Peekaboo A New Look at Infants' Perception of Emotion Expressions

Author: Montague, Diane P. F.; Walker-Andrews, Arlene S. Source: Developmental Psychology November 2001 Vol. 37, No. 6, 826-838 ISSN: 0012-1649 Number: dev376826 Copyright: For personal use only-not for distribution

Affective communication is an important aspect of an infant's interactions with caregivers and others, and the ability to discriminate expressions is a fundamental part of that communication. Parents and other caregivers spend much time in face-to-face interactions with their young infants, beginning immediately after birth. These interactions are predominantly affective interchanges; therefore, one might expect discrimination of affective expressions to occur early on. By the second half of the first year, emotion discrimination seems well established (Nelson, 1987); however, whether younger infants have the ability to discriminate emotion expressions is still under debate. In part the debate has continued because of differences across studies in methodology, including differences in procedures, selection of stimulus materials, measures, and the precise questions being asked (Walker-Andrews, 1997). Initial research on emotion perception with 4- and 5-month-old infants failed to demonstrate discrimination of expressive behaviors (e.g., Charlesworth & Kreutzer, 1973; Spitz & Wolf, 1946). In brief, in the study described by Charlesworth and Kreutzer (1973), 4-, 6-, 8-, and 10-month-old infants were videotaped while viewing a live presentation of several expressions. Only global ratings of attention (e.g., "attends," "attends briefly," or "alternates attention"), activity ("inactive," "still," "moderately active," or "very active"), and affect (positive or negative) were collected. Not until 6 months of age did infants respond differentially to the posed expressions. Spitz and Wolf (1946) examined infants' smiles in response to facial expressions either presented by an experimenter or portrayed on a mask. These authors found no evidence that infants ranging in age from 2 to 6 months could differentiate the expressions.

As additional methods for exploring the development of perception in infancy have emerged, researchers have incorporated them into the study of affective perception by young infants. One technique for assessing infants' perception, the visual preference method, has been used by a number of researchers (e.g., Barrera & Maurer, 1981; Kuchuk, Vibbert, & Bornstein, 1986; LaBarbera, Izard, Vietze, & Parisi, 1976). Typically, in this procedure, experimenters show infants two displays, either side by side or sequentially, and monitor infants' looking time at each display. In one of the earliest studies that used this technique to assess infants' discrimination of expressions, Wilcox and Clayton (1968) showed 5-month-old infants films of static or moving facial expressions sequentially. In the first experiment, infants looked longer at smiling expressions, but in a second experiment they looked longer at moving facial expressions regardless of the emotion portrayed. In another study, Kuchuk et al. (1986) asked whether 3-month-old infants could discriminate various intensities of happy expressions. These authors found that infants looked longer at more pronounced smiles than at neutral faces. Similarly, Schwartz, Izard, and Ansul (1985) showed 5-month-old infants a photographed facial expression during a familiarization interval and then paired this now-familiar facial expression with a novel expression at test. In one experiment, infants discriminated pairs of negative expressions (fear, anger, and sadness), as indicated by their longer looking at the novel expression. In a second experiment, infants failed to show consistent discrimination of negative (anger) and positive (interest and enjoyment) expressions, leading the authors to question the efficacy of this procedure for studying emotion perception, which is discussed further later in this article.

Another technique for assessing the discrimination by infants of various exemplars of facial expressions is the visual habituation procedure (e.g., Horowitz, 1975; Peeke & Herz, 1973). Field et al. (1983), for example, reported discrimination by newborns of live facial expressions presented by an experimenter who actually held the infants while portraying an expression. There is evidence that by 3 months of age infants can discriminate photographed surprised and happy expressions when a visual habituation procedure is used (Young-Browne, Rosenfeld, & Horowitz, 1977). More recently, Serrano, Iglesias, and Loeches (1992, 1995) reported that infants as young as 4 months of age could discriminate and generalize across the angry, happy, and neutral photographed expressions of several female models. Similarly, A. J. Caron, Caron, and MacLean (1988), using an infant-controlled habituation procedure, reported that infants at 4 and 5 months discriminated videotaped sad expressions from happy ones.

Although the results obtained with these looking-time methods suggest that infants younger than 6 months may discriminate emotion expressions, there are inconsistencies in the patterns of results both within and across studies. Given these anomalous findings, some investigators have concluded that (a) the discrimination of emotions is tenuous (Nelson, 1987), (b) infants are making discriminations based on nonaffective information such as featural differences (R. F. Caron, Caron, & Myers, 1982), or (c) affect-specific information intrinsic to each emotion modifies infants' responsiveness in these procedures (e.g., Schwartz et al., 1985; Serrano et al., 1995). Patterns of results that contribute to the ambiguity include the following. First, many researchers have found that some expressions are discriminated only when they are presented in specific orders (e.g., A. J. Caron et al., 1988; Serrano et al., 1995; Young-Browne et al., 1977). For example, A. J. Caron et al. (1988) reported that infants discriminated sad from happy expressions only when they were first familiarized with the sad expression. In addition, infants fail to discriminate some emotion pairs. A. J. Caron et al. (1988) found that happy expressions were discriminated from sad expressions but not angry expressions. In contrast, Young-Browne et al. (1977) found that 3-month-olds did not discriminate happy from sad expressions, although they were successful discriminating happy from surprised expressions. Infants may also respond to nonaffective information, such as featural differences or other extraneous information. For example, as indicated above, Field et al. (1983) reported that newborns could visually discriminate facial expressions, but in this study the infants may have responded to confounding haptic information because the experimenter held the infant. The influence of nonaffective information is exacerbated when static expressions are used. R. F. Caron, Caron, and Myers (1985) showed convincingly that infants would use featural differences such as the shape of the mouth or the presence of teeth to discriminate angry and happy expressions, rather than respond to the emotions themselves, when such featural information was available (see also Kestenbaum & Nelson, 1990).

