DOI: 10.1111/cdev.

14136

EMPIRICAL ARTICLE

Understanding allostasis: Early-­life self-­regulation involves both

up-­and down-­regulation of arousal

S. V. Wass1 | F. U. Mirza1 | C. Smith2

1
Department of Psychology, University of Abstract
East London, London, UK
2
Optimal performance lies at intermediate autonomic arousal, but no previous
Institute of Psychiatry, Psychology &
Neuroscience, King's College London,
research has examined whether the emergence of endogenous control associates
London, UK with changes in children's up-­regulation from hypo-­arousal, as well as down-­
regulation from hyper-­arousal. We used wearables to take day-­long recordings
Correspondence
S. V. Wass, Department of Psychology,
from N = 58, 12-­month-­olds (60% white/58% female); and, in the same infants, we
University of East London, London E15 measured self-­regulation in the lab with a still-­face paradigm. Overall, our findings
4LZ, UK. suggest that infants who showed more self-­regulatory behaviors in the lab were
Email: [email protected]
more likely to actively change their behaviors in home settings moment-­by-­moment
Funding information “on the fly” following changes in autonomic arousal, and that these changes
H2020 European Research Council, result in up-­as well as down-­regulation. Implications for the role of atypical self-­
Grant/Award Number: ONACSA 853251;
Economic and Social Research Council,
regulation in later psychopathology are discussed.
Grant/Award Number: ES/N017560/1

Freud first defined the central nervous system as: “an of self-­regulation assess how quickly a child soothes fol-
apparatus which has the function of getting rid of the lowing crying or distress (Derryberry & Rothbart, 1988);
stimuli that reach it, or of reducing then to the lowest pos- and lab-­based assessments measure a child's behavioral
sible level; or which, if it were feasible, would maintain and physiological responses to a stressor such as arm re-
itself in an altogether unstimulated condition” (Freud straint, or the still-­face procedure (Gagne et al., 2011)—
& Strachey, 1900). Since then, many researchers who both of which are responses to hyper-­ arousal. But
examine the development of regulation have, implicitly, neither questionnaires nor laboratory assessments typ-
followed Freud's approach. Hypo-­arousal as a trait-­level ically measure how children change their behaviors in
feature has been discussed in the context of developmen- response to hypo-­arousal.
tal psychopathologies such as attention deficit hyper- There are, of course, a number of reasons why this is
activity disorder (Bellato et al., 2020) and psychopathy the case. Focusing on behavioral down-­ regulation has
(Verona et al., 2004). But researchers who study active been fruitful, as relevant questionnaire measures based on
regulatory processes, defined as “the ongoing, dynamic, this approach reliably predict later-­ life psychopathology
and adaptive modulation of internal state (emotion, cog- and cognitive outcomes (Eisenberg et al., 2019; Kostyrka-­
nition) or behaviour, mediated by central and peripheral Allchorne et al., 2020; Nigg, 2017). But this concentration
physiology” (Nigg, 2017), mainly tend to study how chil- on behavioral responses to hyper-­arousal is nevertheless
dren down-­regulate arousal in response to hyperstimu- striking given that animal and human research suggests
lation of the autonomic nervous system (ANS), through that, in fact, optimal cognitive performance lies at inter-
behaviors such as gaze aversion, physical self-­soothing, mediate level of ANS activity, with both hypo-­(under)
and distraction (Kopp, 1982; Posner & Rothbart, 2000). as well as hyper-­(over) arousal associated with poorer
For example, most widely used questionnaire measures cognitive performance (Aston-­ Jones & Cohen, 2005;

Abbreviations: ANS, autonomic nervous system; ECG, electrocardiogram; PACF, partial autocorrelation function; RMSSD, root mean square of successive
differences.

This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium,
provided the original work is properly cited.
© 2024 The Author(s). Child Development published by Wiley Periodicals LLC on behalf of Society for Research in Child Development.

2000 | 
wileyonlinelibrary.com/journal/cdev Child Development. 2024;95:2000–2014.
 14678624, 2024, 6, Downloaded from https://srcd.onlinelibrary.wiley.com/doi/10.1111/cdev.14136 by National Health And Medical Research Council, Wiley Online Library on [09/06/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
SELF REG AND UP REG    | 2001

Porges, 2007; Porges & Furman, 2011; Wass, 2021; Yerkes & measured traditionally assessed in the lab. To do this,
Dodson, 1908). Historically, theorists have suggested, con- we first obtained day-­long naturalistic recordings using
sistent with this, that there is an optimal level of stimulation wearable sensors microphones, cameras, and physiologi-
toward which organisms strive (Berlyne, 1960; Dunn, 2002; cal monitors (measuring heart rate, heart rate variability,
Fiske & Maddi, 1961; Piccardi & Gliga, 2022; Porges, 2007; and movement). Then, to the same infants, we admin-
Schneirla, 1959; Wachs, 1977; Zentall & Zentall, 1983; istered a standard mild stressor (a still-­face procedure,
Zuckerman, 1979), with any deviation from optimal stim- Weinberg & Tronick, 1996) to a cohort of typical infants,
ulation (hypo as well as hyper) leading to aversive states for and coded self-­regulation in the traditional manner, by
which individuals need to compensate via allostasis (the ac- video-­coding down-­regulatory behaviors such as clos-
tive process through which homeostasis is maintained by an ing eyes, gaze aversion, and physical self-­soothing (see
organism, Cannon, 1929; McEwen & Wingfield, 2003). For “Methods” section for more details).
example, Gardner, Karmel and colleagues measured how In comparing, for the first time, associations between
young infants' preference for less-­arousing, low-­frequency lab-­assessed self-­regulation and self-­regulation and auto-
visual stimuli versus more-­arousing, high-­frequency visual nomic arousal fluctuations in real-­world settings, we had
stimuli varied contingent on their own arousal (Gardner three main questions. First, is better self-­regulation (as
& Karmel, 1984, 1995). They found that highly aroused assessed using a traditional approach in a lab setting) as-
1-­month-­old infants preferred less arousing, low-­frequency sociated with lower arousal overall in real-­world settings,
stimuli, whereas less-­ aroused infants preferred more-­ and changes in affect and intensity? Second, is better
arousing, high-­frequency stimuli (Gardner & Karmel, 1984, self-­regulation in lab settings associated with differences
1995). These results suggest that typically developing young in the rate of change of affect, intensity, and arousal in
infants dynamically recalibrate their attentional behaviors real-­world settings (see Somers & Luecken, 2022; Somers,
to up-­regulate their own arousal when it is low, as well as Luecken, et al., 2021)? Third, is better self-­regulation in
to down-­regulate their own arousal when it is high (see also lab settings associated with greater evidence for the exis-
Cole, Ram, et al., 2019; Cole, Ramsook, et al., 2019; Morales tence of allostatic processes—and, specifically, increased
et al., 2018). Other research has discussed similar allostatic up-­regulation from hypo-­arousal in real-­world settings?
mechanisms in the context of caregiver–child dyads One specific behavior that we examined in detail is
(Lunkenheimer et al., 2020; Somers, Curci, et al., 2021; spontaneous infant vocal behaviors. Previous research
Somers, Luecken, et al., 2021; Wass et al., 2019). based on the same dataset has suggested that, consis-
An approach that studies up-­as well as down-­ tent with theoretical predictions (Kopp, 1982), clusters of
regulation of arousal will allow us to study how optimal vocalizations (both cries and speech-­like vocalizations)
levels of autonomic arousal may differ between individu- tend overall to occur during periods of elevated infant
als, a possibility that has been suggested (Berlyne, 1960; arousal (Wass et al., 2019, 2022). Cries occur following re-
Fiske & Maddi, 1961; Wass, 2021; Zuckerman, 1979) duced infant arousal stability and elicit changes in care-
but not, to our knowledge, studied (although see, e.g., giver behavior (Wass et al., 2019), which lead to arousal
Somers, Luecken, et al., 2021). If true, this would have down-­regulation via co-­regulation (Wass et al., 2022).
fundamental consequences for how we understand self-­ Speech-­like vocalizations, which are neutral in affect,
regulation. For example, if a child with adverse life expe- also occur at elevated arousal, but lead to longer-­lasting
riences is assessed via questionnaire or laboratory task increases in arousal, and elicit more parental verbal re-
as not soothing quickly following an external stimulus sponses (Wass et al., 2022).
which induces hyper-­arousal, this may be because they In part 1 of our analyses, we examine our data overall.
are trying to down-­regulate and failing; or, it may be be- We predicted that increased lab-­assessed self-­regulation
cause they are using allostasis effectively, but to main- would associate with: reduced overall arousal in natural-
tain the higher level of ANS arousal that is, for them, istic settings; reduced vocal affect and intensity in real-­
optimal (see also Bellato et al., 2020). world settings; a slower rate of change of arousal and
Previous observational and experimental research has reduced variability in vocal affect and intensity in nat-
suggested even neonates can show some down-­regulatory uralistic settings; and faster recovery following “peak”
behaviors (Brazelton, 1983; Stifter & Braungart, 1995). arousal events.
But little previous research has examined whether, and In part 2 of our analyses, we examine how infants' re-
if so how, the capacity to actually use allostatic self-­ sponses and behaviors differ contingent on hypo-­versus
regulation in real-­world settings develops across early de- hyper-­arousal. First, we examined how the stability of
velopment (Cole et al., 2020; Cole, Ramsook, et al., 2019; arousal (i.e., the likelihood of being in the same state at
Wass, 2021). Here, we examine this, concentrating in time t + 1 as at time t) varies contingent on a child's arousal
particular on examining up-­as well as down-­regulatory level at time t. Furthermore, we examined how these rela-
patterns at 12 months, which is thought to be the age tionships differ contingent on lab-­assessed self-­regulation.
when endogenous control can be robustly measured for We predicted that children with better self-­regulation (as
the first time (Colombo & Cheatham, 2006; Wass, 2021). assessed using conventional approaches in the lab) would
Furthermore, we examine how the prevalence of real-­ show both more down-­regulation from hyper-­arousal and
world allostatic processes relates to self-­regulation as more up-­regulation from hypo-­arousal. Second, we tested
 14678624, 2024, 6, Downloaded from https://srcd.onlinelibrary.wiley.com/doi/10.1111/cdev.14136 by National Health And Medical Research Council, Wiley Online Library on [09/06/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2002 |    WASS et al.

