive statistics for each measure of locomotor experience are presented in Table 1. Infants’ maximum crawling speed was associated with both their days of locomoting to criterion, r(112) = .28, p = .002, and their crawling during free play, r(111) = .36, p < .001. However, infants’ days of locomoting to criterion and crawling during free play were unrelated, r(119) = .14, p = .12.

Table 1.

Descriptive Locomotion Data in Study 2.

Variable	Mean	SD	Range
1. Days of locomoting to criterion	27.5	18.5	4-121
2. Total number of strides during free play	23.7	26.8	0-121
3. Maximum crawling speed (feet per second)	0.91	0.38	0.26-1.91

3.4.2. Does locomotor experience predict failed intentional action understanding?

The results of the series of regression analyses predicting infants’ intentional action understanding from age and locomotor status can be seen in Table 2. None of the measures of locomotor experience emerged as significant predictors of infants’ proportion of anticipatory looks (see Figure 3) or their mean latency of looking to the AOI (see Figure 4) (all ps > .36)⁷

Table 2.

Results of Regression Analyses Predicting the Proportion of Anticipatory Looks and Mean Latency of Looking to the AOI from Age and Each Measure of Locomotor Experience in Study 2.

	Prop. of Anticipatory Looks			Mean Latency to the Ball AOI
	F	R2	β	F	R2	β
Model 1: Days of Locomoting	3.48*	.055		2.19	.040
Age			.227*			−.199*
Days of locomoting			.082			−.058
Model 2: Total Strides	3.79*	.060		2.29	.042
Age			.232*			−.209*
Total locomotion strides			−.052			−.033
Model 3: Crawling Speed	3.01	.051		2.23	.044
Age			.223*			−.185
Maximum crawling speed			.024			−.090

^*p < .05.

Figure 3.

Proportion of anticipatory looks by locomotor experience. Error bands reflect 95% CIs. Dots reflect individual data points.

Figure 4.

Mean latency of looking to the AOI relative to the anticipatory looking cut-off (in ms) by locomotor experience. Error bands reflect 95% CIs. Dots reflect individual data points.

Bayesian analyses examining the proportion of anticipatory looks measure revealed BFs (with 95% CI) of 0.21 (0.01, 0.24), 0.06 (0.002, 0.15), and 0.18 (0.004, 0.24) for days of locomoting to criterion, total strides during play, and maximum crawling speed, respectively. For the mean latency of looking to the AOI, BFs were 0.09 (0.002, 0.19), 0.13 (0.003, 0.22), and 0.07 (0.002, 0.17), respectively. Although the strength of the evidence varies by analysis, these Bayes factors suggest that the observed data are between 6.0 and 16.2 times more likely under the null hypothesis (locomotor experience is not associated with better intentional action understanding) than under the alternative hypothesis (locomotor experience is associated with better intentional action understanding). These values can be interpreted as substantial to strong positive evidence for the null.

3.4.3. Does locomotor experience predict triadic engagement?

The results of the regression analyses predicting infants’ triadic engagement from age, gender, and locomotor status are presented in Table 3. No measures of locomotor experience emerged as a significant predictor of infants’ triadic engagement (all ps > .09; see Figure 5).

Table 3.

Results of Regression Analyses Predicting Bouts of Triadic Engagement from Age, Gender, and Each Measure of Locomotor Experience in Study 2.

	Bouts of Triadic Engagement
	F	R2	β
Model 1: Days of Locomoting	1.32	.032
Age			.075
Gender			.121
Days of locomoting			.083
Model 2: Total Strides	2.06	.050
Age			.069
Gender			.095
Total locomotion strides			−.156
Model 3: Crawling Speed	1.50	.040
Age			.052
Gender			.111
Maximum crawling speed			.131

Figure 5.

Bouts of triadic engagement by locomotor experience. Error bands reflect 95% CIs. Dots reflect individual data points.