In addition, any intrinsic meaning of expressions may influence results in other ways. That is, although the infants in the R. F. Caron et al. (1985) study failed to respond to affective information in making discriminations, in other studies, infants' responses may have been driven by the affect-laden information. In particular, Schwartz et al. (1985) discussed the importance of choosing an appropriate paradigm when examining responses to different emotion expressions. These authors showed 5-month-old infants a photographed facial expression for familiarization and then paired the familiar facial expression and a novel expression at test. Infants did not look preferentially in five of six comparisons. Rather than concluding that infants failed to discriminate expressions, Schwartz et al. speculated that infants' looking was influenced by some aspect of a positive emotion expression and by the "relative aversiveness" of an expression such as anger and that these led to specific looking preferences. They concluded that the methods typically used in infant perception work "may underestimate the infant's ability to detect differences between stimuli of widely differing social- or emotion-signal value" (Schwartz et al., 1985, p. 76). Others (e.g., Nelson & Dolgin, 1985) have made similar arguments in accounting for order effects in studies that used visual habituation, suggesting that the emotional valence of the first expression influences patterns of looking at the second expression.

Moreover, as Schwartz et al. (1985) pointed out, investigators use differing criteria for defining expressions. This point raises important questions with respect to the validity of the posed emotion expressions that are used in research and highlights the importance of objectively defining and ensuring consistency in expressive displays within and across studies. A sad expression, for example, is not merely a frown-it entails a full configuration of the face. Ekman and Friesen (1975) described a universal sad expression as having distinct muscle characteristics in the brows (inner corners raised with a triangular furrow), the eyes (small eyes, raised lower lid, raised inner corner of upper lid), and the mouth (corners drawn down or trembling). The challenge for the investigator is to present affective expressions that are both accurate (i.e., meet standards of emotion coding; Ekman & Friesen, 1975; Izard, 1979, 1995) and ecologically valid (i.e., are representative of the infant's experience). In most cases (e.g., Serrano et al., 1995; Young-Browne et al., 1977), researchers have used unfamiliar, static stimulus materials such as schematic faces and photographs-typically because such materials allow more stimulus control. But such materials are not representative of the infant's everyday experience. Furthermore, static displays do not contain dynamic qualities of expressions such as tempo (Fogel, 1993), nor do they provide intermodal correspondences. Information that specifies emotion is found not only in the facial musculature but also in the voice and body gestures (R. F. Caron et al., 1985; Walker-Andrews, 1997).

Finally, some researchers have emphasized the importance of using multiple measures to examine infant's emotion perception (e.g., Walker-Andrews, 1988, 1997). In addition to studies that have used visual attention measures, some researchers have supplemented the looking-time data with measures of affective responsiveness. For example, Serrano et al. (1995; see also Termine & Izard, 1988) had an observer determine whether infants displayed either positive or negative behaviors in response to happy and angry facial expression slides. Serrano et al. reported that infants displayed more positive behaviors (smiling, leaning toward) in response to happy than to angry expressions and more negative behaviors (avoidance movements and precries) in response to angry than to happy expressions. Kahana-Kalman and Walker-Andrews (2001) looked at infants' affective responses to their mothers' and a stranger's affective expressions and found that 3.5-month-olds were more affectively expressive in response to their own mothers. It should be noted, however, that the number of investigators who have used both looking-time and affect measures is small and that in most cases, the infants were responding to affective expressions presented in photographs or videotapes rather than to live presentations.

Toward a New Methodology

If we are to expand and refine our knowledge of the capacity of young infants to perceive others' emotion expressions, greater definition, converging methodologies, and more naturalistic approaches are needed (Walker-Andrews, 1988, 1997). To this end, in the present study we used a peekaboo game to assess infants' perception of others' expressive behaviors. The strategy was to introduce emotion expressions in a game familiar to infants in the targeted age group (younger than 6 months) and to allow for the presentation of carefully posed expressions in a naturalistic context.

A number of researchers (e.g., Fernald & O'Neill, 1993; Fogel et al., 1997; Greenfield, 1972; Parrott & Gleitman, 1989; Rochat, Querido, & Striano, 1999) have studied peekaboo, examining both the structure of the game and the development of infants' social and cognitive abilities. Peekaboo is a game in which one member of a dyad hides and reappears (Gustafson, Green, & West, 1979), a natural occlusion study. Studies suggest that the earliest version of peekaboo is a type of looming game in which the adult looms closer to the infant and is always in view (Fernald & O'Neill, 1993). This form may begin at about the 3rd month. The prototypical hiding version of the game, in which the caregiver disappears and reappears, seems to begin approximately in the 5th month. The hiding component usually is accompanied by the caregiver's vocalizations to the child, and reappearance usually is paired with a "peekaboo" verbalization.

Several features of the peekaboo game make it a viable paradigm for use with infants younger than 6 months. First, it is a game with which babies become very familiar-a source of universal delight for mothers and infants of many diverse cultures (Fernald & O'Neill, 1993). Because parents frequently play this game with their infants, it provides an opportunity to present exaggerated, prototypical expressions in a familiar context. Second, the peekaboo game is an effective elicitor of infant attention (Fernald & O'Neill, 1993) and enjoyment (Sroufe & Waters, 1976), so it is likely to engage the infant and elicit a response. Finally, infants develop specific expectations about how the game is played (Bruner & Sherwood, 1976). For example, Parrott and Gleitman (1989; see also Charlesworth, 1966) varied the location and the identity of the reappearing person. Results showed that infants' enjoyment of the game was influenced by violations of expectations. Similarly, Rochat et al. (1999) examined 2-, 4-, and 6-month-old infants' responses to a peekaboo game that was presented in either its typical fashion or in a scrambled order. Infants at 4 and 6 months responded differentially to the disorganized and organized peekaboo games. Specifically, infants at both 4 and 6 months smiled more and gazed less at the organized game; 2-month-olds did not show sensitivity to modifications in the game, although they smiled and interacted with the adult stranger. Beginning at about 4 months, then, infants have attained social expectancies such as those present in a peekaboo interaction. Thus, the peekaboo paradigm provides a context that is highly naturalistic and one in which babies often participate. To reiterate, it captures their attention, elicits enjoyment, and produces specific expectations.