how infants' responses to hyper-­and hypo-­stimulation Infants taking part in the study were recruited from
differ contingent on lab-­assessed self-­regulation by using the South-East regions of the UK. Usable data from both
dynamical generative models. Our predictions were the components of the study—the lab visit and home visit—
same as for the first hypothesis. Third, we examined how were obtained from 58 infants (mean (SE) age in days:
the relationship between vocalization likelihood and 317.8 (4.4)). Demographic details for the sample, includ-
arousal, and the effects of vocalizing on arousal, differed ing the entire sample and the sample from whom usable
contingent on lab-­assessed self-­regulation. We predicted data were obtained, are given in Table S1. A further N = 15
that, in children with better lab-­assessed self-­regulation, took part but failed to provide enough data to take part
their autonomic arousal would be more predictive of in the study (a full comparison of participants who did
vocal behaviors. and did not provide usable data is included in Table S2).

M ET HOD S Equipment and procedures

Participants Home visit

Participants were recruited by researchers visiting in- Protocol

fant–parent groups, both privately organized and The lab and home testing sessions were scheduled within
through local councils, as well as by mail by purchas- median (SE) of 17.5 (3.5) days of one another. For the
ing names and addresses of families in specific income home visit, parents selected a day for which they would
brackets and areas of the South-­East area of the UK. be spending the entire day with their child but that was
Exclusion criteria were as follows: complex medical con- otherwise, as far as possible, typical. The researcher vis-
ditions, skin allergies, heart conditions, parents below ited the participants' homes in the morning (between
18 years of age, and parents receiving care from a mental 7:30 a.m. and 10:00 a.m.) to fit the equipment and ex-
health organization or professional. We also excluded plain its use, and then returned in the late afternoon (be-
families in which the primary day-­time care was per- tween 4:00 p.m. and 7:00 p.m.) to remove it. Mean (SD)
formed by a male parent because the numbers were insuf- recording time per day was 7.3 (1.4) hours. Participants
ficient to provide an adequately gender-­matched sample. reported no discomfort and disruption to their normal
To assess demographics and maternal characteristics, routines from wearing the equipment.
we asked parents to self-­report occupation, education,
and income. To assess maternal anxiety and depression, Equipment
we administered the general anxiety disorder-­7 (GAD-­7) The equipment consisted of two wearable layers (see
(Spitzer et al., 2006) and patient health questionnaire-­9 Figure 1b). A specially designed baby grow was worn next
(PHQ-­9) (Kroenke et al., 2001) questionnaires. Data col- to the skin, containing a built-­in electrocardiogram (ECG)
lection was conducted between 2018 and 2021. recording device (recording at 250 Hz), accelerometer

F I G U R E 1 (a) A sample plot of raw data collected from a single participant showing (from top to bottom): the home/awake coding
(conducted as described in the “Methods” section); the arousal composite score (calculated as described in the “Methods” section); the arousal
composite score after removal of the autocorrelation (as described in the “Methods” section); the vocal intensity coding; and the vocal affect
coding. (b) Photographs showing the equipment (top) and the same equipment being worn by some participants (bottom).
 14678624, 2024, 6, Downloaded from https://srcd.onlinelibrary.wiley.com/doi/10.1111/cdev.14136 by National Health And Medical Research Council, Wiley Online Library on [09/06/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
SELF REG AND UP REG    | 2003

(30 Hz), GPS (1 Hz), and microphone (11.6 kHz). A T-­ amount of data entered into the analyses was 3.7 h (1.7 h)
shirt, worn on top of the device, contained a pocket to per participant.
hold the microphone and a miniature video camera (a
commercially available Narrative Clip 2). The clothes were Vocal affect and intensity coding
comfortable when worn and, other than a request to keep For reasons of bandwidth, the microphone recorded
the equipment dry, participants could behave exactly as just a 5-­s snapshot of the auditory environment every
they would on a normal day. No discomfort in wearing 60 s. To assess the impact of this “sparse sampling” on
the equipment was generally reported. To ensure good-­ our data, we performed a separate control analysis (see
quality recordings, the ECG was attached using standard Section S1.6; Figure S4), which suggested that the tem-
Ag-­Cl electrodes placed in a modified lead II position. poral structure of our vocalizations was maintained
despite this “sparse sampling” approach. Furthermore,
Autonomic monitoring since our analyses examine average patterns of arousal
Autonomic arousal was indexed by recording electrocar- change around observed vocalizations, we reasoned that
diography, from which heart rate and heart rate variabil- any observed changes would be weakened (not strength-
ity were derived, and actigraphy. ened) by the fact that the vocalization data were sparsely
Data were parsed to identify the time intervals be- sampled (because signal to noise could only be increased
tween the R peaks of the ECG signal using custom-­built by missing vocalizations).
Matlab scripts, employing an adaptation of a standard Post hoc, trained coders listened to each recording
thresholding procedure (Wass et al., 2015), and verified to identify samples in which the infant was vocalizing.
post hoc via visual inspection. Further details of the pars- The infant vocalizations were also coded for vocal affect
ing procedure are given in the Supporting Information on a scale from 1 (fussy and difficult) to 5 (happy and
(Section S1.2; Figures S1 and S2). engaged), and for vocal intensity on a scale from 1 (mild)
Heart rate variability was calculated using the to 5 (intensely engaged). In order to assess inter-­rater
PhysioNet Cardiovascular Signal Toolbox (Vest et al., reliability, 24% of the sample was double coded; Cohen's
2018). A 60-­s window with an increment of 60 s was im- kappa for this coding was .60, which is acceptable
plemented, and the default settings were used with the (McHugh, 2012). All coders were blinded to all intended
exception that the min/max inter-­beat interval was set analyses. The relationship between vocal affect and in-
at 300/750 ms for the infant data and 300/1300 ms for the tensity among all vocalizations is shown in Figure 3d.
adult data. Heart rate variability was calculated from
the root mean square of successive differences (RMSSD) Vocalization type coding
(see further details in Section S1.3.1). A morphological coding scheme (Oller et al., 2013) was
To parse the actigraphy data, the data were first applied with the following categories: cry, laugh, squeal,
manually inspected, then corrected for artifacts spe- growl, quasi-­resonant vowel, fully resonant vowel, mar-
cific to the recording device used. Following that, a ginal syllable, and canonical syllable. Overall, 29% of vo-
Butterworth low-­pass filter with a cut-­off of 0.1 Hz was calizations were cries; 1% laughs; 1% squeal; 3% growl;
used to remove high-­frequency noise (see further details 18% quasi-­ resonant vowel; 18% fully-­ resonant vowel;
in Section S1.3.2). 6% marginal syllable; and 23% canonical syllable. These
The preliminary analyses suggested that the three au- were collapsed into cries and speech-­like vocalizations,
tonomic measures showed strong patterns of tonic and which included the following non-­cry categories: quasi-­
phasic covariation, consistent with previous research resonant vowel; fully resonant vowel; marginal syllable;
(Wass et al., 2015, 2016). See further details, and dis- and canonical syllable. Laughs, squeals, and growls were
cussion, in the Supporting Information (Section S1.4; excluded due to rarity. In order to assess inter-­rater reli-
Figure S4). Motivated by this, and by analogous find- ability, 11% of the sample was double coded, to assess
ings in the animal literature (Calderon et al., 2016; the consistency of the cries (1) versus speech-­like vocali-
Pfaff, 2018), we collapsed the autonomic indices into a zations (0) distinction on which the analysis was based;
single composite measure. Cohen's kappa was .70, which is considered substantial
Our analyses suggested that when infants were out- agreement (McHugh, 2012).
side they were often strapped into a buggy or car seat,
which strongly affected their autonomic data. Because Permutation-­based temporal clustering analyses
of this, only segments of the data when infants were To estimate the significance of time-­series relationships,
at home were included. Since all aspects of autonomic a permutation-­based temporal clustering approach was
arousal differ markedly between waking and sleeping used. This procedure, which is adapted from neuroim-
(see, e.g., Figure 1), sections where infants were asleep aging (Maris, 2012; Maris & Oostenveld, 2007), allows
were also excluded. Details of the criteria through us to estimate the probability of temporally contiguous
which home/not-­ home and sleeping/waking segments relationships being observed in our results, a fact that
were identified are given in the Supporting Information standard approaches to correcting for multiple com-
(Section S1.5). Following these exclusions, the mean (SD) parisons fail to account for (Maris, 2012) (see also Oakes
 14678624, 2024, 6, Downloaded from https://srcd.onlinelibrary.wiley.com/doi/10.1111/cdev.14136 by National Health And Medical Research Council, Wiley Online Library on [09/06/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2004 |    WASS et al.

