RESEARCH

ABDELRAHMAN ANWAR ATTIA RAGAB RAMADAN
200302611
Faculty of dentistry
“ Developmental psychology “
i. Summary
The article investigated infants’ reactions to two chosen melodies, with the conclusion
that infants seem to attribute social meaning to musical pieces played. Particularly, it appears
that familiar songs previously sung by parents lead to infants paying more attention at
unknown presenters compared to unfamiliar (or socially irrelevant) songs performed by
strangers. Interestingly, when a singing toy was used for training instead of parents singing, no
social effect was observed, which confirms the social meaning being assigned to or ¨encoded¨
in the learned songs. It is consistent with other reports about infants learning abilities and their
emotional (Anqulo-Barroso, 2016; Jones, 2011) and social (Moe, 2016; Walle, 2016) aspects.
ii. Critical analysis

a. Consistency of experimental conditions
This research claims to study infants’ selective attention to musical pieces performed by
unfamiliar persons. When one looks at the study design though, it becomes obvious that in the
first two experiments only parents appear in person during training (pre-experimental)
sessions, whereas only videos of unfamiliar people singing are presented during the experiment
per se. That discrepancy (singing in person vs. watching a video of somebody singing) may have
actually biased the results to a certain extent, as training settings differ from testing
environment and taken into account the young age of studied infants which corresponds to
preferring concrete objects over abstract features (Lavigne, 2012; Piaget, 1966, Trehub, 1998).
On the other hand, one could also expect some amount of skill transfer as toddlers’ learning by
using different media tools may transfer to real environment given social meaning of video
characters (Lauricella, 2011).
In the third experiment, a live person appeared in the lab singing before infants during
the selective attention task, after which the person was regularly interacting with those infants
via a Skype voice chat over an iPad. That would however require a control condition of some
infant receiving a live lab performance but no further Skype calls, to identify the possible effects
of a live performance on infants’ selective attention with regards to performed familiar songs.
That control was not implemented in this study which potentially biased the results of the third
experiment. Moreover, mixing speech and singing during Skype sessions is a confusing element
of the experimental setup as this interactional element could not be controlled thus included
between-subject variations. The result of infants being generally enjoying Skype performance is
also a bit confusing as infants’ reactions were not systematically recorded, and data analysis for
this qualitative experimental part did not seem to be consistently arranged and/or properly
described.
200302611
There were also further inconsistencies such as song exposure being significantly less in
the third experiment compared to the first two, although experimental duration was
comparable across all the three experiments. Such discrepancies give rise to questions about
the plausibility of any direct comparison of the results obtained in such varying experimental
conditions.
In the two follow-up experiments (nr. 5 and 6), the infants were retested after a delay of
8 months. The aim was to check their ability to attend to and discriminate between familiar and
unfamiliar songs performed either by toys (in the experiment 4) or the same live performer as
in the third experiment (experiment 5). The follow up study showed that infants pay more
attention (or attend longer) to familiar songs compared to unfamiliar ones in both conditions,
i.e. when played by toys and also when performed live on Skype by a familiar singer. These
effects are consistent with the initial results obtained in the first three experiments, and
demonstrate presence of certain long-term memory mechanisms. This sounds logical;
however, as live performance in the third experiment did not lead to any discrimination or
selective attention effects, a question arises why such effects appear later on when the same
singer performs ¨live¨ over Skype. The presented mix of results from lab live performance,
Skype live singing, Skype voice chatting, and toy ¨singing¨ seems to actually rise more questions
than answer the initial research questions. To solve this mystery, one would need to make sure
those experiments are compatible with each other as in present state experimental conditions
varied across all the experiments, and re-run them again on a bigger amount of subjects. For
the time being, it is very difficult to navigate through results of seemingly similar but in fact
different experimental settings and corresponding outputs.
There is an open discussion about the effectiveness of skills transfer from digital (video,
Skype) and physical toys (singing or talking) to real life, as opposed to traditional social learning
(Lovato, 2016). Some studies report relatively low success rate of around 30% when children
attempt at using their newly learned digital skills to the real physical environment (Troseth,
2006). Other studies refer to the so called ¨video deficit¨ by demonstrating that digital skills
easier transfer to real life when the same person who was interacted with over computer
screen appears in live communication settings (Anderson, 2005; Bar, 2010; Schmitt, 2002).
200302611
b. Considerations about the chosen experimental sample and data analysis
Mostly women served as singing parents as mentioned in the report. That wording
implies however that some chosen parents were women, and some were men. Mother-child
attachment has been shown to be stronger than those of fathers on this early developmental
stage (Lamb, 1977) therefore adjustment upon singing parents’ gender would be useful.
On the other hand, a very small participant sample used (30+ subjects in experiments
one, two and three; 20+ subjects in experiments four and five) clearly does allow generalizing
the results to the general population. Nevertheless, the second experiment which used a
stuffed toy ¨singing¨ the same songs compared to parents singing in the training condition may
actually show a trend which can be confirmed when using a bigger experimental sample. In
fact, in addition to a stuffed toy singing serving as a control condition in the first experiment,
the second experiment per se can be considered as a control to the first one. That makes the
experimental results and the derived conclusions stronger.
Finally, parametric tests only were used in this study, namely t-test and parametric
regression analysis. It is not clear why on such small samples the authors did not go for non-
parametric analysis which is more advisable in case of small samples (Whitley, 2002). There was
also no mention of data distribution for any variables, so it is impossible to judge if parametric
assumptions were met. However, even if parametric assumptions would actually be
consistently met across all the conditions and experimental parts, it would still be more
convincing if non-parametric counterparts of parametric statistical tests were used and their
results described.
iii. Conclusions
Even though the study reports an interesting pattern of social differentiation of musical
sounds and long-term memory effects in infants, the study introduces some confusion due to
certain inconsistencies in study design and due to small samples used. The selective attention
effects which the authors actually reported as the main study effects were not present in the
third experiment where live performance was used instead of video recording during the
selective attention task. In the third experiment, the infants’ attention towards the novel
person singing the familiar song did not differ from chance, whereas the amount of song
exposure also did not predicted the infants’ change in attentional preference towards the
singer. This actually means that the results of the third experiment totally contradict to the
results of the first experiment. It is particularly intriguing because the third experiment used
live lab performance instead of visual recordings during the selection attention task. In this
light, combined with the lack of a representative sample, the main reported result based on the
first two experiments may actually represent a meaningless result regardless demonstrated
statistical significance.
200302611
v References
Anderson, D. R., & Pempel. T. A. (2005). Television and very young children. American Behavioral Science,
48, 505-522.
Bar, R. (2010). Transfer of learning between 2D and 3S sources during infancy: Informing theory and
practice. Developmental Review, 30, 128-154.
Anqulo-Barroso, R. M., Pecina, S., Lin, M., Sturza, J. et al. (2016). Implicit learning and emotional responses in
nine-month-old infants. Cognition and Emotions, 4, 1-10.
Jones, N. A., Field, T., Fox, N. A. et al. (2011). EEG during different emotions in 10-month-old infants of depressed
mothers. Journal of Reproductive and Infant Psychology, 19(4), 1-19.
Lamb, M. E. (1977). Father-infant and mother-infant interaction in the first year of life. Child Development,
48(1), 167-181.
Lauricella, A. R., Howard Gola, A. A., & Calvert, S. L. (2011). Toddlers’ learning from socially meaningful video
characters. Media Psychology, 14, 216-232.
Lavigne, H. (2012). The priming effects of video viewing on preschoolers’ play behavior. MSc Thesis presented at
the University of Massachussetts Amherst Ed.M., Harvard University. Boston: Harvard University Press.
Lovato, S. B., & Waxman, S. R. (2016). Young children learning from touch screes: Taking a wider view. Frontiers
in Psychology, 7, doi: 10.3389/fpsyg.2016.01078. eCollection 2016.
Mehr, S: A., Song, L. A., & Spelke, E. S. (2016). For 5-month-old infants, melodies are social. Psychological
Science, 27(4), 486-501.
Moe, V., Braarud, h. G., Wentzel-Larsen, T., Slinning, k. et al. (2016). Precursors of social emotional functioning
among full-term and preterm infants at 12 months: Early infant withdrawal behavior and symptoms of maternal
depression. Infant Behavior and Development, 44, 159-168.
Piaget, J. (1966). The psychology of the child. Paris, France: Presses Universitaires de France.
Schmitt, K.L., & AnAnderson, D. R. (2002).Television and reality: Toddlers‘ use of visual information from video
to guide behavior. Media Psychology, 4, 51-76.
Trehub (1998). Singing to infants: Lullables and playsongs. Advances in Infancy Research, 12, 43-77.
Troseth, G. L., Saylor, M. M., & Archer, A. H. (2006). Young children’s use of video as a source of socially
relevant information. Child Development, 77,786-799.
Walle, E: A. (2016). Infant social development across the transition from crawling to walking. Frontiers in
Psychology, 7, doi: 10.3389/fpsyg.2016.
Whitley, E., & Ball, J. (2002). Statistics review 6: Nonparametric methods. Critical Care, 6(6), 509-513.
626691
research-article2016
PSSXXX10.1177/0956797615626691Mehr et al.For 5-Month-Old Infants, Melodies Are Social
Research Article
Psychological Science
For 5-Month-Old Infants, Melodies 2016, Vol. 27(4) 486–501