Bayesian analyses revealed BFs (with 95% CI) of 0.26 (0.01, 0.26), 0.04 (0.001, 0.11), and 0.70 (0.02, 0.32), for days of locomoting to criterion, total strides during play, and maximum crawling speed, respectively. These Bayes factors suggest that the observed data are 3.8, 25.5, and 1.4 times more likely under the null hypothesis (locomotor experience is not associated with more triadic engagement) than under the alternative hypothesis (locomotor experience is associated with more triadic engagement) for days of locomoting to criterion, total strides during play, and maximum crawling speed, respectively. These values range from anecdotal to strong; nevertheless, they indicate consistent positive evidence for the null hypothesis.

3.4.4. Does triadic engagement mediate the relation between locomotor experience and intentional action understanding?

Finally, bootstrapping analyses using PROCESS (Version 3) failed to provide support for an indirect effect through triadic engagement of any measure of locomotor experience (days of locomoting to criterion, crawling during free play, maximum crawling speed) on either measure of intentional action understanding (proportion of anticipatory looks, mean latency of looking to the AOI). In each of the six models tested, the 95% confidence interval for the indirect effect coefficient included zero. Thus, although triadic engagement was significantly related to both infants’ proportion of anticipatory looks, r(125) = .234, p = .009, and their mean latency of looking to the AOI, r(109) = −.193, p = .046, there was no evidence for an indirect effect of locomotor experience through triadic engagement.

3.5. Discussion

Overall, Study 2 provides no evidence for the hypothesis that locomotor experience matters for infants’ emerging social-cognitive abilities. Variability in locomotor infants’ crawling experience assessed in three distinct ways—days of locomoting to criterion, total strides during play, and maximum crawling speed—failed to predict infants’ intentional action processingortheir tendency to participate in triadic engagement. Moreover, Bayesian analyses revealed consistent positive evidence (ranging from substantial to strong) in favor of the null hypothesis—that increased locomotor experience is not associated with increased intentional action understanding. Thus, as in Study 1, the current findings fail to find support for either direct or indirect effects of the transition to self-locomotion on infants’ intentional action understanding.

4. General Discussion

To better understand the role of motor transitions in infants’ emerging social cognition, we conducted two studies exploring whether and how the experience of self-locomotion impacts the development of intentional action understanding. Two complementary approaches to assessing the role of self-locomotion were used: comparing age-matched locomotor and prelocomotor infants (Study 1) and exploiting the natural variability that exists among locomotor infants (Study 2). Despite considerable evidence supporting the influence of self-locomotion on infants’ social and cognitive development (e.g., Campos et al., 1997, 2000; Clearfield, 2011; Walle, 2016) and prior evidence of specific associations between self-locomotion and failed intentional action understanding (Brandone, 2015), the current studies revealed null results. Specifically, locomotor and prelocomotor infants did not differ in their intentional action understanding (Study 1) and individual differences in infants’ crawling experience failed to predict their intentional action understanding (Study 2). Results also offered no support for the hypothesis that the effects of self-locomotion on intentional action understanding occur indirectly via influences on infants’ triadic engagement with social partners. Specifically, although triadic engagement predicted infants’ intentional action understanding, self-locomotion was not associated with higher levels of triadic engagement, and triadic engagement did not mediate links between self-locomotion and intentional action understanding. Finally, Bayesian analyses produced consistent positive support for the null hypothesis that locomotor experience is not associated with increases in intentional action understanding or triadic engagement.

What explains these null findings? In particular, why did Brandone (2015) find support for the role of self-locomotion in infants’ intentional action processing and the present studies fail to do so? One possibility is that crawling experience does, in fact, influence intentional action understanding and the current studies failed to detect this effect. We argue that this is unlikely for two reasons. First, in contrast to the coarse parent-report measure of self-locomotion used in Brandone (2015), the current studies assessed self-locomotion more thoroughly and systematically—considering several measures of crawling experience, including multiple observed measures. Although these behaviors were observed in the lab rather than a familiar context and thus may not fully represent infants’ typical locomotor behavior, the high levels of variability in observed crawling behavior and the significant correlations among the locomotion measures lend support to the validity of these measures. The fact that crawling status (crawling or not), the number of days of crawling experience, the spontaneous tendency to crawl during play, and maximum crawling speed were consistently unrelated to infants’ intentional action processing is strong evidence against the effect of self-locomotion on this ability.