et al., 2013). In each case, the test statistic (always speci- and child were seated across an 80-­c m-­w ide table, and
fied in the text) was calculated independently for each instructed to play naturally with four toys positioned on
time window. Series of significant effects across contigu- the table (see Figure 2a). After 4 min, on an instruction
ous time windows were identified using an alpha level from the experimenter, the parent was instructed not to
of .05. A total of 1000 random datasets were then gener- respond to the infant and to hold a neutral face for 2 min.
ated with the same dimensions as the original input data. On a further instruction from the experimenter, the play
Then, the same sequence of analyses was repeated, and resumed for a further 2 min. If the infant become dis-
the longest series of significant effects across contiguous tressed during the still-­face period, as judged using the
time windows was identified. The results obtained from standard guidelines (Weinberg & Tronick, 1996), the ex-
the random datasets were used to generate a histogram, periment was curtailed.
and the likelihood that observed results had been ob- Self-­regulation behaviors were coded from the NDN
tained by chance was calculated by comparing the ob- Physical Pilot coding scheme. Data were coded in 5-­s
served values with the randomly generated values using bins. Self-­regulatory behaviors were coded as the fol-
a standard bootstrapping procedure. Thus, a p value of lowing categories: gaze aversion (looks to experiment/
<.01 indicates that an equivalent pattern of temporally room)/gaze aversion (looks to toy)/closes eyes/plays with
contiguous group differences was observed in 10 or fewer object/physical self-­touch/covers face/turns away (see
of the 1000 simulated datasets created. Figure 2b). In addition, facial affect was coded on a 5-­
point scale from positive (happy and engaged) to neg-
ative (fussy and irritable). Ten percent of the sample
Lab visit was double coded to test inter-­rater reliability. This was
Cohen's kappa .71, which is acceptable.
The self-­regulation task was a standard version of the Self-­regulation was coded as the mean number of
still-­face protocol (Weinberg & Tronick, 1996). Parent self-­regulatory behaviors per time bin. Figure 1c shows

F I G U R E 2 (a) Photographs showing the setup for the still-­face procedure; (b) examples of the coding of self-­regulatory behaviors; (c) bar
chart showing the proportion of different self-­regulatory behaviors observed in our data. Error bars show between-­participants standard error;
(d) plot showing the change in mean facial affect during the 2 min of the still-­face paradigm, subdivided by median split into the high and low
self-­regulatory groups; (e) cross-­c orrelation showing the cross-­c orrelation between self-­regulatory behaviors and facial affect. Red line shows
the area identified as showing a significant difference from zero in the cluster-­based permutation analysis.
 14678624, 2024, 6, Downloaded from https://srcd.onlinelibrary.wiley.com/doi/10.1111/cdev.14136 by National Health And Medical Research Council, Wiley Online Library on [09/06/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
SELF REG AND UP REG    | 2005

F I G U R E 3 (a, b) Histograms showing the distributions of vocal intensity (a) and vocal affect (b) across all infant vocalizations recorded,
subdivided by self-­regulation. (c) Stacked bar chart showing the relationship between vocal intensity (x-­a xis) and vocal affect (color bar scale).

a histogram of the results obtained. For the time-­series Part 1—Associations between
analyses, where it is not possible to look at group dif- self-­regulation and arousal, and vocal affect and
ferences based on a continuous variable, data have been vocal intensity in home settings
split using a median split. The median was 3.84. Table S1
provides a break-­down of the demographic data split Part 1 is in three subsections. First (part 1a), as a pre-
by high/low self-­regulation. Associations between self-­ liminary analysis, we present associations between
regulation and other demographic variables are reported self-­
regulation and demographics and maternal char-
in the results. acteristics. Then, we examine associations between self-­
In order to confirm the validity of our self-­regulation regulation and vocal affect and intensity (part 1b) and
measure, we conducted two additional analyses. First, autonomic arousal in home settings (part 1c).
we divided our group (based on a median split as defined
above) into low/high self-­regulation based on the self-­
regulatory behaviors described above, and examined 1a—Demographics and maternal
how the two groups differed on facial affect (Figure 2d). characteristics
Second, we calculated a cross-­correlation (Figure 2e)
to examine the temporal association between facial af- For all correlations and other statistical tests, more con-
fect (coded dimensionally from negative to positive, as servative non-­parametric statistics are used throughout
shown in Figure 2b) and regulatory behaviors (as also because not all variables were parametrically distrib-
shown in Figure 2b). Areas where the cross-­correlation uted. Infant self-­regulation showed a significant positive
differed significantly from chance were identified and correlation with maternal occupation (ρ = .44, p = .008)
corrected for multiple comparisons using the same but not with other demographic assessments such as ma-
permutation-­based temporal clustering procedure as de- ternal education (p = .86). No associations were observed
scribed above (see also Wass et al., 2019). A significant between infant self-­ regulation and maternal anxiety
positive cross-­correlation was observed around lag 0, (p = .64) and depression (p = .09).
indicating a time-­specific association between increased Poorer infant self-­ regulation was associated with
self-­regulatory behaviors and less negative/more positive higher maternal vocal intensity (ρ = −.52, p = .002) but
affect (Figure 2e). not vocal affect (p = .65). Better infant self-­ regulation
Confirmatory/exploratory statement: analyses de- was associated with higher maternal autonomic arousal
scribed below are relatively exploratory in nature. (ρ = .31, p = .04).
Analyses 1c, 2a, 2b, and 2c use methods that have not, to
our knowledge, been conducted before.
1b—Vocal affect and intensity

R E SU LT S No associations were observed between infant self-­

regulation and infant vocal intensity (p = .62, Figure 3a)
Our results section is in two parts. In part 1, we examine and affect (p = .45, Figure 3b) in home settings. Figure 3c
associations between lab-­ assessed self-­
regulation and shows the association between vocal affect and vocal in-
arousal, and vocal affect and vocal intensity in home set- tensity. We also calculated the variability in vocal affect
tings. In part 2, we specifically examine how responses and intensity by concatenating all vocalizations and cal-
to hyper-­and hypo-­arousal in home settings differ con- culating the RMSSD. This analysis was based on a 60-­s
tingent on lab-­assessed self-­regulation. epoch duration (which is the duration that we used for all
 14678624, 2024, 6, Downloaded from https://srcd.onlinelibrary.wiley.com/doi/10.1111/cdev.14136 by National Health And Medical Research Council, Wiley Online Library on [09/06/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2006 |    WASS et al.