An additional strength of the peekaboo game is its incorporation of both visual and auditory modalities. Infants' perception of emotion expressions seems enhanced by multimodal presentations (e.g., Soken & Pick, 1999), and beginning at as young as 7 weeks of age, infants attend more to faces accompanied by speech (Haith, Bergman, & Moore, 1977). With respect to peekaboo, Greenfield (1972) suggested that the auditory component provides information that guides the infant's attention. She observed that 4-month-old infants displayed different response patterns to auditory/visual cues than to visual cues alone while engaged in peekaboo in an unfamiliar setting. Similarly, according to Fernald and O'Neill (1993), the "vocal melody" that accompanies the peekaboo game captures infants' attention and influences their emotions. Thus, particularly for young infants, vocalizations that accompany visual displays during peekaboo enhance infants' responsiveness.

Recently, a number of investigators have used peekaboo to examine infants' affect regulation. Stifter and Moyer (1991) looked at 5-month-olds' gaze aversion coupled with their positive affect during a peekaboo game with their mothers. Gaze aversion was associated with infants' smiles in this study-infants who exhibited more high-intensity smiles during the mother peekaboo interaction averted their gaze more often than did those infants who exhibited low-intensity smiles during the game. These authors also concluded that a moderately active peekaboo play style is optimal for creating high levels of positive arousal, given that moderately active mothers elicited more smiles of longer duration and intensity as well as more gaze aversion than did other mothers in the study. Eckerman, Hsu, Molitor, Leung, and Goldstein (1999) used a standardized peekaboo game enacted by an experimenter to determine whether very low birth weight infants differed from healthy full-term 4-month-olds in their arousal in a social setting. The very low birth weight infants showed less positive arousal and more negative arousal than did the full-term infants. The findings demonstrated that the differences found among infants occurred because of differences in the way preterm versus full-term infants respond to the same forms of social stimulation rather than differences in the ways in which parents interact with their infants. As Eckerman et al. (1999) concluded, "The use of a standardized protocol of stimulation with a trained examiner enabled us to establish clear differences in infant arousal to the same stimulation" (p. 290).

In the present study, we adapted the peekaboo paradigm to look at infants' perception of emotion. We targeted 4-month-olds for two reasons: First, there is wide agreement that infants discriminate emotion expressions after 6 months of age (Nelson, 1987), but the findings regarding emotion perception by infants at 4-5 months are ambiguous (e.g., Schwartz et al., 1985; Walker-Andrews, 1997). Second, infants seem to appreciate the rule-based aspects of the peekaboo game beginning at about 4 months (Rochat et al., 1999). Thus, we anticipated that this naturalistic context would provide a sensitive measure of young infants' discrimination of emotion expressions. Our aim in the present study was to determine whether infants detect changes in expressions embedded in a game of peekaboo and respond to them in differential ways. Four interrelated predictions were made, two focusing on looking time and two related to affective responsiveness. We included measures of looking and affective responsiveness by infants, and both between-groups and within-group differences were assessed.

First, we hypothesized that if infants discriminated a change in expression, one unexpected within a game of peekaboo, they would alter their looking patterns on the trial in which a change occurred. Second, given that expressions differ in social-signal value (e.g., Schwartz et al., 1985) or may induce differential emotional states (e.g., Haviland & Lelwica, 1987), we predicted that if infants recognized angry, fearful, and sad expressions as meaningful events, they would display different amounts of looking at each, beyond any response to a change. Specifically, we anticipated that infants would increase their looking time for fear and anger (e.g., Nelson & Dolgin, 1985; Rochat et al., 1999) and decrease it for sadness (e.g., A. J. Caron et al., 1988; Termine & Izard, 1988). Third, we anticipated that the violation of the typical peekaboo game introduced by the change in affect would result in different affective behaviors for infants who received a change in expression on the target trial. Specifically, we hypothesized that infants who received a change in expression would display more interest/surprise, greater lability of affect, and more varied expressiveness than would infants who continued to receive the typical expression (cf. Haviland & Lelwica, 1987; Kahana-Kalman & Walker-Andrews, 2001). Fourth, we expected different patterns of affective responsiveness across trials depending on the valence or meaning of the discrete emotion expression on the first change trial. Specifically, we anticipated that infants within a change group would display increased interest/surprise, lability of affect, and variability in expressiveness on trials in which a change in expression occurred, whereas infants in the no-change condition would show reductions on these measures. Finally, because others have reported that infants sometimes match the expression of an experimenter (see Field et al., 1983), we planned to examine any patterns in affective responsiveness that might reflect a matching or imitative response.

Method

Participants

Forty infants (23 boys and 17 girls) participated in the study at 4 months of age (mean age = 129 days, SD = 10.70). Data from 10 additional infants were eliminated because of infants' fussiness or equipment failure. All infants were full term at birth (defined as having a birth weight greater than 2,500 g and a gestational age of 38 weeks or greater based on maternal report). Infants had no apparent visual or auditory deficits and were in good health at the time of testing. Most of the infants were Caucasian (38 White and 2 Hispanic). Infants were recruited through the use of published local birth announcements and by referral. Mothers were informed with respect to the procedure and goals of the study and were asked to sign the consent form prior to testing. Sessions were scheduled at a time when the infants were reported by their mothers to be alert and playful. Infants were observed either in their respective homes (n = 24) or in the laboratory (n = 16), depending on the parent's preference.

Procedure

Prior to the study, we observed each mother playing peekaboo with her infant to assess whether each infant was familiar with the prototypical hiding version of peekaboo to be used in the study. Each mother was using this form of hiding (i.e., covering her own face), so no infants were disqualified because of this restriction.