Second, it is also unlikely that the current studies simply failed to detect an effect of crawling experience given the results of the Bayesian analyses. Bayes factors can be used to evaluate the strength of observed evidence for the alternative hypothesis relative to the null hypothesis (Dienes, 2014; Jeffreys, 1961). Across both studies, Bayes factors consistently fell in the area of positive support for the null hypothesis. Specifically, 7 of the 8 analyses examining measures of self-locomotion and intentional action understanding yielded Bayes factors less than .33— indicating clear positive support for the null. Considered another way, these values show that the observed data are 2 to 16 times more likely under the null hypothesis. Together, these results provide firm evidence against the possibility that we failed to detect an actual direct effect of self-locomotion on infants’ intentional action processing.

It is also important to consider the possibility that the current studies failed to reveal a true indirect effect—that is, self-locomotion impacting intentional action understanding indirectly by way of the changes it promotes in infants’ social interactions. Although we found support for the role of triadic engagement in intentional action understanding (consistent with prior evidence from the same sample; Brandone et al., 2019), we did not find systematic links between self-locomotion and triadic engagement, and Bayesian analyses suggested that the observed data are considerably more likely under the hypothesis that locomotor experience is not associated with more bouts of triadic engagement. Furthermore, despite prior evidence that the link between self-locomotion and intentional action understanding was partially mediated by infants’ tendency to initiate joint attention (Brandone, 2015), we did not find support for significant indirect effects in the current studies. Given established findings showing links between the onset of self-locomotion and changes in infants’ socioemotional development (e.g., Campos et al., 1992, 1997; Gustafson, 1984), it remains possible that self-locomotion facilitates the kinds of social interactions necessary to build an understanding of intentional actions and that the current studies were limited by our measure of triadic engagement. For example, in Brandone (2015), it was parent-reports of infants’ tendency to initiate joint attention (e.g., pointing to direct attention, showing objects) that mediated links between self-locomotion and intentional action processing. In the current measure of triadic engagement, infant- vs. researcher-initiated bouts were not distinguished. Thus, it may be that self-locomotion is related to the more purposeful use of behaviors such as eye contact or gestures to initiate shared attention with a social partner rather than to the tendency to engage in triadic interactions more generally. In addition, assessments in more naturalistic contexts, such as with a familiar caregiver (rather than an experimenter) and in free play on the floor where self-locomotion is possible (rather than seated at table), may provide more precise measures of the kinds of rich interactions with social partners that are linked to crawling experience. Thus, although the current studies cast doubt on this hypothesis, future research should consider further the possibility of indirect effects of self-locomotion on intentional action understanding through triadic social interactions.

Another possible explanation for the inconsistency between the current findings and those reported in Brandone (2015) is our focal emphasis on crawling as the influential form of motor development. The parent-report measure applied in Brandone (2015) included a wider range of gross motor abilities, including pulling to a standing position, cruising or walking with support, and walking independently. We examined crawling, specifically, in the present studies as it represents the most common form of early, successful self-locomotion (Bayley, 1969), has been shown to coincide with a number of other cognitive, social, and emotional advances in infancy (see Campos et al., 2000 for a review), and is temporally aligned with the acquisition of failed action understanding (Brandone et al., 2014). However, other motor milestones may be contributing to the development of intentional action understanding. In particular, both pulling to stand (Atun-Einy, Berger, & Scher, 2012) and cruising (Adolph, Berger, & Leo, 2011) typically emerge in the second half of the first year alongside the development of intentional action understanding. These milestones involve upright postures that enable infants to visually attend to the environment differently, including allowing a wider field of view with increased looking to distal objects and people (Franchak, Kretch, & Adolph, 2018; Kretch, Franchak, & Adolph, 2014). Thus, additional research should consider whether pulling to stand and cruising may have cascading implications for infant’s social engagement and understanding of intentional actions.