analyses except those that specifically compare between Figure 4c shows the lag terms of the PACF at different
epoch durations). Here, we observed that poorer infant epoch durations. At low epoch durations the lag 1 term
self-­regulation associated with more variability in vocal of the PACF is higher, indicating that the arousal data
affect (ρ = −.48, p = .003) and poorer infant self-­regulation show more autocorrelation at shorter epoch durations,
was marginally associated with greater variability in as expected. (Of note, at longer epoch durations (20 and
vocal intensity (ρ = −.32, p = .06). 30 min) the lag 2 term is significantly negative, pointing
to low-­frequency oscillatory changes that have not to our
knowledge been documented.)
1c—Autonomic arousal Figure 4d shows just the lag 1 term of the PACF,
with the different epoch durations used for the down-­
No associations were observed between infant self-­ sampling the data prior to calculating the autocorrela-
regulation and infant autonomic arousal in home settings tion shown on the x-­axis. The data have been subdivided
(ρ = .09, p = .58). In addition, to examine variability in au- into high and low self-­regulation groups. Permutation-­
tonomic arousal across multiple time scales of change in based temporal clustering analyses were used to test for
arousal, we conducted the following analysis: first, we group differences while correcting for multiple compar-
down-­sampled the arousal data using multiple epoch du- isons. Significant (p < .05) group differences were ob-
rations, from 1-­s epochs through to 30-­m in epochs (see served at the 1-­and 5-­m in epoch durations, indicating
Figure 4a). Next, we calculated the partial autocorrela- that the high self-­regulation group showed a slower rate
tion function (PACF). The lag 1 term of the PACF in- of change of arousal across these time scales.
dexes the autocorrelation in the data: a high PACF term
indicates greater autocorrelation (i.e., a slower rate of Part 1—Summary
change in the data) (see Figure 4b). The lag 2 term in- Overall, these results suggest that: (i) children with
dexes the autocorrelation at 2 epochs distance after the higher self-­regulation show a slower rate of change of
lag 1 autocorrelation has been controlled for, and so on. vocal intensity and affect in home settings; (ii) children

F I G U R E 4 (a, b) Raw data samples to illustrate the analysis. (a) Raw autonomic arousal data from a single participant, down-­sampled
using multiple different epoch durations from 1 s to 30 min. (b) Raw data samples to illustrate the partial autocorrelation function (PACF)
that was calculated to index variability in the arousal data; top plot shows an example of a participant with high PACF and bottom shows an
example with low PACF. (c) Line chart showing how the different lag terms of the PACF against the different epoch durations used for down-­
sampling. Dotted red lines show the thresholds for significant PACF terms (positive and negative). (d) Line plot showing how the PACF lag 1
term differs between the high and low self-­regulation groups, across different epoch durations. Error bars indicate standard error of the means.
*Indicates significant group difference after correcting for multiple comparisons.
 14678624, 2024, 6, Downloaded from https://srcd.onlinelibrary.wiley.com/doi/10.1111/cdev.14136 by National Health And Medical Research Council, Wiley Online Library on [09/06/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
SELF REG AND UP REG    | 2007

with higher self-­regulation show a slower rate of change We predicted that, in children with better lab-­assessed
of autonomic arousal, across the 1-­to 5-­m in time scale, self-­regulation, their autonomic arousal would be more
in home settings; and (iii) we found no evidence that chil- predictive of spontaneous vocalizations, indicating that
dren with higher lab-­assessed self-­regulation differ on they are using vocalizations to self-­regulate.
mean vocal affect, intensity, and autonomic arousal in
home settings.
2a—Adapted Poincaré plots

Part 2—Responses to hypo-­and hyper-­arousal First, we used adapted Poincaré plots to examine how
the stability of arousal (i.e., the likelihood of being in the
In order to examine how responses to hypo-­and hyper-­ same arousal bin at time t + 1 as at time t) varies as a func-
arousal differed between the low and high self-­regulation tion of arousal at time t. We examined the likelihood of a
groups we conducted three analyses. First (part 2a), we decrease in arousal occurring at time t + 1 when arousal
used adapted Poincaré plots to examine how the stabil- at time t was above the mean and compared it with the
ity of arousal (i.e., the likelihood of being in the same likelihood of a decrease in arousal occurring at time
arousal bin at time t + 1 as at time t) varies as a func- t + 1 when arousal at time t was below the mean. We also
tion of arousal at time t. We predicted that children examined how these likelihoods differed contingent on
with better lab-­assessed self-­regulation would be more children's lab-­assessed self-­regulation. We predicted that
likely to show decreases in arousal following hyper-­ children with better lab-­assessed self-­regulation would
arousal and increases in arousal following hypo-­arousal. be more likely to show increases in arousal following
Second (part 2b), we examined the same question in a hypo-­arousal as well as decreases in arousal following
different way, by using modeling to generate multiple hyper-­arousal.
simulated datasets with different parameter settings To test this, we first down-­sampled the continuous
to measure which parameter setting shows the best fit arousal data using variable epoch durations (1, 60, and
for the observed data. We predicted that children with 600 s) and binned it, participant by participant, into five
better lab-­ assessed self-­regulation would show more equally sized bins (Figure 5a). This binning was per-
“mean-­ c entering” following fluctuations above mean formed participant by participant in order to correct for
arousal and below mean arousal in the naturalistic home individual differences in mean arousal between partici-
data. Third (part 2c), we examined how arousal patterns pants. These data were then visualized using an adapted
change around vocalizations, and how these relation- Poincaré plot with arousal at time t on the x-­axis and
ships differed contingent on lab-­assessed self-­regulation. arousal at time t + 1 on the y-­axis. Data points on the 1:1

F I G U R E 5 (a) Raw data sample from an individual participant, which has been down-­sampled to 1-­m in epochs and binned into five
equally sized bins. (b) Magnified example showing one 9-­m in excerpt from the day-­long recording shown in (a). (c) Adapted Poincaré plot
showing arousal at time t (x-­a xis) against arousal at time t + 1 (y-­a xis). (d-­f ) Bar charts showing how the likelihood that arousal at time
t = arousal at time t + 1 varies as a function of arousal at time t. Error bars show standard error of the means.The three charts show the identical
analysis, but based respectively on 1 second epochs (d), 60 second epochs (e) and 600 second epochs (f). *Indicates significance of group
comparisons *p < .05, (*)p < .1.
 14678624, 2024, 6, Downloaded from https://srcd.onlinelibrary.wiley.com/doi/10.1111/cdev.14136 by National Health And Medical Research Council, Wiley Online Library on [09/06/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2008 |    WASS et al.