© The Author(s) 2016
Reprints and permissions:
Are Social sagepub.com/journalsPermissions.nav
DOI: 10.1177/0956797615626691
pss.sagepub.com
Samuel A. Mehr1,2, Lee Ann Song2, and Elizabeth S. Spelke2

1
Graduate School of Education and 2Department of Psychology, Harvard University
Abstract
For 1 to 2 weeks, 5-month-old infants listened at home to one of two novel songs with identical lyrics and rhythms, but
different melodies; the song was sung by a parent, emanated from a toy, or was sung live by a friendly but unfamiliar
adult first in person and subsequently via interactive video. We then tested the infants’ selective attention to two novel
individuals after one sang the familiar song and the other sang the unfamiliar song. Infants who had experienced a
parent singing looked longer at the new person who had sung the familiar melody than at the new person who had
sung the unfamiliar melody, and the amount of song exposure at home predicted the size of that preference. Neither
effect was observed, however, among infants who had heard the song emanating from a toy or being sung by a
socially unrelated person, despite these infants’ remarkable memory for the familiar melody, tested an average of more
than 8 months later. These findings suggest that melodies produced live and experienced at home by known social
partners carry social meaning for infants.
Keywords
music, social cognition, memory, infant development, open data, open materials
Received 10/13/14; Revision accepted 12/17/15
Music is a putative human universal (Brown, 1991) that is (Schachner & Hannon, 2011), or speak their parents’ lan-
ancient (Conard, Malina, & Münzel, 2009), highly variable guage (Kinzler, Dupoux, & Spelke, 2007). Such people
(Lomax, 1968), and captivating throughout the life span, are likely members of an infant’s social group, and infants
including in infancy (Patel, 2008; Trehub, 2003). Human might be sensitive to this information as a result of innate
infants learn and remember melodies (Trainor, Wu, & dispositions, early learning, or both. We therefore exam-
Tsang, 2004), are sensitive to rhythmic detail (Winkler, ine whether infants selectively attend to novel singers of
Háden, Ladinig, Sziller, & Honing, 2009), and move spon- melodies learned in social settings.
taneously to music (Zentner & Eerola, 2010). Parents sing Melody could signal social affiliation for two reasons.
frequently to their infants and children (Custodero & First, before the advent of sound recordings, melodies
Johnson-Green, 2003; Mehr, 2014), in a style that is iden- were learned only from other people. Because melodies
tifiable across many cultures (Trehub, Unyk, & Trainor, are complex and highly variable, members of disjoint
1993). Why do parents sing to their infants? Why do groups are unlikely to invent the same melodies. The
infants engage with songs sung by their parents? These ability and propensity to sing a familiar song could there-
questions echo long-standing debates over music’s psy- fore carry social information: A novel person who sings a
chological functions in modern environments and origins known melody is more likely to be connected to one’s
in human ancestry (e.g., Darwin, 1871; James, 1890; social group than is a novel person who sings an
Spencer, 1857). Here, we explore the hypothesis that unknown melody. Second, in many cultures, people sing
melody—a salient feature of vocal music—conveys social together in social contexts (Savage, Brown, Sakai, &
information to infants.
As do adults, infants regularly choose whether and
Corresponding Author:
how to engage with other individuals, selectively attend- Samuel A. Mehr, Harvard University, Department of Psychology, 33
ing to novel people who look at them (Farroni, Csibra, Kirkland St., Cambridge, MA 02138
Simion, & Johnson, 2002), produce infant-directed speech E-mail: [email protected]
Downloaded from pss.sagepub.com at University of Bristol Library on May 16, 2016

For 5-Month-Old Infants, Melodies Are Social 487
Currie, 2015). Infants may be attentive to this fact, and Experiment 1

these social experiences may lead them to endow music
with social meaning. Method
Thus, we hypothesized that infant-directed songs con- In Experiment 1, parents with little or no musical exper-
vey social meaning to infants. In the experiments reported tise learned to sing one of the two songs. They sang that
here, we examined infants’ preferences for novel indi- song to their infants on a regular basis and visited a Web
viduals, one who had sung a familiar melody and another site where they could listen to the song, to help them
who had sung an unfamiliar melody. The infants were faithfully reproduce it at home. To maximize ecological
familiar with one of the songs as a result of exposure to validity, we instructed the parents to “sing the song to
it in one of three forms: (a) live song from a parent in the your baby as much as you like.” After 1 to 2 weeks of
home, (b) recorded song emanating from a toy presented song exposure, the infants returned to the lab for a selec-
and activated by a parent in the home, or (c) live song tive-attention test (described later in this section). Finally,
presented by a novel adult both in person in the lab and the parents were recorded as they sang to their infants,
via live, interactive video at home. so that expert raters could judge their pitch accuracy.
These three forms of song exposure have different
social implications. A musical toy presented by a parent Participants. We recruited 38 full-term infants and
divorces the source of music from the source of social their parents from the greater Boston area. The parents
interaction, while maintaining a socially engaging con- were given a $5 travel reimbursement at each lab visit,
text provided by the parent, who activates the toy and and the infants were given a toy or other small reward
modulates the infant’s engagement with it. Live, interac- after completing the study. All testing took place at the
tive song presented by an unfamiliar adult involves social Laboratory for Developmental Studies at Harvard Univer-
interaction between the singer and infant but divorces sity. Data from 6 infants were excluded because they
the source of music from any history of social interaction; were fussy (n = 1) or inattentive (n = 1), or because their
when the singing episodes occur at home over interac- parents failed to produce a recognizable song (n = 4).
tive video, they are also divorced from other social inter- Thus, analyses included 32 infants (17 females; mean
actions in the home. age = 5.61 months, SD = 0.31, range: 5.06–6.11). Typi-
Nevertheless, each of these forms of song exposure cally, the infants were brought to the lab by one parent,
may carry social meaning. Infants reliably learn from who participated in the experiment; when both parents
both musical recordings (e.g., Trainor et al., 2004) and were present, one elected to serve as the participating
socially contingent nonmusical video interactions (e.g., parent, who completed all surveys, was given the music
Roseberry, Hirsh-Pasek, & Golinkoff, 2014). Many infants lesson, and sang to the infant at home and in the
in modern societies encounter music produced by sound lab. Participating parents were predominantly female
recordings as frequently as music produced by a parent (26 female). Some parents reported that other adults
(e.g., Mehr, 2014). Indeed, parents may believe that learned the song and sang it to their infant as well.
recordings provide higher-quality musical experiences
than their own singing, given widespread beliefs that Statistical power. The target sample size of 32 was
music listening and music education can have a positive determined before the experiment began, to ensure ade-
influence on cognitive skills (see Mehr, 2015). Thus, we quate power to detect a positive selective-attention effect.
tested for social effects of music under all three learning A similar experiment testing effects of language rather
conditions and investigated whether the magnitude of than music (Kinzler et al., 2007) obtained an effect size
these effects varied by condition. (d) of 0.54, and a sample of 32 had .84 power to detect
The infants in our experiments learned one of two lul- an effect of this magnitude.
labies that were adapted to equate their rhythms and lyr-
ics; they differed only in melody and sounded highly Musical content. We adapted two obscure lullabies
similar (see Fig. 1a and Videos S1 and S2 in the Supple- from folk repertoires (“Babushka Baio” and “Shady
mental Material available online). The infants were ran- Grove, My Little Love”; Feierabend, 2000) to create two
domly assigned to a 1- to 2-week familiarization with one songs with identical rhythms and lyrics but different mel-
song or the other, presented by one of the methods just odies (see Fig. 1a for the music notation and Videos S1
described. At test, we measured the duration of the and S2 in the Supplemental Material for the performances
infants’ looking at two novel people before and after they of the songs used in the selective-attention test). In pilot
each sang one of the songs. If melodies convey informa- testing, the ratings of approximately 30 adults demon-
tion about social affiliation, then the infants would be strated that the songs were equally pleasant; half pre-
expected to attend selectively to the new singer of the ferred one song, and half preferred the other. Additionally,
familiar song. twelve 5-month-old infants in a pilot study attended