Eight trials were presented, two of which were emotion-change trials (Trials 4 and 8). The trial sequence was as follows: three typical happy/surprised peekaboo trials (baseline Trials 1-3), followed by the target (emotion-change) trial (Trial 4). This sequence was repeated over the next four trials (Trials 5-7 were typical happy/surprised trials, and Trial 8 was the target emotion-change trial). We adapted this procedure from Parrott and Gleitman (1989) to maximize the infants' potential for detecting a change from the familiar peekaboo expression and to allow time for the infant to become reestablished in the game after a switch. Infants were randomly assigned to one of four emotion groups (10 infants per group) for the target trials: the sad, anger, fear, or continued happy/surprised (alternately referred to as happy) groups. The happy/surprised blend (i.e., smile with raised brows) was selected because that is the expression generally displayed during peekaboo play.

Each infant was placed in an infant seat on a table facing the seated experimenter. The infant's head was approximately 45 cm from the experimenter's face. The experimenter placed her chair in front of a blank wall or white curtain so that no distracting objects were in view behind her. For both the laboratory and home visits, the only additional person in the room was the infant's mother, who sat behind the infant.

Prior to the start of the experiment, the investigator established eye contact with the infant and called the child's name to vocally invite participation. The experimenter's facial and vocal expressions and the infant's behaviors were recorded independently with two videorecorders. For each 10-s trial, there were two phases during which the experimenter (a) covered her face for 3 s by holding up a red cloth and called the child's name (hiding phase) and (b) reappeared for 7 s with an emotion expression and its affectively matched vocalized "peekaboo" expression (presentation phase). The expression was held for 7 s with no change in intensity or affect. In this respect it was similar to a still-face procedure; however, the still-face phenomenon reflects infants' responses to cessations of interactions for much longer durations (> 1 min; e.g., Gusella, Muir, & Tronick, 1988). Verbalizations were consistent across trials and participants. During the hiding phase, the experimenter said, "Where's [child's name]?" During the presentation phase, only the word "peekaboo" was spoken. There were no delays between trials-each hiding phase immediately followed each presentation phase of the previous trial. The experimenter wore a miniature earphone that provided a signal to initiate a trial so as to ensure timing accuracy.

To ensure the consistency and accuracy of facial expressions, the experimenter was trained in both Izard's (1979, 1995) Max system (described in the next section) and Ekman and Friesen's (1975) technique for portraying emotion expressions with specific muscle movements. In addition, the experimenter positioned a small mirror (5 cm × 7 cm) behind the cloth for use during the hiding phase. This permitted configuration of the expression so as to avoid the potential for transitional expressions at the onset of presentation. A second mirror (5 cm × 7 cm) was attached to the infant seat just behind and at the top of the infant's head for use by the experimenter in controlling the precision of the target expressions during the 7-s presentation phase. To maintain the integrity of each expression, the experimenter looked at her mirror image rather than at the infant on all trials in all conditions. Therefore, the experimenter was looking approximately 8° above the infant's eyes. Research indicates that infants are sensitive to direction of gaze, but most studies use much larger deviations from 0 degrees at longer durations (e.g., Hains & Muir, 1996, used 20° and 40° over 1-min periods). Smaller vertical shifts in gaze have not produced the same effect. For example, Symons, Hains, and Muir (1998) found that infants were not sensitive to 5° vertical shifts (i.e., looking at the top of the infant's head or looking at the chin). Finally, the experimenter's presentations were videotaped and later coded for fidelity of expressions. For the vocal expression, the experimenter was asked to say "peekaboo" in a tone of voice that matched the emotion in the facial expression for each trial. These vocal expressions were later coded for fidelity and intensity of emotion.

Measures and Coding

We assessed both the experimenter's expressions and the infants' responses to those expressions. For the experimenter, we examined both facial and vocal expressions to determine their accuracy. For the infant, we examined several behaviors: looking time, frequency of discrete emotion expressions, lability of affect, and facial expression time. Each of these is discussed below.

Experimenter's Expressions

To assess the fidelity of the experimenter's facial expressions, two individuals coded tapes using Izard's (1979, 1995) Maximally Discriminative Facial Movement Coding System (Max), which distinguishes eight fundamental emotions: joy, interest, surprise, anger, disgust, contempt, fear, and sadness/distress. Rather than using global and subjective labels to identify each expression, this system uses numerical codes that correspond to specific muscle movements in the brow, eye, and mouth regions of the face. As required, each coder viewed each of the selected facial expressions three times in order to code each region of the face separately. For example, a fear expression would be characterized by the codes 22 (brow) + 31 (eyes) + 53 (mouth). A primary observer coded all the tapes; a second observer coded 25% of the tapes (90 of 360 posed expressions) for reliability. There was 100% agreement between coders for all of the numerical codes for all of the facial expressions. Further, for each modeled expression, the numerical codes met the strict criteria specified in the coding manual for defining the targeted emotions (Izard, 1995, pp. 16-17). The codes for the exact muscle movements used to portray the emotions for each brow, eye, and mouth region, respectively, were as follows: sadness = 23 + 33 + 56, anger = 25 + 33 + 54, fear = 22 + 31 + 53, happiness/surprise = 20 + 30 + 52. Therefore, the experimenter's expressions were judged to be accurate depictions of happiness/surprise, anger, fear, or sadness.

To assess the validity of the experimenter's vocal expressions, naive raters judged the emotion type and intensity of the vocalizations for each emotion. Specifically, the vocalizations heard by the infants on the target trials were randomly assembled onto an audiocassette tape. This tape was then coded in a classroom setting by 50 undergraduate students who were naèive to the design of the study and were not trained in emotion coding. Raters were asked to (a) identify each emotion heard on the tape by selecting one of four labels (happy, scared, angry, sad) for each and (b) rate the intensity of each vocal expression using a 5-point scale (1 = low intensity; 5 = high intensity). Overall, raters accurately identified each vocal expression. Across target trials, the appropriate label was selected, on average, by 91% of the raters (range = 77% to 99%). The intensity ratings were comparable for all emotion types (range = 3.03 to 3.40), F(3, 30) = 0.536, p = .66.