line (highlighted in red) are at the same arousal level at across the day, 𝜎 is the variance in arousal for that partic-
time t + 1 as at time t. ipant, and W is a random Wiener process.
Figure 5d shows how the likelihood that arousal Thus, simulated arousal levels at time t + 1 are gen-
at time t = arousal at time t + 1 varies as a function of erated by combining two terms: first, random noise
arousal at time t. All of these plots have a U-­shaped (generated based on the variance in arousal observed
pattern, indicating that extreme high and low levels of for that participant); and second, a “mean-­c entering”
arousal are “sticky,” as noted previously within this term generated by calculating the difference between
dataset (Wass et al., 2020). Blue and red bars show the the current arousal level and the average arousal lev-
low and high self-­ regulation groups. Group differ- els observed for that child that day. When the current
ences were calculated using the non-­parametric Mann– arousal levels are above the average for that child that
Whitney U test and corrected for multiple comparisons day, the “mean-­c entering” term will be negative, and
using the Benjamini–Hochberg procedure (Benjamini vice versa. The strength of the “mean-­c entering” term
& Hochberg, 1995). Results indicate that no significant is controlled by the variable 𝛽 . Two 𝛽 terms were used,
differences are observed at 1-­and 60-­ s epoch dura- and varied independently: βPOS for samples where the
tions, but that at longer epoch durations (600 s), hypo-­ current arousal levels are above the average for that
aroused states are less stable in the high self-­regulation child, and a βNEG for samples where the current arousal
group. Overall, these results suggest that, across longer levels are below the average for that child. A total of
time scales, low arousal levels were more likely to be fol- 1000 random simulated datasets were generated for
lowed by increases in arousal in children with better lab-­ each level of βNEG from 0 to 1 (in increments of .1) and
assessed self-­regulation, but no differences were found for βPOS in the same increments.
around high arousal levels. Then, separately for each participant, we calculated
the goodness of fit between the simulated data and the
observed data, using the following procedure. The av-
2b—Modeling erage duration of “hyper-­arousal” and “hypo-­arousal”
episodes was measured by classifying each arousal
Next, we used modeling to generate multiple simulated epoch into five equally sized bins. This was done par-
datasets with different parameter settings to test which ticipant by participant in order to control for individual
parameter setting shows the best fit for the observed differences in arousal between bins. The durations of
arousal data. Using dynamical modeling (cf., Cole, “hyper-­arousal” episodes were quantified by calculat-
Ramsook, et al., 2019; Morales et al., 2018), we gener- ing the time intervals between arousal first exceeding
ated simulated datasets with differing levels of “mean-­ the mean and returning to it, and the figure was divided
centering” (i.e., the tendency to return to the mean by the average duration of “average arousal” episodes
following fluctuations above or below it), using differ- to control for differing levels of autocorrelation in the
ent parameters to capture “mean-­ c entering” follow- data. The same analysis was then repeated for each of
ing fluctuations above mean arousal (i.e., how quickly the simulated datasets, and the average difference in
above-­average arousal tended to return back down to “hyper-­arousal episode duration” between the observed
mean levels) and below mean arousal (i.e., how quick and simulated data was calculated. The same procedure
below-­average arousal tended to return back up to mean was then repeated for “hypo-­arousal episode duration.”
levels). We then assessed how each of these parame- In Figure 6a,b,d, green indicates a good fit (i.e., that the
ters affected the goodness of fit between the simulated average duration of hyper-­or hypo-­arousal episodes was
data and our actual observed data, separately for each similar between the observed and simulated data); red
participant. And, we compared how the best-­ f itting indicates that the duration of hyper-­or hypo-­arousal ep-
parameter settings from our simulations related to in- isodes was longer in the real than the observed data; blue
dividual differences on the lab self-­regulation measure, indicates the opposite. As expected, visual inspection
in order to assess whether children who showed more showed that βPOS (the x-­axis) was more influential in de-
self-­regulatory behaviors on the lab task showed greater termining hyper-­arousal episode duration, as shown by
“mean-­ c entering”: following fluctuations above mean the fact that the colored lines on Figure 6a (left plot) are
arousal and below mean arousal in the naturalistic home primarily vertical. βNEG (the y-­axis) was more influential
data. in determining hypo-­arousal episode duration, as shown
This analysis was conducted on data down-­sampled by the fact that the lines on Figure 6a (right plot) are pri-
into 60-­s epochs. The simulated datasets were generated marily horizontal. For hyper-­arousal episode durations
using the following equation: (Figure 6a, left plot), the best-­fitting βPOS value was .5.
ΔΘ(t) = 𝛽(𝜇 − Θ(t)) + 𝜎W(t), For values less than .5, the model tended to underesti-
mate hyper-­arousal episode durations (blue values); for
where β = βPOS if 𝜇 − Θ(t) > 0 and β = βNEG if 𝜇 − Θ(t) < 0. values less than .5, it tended to overestimate them (red
Θ(t) is the arousal at time t, βPOS and βNEG are weight- values). For hypo-­arousal episode durations (Figure 6a,
ing terms, 𝜇 is the mean arousal level for that participant right plot), the best-­f itting βNEG value was .2. For values
 14678624, 2024, 6, Downloaded from https://srcd.onlinelibrary.wiley.com/doi/10.1111/cdev.14136 by National Health And Medical Research Council, Wiley Online Library on [09/06/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
SELF REG AND UP REG    | 2009

F I G U R E 6 (a) Goodness of model fit for the duration of hyper-­arousal and hypo-­arousal episodes. (b) Goodness of model fit for hyper-­
arousal episode duration (top) and hypo-­arousal episode duration (bottom), subdivided between the low self-­regulation group (left) and the
high self-­regulation group (right). (c) Bar chart showing the best-­f itting βNEG and βPOS values from (b), divided by self-­regulation group. Error
bars indicate standard errors, and *p < .05.

higher than this, it tended to overestimate hypo-­arousal low arousal (bin 1, Figure 7a) and more likely to vocalize
episode durations. at elevated arousal (bin 10, Figure 7a); but that overall,
Figure 6b shows the same plots split by self-­regulation the relationship between arousal and vocalization likeli-
group, and Figure 6c shows just the best-­f itting βNEG val- hood was stronger (significant from bins 6 to 10) in the
ues for the low and high self-­regulation groups. A Mann– high self-­regulation group. Cries were more likely at ele-
Whitney U test indicated that no significant differences vated arousal in both groups (Figure 7b). Speech-­like vo-
were observed for βPOS (“mean-­ c entering” following calizations were less likely at low arousal in both groups
hyper-­arousal) contingent on self-­regulation Z = −0.99, (Figure 7c), but more likely at intermediate-­ to-­
h igh
p = .32. However, the best-­fitting βNEG values (“mean-­ arousal in the high self-­regulation group only (bins 6–9).
centering” following hypo-­ arousal) were significantly Next, we examined how arousal levels change around
higher in the high self-­regulation group, Z = 2.6, p = 009. vocalizations, by excerpting each participant's arousal
Overall, these results suggest that “mean-­c entering” level during the 20 min before and after each vocaliza-
(i.e., the tendency to return to mean following fluctu- tion. No significant differences were identified when
ations above or below mean arousal) is observed more considering arousal levels around cries (Figure 7d). For
strongly in response to hypo-­arousal in the group with speech-­like vocalizations, however, a sustained increase
better lab-­assessed self-­regulation. No differences were in arousal was identified during the time period after
observed around hyper-­arousal. a speech-­like vocalization, in the high self-­regulation
group only (Figure 7e).
Finally, we examined the likelihood that an infant's
2c—Vocalizations speech-­like vocalization would be preceded by a care-
giver response, or elicit one in response (Figure 7f).
Finally, we examined how predictive a child's autonomic To measure this, we examined how the observed rate
arousal was of how likely they were to produce a cry or a of caregiver vocalizations changed during the 20 min
speech-­like vocalization at a given moment in time. We before and after an infant speech-­like vocalization. In
also examined how these relationships differed contin- the low self-­regulation group, adult vocalization rates
gent on lab-­assessed self-­regulation. We predicted that, were significantly above chance during the period from
in children with better lab-­assessed self-­regulation, their 2 mins before the infant vocalization to 6 mins after;
autonomic arousal would be more predictive of sponta- in the high self-­ regulation group, they were above
neous vocalizations, indicating that they are using vo- chance from 0 to 6 mins after (Figure 7f). Thus, infant
calizations to self-­regulate. speech-­like vocalizations were more likely to be pre-
Vocalization likelihoods were compared with chance ceded by adult child-­d irected vocalizations in the low-­
by comparing arousal levels at the time of observed self-­regulation group, but not the high self-­regulation
vocalizations with arousal levels during randomly se- group. In both groups, infant speech-­ l ike vocaliza-
lected moments within each participant's data where tions were likely to be followed by adult child-­d irected
no vocalization was taking place. Observed results were vocalizations.
compared with chance using t-­tests, and corrected for Overall, these results suggest that the relationship be-
multiple comparisons using a permutation-­based tempo- tween vocalization likelihood and arousal is stronger in
ral clustering (see Section S1.6). Results suggested that the high self-­regulation group, and that this difference is
both groups were less likely than chance to vocalize at primarily caused because speech-­like vocalizations are
 14678624, 2024, 6, Downloaded from https://srcd.onlinelibrary.wiley.com/doi/10.1111/cdev.14136 by National Health And Medical Research Council, Wiley Online Library on [09/06/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2010 |    WASS et al.

F I G U R E 7 (a) Relationship between vocalization likelihood and arousal, relative to a control likelihood (calculated as described in the
“Results” section). Blue and red lines above the plot show the areas of significant difference from zero in each group, after correcting for
multiple comparisons. (b) Relationship between the likelihood of cries and arousal, calculated in the same way as for (a). (c) Relationship
between likelihood of speech-­l ike vocalizations and arousal, calculated in the same way as for (a). (d) Average change in infant arousal around
cries. (e) Average change in infant arousal around speech-­l ike vocalizations. Area highlighted with black line indicates the area of significant
group difference after correcting for multiple comparisons using permutation-­based temporal clustering. (f) Change in the likelihood of
an adult child-­d irected vocalization around infant speech-­l ike vocalizations. Blue and red lines above the plot show the areas of significant
difference from the baseline vocalization rates.