488 Mehr et al.
a b
Song 1
Lul - la - by, my lit - tle ba - by, soft - ly sleep, my child.
Order of Events
Close your eyes and fly a - way, I'll wake you with a smile.
Song 2
Lul - la - by, my lit - tle ba - by, soft - ly sleep, my child.
Close your eyes and fly a - way, I'll wake you with a smile.
Fig. 1. The lullabies (a) and testing procedure (b) used in Experiments 1 through 3. During the baseline trial of the selective-attention test, two
novel individuals stood silently smiling, directly gazing at the infant, for 16 s. Next, during the two familiarization trials, each individual sang one of
the two lullabies in turn while gazing at the infant (22 s per trial; i.e., one person sang the lullaby that was familiar to the infant, and the other sang
the unfamiliar lullaby). Finally, during the test trial, the two individuals again stood silently smiling, directly gazing at the infant.
comparably to videos of novel individuals singing the (Satterthwaite’s t test). Thus, the two songs appear to
two songs. have been comparably easy for untrained musicians to
learn.
Music instruction and assessment of music apti- Parents also completed a standardized assessment of
tude. At the first lab visit, each participating parent was auditory perception skill, the Advanced Measures of
randomly assigned to learn one or the other song (using Music Audiation (AMMA; Gordon, 1989), so that we
the True Random Number Generator at www.random could test for effects of parents’ music aptitude on their
.org) and was given a 10- to 30-min music lesson, con- infants’ behavior.
ducted without music notation and with the aid of a
keyboard. The lesson concluded when the parent indi- Assessment of singing accuracy. At the second lab
cated that he or she felt confident in being able to visit, we recorded the parents singing the song from the
reproduce the song without assistance (for an example experiment to their infants during a free-play session. The
of a parent singing to her infant, see Video S3 in the audio recordings were presented to three professional
Supplemental Material). Parents were given access to a musicians, who independently judged the number of
Web site that provided two recordings of their song accurate pitches (of 25 total pitches) in each performance.
(with and without lyrics), as well as the printed text of Interrater reliability (computed from the raw number of
the lyrics; they were encouraged to practice by singing accurate pitches) was high (Cronbach’s α = .90). When
along with these recordings, but were also told that any one rater was unable to identify the song from a par-
their infant should hear the song only from live indi- ent’s performance, the data from the parent and infant
viduals (i.e., that they should never play the recorded were excluded from analyses. This occurred in four cases.
music on the Web site to their infants or record their With the exception of these participants, parents’ pitch
own voice and play the recording for their infants). accuracy (proportion of correct pitches) was comparable
Compliance with these instructions was high: Thirty- across the two song conditions (Song 1: M = .87, SD = .13,
one of the 32 parents visited the Web site at least once range: .56–1; Song 2: M = .83, SD = .15, range: .52–1),
and spent enough time on the site listening to the song, t(29.7) = 0.76, p = .45 (Satterthwaite’s t test).
on average, to listen to it a total of approximately 15
times over the course of the study. Total time spent on Survey. After the first lab visit, we e-mailed parents
the site listening to the song was comparable across the daily with a brief survey, in order to determine the
two song conditions (Song 1: M = 5.73 min, SD = 5.62; approximate number of times each infant heard the song
Song 2: M = 5.23 min, SD = 6.02), t(29.9) = 0.24, p = .81 each day. The rate of survey completion was high

(92.3%). To estimate the amount of the infants’ song baseline. The proportion of time that they looked toward
exposure, we took the mean of each parent’s responses the person who sang the familiar song did not differ from
to the question “About how many times did you sing chance (.5; M = .521, SD = .177, 95% confidence interval,
your new lullaby to your baby today?” and multiplied it or CI = [.457, .585]), t(31) = 0.67, p = .51 (one-sample t
by the number of days of that family’s participation in test; see Fig. 2a). Moreover, the infants attended highly
the study. These estimates thus accounted for both and equally to the two singers during the familiarization
incomplete survey responses and variability in study trials, as each singer appeared by herself and sang a song
length across participants. (duration of looking toward the singer of the familiar
song: M = 15.6 s, SD = 5.07; duration of looking toward
Selective-attention test. At the second lab visit, we the singer of the unfamiliar song: M = 15.3 s, SD = 5.10),
tested each infant’s attention to two novel individuals, t(31) = 0.28, p = .78 (paired t test). At test, however, the
one who sang the song that was familiar to the infant and infants selectively attended to the now-silent singer of the
another who sang the unfamiliar song. The infant sat on song with the familiar melody; the proportion of time dur-
his or her parent’s lap, approximately 5 ft away from a ing which they looked toward her was greater than
55- × 40-in. projection screen. The parent closed his or chance (.5; proportion of looking: M = .593, SD = .179,
her eyes and wore noise-canceling headphones that 95% CI = [.529, .658]), t(31) = 2.96, p = .006, d = 0.52 (one-
played masking music throughout the experiment. The sample t test), and greater than the proportion at baseline
selective-attention test had four trials (see Fig. 1b). First, (difference in proportion of looking: M = .072, SD = .169,
the infant viewed side-by-side high-definition video 95% CI = [.011, .133]), t(31) = 2.42, p = .022, d = 0.43
recordings of the two unfamiliar individuals, smiling with (paired t test; see Fig. 2a).
direct gaze at the infant, for 16 s (baseline trial). Then, the We used simple linear regression to investigate
infant viewed, in turn, one 22-s video of each of the two whether the degree of infants’ increase in attention to the
individuals singing one of the two songs while continu- singer of the familiar song from baseline to test depended
ing to look and smile at the infant (familiarization trials; on their level of in-home exposure to that song. Parents
see Videos S1 and S2). Finally, the infant viewed a silent reported singing regularly to their infants (median of 9
16-s test trial that was identical to the baseline trial. A performances per day, interquartile range = [4, 11]; esti-
looming object with an attractive sound effect brought mated total number of song performances: M = 76,
the infant’s eyes to the center of the screen before the SD = 56). After a log2 transformation (because of strong
baseline and test trials. curvature), song exposure was a significant predictor of
The videos were dubbed so that the two unfamiliar the within-subjects main effect (Fig. 2b), χ2(1) = 7.53,
individuals sang in the same voice, to control for potential p = .006, R2 = .14 (Wald test).1 A doubling of the approxi-
differences in singing quality across the two song condi- mate number of parental performances corresponded to
tions. The song-to-singer pairing, order of the familiariza- an estimated 0.37-SD increase in attention to the novel
tion trials, and presentation location (left or right side of person who sang that song.
the screen) were fully counterbalanced. Each infant’s gaze We also tested the predictive power of two character-
in each of the four trials was recorded with a hidden high- istics of parents’ musical abilities: the objective accuracy
definition camera and was coded frame by frame, at 30 of their song performances (proportion of correct pitches,
frames per second, by a coder who was blind to which as judged by the expert raters) and their music percep-
song the infant was familiar with and to the presentation tion skills (as measured by the AMMA). Both measures
location of each song. A second person recoded gaze for showed considerable variation across the parents in the
all the infants, and interrater reliability (computed by cor- sample, but neither predicted infants’ attentional prefer-
relating the raw proportion of looking toward the singer ences—singing accuracy: χ2(1) = 0.25, p = .61; music
of the familiar song) was high (r = .91). The coders viewed aptitude: χ2(1) = 0.85, p = .36 (Wald tests).
the baseline and test trials before the familiarization trials,
so that no differences in the infants’ behavior during the
Discussion
singing could influence the coding of attention to the
individuals during the test trial. The infants selectively attended to the novel person who
previously sang the song they had learned from their
parents, and the amount of in-home exposure to the
Results song predicted the size of this effect. In Experiment 2, we
No variables differed between infants exposed to Song 1 asked whether infants would show a similar effect if their
and those exposed to Song 2 (ps > .1); thus, we present parents presented them with a novel melody via a sound
the data in aggregate. In the selective-attention test, the recording, emanating from a toy—that is, when the
infants showed no preference for either individual at source of the melody was not a known social partner.

490 Mehr et al.
a b
1.0
* .6
Increase in Proportion of Looking From Baseline to Test

Proportion of Looking Toward Singer of Familiar Song
.9
.5
.8
.4
.7
.3
.6
** .2
.5
.1
.4
.3 0
.2 –.1
.1 –.2
0 –.3
Before After 10 20 40 80 160 320