Infants' Looking Time

Each infant's looking times were coded from the videotapes. Looking time was calculated as the total amount of visual attention (in seconds) to the experimenter's facial expressions during each 7-s presentation phase, beginning at the point of reappearance and ending when the experimenter's face was hidden. Because live presentations may vary, we calculated the precise durations of all trials. They ranged from 6.2 s to 7.3 s; therefore, only data from the first 6.2 s of each trial were used in the looking-time analyses. For the actual coding, videotapes were copied and edited to remove the auditory track, and coders were naive with respect to emotion changes. Training of coders involved their viewing and coding several videotapes from our existing library of infants' responses prior to coding those from this study. A primary observer, naive to the design of the study, coded the data for all of the infants. Two additional observers coded the data for reliability only. The average correlation between all three raters was high (Pearson r = .93). In terms of actual looking-time scores, no difference between any two raters for an infant's looking time across all trials was greater than 1 s (range = 0.00-0.91 s).

Infants' Affective Expressions

Each infant's affective facial expressions were coded with the Max system to determine whether infants responded differentially to the change in the peekaboo expressions. Two observers coded the infants' expressions: The primary observer coded 100% of the targeted trials, and a second observer independently coded 33% of the targeted trials to assess reliability. Percentage agreement was calculated for movement codes in each region of the face for each baby. Overall, coders achieved 92% agreement on the movement codes. The ranges for each region of the face were as follows: brow region = 92%-98% (M = 95%); eye region = 79%-100% (M = 91%); mouth region = 80%-100% (M = 92%).

We examined the data obtained with the Max system in terms of both the discrete emotions portrayed by the infants and the changes in expression captured by the measures of lability and facial expression time. For the discrete emotions, we focused on infants' interest and surprise expressions as responses to the violation of the typical peekaboo sequence, and we also examined the data for evidence of infants' matching of the portrayed expressions. The interest/surprise measure was obtained by totaling the number of codes that corresponded to interest and surprise in the Max system. Matching was defined as the frequency with which the infant's expression contained affect codes that corresponded to the emotion expressed by the experimenter. The two additional measures of affective responsiveness were lability and facial expression time. Lability was obtained by totaling the number of movement codes assigned by an observer during a 7-s target trial. This indicated the number of times a baby's expression changed during the interval. For the measure of facial expression time, we calculated the percentage of the coding interval in which an infant showed a facial expression different from neutral (e.g., Kahana-Kalman & Walker-Andrews, 2001). This measure was obtained using the codes derived from the brow region (for 4 infants, the brow area was obscured during part of the interval, so these infants were not included in this portion of the analyses).

Therefore, each infant's responses to the peekaboo expressions presented during target Trials 4 and 8 were compared to his or her response during Trial 3 (the trial prior to the introduction of the first change in expression) to produce difference scores for within-subject measures. In this way each baby served as his or her own baseline. In addition, responses were compared across groups to yield between-subjects measures.

Results

The major questions motivating the present research were (a) whether young infants would discriminate emotion expressions that were embedded in a familiar context and (b) whether infants would respond differentially to expressions depending upon the emotional valence. We examined these questions using two types of data, looking time and affective responsiveness, as described below.

Looking Time

Overall Attentiveness

Throughout the session, infants looked at the experimenter's facial expressions for about two thirds of the time available. Across groups, the amount of time spent looking at the experimenter's facial/vocal expressions averaged 66% on the first trial and 61% on the last trial. Preliminary analyses were conducted to assess the effects of sex, setting (home or laboratory), and age of the infant (age in days by median split). A 2 (sex) × 2 (setting) × 2 (age) analysis of variance (ANOVA) revealed no significant effects (ps > .10).

In addition, before comparing the separate measures of infants' looking time and affective responsiveness, we explored the relationship between these measures. That is, we conducted two analyses, one for each target trial. The amount of looking time, the number of interest/surprise codes, the number of movement codes (lability), and the number of non-neutral codes (facial expression time) were correlated for Trial 4 and Trial 8. We found that for Trial 4, the measures of affective responsiveness were significantly correlated: interest/surprise with lability, r = .51, p < .001; interest/surprise with facial expression time, r = .53, p < .001; and lability with facial expression time, r = .36, p < .025. Similarly, lability and facial expression time were correlated with interest/surprise on Trial 8 (r = .40, p < .012 and r = .40, p < .016, respectively). The relationship between lability and facial expression time was not significant (p < .34). Looking time was not correlated with any of the affective measures across Trials 4 and 8.

Visual Discrimination of Expressions

Our first hypothesis was that if infants discriminated a change in expression, they would alter their looking time on the target trial. Therefore, to determine whether infants looked differentially on the first target trial (Trial 4), a 4 (emotion) × 3 (Trials 3, 4, and 5) ANOVA with repeated measures on trial was conducted. A significant quadratic effect was found for trial, F(1, 36) = 8.53, p < .006, and a main effect was found for emotion, F(3, 36) = 4.34, p < .01. No significant interaction was found. Means and standard deviations for looking times on Trial 4 were as follows: happy, M = 3.3, SD = 1.62; sad, M = 3.67, SD = 2.14; fear, M = 5.39, SD = 1.07; and anger, M = 5.2, SD = 1.35. To determine the extent of differences in looking time related to emotion, we performed pairwise comparisons (one-tailed). These revealed significant differences in looking time on the target trial (Trial 4) for happiness versus fear (p < .002), for happiness versus anger (p < .006), for sadness versus anger (p < .036), and for sadness versus fear (p < .018). In each case, looking time in response to anger and fear was greater than that in response to either sadness or happiness.