more concentrated around intermediate-­to-­h igh arousal self-­regulation would associate with: reduced overall
in the high self-­regulation group. Speech-­like vocaliza- arousal in naturalistic settings; reduced vocal affect and
tions are more likely to be self-­generated (i.e., not pre- intensity in real-­world settings; a slower rate of change
ceded by an adult child-­ directed vocalization) in the of vocal affect, intensity, and arousal in naturalistic
high self-­regulation group. settings; and faster recovery following “peak” arousal
events. Overall, we found evidence that better lab-­
assessed self-­regulation associated with differences in
DI SC US SION variability and the rate of change of affect and arousal
in naturalistic settings (compare Somers, Luecken,
We used specially designed home-­wearable sensors to et al., 2021; Somers & Luecken, 2022). Specifically,
examine differences in how children spontaneously children with better lab-­ assessed self-­regulation
manifest arousal changes in home settings contingent on showed more variability in affect in their spontaneous
lab-­assessed self-­regulation. In part 1 of our analyses, vocal behaviors and a slower rate of change of auto-
we examined our data overall. We took a standard, lab-­ nomic arousal across multiple time scales (Figure 4d).
based measure, the still-­face protocol, and video-­coded However, we found no evidence for associations be-
known down-­regulatory behaviors such as gaze aversion tween self-­regulation in the lab and average arousal
and physical self-­touch (Figure 2b,c). We split our group levels in real-­world settings. We also found no evidence
using a median split into one group who showed more that children with better self-­regulation show overall
self-­regulatory behaviors and another group who showed differences in vocal affect (e.g., reduced negative affect),
fewer, and we validated this approach by showing that or reduced vocal intensity, in home settings.
children who used fewer self-­regulatory behaviors also In part 2 of our analyses, we examine how infants'
showed more negative facial affect (Figure 2d). In ad- responses and behaviors differ contingent on hypo-­
dition, we used cross-­correlations to show that, during versus hyper-­arousal. We predicted that children with
times when the child was using more self-­regulatory be- better lab-­ assessed self-­regulation would show both
haviors, they showed decreased negative/increased posi- more down-­ regulation from hyper-­ arousal and more
tive affect (Figure 2e). up-­regulation from hypo-­ arousal. We also examined
We then examined how children with better self-­ how the relationship between vocalization likelihood
regulation in lab settings showed altered behaviors in and arousal differed contingent on lab-­ assessed self-­
home settings. We predicted that increased lab-­assessed regulation. We predicted that, in children with better
 14678624, 2024, 6, Downloaded from https://srcd.onlinelibrary.wiley.com/doi/10.1111/cdev.14136 by National Health And Medical Research Council, Wiley Online Library on [09/06/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
SELF REG AND UP REG    | 2011

lab-­ assessed self-­

regulation, their autonomic arousal in the high self-­regulation group (Figure 7f). When we
would be more predictive of vocal behaviors. examined how infant arousal changed during the time
In Analysis 2a we examined how the stability of windows before and after vocalizations, we found that
arousal (i.e., the likelihood of being in the same state speech-­like vocalizations were associated with longer-­
at time t + 1 as at time t) varied contingent on a child's lasting increases in arousal in the high self-­regulation
arousal level at time t, and their lab-­ assessed self-­ group (Figure 7e).
regulation (Figure 5). Strikingly, we found no evidence Importantly, this is not an allostatic “corrective” pro-
across any time scale that episodes of high arousal were cess, insofar as vocalizations are triggered at medium-­
more likely to be followed by decreases in arousal in the to-­h igh arousal and are followed by increases in arousal.
children who showed high lab-­assessed self-­regulation Rather, it is evidence that, even from early infancy, chil-
(Figure 5d–f). Across longer time scales, we did, how- dren use active processes to maintain arousal at an el-
ever, find that low arousal levels were more likely to be evated state. For example, research has suggested that
followed by increases in arousal in children with better information that is actively elicited by children (by vo-
lab-­assessed self-­regulation (Figure 5d–f). In Analysis calizing, or pointing) tends to be better retained (Begus
2b, we tested the same idea using dynamical generative & Bonawitz, 2020; Begus & Southgate, 2018). One pos-
models and found a similar result. We produced simu- sibility is that, if medium-­to-­h igh arousal is optimal for
lated datasets with differing levels of “mean-­c entering” learning (Aston-­Jones & Cohen, 2005; Wass, 2018), then
(i.e., the tendency to return to the mean following fluc- children with better lab-­assessed self-­regulation may be
tuations above or below it). We evaluated how “mean-­ more likely to produce speech-­like vocalizations to elicit
centering” differed between fluctuations above mean caregiver interactions at optimal times. Future work
arousal and below it, and how this differed contingent should also investigate further the differences between
on lab-­assessed self-­regulation. We found no differences prolonged expressions of high arousal associated with
contingent on self-­regulation when we examined mean-­ positive affect and prolonged high arousal associated
centering following hyper-­arousal (Figure 6b,c); but we with negative affect (Lunkenheimer et al., 2011; Somers,
found increased mean-­c entering following hypo-­arousal Luecken, et al., 2021; Somers & Luecken, 2022).
in children with better lab-­ assessed self-­
regulation Studying early development is essential for under-
(Figure 6b,c). Thus, children with different levels of standing early causative pathways in developmental
behaviorally-­assessed self-­regulation in the lab showed psychopathology, given the associations noted between
clearer differences in up-­rather than down-­regulation in early self-­regulatory behaviors and long-­term outcomes
naturalistic settings. (Eisenberg et al., 2019; Kostyrka-­A llchorne et al., 2020;
Any approach to studying homeostasis and allostasis Nigg, 2017). But one intrinsic limitation in studying nat-
in naturalistic settings faces the challenge of distinguish- uralistic data in this age range is the difficulty of dis-
ing active allostatic processes from the mere dissipation tinguishing co-­from self-­ regulatory processes, given
of arousal states (Brooks et al., 2021; Davis, 1958). Here, the close interdependency between caregivers and in-
it is relevant to note that our cohort was defined from fants during this period (Feldman, 2007; Kopp, 1982;
a lab self-­regulation measure that explicitly examines Lunkenheimer et al., 2020; Somers, Curci, et al., 2021;
active down-­regulatory behaviors such as gaze aversion Somers, Luecken, et al., 2021; Wass et al., 2019, 2024).
and physical self-­soothing. Perhaps most informative, For example, in the still-­face procedure that we used
however, were our analyses that examined how the like- to characterize our sample, evidence suggests that the
lihood of infants vocalizing varied as a function of their caregiver's behavior toward their infant influences the
arousal, and how these associations differed contingent infant's behaviors toward the parent during the still-­
on lab-­assessed self-­regulation. Vocalizations are known face (Field, 1994; Mesman et al., 2009). This questions
to be a mechanism for eliciting caregiver engagement and whether the still-­face procedure is a test of an infant's
arousal co-­regulation (Kopp, 1982; Wass et al., 2022). self-­regulatory capacity per se, or whether it is better
We found that all infants were more likely to produce thought of as measuring infant–caregiver communica-
cries at elevated arousal (Figure 7b), and that changes tion. Against this, and in favor of treating the infants'
in autonomic arousal around cries did not significantly self-­regulatory behaviors during the still-­face as a mea-
differ between groups (Figure 7d). However, we found sure of the infant's self-­regulatory capacity, are our anal-
that, whereas the low lab-­assessed self-­regulation group yses in Figure 2d which show how the change in facial
generally showed no association between arousal and affect during the still-­face differs contingent on the de-
the probability of producing a speech-­ like vocaliza- gree of self-­regulatory behaviors shown by the child, and
tion, the high lab-­ assessed self-­
regulation group did Figure 2e, which shows a time-­specific association be-
show an association, such that they were more likely to tween increased self-­regulatory behaviors and less nega-
produce speech-­ like vocalizations at medium-­ to-­
h igh tive/more positive affect.
arousal (bins 6–9 of 10) (Figure 7c). Speech-­like vocal- The same point, that self-­and co-­regulatory processes
izations were also more likely to be self-­generated (i.e., are hard to disentangle in this age range, also applies to
not preceded by an adult child-­directed vocalization) our analyses of the home data, where it is impossible
 14678624, 2024, 6, Downloaded from https://srcd.onlinelibrary.wiley.com/doi/10.1111/cdev.14136 by National Health And Medical Research Council, Wiley Online Library on [09/06/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2012 |    WASS et al.