Singing Singing Estimated Number of Times Infant Heard Song
(Baseline) (Test)
Fig. 2. Main effects in Experiment 1. The box plots in (a) show the proportion of time in the baseline and test trials during which
the infants looked toward the singer of the familiar song. The dotted line indicates chance level (.5), the Xs indicate the means, the
horizontal lines indicate the medians, the boxes indicate the interquartile ranges, and the vertical lines indicate the full ranges. Aster-
isks indicate significant differences (*p < .05, **p < .01). The scatterplot in (b) shows each infant’s increase in proportion of looking
toward the singer of the familiar song from baseline to test, along with the predicted effect of the amount of song exposure on this
within-subjects main effect, from a bootstrapped model with 40,000 replications. The dashed lines represent ±2 bootstrap standard
errors, and chance (0) is indicated by the dotted line. Note that the x-axis is displayed on a log2 scale.
Experiment 2 infants at home. (Parent’s pitch accuracy and music apti-

tude were not measured, because the parents in this
Method experiment never sang the song to their infants.)
We repeated Experiment 1 with a second cohort of
infants, altering the means of song exposure. In Experi- Participants. We recruited 35 full-term infants. Data
ment 2, parents were instructed never to sing the song to from 3 infants were excluded because they were fussy
their child; instead, they were given a stuffed animal that (n = 1), failed to attend to the test stimuli (n = 1), or had
produced music. When a parent squeezed this toy, it prior familiarity with an actor in the test stimuli (n = 1).
played a recording of an expert vocalist singing one of Analyses included 32 infants (18 females; mean age =
the two songs in an infant-directed fashion. Thus, song 5.49 months, SD = 0.31, range: 4.99–6.14). Participating
exposure in Experiment 2 was highly compelling and parents were predominantly female (27 females).
occurred in a social context, but the source of the song
was an inanimate object. The procedure for recruiting Song exposure: musical toy. We adapted an age-
participants, target sample size and statistical power, appropriate stuffed animal ( Jellycat Inc., London, United
musical content, daily surveys (with minor changes in Kingdom) to play a recording of the song (for an exam-
wording to account for the difference in exposure type, ple of a parent playing the toy for her infant, see Video
and a comparably high completion rate of 90.6%), selec- S4 in the Supplemental Material). The toy was no longer
tive-attention test, and analyses were the same as in publicly available for purchase, and no parent indicated
Experiment 1. Thus, only the type of song exposure dif- that his or her infant was familiar with the toy at the out-
fered: In lieu of taking a music lesson in the lab and then set of the experiment. We removed stuffing from each toy
singing the song in the home, parents in Experiment 2 and inserted a sound module (Invite By Voice LLC, Eden
were given a musical toy in the lab and used it with their Prairie, MN) that played a recording of the song when the

toy was squeezed. The recorded singers were two profes- p = .70 (one-sample t test; see Fig. 3a). In addition, the
sional vocalists (gender matched to the participating par- infants attended equally to the two singers during famil-
ent), instructed to sing in an infant-directed fashion; iarization (duration of looking toward the singer of the
neither vocalist was the singer in the selective-attention familiar song: M = 16.2 s, SD = 5.18; duration of looking
test. The toy played the song at approximately 55 dB at a toward the singer of the unfamiliar song: M = 15.2 s, SD =
distance of 5 in., a comfortable volume comparable to 6.69), t(31) = 0.95, p = .35 (paired t test). The levels of
that of other commercially available musical toys intended attention to the two novel adults in Experiment 2 were
for young infants. The infants lacked the dexterity to acti- comparable to the levels in Experiment 1 both during the
vate the toy themselves, and thus required a parent or baseline period, t(62.0) = 0.20, p = .84 (Satterthwaite’s t
another individual to do so. As in Experiment 1, the test), and during the familiarization trials—familiar song:
infants were randomly assigned to learn one song or the t(62.0) = 0.49, p = .63; unfamiliar song: t(57.9) = 0.06, p =
other. .95 (Satterthwaite’s t tests).
At the first visit, we introduced the infants and parents Despite the high appeal of the toy and professional
to the musical toy and demonstrated its use. We instructed quality of the singing, the selective-attention effects found
parents to “treat the toy as if you purchased it and use it in Experiment 1 were not present in Experiment 2. At
as much as you like,” indicating that the amount of song test, the proportion of time during which the infants
exposure and the nature of the infants’ toy-directed looked toward the singer of the familiar song (M = .537,
actions were at the parents’ discretion. Parents were SD = .181, 95% CI = [.472, .602]) did not differ from
instructed never to sing the song to their infants, and to chance (.5), t(31) = 1.17, p = .25 (one-sample t test), or
relay this instruction to any other individuals who would from the proportion at baseline (difference in proportion
come into contact with the toy. Compliance with this of looking: M = .025, SD = .165, 95% CI = [−.034, .085]),
instruction was high. We asked parents at the second lab t(31) = 0.86, p = .40 (paired t test). Further, the amount of
visit if they had sung the song during the exposure song exposure did not predict the infants’ change in
period. At most, parents reported a total of three or four attentional preference toward the new individual who
instances of singing or humming the song, usually “by sang the familiar song (Fig. 3b), χ2(1) = 0.01, p = .92
accident.” Parents completed online surveys to report the (Wald test).
amount of the infants’ song exposure, as in Experiment 1.
Comparison with Experiment 1. We used multiple
Selective-attention test. Parents returned the toys to regression to test the effects of exposure type (via a par-
the lab at the second visit, when the infants participated ent, in Experiment 1, vs. via a toy, in Experiment 2) and
in a testing session identical to that of Experiment 1. amount of song exposure on the within-subjects increase,
Interrater reliability, computed in the same fashion as in from baseline to test, in looking toward the singer of the
Experiment 1, was high (r = .96). familiar song. As in Experiment 1, we performed a log2
transformation on the parents’ self-reported number
of song performances. We used no-constant models
Results because the population intercept is known to be 0: If
Main analyses. The infants’ degree of song exposure infants have never heard either song (i.e., 0 song perfor-
was comparable across the two experiments: The esti- mances), on average, they should not show any increase
mated total number of song performances was similar in in looking toward either singer from baseline to test.
Experiment 1 (M = 76.3, SD = 56.2) and Experiment 2 Model building began with a simple no-constant model
(M = 81.8, SD = 50.5), t(61.3) = 0.41, p = .68 (Satter predicting the overall increase in looking from exposure
thwaite’s t test), and the exposure periods were of com- quantity only; we then added the predictor of exposure
parable duration (in both experiments, Mdn = 7 days, type, and finally an Exposure Type × Exposure Quantity
range: 7–14). Given that a song performance lasted 22 s, interaction.
we estimated that the infants in Experiment 2 received The final model was statistically significant, χ2(3) =
about 30 min of song exposure, on average; this was 12.0, p = .0076, R2 = .16 (Wald test), as was the Exposure
more than the exposure in a previous study showing that Type × Exposure Quantity interaction, b = 0.059, 95%
6-month-olds remembered songs presented to them via CI = [0.014, 0.105], z = 2.53, p = .011 (z test of the boot-
audio recordings (21 min in Trainor et al., 2004). strapped regression coefficient). The inclusion of expo-
In the selective-attention test, the infants looked sure type and the interaction term significantly increased
equally toward the two unfamiliar adults at baseline. The model fit, χ2(2) = 6.63, p = .036 (nested test). At the grand
proportion of time that they looked toward the person mean of exposure quantity (79 song performances),
who sang the familiar song did not differ from chance infants in Experiment 1 showed a larger increase in atten-
(.5; M = .512, SD = .175, 95% CI = [.449, .575]), t(31) = 0.39, tion toward the singer of the familiar song from baseline

492 Mehr et al.
a b
1.0
n.s. .4

.9
.8 .3
.7
.2
.6
n.s. .1
.5
.4 0
.3
–.1
.2
–.2
.1
0 –.3
Before After 20 40 80 160 320
Singing Singing Estimated Number of Times Infant Heard Song
(Baseline) (Test)
horizontal lines indicate the medians, the boxes indicate the interquartile ranges, and the vertical lines indicate the full ranges. The
scatterplot in (b) shows each infant’s increase in proportion of looking toward the singer of the familiar song from baseline to test,
along with the predicted effect of the amount of song exposure on this within-subjects main effect, from a bootstrapped model with
40,000 replications. The dashed lines represent ±2 bootstrap standard errors, and chance (0) is indicated by the dotted line. Note that
the x-axis is displayed on a log2 scale.
to test than did infants in Experiment 2, b = 0.072, 95% Live and recorded singing differ in many respects,
CI = [−0.013, 0.156], z = 1.67, p = .047, β = 0.43 (general however. Although parents in Experiment 2 activated the
linear hypothesis test, one-tailed). At the 75th percentile toys, the toys themselves were inert: Unlike live singers,
of exposure quantity (101 song performances), the differ- they did not interact contingently with the infants, vary
ence between the two experiments was larger, b = 0.093, the style and content of their song production, fine-tune
95% CI = [−0.001, 0.187], z = 1.95, p = .025, β = 0.55 (gen- their singing to the infants’ affective state, or move in
eral linear hypothesis test, one-tailed). synchrony with their singing. In Experiment 3, we pre-
sented infants with singing that incorporated these char-
acteristics, but the source of the song was an unfamiliar
Discussion adult who sang to the infants primarily over live, interac-
Despite the use of recorded music, a common form of tive video.
musical exposure for infants, the results of Experiment 2
differed markedly from those of Experiment 1. Infants
selectively attended to a novel singer of a song originally Experiment 3
sung by a parent, but not to a novel singer of a song
originally played via a musical toy. This contrast suggests
Method
that infants do not prefer novel individuals who are asso- We repeated Experiment 1 with a third cohort of infants,
ciated with any familiar melody presented in a positive again altering the means of song exposure. As in Experi-
social context, despite long-standing findings that people ment 2, parents were asked never to sing the song to their
prefer familiar sounds, objects, and patterns over unfa- infants, but the parents and infants were introduced to a
miliar ones, in both adulthood (Zajonc, 2001) and infancy friendly adult (a university student with musical training)
(Bornstein, 1989). who sang the song to the infants in the lab. Parents then