Affect-Specific Patterns of Looking

We also hypothesized that infants would show different looking patterns in response to the emotion expressions that would be related to the valence or meaning of those expressions. Haviland and Lelwica (1987) and Schwartz et al. (1985) argued that infants' responsiveness to expressions may be influenced by the signal value of the emotion depicted. That is, infants might look less at a sad expression than at a happy expression because they prefer happy expressions. In addition, some studies (e.g., Nelson, Morse, & Leavitt, 1979; Walker & Grolnick, 1983) have found order effects related to emotion. Such order effects may indicate that infants' looking time on trials subsequent to a particular expression is influenced by the valence of that expression. For that reason, in the present experiment, infants were provided two target (change) trials (Trials 4 and 8). This design allowed us to compare looking times on the happy/surprised trial (Trial 3) with looking times on the first target trial (Trial 4). It also allowed us to examine looking times on subsequent trials to determine whether there were any carryover effects from the first target trial.

To assess specifically whether infants showed order effects that could be linked back to the emotional valence of a target trial, as Nelson and others have suggested (e.g., Nelson et al., 1979), we performed additional analyses. We conducted a 4 × 8 repeated measures ANOVA with emotion as the between-subjects factor and trial as the repeated measure. The interaction between emotion and trial was significant, F(3, 36) = 5.58, p = .003. Therefore, trend analyses were conducted separately for each emotion category for all trials. For infants who observed only the typical happy/surprised expression throughout the experimental session, no trial effects were found (ps > .10). For infants who observed angry expressions on the target trials, there was a linear trend, F(1, 9) = 8.59, p = .017; their looking time increased on the first target trial and on subsequent trials. Those infants who saw fearful expressions on the target trials showed a significant quartic trend, F(1, 9) = 5.05, p = .051; their looking time increased on the first target trial, decreased on typical Trials 5-7, and increased again on target Trial 8. The infants who observed sad expressions on the target trials showed a linear trend, F(1, 9) = 5.66, p = .041; their looking time decreased on the first target trial and thereafter. Figure 1 depicts the significant patterns of looking time for the change groups. These patterns suggest that infants modified their looking time across subsequent trials in response to the valence of the emotion presented on the first target trial.

Affective Responsiveness

Infants' affective responsiveness was examined for type of affect (i.e., interest/surprise and "matching" expressions), for lability, and for facial expression time. To assess overall group differences, we first conducted 4 (emotion) × 2 (target Trials 4 and 8) ANOVAs. In addition, given that we anticipated that infants who received a change in emotion would increase their reactivity compared to that of infants who continued to view the typical sequence, we conducted planned comparisons using t tests (one-tailed) between the no-change group and each change group. Finally, to address the hypothesis that infants' responsiveness would be influenced by the valence of a particular emotion presented on the target trials, we conducted matched-sample, one-tailed t tests using data from Trials 3, 4, and 8.

Interest/Surprise

Group differences.

To assess whether infants expressed different amounts of interest/surprise according to group, a 4 × 2 repeated measures ANOVA was conducted on the difference scores that were calculated for the number of codes corresponding to interest/surprise facial expressions. A marginal group effect was found, F(3, 36) = 2.71, p = .058; no other effects were significant.

To test the hypothesis that only infants in the change groups would display increased interest/surprise expressions in response to the violation of expectancy, we conducted planned comparisons between the no-change group and each change group using the difference scores. These revealed significant differences between the happy (no-change) and sad groups for both the first target trial, t(18) = 2.04, p = .028, and the second target trial, t(18) = 3.25, p = .002; between the happy and fear groups for the first target trial only, t(18) = 2.25, p = .019; and between the happy and anger groups for the first target trial, t(18) = 1.99, p = .031. Infants who were presented a change on the target trials showed larger increases in the frequency of interest/surprise codes than did infants who continued to view happy expressions. Table 1 provides the actual means and standard deviations by group for the interest/surprise codes as well as the difference scores.

Trial differences.

For the within-group measure, we tested the hypothesis that infants in each of the change groups would display an increase in interest/surprise when they viewed the change in expression. Infants in the no-change group, however, were predicted to show a decrease in interest/surprise. In general, infants assigned to the change groups maintained or increased their expressions of interest/surprise on the target trials (see Figure 2). The analyses revealed that infants in the happy (no-change) group expressed decreased interest/surprise on Trial 4, t(9) = −2.12, p = .032, and on Trial 8, t(9) = −2.41, p = .020, which suggests that infants shown continued happy expressions across the duration of the session lost interest in the happy expression. In contrast, infants in the sad group expressed increased interest/surprise, especially on Trial 8, t(9) = 2.23, p = .027. No significant differences in interest/surprise expressions were found for the anger or fear groups.

Matching of Discrete Emotions

Given that infants have been found to match the facial expressions portrayed to them in some studies (e.g., Haviland & Lelwica, 1987, but see Serrano et al., 1995), we examined the Max codes to determine whether infants in a group displayed the same emotion portrayed by the experimenter on the target trials. As a first step, we counted the number of infants in each group who showed any component (eye, brow, or mouth) of the emotion expression depicted by the experimenter. Few discrete expressions were observed. In addition, there was no strong evidence of matching: 2 of 10 infants in the sad group displayed a sad brow (Code 23); no infants in the fear group or in the anger group displayed any matching codes. For the happy (no-change) group, 6 of 10 infants displayed matching components; however, happy expressions were shown by all infants in all groups, as would be expected in a social interaction (e.g., Tronick, Ricks, & Cohn, 1982). Specifically, 4 sad-group, 5 fear-group, and 7 anger-group infants showed components of the happy expression on the target trials.

Lability

Group differences.

To investigate whether, overall, infants showed differential amounts of affect lability, we conducted a 4 × 2 ANOVA with repeated measures on the difference scores calculated for number of movement codes. No significant differences were found.

To determine whether infants who received a change in expression displayed greater amounts of lability than those who did not receive a change, we conducted planned comparisons. Significant differences were found between the happy and sad groups on both the first target trial, t(18) = 2.62, p = .009, and the second target trial, t(18) = 2.09, p = .025. Infants who received a sad expression on the target trials displayed more movement changes than did infants who continued to receive the happy/surprised expression. No other between-group differences in lability were found.