to be sure whether group differences in how infants' therapeutic work should consider the former as well as
arousal changes over time (that we documented, e.g., in the latter.
Figures 4d, 5 and 6) arise from differences in the infants'
behavior per se, or in how the caregiver interacts with AC K NOW L E D G M E N T S
the child, moment by moment. Of note, however, when This project has received funding from the European
we examined behaviors such as infant cries, which are Research Council (ERC) under the European Union's
known to elicit co-­regulatory behaviors in caregivers Horizon 2020 research and innovation program (grant
(Wass et al., 2019, 2022), we did not observe significant agreement No. [853251—ONACSA]). It has also re-
differences between the low and high self-­ regulation ceived funding from grant number ES/N017560/1 from
groups (Figure 7b,d). And when we examined how likely the Economic and Social Research Councils, UK.
infant speech-­like vocalizations were either to follow, Thank you to all participating infants and parents, to
or to be followed by, an adult infant-­directed vocaliza- Kaili Clackson for additional data collection, and to
tion we did not find a significant difference contingent members of the UEL BabyDev Lab for countless useful
on child self-­regulation (Figure 7f). This suggests that discussions.
the individual differences we observed between infants
were not measurably contingent on observable caregiver DATA AVA I L A B I L I T Y STAT E M E N T
behaviors. Given the sensitive nature of the data included in the
One further intrinsic limitation of our approach manuscript (personally identifiable home audio record-
is our use of a lab-­based measure of self-­regulation to ings of parent–child interactions), the data on which
identify individual differences in self-­regulation—which these analyses are based are only available on personal
assumes, for example, that individual differences in request to the corresponding author. Sharing these data
self-­regulation are entirely trait—rather than state-­level will require additional approval from our ethics board
features (see Baumeister et al., 2019). Future work using before sharing. The analyses here were not preregistered.
more advanced dynamical modeling techniques, includ- All analytical code is available freely on request to the
ing phase space analyses (Dezhina et al., 2023; Lazarus corresponding author.
et al., 2023), could directly identify individual differences
in the stability of naturally occurring arousal states; and, ORC I D
through that, classify self-­regulation based on natural- S. V. Wass https://orcid.org/0000-0002-7421-3493
istic data, without reference to a separate experimental
measure. R EF ER ENCE S
Overall, our results suggest that research into how Aston-­Jones, G., & Cohen, J. D. (2005). An integrative theory of locus
children coordinate internal and behavioral responses in coeruleus-­ norepinephrine function: Adaptive gain and opti-
mal performance. Annual Review of Neuroscience, 28, 403–450.
response to changing environmental demands needs to
https://​doi.​org/​10.​1146/​annur​ev.​neuro.​28.​061604.​135709
address two distinct areas of individual difference: first, Baumeister, R. F., Wright, B. R. E., & Carreon, D. (2019). Self-­c ontrol
what an individual's optimal state is (which may differ “in the wild”: Experience sampling study of trait and state self-­
from individual to individual, although we did not ex- regulation. Self and Identity, 18(5), 494–528.
amine this here); and second, how effective an individual Begus, K., & Bonawitz, E. (2020). The rhythm of learning: Theta os-
cillations as an index of active learning in infancy. Developmental
is at maintaining that optimal state through allostasis
Cognitive Neuroscience, 45, 100810.
(Cole et al., 2020; Kopp, 1982), and in self-­generating Begus, K., & Southgate, V. (2018). Curious learners: How infants' mo-
learning opportunities that are concentrated around tivation to learn shapes and is shaped by infants' interactions
times when autonomic arousal is optimal for learning (see with the social world. In M. Saylor & P. Ganea (Eds.), Active
Porges, 2007). Questionnaire measures of self-­regulation learning from infancy to childhood (pp. 13–37). Springer.
Bellato, A., Arora, I., Hollis, C., & Groom, M. J. (2020). Is autonomic
may measure individual differences in the former (i.e.,
nervous system function atypical in attention deficit hyperac-
differences between children's optimal states), or the lat- tivity disorder (ADHD)? A systematic review of the evidence.
ter (i.e., differences between how well children are able to Neuroscience & Biobehavioral Reviews, 108, 182–206.
maintain their optimal state)—but do not differentiate Benjamini, Y., & Hochberg, Y. (1995). Controlling the false discov-
between the two (Cole, Ram, et al., 2019; Cole, Ramsook, ery rate: A practical and powerful approach to multiple testing.
Journal of the Royal Statistical Society: Series B: Methodological,
et al., 2019).
57, 289–300.
These findings also have potential therapeutic impli- Berlyne, D. E. (1960). Conflict, arousal, and curiosity. Martino Fine
cations. They suggest that children who perform less well Books.
on laboratory measures of self-­regulation may show ei- Brazelton, T. B. (1983). Precursors for the development of emotions in
ther reduced interoception or a reduced ability to gener- early infancy. In R. Plutchik & H. Kellerman (Eds.), Emotions in
early development (pp. 35–55). Elsevier.
ate behaviors that modulate the internal state contingent
Brooks, J., Crone, J. C., & Spangler, D. P. (2021). A physiological
on changing environmental demands (Geva et al., 2013; and dynamical systems model of stress. International Journal of
Porges, 2007); which, in real-­world settings, affects how Psychophysiology, 166, 83–91.
children up-­regulate following hypo-­arousal more than Calderon, D. P., Kilinc, M., Maritan, A., Banavar, J. R., & Pfaff, D.
how they down-­regulate following hyper-­arousal. Future (2016). Generalized CNS arousal: An elementary force within
 14678624, 2024, 6, Downloaded from https://srcd.onlinelibrary.wiley.com/doi/10.1111/cdev.14136 by National Health And Medical Research Council, Wiley Online Library on [09/06/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
SELF REG AND UP REG    | 2013