were given an iPad, by means of which this adult inter- parent’s lap, others sat in a high chair with the iPad on
acted daily with the infants, using interactive video chat the stand, and so on.
(Skype). The procedure for recruiting participants, sample The infants were randomly assigned to learn one song
size and statistical power, musical content, daily surveys or the other, and the parents were instructed that they
(with minor changes in wording to account for the differ- should never sing the song to their infants and should
ence in exposure type, and a comparably high comple- relay this instruction to other individuals who were pres-
tion rate of 90.0%), and selective-attention test were the ent during the video sessions (who might learn the song
same as in Experiment 1. incidentally). Compliance with this instruction was high,
with only two parents in the study reporting any live
Participants. We recruited 39 infants. Data from 7 singing of the song.
infants were excluded because they were fussy (n = 1),
failed to attend to test stimuli (n = 1), or did not complete Selective-attention test. Parents returned the iPad to
all the interactive video sessions (n = 2), or because of the lab at the second visit, when the infants participated
experimenter error (n = 3). Analyses included 32 infants in a selective-attention test identical to that of Experi-
(12 females; mean age = 5.82 months, SD = 0.49, range: ment 1. Interrater reliability, computed in the same fash-
5.03–6.47). Participating parents were predominantly ion as in Experiments 1 and 2, was high (r = .95).
female (24 females).
Assessment of the infants’ responses to song expo-
Song exposure: interactive video sessions. Parents sure. The presentation of songs via interactive video
were provided with an iPad equipped with the Skype allowed us to analyze the infants’ responses to song
application and a flexible stand, a carrying bag, and a exposure in a fashion that was not possible in Experi-
charger, all of which they kept for the duration of the ments 1 and 2: In Experiment 3, all video sessions were
study. At the first lab visit, the infants met the new adult, recorded (for a typical song performance, see Video S5 in
who played with them and sang the song directly to the Supplemental Material). These videos enabled us to
them six to eight times. The adult and parents then compare the infants’ responses to the song and speech in
scheduled daily appointments for singing over interac- a natural setting and to test for the relation between the
tive video. The parents were asked to choose times dur- infants’ engagement with the song during learning and
ing which they expected the infants to be comfortable their selective attention to a novel singer at test.
and attentive; across the cohort, appointment times We analyzed three video sessions per infant, from the
were spread throughout the day, but within subjects beginning, middle, and end of the exposure period. Vid-
they tended to occur consistently at the same time (e.g., eos in which the infants were fussy were not used (exclu-
every day after breakfast). The adult kept in regular sion decisions were made before any coding was
contact with each parent via e-mail so as to ensure that performed, to avoid confirmation bias). Two coders
the infant and parent were both present for each video viewed the videos at 10 frames per second and recorded
appointment. the durations of four categories of events: infant smiling,
The infants participated in 6 to 11 interactive video infant gazing toward the screen (where the adult was
sessions at home (Mdn = 7), each lasting roughly 10 min, always visible), adult singing, and adult speaking. One of
during which the singer sang the song 4 to 11 times the three videos from each infant was randomly selected
(Mdn = 7). Because the singing was live, there was natu- for double coding, and interrater reliability (computed as
ral variability in performance both within and between the frame-by-frame percentage of agreement across the
sessions. When the adult was not singing, she talked to two coders) was high (82%–99%).
the infants and parents. At the beginning of each session, Although infants’ reactions to songs performed via
she confirmed that the infant could see and hear her, and video chat have never been systematically assessed,
during the session, she asked the parent to reposition the infants find singing highly enjoyable (for a review, see
iPad to maintain the infant’s line of sight, as needed. The Patel, 2008). We expected that if the infants were suffi-
parents were invited to use their discretion in positioning ciently engaged with the adult during the interactive video
the infants; we chose to encourage this because, during sessions, they would smile more and attend longer to her
pilot testing, we determined that most families were when she was singing than when she was speaking.
already accustomed to using interactive video with their
infants; indeed, 86% of the infants in Experiment 3 had
Results
previously used Skype, FaceTime, Hangouts, or some
other form of interactive video chat before participating Main analyses. Parents’ estimates of the total number
in the study. Thus, there was variability in the infants’ of song performances (M = 47.7, SD = 17.0) were slightly
positioning during the sessions: Some infants sat in a lower than the actual number, as determined from the

494 Mehr et al.
recorded video sessions (M = 48.7, SD = 16.9); this differ- reason, we used somewhat different methods to compare
ence was not significant, t(31) = 0.59, p = .56 (paired t the results of Experiments 1 and 3 than we used to com-
test). The total duration of song exposure (M = 17.5 min, pare the results of Experiments 1 and 2. We began by
SD = 6.23; extrapolated from the three videos that were testing for differences between the experiments (expo-
coded for singing duration) was slightly lower than in a sure type) without adjusting for differences in the amount
previous study in which 6-month-old infants were taught of song exposure. Then we proceeded with a regression
songs and subsequently remembered them (21 min; model that differed in one respect from that used to com-
Trainor et al., 2004). Parents reported significantly less pare Experiments 1 and 2: Instead of predicting selective
song exposure in Experiment 3 than in the other two attention from the estimated total number of song perfor-
experiments—comparison with Experiment 1: t(36.6) = mances, after a log2 transformation, we standardized
2.75, p = .009; comparison with Experiment 2: t(38.0) = those values by dividing them by the standard deviation
3.6, p = .001 (Satterthwaite’s t tests)—although the study in each experiment in an effort to reduce the differences
duration was comparable across all three experiments in parents’ self-reporting bias across the two experiments.
(Experiment 1: Mdn = 7 days, range: 7–14; Experiment 2: The predictor can thus be interpreted as standard-devia-
Mdn = 7 days, range: 7–14; Experiment 3: Mdn = 7 days, tion units of song exposure.
range: 7–12). We address possible bias in these reports Both approaches revealed differences between the
later in this section. two experiments. The first showed that at test, the infants
In the selective-attention test, the infants showed no attended significantly longer to the singer of the familiar
preference for either of the two unfamiliar adults at base- song in Experiment 1 than in Experiment 3, t(58.2) =
line. The proportion of time that the infants looked 2.04, p = .046 (Satterthwaite’s t test). The difference
toward the person who sang the familiar song did not between the experiments in the increase in proportion of
differ from chance (.5; M = .479, SD = .183, 95% CI = looking toward the singer of the familiar song from base-
[.413, .545]), t(31) = 0.64, p = .53 (one-sample t test; see line to test was smaller, but in the same direction, t(60.5) =
Fig. 4a). Similarly, the infants attended equally to the two 1.39, p = .086 (Satterthwaite’s t test, one-tailed). We con-
singers during the familiarization trials (duration of look- tinued with the second approach, which included the
ing to the singer of the familiar song: M = 18.4 s, SD = standardized measure of song exposure as a predictor.
4.75; duration of looking to the singer of the unfamiliar The model’s omnibus test yielded a significant result,
song: M = 17.2 s, SD = 4.93), t(31) = 1.28, p = .21 (paired χ2(3) = 11.0, p = .01, R2 = .14 (Wald test), as did a test of
t test). The infants in Experiments 1 and 3 exhibited com- the Exposure Type × Exposure Quantity interaction, b =
parable levels of attention during the baseline period, 0.061, 95% CI = [0.006, 0.115], z = 2.18, p = .029 (z test of
t(61.9) = 0.93, p = .36, and during the familiarization trial the bootstrapped regression coefficient). The inclusion of
for the unfamiliar song, t(61.9) = 1.47, p = .15. However, exposure type and the interaction term significantly
the infants in Experiment 3 attended longer to the famil- increased model fit, χ2(2) = 6.87, p = .032 (nested test).
iarization trial for the familiar song than did the infants in Thus, although the comparison of the experiments was
Experiment 1, t(61.7) = 2.26, p = .027 (all Satterthwaite’s t complicated by the introduction of bias in the measure of
tests). song exposure, the infants’ behaviors at test clearly dif-
The selective-attention effects found in Experiment 1 fered between Experiments 1 and 3.
were not present in Experiment 3. The proportion of time
that the infants looked toward the novel person who Infants’ responses to song and speech presented
sang the familiar song did not differ from chance (.5; M = via interactive video. Across the three sessions that
.488, SD = .232, 95% CI = [.404, .571]), t(31) = 0.30, p = .77 were coded for each infant, we computed four variables:
(one-sample t test), or from the proportion at baseline duration of smiling while the adult sang, duration of
(difference in proportion of looking: M = .009, SD = .198, looking at the video screen while the adult sang, dura-
95% CI = [−.063, .080]), t(31) = 0.24, p = .81 (paired t test; tion of smiling while the adult spoke, and duration of
see Fig. 4a). Further, the amount of song exposure did looking at the video screen while the adult spoke. Each
not predict the infants’ change in attentional preference variable was computed as a proportion of singing time or
toward the new singer of the familiar song (Fig. 4b), speaking time. We then used planned comparisons to
χ2(1) = 0.46, p = .50 (Wald test). test whether the infants smiled and gazed at the video
differentially, within subjects, depending on whether the
Comparison with Experiment 1. Because parents adult was singing or speaking.
may have paid greater attention to songs they themselves When the adult sang, the infants smiled at nearly twice
sang than to songs sung by a toy or another person, their the rate (M = .098, SD = .106, 95% CI = [.059, .136]) that
reports of song exposure may have been depressed in they did when she spoke (M = .054, SD = .062, 95% CI =
Experiment 3, relative to Experiments 1 and 2. For this [.032, .077]), t(31) = 2.85, p = .008 (paired t test). Likewise,