A final factor to consider in explaining the discrepancy between the current findings and those reported in Brandone (2015) is the prior study’s reliance on parent reports. Parent-report measures are widely used in infancy research based on both their ease of implementation and the rich information available to parents about infants’ everyday behavior (e.g., Rothbart & Goldsmith, 1985). However, parent-report measures have limitations. Most important for the current discussion is the fact that these reports are filtered through the lens of the parents and thus may represent characteristics and beliefs of the parents as much as descriptions of their infants (e.g., Reznick, 1999; Rubin, Provenzano, & Luria, 1974; Vaughn, Taraldson, Crichton, & Egeland, 1981). Research has documented that, even within the realm of motor development, parents can be biased in their reporting. For example, Mondschein, Adolph, and Tamis-LeMonda (2000) revealed a gender bias in parent reports of infants’ crawling despite no gender differences in actual performance. In addition, there is evidence to suggest that some of the reported changes in infant behavior following the onset of locomotion are more a function of parental perception than actual behavior (Hendrix & Thompson, 2011). Nevertheless, the fact that parental reports did map onto observed measures of failed action understanding in Brandone (2015) suggests that these measures may be capturing some meaningful variability. Perhaps parents are attuned to a key aspect of infant behavior that the current studies are not capturing. Alternatively, these parent-report measures may be tapping into meaningful differences between parents and their interactions with their infants that have implications for the development of intentional action understanding. For example, research suggests that parents differ in their perceptions of infants’ intentionality (Reznick, 1999) and internal experience (Koren-Karie, Oppenheim, Dolev, Sher, & Etzion-Carasso, 2002; Meins, Fernyhough, Fradley, & Tuckey, 2001) and that these differences have implications for how parents assess their infants’ behavioral capacities and how they interact with their infants (Koren-Karie et al., 2002; Meins et al., 2001). Given studies demonstrating links between caregiver behaviors and infants’ action understanding (Hofer, Hohenberger, Hauf, & Aschersleben, 2008; Licata et al., 2014), more work is needed to unpack how parental perceptions and behaviors may influence the development of intentional action understanding around the onset of self-locomotion.

In sum, despite the complementary and systematic ways of assessing self-locomotion in the current studies, we found no clear evidence for the role of self-locomotion in the development of intentional action understanding. These findings cast serious doubt on the possibility of direct effects of crawling on intentional action understanding. However, they also raise further questions about the cascading implications of motor skill acquisition in infancy, including (1) the potential indirect effects of crawling through changes in joint-attentive social interactions, (2) the effects of subsequent locomotortransitions, such as pulling to stand and cruising, and (3) the role of parental eliefs and behaviors around the transition to self-locomotion.

Finally, given the current lack of support for a transformative role of self-locomotion, it is important to consider what kinds of experiences do, in fact, matter for the development of intentional action understanding. Existing research lends support to two types of experience. First, infants’ first-hand experience producing intentional actions influences their understanding of others’ actions (Gallese et al., 2009; Gredebäck & Falck-Ytter, 2015; Woodward et al., 2009). These first-hand effects appear to be highly specific in nature: that is, producing a particular motor behavior (e.g., reaching for an object, placing an object in a container) shapes infants’ understanding of that same behavior in others (Gerson & Woodward, 2013; Sommerville et al., 2005). Second, infants’ experience in triadic interactions in which they coordinate attention between a social partner and an object of mutual interest create unique opportunities for learning about others’ actions (Brandone et al., 2019; Brune & Woodward, 2007; Dunphy-Lelii et al., 2014). Examining how these mechanisms work together to promote infants’ intentional action understanding, as well as whether and how self-locomotion plays a facilitative role, will be important goals for future research.

Research Highlights.

• Self-locomotion has been proposed to impact infants’ understanding of others’ actions.

• Age-matched crawling and non-crawling infants did not differ in intentional action understanding.

• Variability in crawling experience did not predict infants’ intentional action understanding.

• Results cast doubt on the role of crawling in infants’ emerging understanding of intentional actions.