the vertebrate nervous system. Neuroscience & Biobehavioral series: Towards a science of moments. https://​osf.​io/​prepr​i nts/​
Reviews, 68, 167–176. psyar ​x iv/​4nxqh​
Cannon, W. B. (1929). Organization for physiological homeostasis. Lunkenheimer, E., Hamby, C. M., Lobo, F. M., Cole, P. M., & Olson,
Physiological Reviews, 9(3), 399–431. S. L. (2020). The role of dynamic, dyadic parent–child processes
Cole, P. M., Lougheed, J. P., Chow, S.-­ M., & Ram, N. (2020). in parental socialization of emotion. Developmental Psychology,
Development of emotion regulation dynamics across early child- 56(3), 566.
hood: A multiple time-­s cale approach. Affective Science, 1, 1–14. Lunkenheimer, E. S., Olson, S. L., Hollenstein, T., Sameroff, A. J., &
Cole, P. M., Ram, N., & English, M. S. (2019). Toward a unifying Winter, C. (2011). Dyadic flexibility and positive affect in par-
model of self-­ regulation: A developmental approach. Child ent–child coregulation and the development of child behavior
Development Perspectives, 13(2), 91–96. problems. Development and Psychopathology, 23(2), 577–591.
Cole, P. M., Ramsook, K. A., & Ram, N. (2019). Emotion dysregu- Maris, E. (2012). Statistical testing in electrophysiological studies.
lation as a dynamic process. Development and Psychopathology, Psychophysiology, 49(4), 549–565.
31(3), 1191–1201. Maris, E., & Oostenveld, R. (2007). Nonparametric statistical test-
Colombo, J., & Cheatham, C. L. (2006). The emergence and basis of ing of EEG-­and MEG-­d ata. Journal of Neuroscience Methods,
endogenous attention in infancy and early childhood. Advances 164(1), 177–190.
in Child Development and Behavior, 34, 283–322. McEwen, B. S., & Wingfield, J. C. (2003). The concept of allostasis
Davis, R. C. (1958). The domain of homeostasis. Psychological Review, in biology and biomedicine. Hormones and Behavior, 43(1), 2–15.
65(1), 8–13. https://​doi.​org/​10.​1037/​h0045358 McHugh, M. L. (2012). Interrater reliability: The kappa statistic.
Derryberry, D., & Rothbart, M. K. (1988). Arousal, affect and atten- Biochemia Medica: Biochemia Medica, 22(3), 276–282.
tion as components of temperament. Journal of Personality and Mesman, J., van IJzendoorn, M. H., & Bakermans-­K ranenburg, M. J.
Social Psychology, 55(6), 958–966. https://​doi.​org/​10.​1037/​0 022-­​ (2009). The many faces of the still-­face paradigm: A review and
3514.​55.6.​958 meta-­analysis. Developmental Review, 29(2), 120–162.
Dezhina, Z., Smallwood, J., Ting, X., Turkheimer, F. E., Moran, R. J., Morales, S., Ram, N., Buss, K. A., Cole, P. M., Helm, J. L., & Chow, S.
Friston, K. J., Leech, R., & Fagerholm, E. D. (2023). Establishing (2018). Age-­related changes in the dynamics of fear-­related reg-
brain states in neuroimaging data. PLoS Computational Biology, ulation in early childhood. Developmental Science, 21(5), e12633.
19(10), e1011571. Nigg, J. T. (2017). Annual research review: On the relations among
Dunn, W. (2002). Infant/toddler sensory profile: User's manual. self-­regulation, self-­c ontrol, executive functioning, effortful con-
Pearson. trol, cognitive control, impulsivity, risk-­taking, and inhibition
Eisenberg, I. W., Bissett, P. G., Zeynep Enkavi, A., Li, J., MacKinnon, for developmental psychopathology. Journal of Child Psychology
D. P., Marsch, L. A., & Poldrack, R. A. (2019). Uncovering the and Psychiatry, 58(4), 361–383.
structure of self-­regulation through data-­d riven ontology dis- Oakes, L. M., Baumgartner, H. A., Barrett, F. S., Messenger, I. M., &
covery. Nature Communications, 10(1), 1–13. Luck, S. J. (2013). Developmental changes in visual short-­term
Feldman, R. (2007). Parent–infant synchrony and the construction memory in infancy: Evidence from eye-­tracking. Frontiers in
of shared timing; physiological precursors, developmental out- Psychology, 4, 697. https://​doi.​org/​10.​3389/​fpsyg.​2013.​0 0697​
comes, and risk conditions. Journal of Child Psychology and Oller, D. K., Buder, E. H., Ramsdell, H. L., Warlaumont, A. S.,
Psychiatry, 48(3–4), 329–354. Chorna, L., & Bakeman, R. (2013). Functional flexibility of in-
Field, T. (1994). The effects of mother's physical and emotional un- fant vocalization and the emergence of language. Proceedings of
availability on emotion regulation. Monographs of the Society for the National Academy of Sciences of the United States of America,
Research in Child Development, 59, 208–227. 110(16), 6318–6323.
Fiske, D. W., & Maddi, S. R. (1961). Functions of varied experience. Pfaff, D. (2018). How brain arousal mechanisms work: Paths toward
Literary Licensing, LLC. consciousness (Vol. 1). Cambridge University Press.
Freud, S., & Strachey, J. (1900). The interpretation of dreams (Vol. 4). Piccardi, E. S., & Gliga, T. (2022). Understanding sensory regulation
Allen & Unwin. in typical and atypical development: The case of sensory seek-
Gagne, J. R., van Hulle, C. A., Aksan, N., Essex, M. J., & Goldsmith, ing. Developmental Review, 65, 101037.
H. H. (2011). Deriving childhood temperament measures from Porges, S. W. (2007). The polyvagal perspective. Biological Psychology,
emotion-­eliciting behavioral episodes: Scale construction and 74(2), 116–143. https://​doi.​org/​10.​1016/j.​biops​ycho.​2006.​06.​0 09
initial validation. Psychological Assessment, 23(2), 337. Porges, S. W., & Furman, S. A. (2011). The early development of the
Gardner, J. M., & Karmel, B. Z. (1984). Arousal effects on visual pref- autonomic nervous system provides a neural platform for social
erences in neonates. Developmental Psychology, 20(3), 374. behavior: A polyvagal perspective. Infant and Child Development,
Gardner, J. M., & Karmel, B. Z. (1995). Development of arousal-­ 20(1), 106–118. https://​doi.​org/​10.​1002/​icd.​688
modulated visual preferences in early infancy. Developmental Posner, M. I., & Rothbart, M. K. (2000). Developing mechanisms
Psychology, 31(3), 473–482. of self-­regulation. Development and Psychopathology, 12(3),
Geva, R., Sopher, K., Kurtzman, L., Galili, G., Feldman, R., & Kuint, J. 427–441.
(2013). Neonatal brainstem dysfunction risks infant social engage- Schneirla, T. C. (1959). An evolutionary and developmental theory of bi-
ment. Social Cognitive and Affective Neuroscience, 8(2), 158–164. phasic processes underlying approach and withdrawal. MIT Press.
Kopp, C. B. (1982). Antecedents of self-­regulation: A developmental Somers, J. A., Curci, S. G., Winstone, L. K., & Luecken, L. J. (2021).
perspective. Developmental Psychology, 18(2), 199–214. Within-­mother variability in vagal functioning and concurrent
Kostyrka-­A llchorne, K., Wass, S. V., & Sonuga-­Barke, E. J. S. (2020). socioemotional dysregulation. Psychophysiology, 58(9), e13855.
Research review: Do parent ratings of infant negative emotion- Somers, J. A., & Luecken, L. J. (2022). Prenatal programming of be-
ality and self-­regulation predict psychopathology in childhood havior problems via second-­by-­s econd infant emotion dynamics.
and adolescence? A systematic review and meta-­analysis of pro- Psychological Science, 33, 2027–2039.
spective longitudinal studies. Journal of Child Psychology and Somers, J. A., Luecken, L. J., McNeish, D., Lemery-­Chalfant, K., &
Psychiatry, 61(4), 401–416. Spinrad, T. L. (2021). Second-­by-­s econd infant and mother emo-
Kroenke, K., Spitzer, R. L., & Williams, J. B. W. (2001). The PHQ-­ tion regulation and coregulation processes. Development and
9: Validity of a brief depression severity measure. Journal of Psychopathology, 34, 1–14.
General Internal Medicine, 16(9), 606–613. Spitzer, R. L., Kroenke, K., Williams, J. B. W., & Löwe, B. (2006).
Lazarus, G., Song, J., Jeronimus, B. F., & Fisher, A. J. (2023). A brief measure for assessing generalized anxiety disorder: The
Delineating discrete generalizable states from intraindividual time GAD-­7. Archives of Internal Medicine, 166(10), 1092–1097.
 14678624, 2024, 6, Downloaded from https://srcd.onlinelibrary.wiley.com/doi/10.1111/cdev.14136 by National Health And Medical Research Council, Wiley Online Library on [09/06/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2014 |    WASS et al.

Stifter, C. A., & Braungart, J. M. (1995). The regulation of negative Wass, S. V., Smith, C., Clackson, K., & Mirza, F. U. (2020). In in-
reactivity in infancy: Function and development. Developmental fancy, it's the extremes of arousal that are ‘sticky’: Naturalistic
Psychology, 31(3), 448–455. data challenge purely homeostatic approaches to studying self-­
Verona, E., Patrick, C. J., Curtin, J. J., Bradley, M. M., & Lang, P. J. regulation. Developmental Science, 24(3), e13059.
(2004). Psychopathy and physiological response to emotionally Wass, S. V., Smith, C. G., Clackson, K., Gibb, C., Eitzenberger, J., &
evocative sounds. Journal of Abnormal Psychology, 113(1), 99. Mirza, F. U. (2019). Parents mimic and influence their infant's
Vest, A. N., Da Poian, G., Li, Q., Liu, C., Nemati, S., Shah, A. J., autonomic state through dynamic affective state matching.
& Clifford, G. D. (2018). An open source benchmarked toolbox Current Biology, 29(14), 2415–2422. e4.
for cardiovascular waveform and interval analysis. Physiological Weinberg, M. K., & Tronick, E. Z. (1996). Infant affective reactions to
Measurement, 39(10), 105004. the resumption of maternal interaction after the still-­face. Child
Wachs, T. D. (1977). The optimal stimulation hypothesis and early de- Development, 67(3), 905–914.
velopment. In Ina Č. Užgiris, F. Weizmann (Eds.), The structur- Yerkes, R. M., & Dodson, J. D. (1908). The relation of strength of
ing of experience (pp. 153–177). Springer. stimulus to rapidity of habit formation. Journal of Comparative
Wass, S. V., De Barbaro, K., & Clackson, K. (2015). Tonic and phasic Neurology and Psychology, 18, 459–482.
co-­variation of peripheral arousal indices in infants. Biological Zentall, S. S., & Zentall, T. R. (1983). Optimal stimulation: A model
Psychology, 111, 26–39. of disordered activity and performance in normal and deviant
Wass, S. V., Clackson, K., & De Barbaro, K. (2016). Temporal dy- children. Psychological Bulletin, 94(3), 446.
namics of arousal and attention in 12-­ month-­ old infants. Zuckerman, M. (1979). Sensation seeking. Wiley Online Library.
Developmental Psychobiology, 58(5), 623–639.
Wass, S. V. (2018). How orchids concentrate? The relationship between
physiological stress reactivity and cognitive performance during S U PP ORT I NG I N F OR M AT ION
infancy and early childhood. Neuroscience & Biobehavioral Additional supporting information can be found online
Reviews, 90, 34–49. in the Supporting Information section at the end of this
Wass, S. V. (2021). The origins of effortful control: How early devel-
article.
opment within arousal/regulatory systems influences attentional
and affective control. Developmental Review, 61, 100978.
Wass, S. V., Greenwood, E. M. G., Esposito, G., Smith, C. G., Necef,
I., & Phillips, E. (2024). Annual research review: “There, the
dance is, at the still point of the turning world.” Dynamic sys- How to cite this article: Wass, S. V., Mirza, F. U., &
tems perspectives on co-­regulation and dysregulation during Smith, C. (2024). Understanding allostasis:
early development. Journal of Child Psychology and Psychiatry, Early-­life self-­regulation involves both up-­and
65(4), 481–507.
down-­regulation of arousal. Child Development,
Wass, S. V., Phillips, E., Smith, C., & Goupil, L. (2022). Vocalisations
and the dynamics of interpersonal arousal coupling in caregiver-­ 95, 2000–2014. https://doi.org/10.1111/cdev.14136
infant dyads. eLife, 11, e77399.