a b
1.0
n.s. .4

.9
.3
.8
.2
.7
.1
.6
n.s. 0
.5
–.1
.4
.3 –.2
.2 –.3
.1 –.4
0 –.5
Before After 20 30 40 50 60 70 80 90
Singing Singing Number of Times Infant Heard Song
(Baseline) (Test)
horizontal lines indicate the medians, the boxes indicate the interquartile ranges, and the vertical lines indicate the full ranges. The
scatterplot in (b) shows each infant’s increase in proportion of looking toward the singer of the familiar song from baseline to test,
along with the predicted effect of the amount of song exposure on this within-subjects main effect, from a bootstrapped model with
40,000 replications. The dashed lines represent ±2 bootstrap standard errors, and chance (0) is indicated by the dotted line.
they attended to the screen at a significantly higher rate than do either musical toys or singing by a minimally
during her singing (M = .623, SD = .184, 95% CI = [.557, familiar adult over interactive video. Did these differing
.689]) than during her speaking (M = .547, SD = .130, 95% effects stem from infants’ differential attention to and
CI = [.500, .594]), t(31) = 4.02, p = .0003 (paired t test). learning of the songs in the three conditions? Although
Given that these positive responses might reflect dif- we know that the infants in Experiment 3 were highly
ferences in infants’ experiences during song exposure attentive to the singer, the infants in Experiments 2 and 3
between Experiments 1 and 3, we conducted two explor- might have failed to learn the song well enough to recog-
atory analyses to test whether these measures predicted nize it at test.
the infants’ preferences regarding the novel individuals at The best test of this possibility would be to compare
test. The results were negative: The rate of smiling during the infants’ later memory for the melody across the three
singing episodes were not related to the within-subjects experiments, but this comparison was not possible. Many
main effect, χ2(1) = 0.23, p = .63 (Wald test), and neither of the parents in Experiment 1 continued to sing the song
was the rate of looking during singing episodes, χ2(1) = after the experiment ended,2 whereas the return of the
0.50, p = .48 (Wald test). Thus, although the infants toy at the end of Experiment 2 and the return of the iPad
responded more positively to songs than to speech, as at the end of Experiment 3 made it unlikely that the
evidenced by two measures, the magnitude of these posi- infants heard the song again. Thus, in Experiments 4 and
tive responses to the singing did not predict the infants’ 5, we investigated long-term retention of the melody
attentional preferences between the novel singers at test. among the participants from Experiments 2 and 3,
respectively.
Three to 12 months after the conclusion of those
Discussion experiments (more than 8 months later, on average), we
The contrasting findings of Experiments 1 through 3 sug- tested whether the infants could discriminate the familiar
gest that parents’ singing has different effects on infants melody from the other melody while listening in the

496 Mehr et al.
same context in which they had originally learned the identity of the singer. Because the two toys were visually
song. In Experiment 4, we tested infants’ attention to two identical, the only basis for distinguishing between them
visually identical toys from Experiment 2; one toy played was the melodies they played. The order of presentation
the familiar song, and the other played the unfamiliar (familiar vs. unfamiliar song first) and the location of
song. In Experiment 5, we tested infants’ attention to two each toy (left side vs. right side) were counterbalanced,
videos of the singer from Experiment 3; she sang the though lower-than-expected recruitment led to a slight
familiar song in one video and the unfamiliar song in the imbalance in presentation order, such that 13 infants
other. heard the familiar song on Trials 1 and 3, and 9 heard it
on Trials 2 and 4 (we report a test for order effects in the
next section). Each infant’s gaze was recorded with a hid-
Experiment 4 den high-definition camera and coded at 30 frames per
second by an experimenter who was blind to the infant’s
Method song familiarity. A second experimenter recoded the vid-
Participants. We attempted to test all 32 infants from eos of all participants, and interrater reliability (computed
Experiment 2, but we failed to reach three families, two by correlating raw looking times, four trials per infant)
families declined to participate, and one family had was high (r = .97).
moved away; in addition, 4 infants participated but were
excluded because they were fussy during the lab session.
Results
Thus, our analyses included 22 full-term infants (11
females; mean age = 13.9 months, SD = 1.50, range: 10.8– Despite the lengthy delay between initial song exposure
17.1), who were tested an average of 8.63 months after and test, the infants looked longer, on average, at the toy
they had last heard the recorded song in Experiment 2 that played the song they had heard at home more than
(SD = 1.51, range: 5.82–11.6 months). During this interim, 8 months earlier than at the toy that played the unfamiliar
some of these infants had returned to the lab for unre- song. This effect was strongest during the first phrase of
lated experiments, but none had received any additional the songs (Fig. 5b; difference in looking time: M = 0.815
exposure to the toy or the recording of the familiar song. s, SD = 1.61, 95% CI = [0.10, 1.53]), t(21) = 2.38, p = .027
We asked the parents if they had sung the toy’s song (paired t test), which is consistent with findings in a pre-
after the original study ended. Parents of 5 of the 22 vious study of musical memory (Saffran, Loman, & Rob-
infants reported singing the song “once or twice” during ertson, 2000), but it remained marginally detectable over
the weeks following Experiment 2, but not afterward. a longer interval (first half of the songs; difference in
looking time: M = 1.28 s, SD = 3.66, 95% CI = [−0.35,
Procedure. The infants sat in a high chair while we 2.90]), t(21) = 1.63, p = .059 (one-tailed). The first-phrase
presented them with toys from Experiment 2, at a dis- effect was no larger for the 5 infants whose parents
tance of approximately 26 in. Parents sat next to their reported singing the familiar song after Experiment 2
infants, facing away from the toys, and wore noise-can- than for the 17 infants whose parents did not report such
celing headphones that played masking music for the exposure (mean difference = 1.46 s, 95% CI = [−1.00,
duration of the experiment. The experiment comprised 3.91]), t(5.13) = 1.51, p = .19 (Satterthwaite’s t test). The
four trials. On each trial, the experimenter, who was size of the main effect did not vary with presentation
unaware of which song had been presented to the order (familiar vs. unfamiliar song first; mean difference =
infant during the original experiment, placed a toy to 0.95 s, 95% CI = [−0.53, 2.43]), t(16.0) = 1.37, p = .19 (Sat-
one side in front of the infant. He said, “Look at this toy, terthwaite’s t test). In an exploratory analysis, we tested
[baby’s name]!” and activated the song (see Fig. 5a). The whether the amount of song exposure in Experiment 2
trials alternated between two visually identical toys, one predicted the strength of the memory effect in Experi-
that played the song the infant had heard in Experiment ment 4. The result was negative: Infants with more song
2 and one that played the other song used in Experi- exposure were no more likely, on average, to show a
ment 2, which the infant had never heard except during greater preference for the toy playing the familiar song
the selective-attention test in that experiment. Thus, over the toy playing the unfamiliar song, χ2(1) = 0.51,
each infant heard the familiar song twice and the unfa- p = .47 (Wald test).
miliar song twice. A given song was presented on the In sum, Experiment 4 provides evidence that the
same side both times. infants in Experiment 2 learned the song from the toy,
On each trial, we measured the infant’s gaze toward discriminated that song from a song with identical lyrics
the toy during the singing (22 s) and afterward in silence and timing but a different melody, and retained this dis-
(16 s). The recordings were identical to those used in tinction between the two highly similar songs for 6 to 12
Experiment 2, and did not differ in lyrics, rhythm, or the months (M > 8 months).

a b
12
*
toy 1
10
Looking Time During First Phrase of Singing (s)

8
toy 2
Order of Events
6
toy 1
toy 2 Toy Playing Toy Playing

Familiar Song Unfamiliar Song
Fig. 5. Procedure and results of Experiment 4. On each of four trials (a), one of two visually identical toys was presented to one side in front
of the infant for the duration of one performance of either the familiar or the unfamiliar song and during a 16-s delay afterward. The two songs
were presented in alternation. The box plots (b) show the duration of looking toward each of the toys during the first phrase of the song. The
Xs indicate the means, the horizontal lines indicate the medians, the boxes indicate the interquartile ranges, and the vertical lines indicate the full
ranges. The asterisk indicates a significant difference (*p < .05).
Experiment 5 from either the singer or interactive video. Parents of 6 of

the 23 infants reported singing the song during the first
Method weeks following Experiment 3, but not afterward. Thus,
Participants. We attempted to test all 32 infants from exposure conditions in Experiment 5 were comparable
Experiment 3, but we failed to reach six families, and to those in Experiment 4.
three families declined to participate. The sample com-
prised 23 full-term infants (8 females; mean age = 5.86 Procedure. The infants sat on the lap of a parent, who
months, SD = 0.50, range: 5.03–6.47), who were tested an had closed eyes and wore noise-canceling headphones
average of 8.48 months after they had last heard the that played masking music for the duration of the experi-
familiar song from Experiment 3 (SD = 2.91, range: 2.99– ment. They viewed high-definition videos of the adult
12.1). During the interim, some of the infants had singer from Experiment 3. On four alternating trials, she
returned to the lab for other experiments, but none had sang the familiar and unfamiliar songs (Fig. 6a). Because
received any additional exposure to the familiar song the same person sang both songs, the familiar song was