Trial differences.

To determine whether infants showed increased amounts of affect lability across the session, we conducted matched-sample t tests. Significant effects for increased lability were found for infants who observed sad expressions on Trial 4, t(9) = 2.44, p = .019, and on Trial 8, t(9) = 2.64, p = .014 (see Figure 3). No other significant differences in lability were found.

Facial Expression Time

Group differences.

To determine whether infants responded with different emotion expressions (i.e., other than neutral), we conducted an analysis using the facial expression time. A 4 × 2 ANOVA with repeated measures was conducted on the difference scores. There was a main effect of group, F(3, 32) = 3.28, p = .035.

To assess the hypothesis that infants in the emotion-change groups would display greater amounts of facial expression time than would infants in the no-change group, we conducted planned comparisons on the difference scores. In general, all groups who received a change in expression on the target trials showed greater expressiveness than did the happy (no-change) group (see Figure 4). Differences were found between the happy and anger groups on Trial 8, t(17) = 2.41, p = .014; between the happy and sad groups on Trial 4, t(15) = 1.99, p = .033; and between the happy and fear groups on Trial 4, t(16) = 1.94, p = .035, and on Trial 8, t(16) = 2.18, p = .022. Table 2 provides the difference scores and the actual means and standard deviations by group for facial expression time.

Trial differences.

To determine whether infants showed greater amounts of expression time across the session, we conducted matched-sample t tests on the difference scores for each group. Infants in the anger group were more expressive on Trial 8, t(9) = 1.98, p = .04. No other significant differences in facial expression time were found.

Discussion

These data indicate that infants as young as 4 months detect changes in facial/vocal expressions and respond to these emotion expressions in meaningful ways. When infants observe naturalistic emotion expressions in a familiar context, they demonstrate an early sensitivity to emotion. We were interested in addressing two broad questions about infants' perception of others' expressions. The first question was whether infants would discriminate expressions, as indicated by changes in looking time in response to changes in expressions. The second, and more complex, question was whether infants would perceive affective information inherent in each of several discrete emotions, especially as emotions function in a social interaction. Answers to this latter question derive from patterns of both looking time and affective responsiveness. We addressed these larger questions with four interrelated predictions: (a) that infants would show different amounts of looking in response to a change in expression; (b) that they would show different amounts of visual attention to each expression, with increases for fear and anger and a decrease for sadness; (c) that infants in the change groups, compared to those in the no-change group, would show increases on measures of affective responsiveness; and (d) that infants within each group would modify their affective responsiveness on the target trials, relative to their baseline. Each of these predictions was supported.

In accordance with our first prediction, infants looked for different amounts of time on the target trial than on the prior and subsequent trials, indicating that they perceived a change in emotion. This finding, that infants as young as 4 months of age can discriminate expressions, is consistent with the work of others who have demonstrated emotion discrimination in infants younger than 6 months of age in naturalistic situations (e.g., Field et al., 1983; Haviland & Lelwica, 1987).

With respect to infants' recognition of emotion per se, measures of the infants' looking time and affective responsiveness converged. Different looking patterns were observed for each emotion. Those infants who viewed a sad expression on the first target trial decreased looking time across the remaining trials. In contrast, infants who observed an angry expression increased looking time on the target trial and thereafter. Those who viewed a fearful expression increased looking time only on the two target trials. Different patterns of affective responsiveness were also observed. Affective responsiveness was assessed with measures of interest/surprise, lability, and facial expression time. As expected, infants showed different patterns in their expressiveness in response to the different emotion displays. Specifically, they displayed less interest/surprise in the happy (no-change) condition than in the change conditions. Moreover, expressions of interest/surprise decreased across the trials for the happy group, which suggests that infants became less attentive and interested in the continuation of the typical happy/surprised peekaboo sequence. In contrast, infants in the other groups tended to maintain or increase their expressions of interest/surprise. There was a significant increase in interest/surprise for the sad group by the second target trial.

The lability measure revealed group differences as well. Infants in the sad condition showed more lability than those infants receiving the typical happy/surprised peekaboo trials. Moreover, introduction of a sad expression was accompanied by significant increases in affect lability on the target trials compared to baseline. For those infants in the happy, fear, and anger groups, although the means for lability were in the predicted direction, the differences were not significant.

Finally, facial expression time was also variable across the groups. Infants who were presented with a change in expression (either sadness, anger, or fear) were more expressive than were those infants who continued to receive happy/surprised expressions. Infants in the sad, anger, and fear conditions did not remain neutral-they responded affectively to the change in expression. In contrast, those who viewed only happy/surprised expressions across trials became more neutral as the session progressed.

We are especially intrigued with the affective responsiveness of the infants to the emotion expressions displayed in the peekaboo sequence. Together with the looking-time data, these patterns suggest that infants may be responding to the meaning or valence of each expression or, alternatively, that they are actually reacting with an emotion through contagion or mood induction (e.g., Campos, Mumme, Kermoian, & Campos, 1994; Feinman, 1982; Haviland & Lelwica, 1987). We reject the latter alternative given that the infants did not mirror the emotion expressions presented to them. It may be that the duration of each trial (7 s) precluded the full expression of a discrete emotion on the part of such young infants. That is, infants did not, for example, show anger in response to the angry expression or sadness in response to the sad expression. Rather, they responded with affective changes across trials-changes in lability, in the amount of expressions other than neutral, and in the frequency of interest/surprise expressions-that differed for each emotion. Coupled with the gaze patterns, their responses were, in a sense, functionally relevant to the particular emotion in the particular context of a peekaboo game (cf. Campos et al., 1994).