498 Mehr et al.
a b
40
*
35
Looking Time During Singing (s)

Order of Events
30
25
20
Familiar Unfamiliar
Song Song
Fig. 6. Procedure and results of Experiment 5. On each of four trials (a), the infants viewed a video of the person who had previously
Skyped with them. She sang either the familiar or the unfamiliar song (on alternating trials) and appeared alongside a green looming object.
The box plots (b) show the duration of looking toward the videos of the singer while she sang the familiar song and while she sang the
unfamiliar song. The Xs indicate the means, the horizontal lines indicate the medians, the boxes indicate the interquartile ranges, and the
vertical lines indicate the full ranges. The asterisk indicates a significant difference (*p < .05).
not identifiable from the identity of the singer’s voice. Results

Because infants are highly attentive to videos of singing,
The infants attended longer to the video of the adult
we presented each song video alongside a video of a
when she sang the same song as in Experiment 3 than
looming object that moved out of synchrony with the when she sang the other song (Fig. 6b; difference in
song, so as to avoid ceiling effects on the infants’ gaze looking time: M = 2.23 s, SD = 4.80), t(22) = 2.27, p = .033
toward the face of the singing adult. The order of the (paired t test). Preferential looking on familiar-song trials
songs (familiar vs. unfamiliar song first) and location of was no higher for the 6 infants whose parents reported
the singer (left vs. right side) were fully counterbalanced. singing the familiar song than for the 17 infants whose
The infants’ gaze was recorded with a hidden high-defi- parents did not report such exposure (mean difference =
nition camera and coded at 30 frames per second by two 2.67 s, 95% CI = [−2.78, 8.12]), t(7.89) = 1.13, p = .29 (Sat-
experimenters, both of whom were unaware of which terthwaite’s t test). Thus, these findings provide evidence
song was familiar to each infant. Interrater reliability that the infants in Experiment 3 learned the song via
(computed by correlating raw looking times, four trials interactive video, discriminated that song from a song
per infant) was high (r = .93). with identical lyrics and timing but a different melody,

and retained this distinction for more than 8 months, on engagement while they learned the song did not predict
average. the degree of the memory effect.
In contrast to Experiment 4, Experiment 5 revealed a
memory effect that was not largest in the first phrase of
General Discussion
the song. Instead, effects were detectable through the
entirety of the song. In addition, the latter two of the Infants selectively attended more to a novel individual
four trials (Trials 3 and 4) drove the overall difference in who sang a song learned from a parent’s singing than to
attention to the singing of the familiar versus unfamiliar a novel individual who sang a contrasting song with the
song (mean difference between Trials 3 and 4 = 2.18 s, same words and rhythms but a different melody. Infants
95% CI = [0.47, 3.89]), t(22) = 2.61, p = .015 (paired t displayed no such preference when they learned the
test). This difference between the effects observed in song from a recording emanating from an inanimate toy
the two experiments is likely due to differences in the or from live video interactions with a singing adult whom
format of the memory test: In Experiment 4, the toy they had met only briefly.
never moved on its own, and its recorded song was Strikingly, the infants in the latter two conditions
always performed identically. Visual attention to the toy remembered the melody to which they had been exposed
therefore dropped off quickly after the beginning of for an average of more than 8 months, in sufficient detail
each test trial. In contrast, in Experiment 5, the infants to discriminate it from a second, highly similar melody
viewed videos of the person who had previously sung (see Fig. 1a and Videos S1 and S2 in the Supplemental
to them in a variable context, and whose behavior var- Material). Moreover, analyses of video recordings of the
ied throughout the test; thus, their attention was better infants as they learned the song in Experiment 3 revealed
sustained throughout each trial. that they exhibited substantially more positive engage-
In an exploratory analysis, we tested whether the ment with the adult while she sang than while she spoke.
amount of song exposure in Experiment 3 predicted the Thus, 5-month-old infants enjoy melodies that are sung by
strength of the memory effect in Experiment 5. As we a variety of people under different conditions, and they
found in the corresponding analysis in Experiment 4, the show long-term retention of melodies learned from a vari-
result was negative: Infants with more song exposure ety of sources. Nevertheless, only a melody produced live
were no more likely, on average, to show a greater pref- by a parent leads infants to display an attentional prefer-
erence for the video of the adult singing the familiar song ence for a new person who sings that melody.
over the video of her singing the unfamiliar song, χ2(1) = The effect of parents’ singing was robust to variation
0.04, p = .84 (Wald test). Finally, we tested whether a rela- in their musical skills: Infants readily identified familiar
tion between song exposure and strength of the memory melodies sung by novel individuals even when their par-
effect might be obtained by collapsing across the two ents’ renditions of those melodies only roughly matched
experiments. In a simple logistic regression, the likeli- the new performances. This finding speaks to parents
hood of a difference in looking time between the two and early-childhood educators who favor high-quality,
songs was unrelated to the degree of prior song expo- professionally recorded music as a source of infants’ song
sure, χ2(1, N = 45) = 0.08, p = .77 (Wald test). Thus, the exposure. Caregivers with low confidence in their musi-
degree of exposure did not contribute to the size of the cal abilities need not worry that the effects of their live
memory effects in Experiments 4 and 5. singing are reduced by their lack of extensive musical
Because song exposure was video-recorded in Experi- training: In our experiments, we could not predict infants’
ment 3, we were able to test whether the infants’ levels of behaviors at test from their parents’ musical abilities.
engagement while they learned the song in that experi- Our findings suggest an early link between live song
ment predicted their memory in Experiment 5. We com- and social engagement that is independent of songs’
puted the size of the memory effect in Experiment 5 as semantic content: Social responses were driven by mel-
the raw difference in duration of looking toward the per- ody alone. This link may be attributable to the early
son singing the familiar song and duration of looking experiences of infants within their families, evolved pre-
toward the person singing the unfamiliar song; we tested dispositions to view songs as signals of a social connec-
for relations between this measure and the measures of tion, or both; our experiments do not distinguish between
smiling and looking during song exposure in Experi- these interpretations.
ment 3. No model showed a clear predictive relation— Moreover, the present experiments do not reveal
smiling only: χ2(1) = 1.32, p = .25; looking only: χ2(1) = whether the selective-attention effect found in Experi-
2.31, p = .13; both smiling and looking: χ2(2) = 2.28, ment 1 was driven by the song being sung by a parent or
p = .32; smiling, looking, and their interaction: χ2(3) = by its being sung in person. It is possible that infants
1.62, p = .66. Although the infants remembered the song show social preferences for new singers of familiar songs
from Experiment 3 after a long delay, their levels of only when those songs have been learned from family

500 Mehr et al.
members. Alternatively, infants may show social prefer- for research assistance; Joseph Anthony and Erica Washburn
ences for new singers of songs learned from any friendly for musical assistance; and Rachel Katz and Adena Schachner
adult, but only when the adult sings to them in person. for assistance with the creation of the selective-attention task.
Further experiments using the same general methods as
the present experiments could distinguish between these Declaration of Conflicting Interests
possibilities. For example, might infants display social The authors declared that they had no conflicts of interest with
preferences for novel singers of songs learned from family respect to their authorship or the publication of this article.
members who sing only over interactive video? We pre-
dict that they will. If so, further analyses of singing over Funding
live video could then determine the particular perfor- This work was supported by the Dana Foundation grant “Arts
mance features that elicit social interpretations of music. and Cognition: Effects of Music Instruction on Cognitive Sys-
For example, songs may convey social information to tems at the Foundations of Mathematics and Science” and by
infants more effectively the more the singers are attuned the Harvard University Mind/Brain/Behavior Interfaculty Initia-
to the infants’ affective states, varying their singing style in tive grant “Caregiving and the Core Functions of Music.”
a fashion coordinated with the infants’ responses.
Whatever the findings of such studies, the present Supplemental Material
research demonstrates that the social information con- Additional supporting information can be found at http://pss
veyed by a melody depends on its original source. We .sagepub.com/content/by/supplemental-data
found that a melody conveys social information about its
singer when it is sung by a parent, during the course of Open Practices
parent-infant interactions. In contrast, such social infor-
mation is not conveyed when the melody is produced by
a musical toy, even if the parent is highly engaged with
All data and materials have been made publicly available via
the infant while playing with the toy, or when the melody
Open Science Framework and can be accessed at http://osf.io/
is sung by an interactive but socially unrelated adult,
y3kzd and http://osf.io/64vqh, respectively. The complete Open
even if the infant is highly engaged with that adult in Practices Disclosure for this article can be found at http://pss
interactive video sessions. .sagepub.com/content/by/supplemental-data. This article has
Why might melodies serve a social function? In small- received badges for Open Data and Open Materials. More infor-
scale human societies, child rearing is conducted by mul- mation about the Open Practices badges can be found at https://
tiple individuals who communicate with one another, osf.io/tvyxz/wiki/1.%20View%20the%20Badges/ and http://pss
sharing language and music (e.g., Hrdy, 2009). Infants .sagepub.com/content/25/1/3.full.
might do well to identify individuals who could care for
them when a parent is not available, and attending to Notes
other people’s speech and singing might facilitate such 1. All regressions reported in this article were bootstrapped
identification. Indeed, some researchers have proposed with 40,000 replications and subsequently validated with sensi-
that the human music faculty evolved in the context of tivity analyses, to ensure that their results were not attributable
child rearing (e.g., Trehub, 2003). Our studies did not test to the presence of influential observations.
any particular evolutionary theory, but this speculative 2. Parents volunteered this information when they returned to
interpretation of our findings is consistent with the pos- our lab to participate in unrelated studies.
sibility that in ancestral environments, infants’ caregivers
reliably produced melodies, and infants reliably listened References
to and remembered them. Bornstein, R. F. (1989). Exposure and affect: Overview and
meta-analysis of research, 1968–1987. Psychological
Author Contributions Bulletin, 106, 265–289. doi:10.1037/0033-2909.106.2.265
Brown, D. E. (1991). Human universals. New York, NY:
S. A. Mehr and E. S. Spelke designed the research. S. A. Mehr
McGraw-Hill.
collected the data for Experiments 1, 2, 4, and 5. L. A. Song col-
Conard, N. J., Malina, M., & Münzel, S. C. (2009). New flutes
lected the data for Experiment 3, under the supervision of S. A.
document the earliest musical tradition in southwestern
Mehr and E. S. Spelke. S. A. Mehr analyzed the data. S. A. Mehr
Germany. Nature, 460, 737–740. doi:10.1038/nature08169
and E. S. Spelke wrote the manuscript.
Custodero, L. A., & Johnson-Green, E. A. (2003). Passing the
cultural torch: Musical experience and musical parenting of
Acknowledgments infants. Journal of Research in Music Education, 51, 102–
We thank the children and parents for their participation; Ellyn 114. doi:10.2307/3345844
Schmidt and Rosa Guzman for administrative assistance; Bryan Darwin, C. (1871). The descent of man, and sexual selection in
Zuluaga, Olivia Sommer, Veri Seo, Amy Bu, and Marina Ebert relation to man. London, England: Watts.

Farroni, T., Csibra, G., Simion, F., & Johnson, M. H. (2002). Saffran, J. R., Loman, M. M., & Robertson, R. R. (2000). Infant
Eye contact detection in humans from birth. Proceedings memory for musical experiences. Cognition, 77, B15–B23.
of the National Academy of Sciences, USA, 99, 9602–9605. doi:10.1016/S0010-0277(00)00095-0
doi:10.1073/pnas.152159999 Savage, P. E., Brown, S., Sakai, E., & Currie, T. E. (2015). Statistical
Feierabend, J. M. (2000). The book of lullabies: Wonderful songs universals reveal the structures and functions of human
and rhymes passed down from generation to generation for music. Proceedings of the National Academy of Sciences,
infants and toddlers. Chicago, IL: GIA. USA, 112, 8987–8992. doi:10.1073/pnas.1414495112
Gordon, E. E. (1989). Advanced measures of music audiation. Schachner, A., & Hannon, E. E. (2011). Infant-directed speech
Reston, VA: GIA. drives social preferences in 5-month-old infants. Develop-
Hrdy, S. B. (2009). Mothers and others: The evolutionary ori- mental Psychology, 47, 19–25. doi:10.1037/a0020740
gins of mutual understanding. Cambridge, MA: Harvard Spencer, H. (1857). The origin and function of music. Fraser’s
University Press. Magazine for Town and Country, 56, 396–407.
James, W. (1890). The principles of psychology. New York, NY: Trainor, L. J., Wu, L., & Tsang, C. D. (2004). Long-term
Henry Holt. memory for music: Infants remember tempo and timbre.
Kinzler, K., Dupoux, E., & Spelke, E. S. (2007). The native Developmental Science, 7, 289–296. doi:10.1111/j.1467-7687
language of social cognition. Proceedings of the National .2004.00348.x
Academy of Sciences, USA, 104, 12577–12580. doi:10.1073/ Trehub, S. E. (2003). The developmental origins of musicality.
pnas.0705345104 Nature Neuroscience, 6, 669–673. doi:10.1038/nn1084
Lomax, A. (1968). Folk song style and culture. Washington, DC: Trehub, S. E., Unyk, A. M., & Trainor, L. J. (1993). Adults
American Association for the Advancement of Science. identify infant-directed music across cultures. Infant
Mehr, S. A. (2014). Music in the home: New evidence for Behavior & Development, 16, 193–211. doi:10.1016/0163-
an intergenerational link. Journal of Research in Music 6383(93)80017-3
Education, 62, 78–88. doi:10.1177/0022429413520008 Winkler, I., Háden, G. P., Ladinig, O., Sziller, I., & Honing,
Mehr, S. A. (2015). Miscommunication of science: Music cogni- H. (2009). Newborn infants detect the beat in music.
tion research in the popular press. Frontiers in Psychology, Proceedings of the National Academy of Sciences, USA, 106,
6, Article 388. doi:10.3389/fpsyg.2015.00988 2468–2471. doi:10.1073/pnas.0809035106
Patel, A. D. (2008). Music, language, and the brain. New York, Zajonc, R. B. (2001). Mere exposure: A gateway to the sub-
NY: Oxford University Press. liminal. Current Directions in Psychological Science, 10,
Roseberry, S., Hirsh-Pasek, K., & Golinkoff, R. M. (2014). Skype 224–228. doi:10.1111/1467-8721.00154
me! Socially contingent interactions help toddlers learn Zentner, M., & Eerola, T. (2010). Rhythmic engagement with
language. Child Development, 85, 956–970. doi:10.1111/ music in infancy. Proceedings of the National Academy of Sci-
cdev.12166 ences, USA, 107, 5768–5773. doi:10.1073/pnas.1000121107

RESEARCH

Uploaded by

Copyright:

Available Formats

RESEARCH

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

RESEARCH

Uploaded by

Copyright:

Available Formats

ABDELRAHMAN ANWAR ATTIA RAGAB RAMADAN

ii. Critical analysis

b. Considerations about the chosen experimental sample and data analysis

For 5-Month-Old Infants, Melodies 2016, Vol. 27(4) 486­–501

Samuel A. Mehr1,2, Lee Ann Song2, and Elizabeth S. Spelke2

Received 10/13/14; Revision accepted 12/17/15

Downloaded from pss.sagepub.com at University of Bristol Library on May 16, 2016

Currie, 2015). Infants may be attentive to this fact, and Experiment 1

Downloaded from pss.sagepub.com at University of Bristol Library on May 16, 2016

Lul - la - by, my lit - tle ba - by, soft - ly sleep, my child.

Lul - la - by, my lit - tle ba - by, soft - ly sleep, my child.

Downloaded from pss.sagepub.com at University of Bristol Library on May 16, 2016

Downloaded from pss.sagepub.com at University of Bristol Library on May 16, 2016

Increase in Proportion of Looking From Baseline to Test

Before After 10 20 40 80 160 320

Experiment 2 infants at home. (Parent’s pitch accuracy and music apti-

Downloaded from pss.sagepub.com at University of Bristol Library on May 16, 2016

Downloaded from pss.sagepub.com at University of Bristol Library on May 16, 2016

Increase in Proportion of Looking From Baseline to Test

Downloaded from pss.sagepub.com at University of Bristol Library on May 16, 2016

Downloaded from pss.sagepub.com at University of Bristol Library on May 16, 2016

Downloaded from pss.sagepub.com at University of Bristol Library on May 16, 2016

Increase in Proportion of Looking From Baseline to Test

Downloaded from pss.sagepub.com at University of Bristol Library on May 16, 2016

Downloaded from pss.sagepub.com at University of Bristol Library on May 16, 2016

Looking Time During First Phrase of Singing (s)

toy 2 Toy Playing Toy Playing

Experiment 5 from either the singer or interactive video. Parents of 6 of

Downloaded from pss.sagepub.com at University of Bristol Library on May 16, 2016

Looking Time During Singing (s)

not identifiable from the identity of the singer’s voice. Results

Downloaded from pss.sagepub.com at University of Bristol Library on May 16, 2016

Downloaded from pss.sagepub.com at University of Bristol Library on May 16, 2016

Downloaded from pss.sagepub.com at University of Bristol Library on May 16, 2016

Downloaded from pss.sagepub.com at University of Bristol Library on May 16, 2016

You might also like

For 5-Month-Old Infants, Melodies 2016, Vol. 27(4) 486–501