These findings suggest some interesting possibilities about infants' rudimentary understanding of others' emotion expressions. These patterns can be examined for each emotion. For example, infants in the sad condition showed a systematic downward trend in looking coupled with increases in interest/surprise expressions, a corresponding increase in expressions other than neutral, and greater affective lability. These findings are consistent with other research that reported gaze aversion in response to sad expressions (e.g., Haviland & Lelwica, 1987; Termine & Izard, 1988) and with indications from research on children with depressed mothers that highlighted, by comparison, that infants with nondepressed mothers may show interest in response to sad expressions (Pickens & Field, 1993). These findings are also consistent with theoretical propositions about the adaptive function of sadness (Izard, 1991; Tomkins, 1962). Stearns (1993) expanded on this idea, suggesting that low attention to sadness may serve a self-protective function. Likewise, the pattern of looking in the anger group is consistent with theoretical views on the unique function of anger. Anger theoretically functions to mobilize and sustain high energy levels (Izard, 1993) and to organize and regulate internal processes needed for self-defense (Lemerise & Dodge, 1993). Responses to anger might include vigilance, high arousal, and other responses tailored to a stressful event. In the present study, infants increased their looking on both anger trials, maintained a high degree of looking across all trials following the first change to anger, and displayed more facial affect. Infants' looking patterns in the fear condition showed an initial dramatic increase in looking at the first fearful expression, followed by decreased looking on subsequent trials, with only a minimal increase in looking at the second fearful expression. As with the sad and anger conditions, this pattern of looking is in keeping with views on the adaptive function of the fear emotion (Izard & Ackerman, 2000; Tomkins, 1991). Infants who observed fearful expressions also showed increased facial expression time.

That emotions may be functionally adaptive is still the topic of considerable debate; nonetheless, the parallels between infants' responses to discrete emotion expressions in this study and the hypothesized functions of discrete emotions are striking. At least when presented in a familiar context, young infants respond to others' affective behaviors in ways predictable from a functionalist perspective on emotion. Although this way of looking at infants' responsiveness is speculative, we suggest that it opens up new avenues and approaches to understanding infants' perception of emotion. Unfortunately, the present experiment does not allow us to determine whether infants were sensitive to facial or vocal expressions or both. Pairing taped voices with the facial expressions might provide useful data on this point.

These results also speak to the role of context in infants' discrimination of and responsiveness to distinct emotion expressions. The peekaboo game proved to be an effective paradigm for assessing infants' perception of others' expressions in several ways. It allowed greater control over the fidelity and accuracy of an expression than would a purely naturalistic design. All of the experimenter's expressions met stringent criteria for depicting discrete emotion expressions with specific muscle movements (Ekman & Friesen, 1975; Izard, 1979, 1995). The peekaboo game also afforded infants a more "interactive" experience than would a more typical visual preference or habituation experiment using slides, photographs, or videotapes. Infants were engaged in the game throughout the session and attended more than two thirds of the time to the researcher's facial and vocal expressions. In addition, this paradigm allowed us to embed facial/vocal expressions in a familiar context, one in which social contingency was prominent (e.g., Hains & Muir, 1996). Accordingly, we found evidence of infants' sensitivity to changes in others' emotion expressions at an earlier age than has been reported in studies that used less naturalistic paradigms. These data suggest that familiarity itself may facilitate infants' sensitivity to expressions (e.g., Kahana-Kalman & Walker-Andrews, 2001; Walker-Andrews, Montague, & Kahana-Kalman, 2000), just as the use of multimodal stimulus materials may make discrimination easier (e.g., Field et al., 1983; Gibson, 1991; Walker-Andrews, 1997). Given the effectiveness of the peekaboo paradigm in the study of emotion perception, it might well be used to study infants' interactions with depressed mothers, the perception of affect by infants with developmental delays, the contributions of acoustic and visual information to the perception of affect (see A. J. Caron et al., 1988; Lewkowicz, 1996; Walker-Andrews, 1997), and as others are already finding, the development of emotion regulation among infants (e.g., Eckerman et al., 1999; Fogel et al., 1997; Rochat et al., 1999). A drawback to the present study was that the woman playing peekaboo was a stranger. Infants might have shown more marked responses to changes in emotion portrayed by their own mothers, as suggested by the intermodal matching shown by infants to their mothers' expressions at 3.5 months in previous studies (Kahana-Kalman & Walker-Andrews, 2001; Montague, 2000).

The successful use of the peekaboo game also speaks to the role of infants' expectations. One of the earlier studies of peekaboo by Parrott and Gleitman (1989) demonstrated that infants appeared to have specific expectations about how the game was played and, in fact, enjoyed the game more when modifications to the location and identity of the person were made. More recently, Rochat et al. (1999) modified the structural aspects of a game of peekaboo and found that infants as young as 4 months of age detected these changes. Looking time was less in the typical peekaboo game, and smiling was greater in the disorganized game. In the present case, the structural violation was more intricate. The sequence and timing remained essentially the same; however, the meaning or affect introduced at each reappearance was unexpected for infants who could discriminate emotion expressions. That is, contingency with respect to structure and timing was preserved, but social contingency was clearly violated. A similar type of violation was reported in the original still-face research (e.g., Brazelton, Tronick, Adamson, Als, & Wise, 1975) in which the mother's face became unresponsive for an extended period and infants also showed emotional reactions. A number of researchers (e.g., Gusella et al., 1988; Mayes & Carter, 1990; Stack & Muir, 1990; Tronick, Als, Adamson, Wise, & Brazelton, 1978) have found that infants decrease visual fixation, show more grimacing, and have greater lability during still-face interactions than during other interactions. Analogous to the case in our condition, infants detect emotion expressions, and their responses are "a reaction to the violation in their expectations" for continued social contingency (Muir & Hains, 1993).

In conclusion, the emphasis of the present research was on infants' sensitivity to others' emotion expressions. A peekaboo paradigm was used to capitalize on the naturalistic, familiar dyadic interactions in the infant's world. We found that dynamic emotion expressions embedded in the rule-governed peekaboo task provided a sensitive context for evaluating infants' perception of emotion expressions. These data provide strong evidence that 4-month-old infants are differentially responsive to composite facial/vocal expressions of emotion in a familiar, naturalistic setting.