Sustained Attention Across Toddlerhood: The Roles of Language and Sleep

Abstract

The present study examined individual differences in the development of sustained attention across toddlerhood, as well as how these individual differences related to the development of language and sleep. Toddlers (N = 314; 54% male) were assessed at 30, 36, and 42 months using multiple measures of attention, a standardized language assessment, and actigraphic measures of sleep. Toddlers were 80% White. Family socioeconomic status was calculated using the Hollingshead Four Factor Index and ranged from 13 to 66 (M = 47.59, SD = 14.13). Aims were 1) to examine associations between measures of attention across situations, informants, and time, 2) to consider the independent and interactive effects of language and sleep on attention, and 3) to test potential bidirectional associations between sleep and attention. Findings showed attention measures were stable across time but were only weakly linked with each other at 42 months. Attention was consistently linked with language. More variable sleep and longer naps were associated with less growth in sustained attention across time. Nighttime sleep duration interacted with language in that sleep duration was positively associated with attention scores among toddlers with less advanced language, even when socioeconomic status was controlled. The findings describe an understudied aspect of how sustained attention develops, involving the main effect of consistent sleep schedules and the interaction effect of amount of sleep and child language development. These findings are relevant to understanding early childhood risk for developing attention problems and to exploring a potential prevention target in family sleep practices.

The concept of sustained attention, defined here as the effortful maintenance of focused attention, is central in accounts of children’s social and academic competencies (Ruff & Rothbart, 2001). Studies are needed to examine how sustained attention changes across toddlerhood, as well as to identify factors that contribute to individual differences in toddlers’ levels and rates of growth in sustained attention. Such research would help efforts aimed at preventing the costly problems associated with poor sustained attention. This study longitudinally followed a large sample of toddlers to examine associations between measures of attention across situations (different lab tasks), informants (parents and secondary caregivers), and time (30, 36, and 42 months). We also considered the independent and interactive effects of language and sleep on attention and tested bidirectional links between sleep and attention.

Measurement of Individual Differences in Sustained Attention

Sustained attention has been measured in multiple ways, but two of the most widely used measures in early childhood involve free play with toys (Ruff & Capozzoli, 2003) and structured play activities, such as sorting objects (Gaertner, Spinrad, & Eisenberg, 2008). In previous research, correlations between sustained attention measures across situations were low to modest (absolute r values ranged from .04 to .36), suggesting that children’s attention may vary with task demands, as well as with children’s abilities and motivations to meet those demands (Ruff, Capozzoli, & Weissberg, 1998). Although laboratory assessments of sustained attention are widely used and validated, they may not adequately capture children’s functional behavior in ordinary life situations (Barkley, 1991). As a way of capturing naturally occurring behavior, and thus serving as a potential complement to laboratory measures of sustained attention, assessment batteries often include parent- and teacher-report measures of attention-relevant, functional behavior in everyday life. To advance measurement of sustained attention in toddlerhood, this study used both lab tasks and caregiver measures.

Child Factors Involved in the Development of Sustained Attention

Previous research has demonstrated the important role of mature, responsive social partners (often parents) in fostering children’s sustained attention (Wass et al., 2018). However, regardless of parental involvement, toddlers’ sustained attention tends to create learning moments that are visually rich, as objects that are held during bouts of sustained attention appear large, centered, and dominant in head-camera views whether a toddler is playing alone, with a directive parent, or with a responsive parent (McQuillan, Smith, Yu, & Bates, 2019). Therefore, although parental responsiveness has been associated with better sustained attention, children themselves play a critical role. The present study seeks to extend this work by considering child factors, namely language ability and sleep, that could explain individual differences in the development of sustained attention. Language ability and sleep were selected because these child factors develop across toddlerhood and have been linked with attention during early childhood.

Language.

Language develops rapidly across toddlerhood (Bloom, 2000), and its connection with attention has been well-established. At a neural level, the language and attention systems of the brain have been shown to be connected via the visual word form area (VWFA; Chen et al., 2019). In a cross-lagged panel model with a large sample of toddlers, maternal reports of language and attention problems were bi-directionally, inversely associated across time (Ribeiro et al., 2011). Another longitudinal study using crossed-lagged panel models spanning early childhood to adolescence found that language deficits more strongly predicted later attention problems, than the converse (Petersen et al., 2013).

Language development may facilitate self-regulation development, which may help a toddler voluntarily direct and maintain attention (Rothbart & Bates, 2006). Vygotsky (1962) posited that children become able to use private, self-directed speech to guide goal-directed behavior and facilitate problem solving. In toddlerhood, when language skills are actively developing, private speech is thought to emerge from the regulatory role of language during social interactions (Vygotsky, 1962). The self-regulatory function of language for attention has been demonstrated in a longitudinal study of toddlers by our research group: Self-regulation, indexed by an inhibitory control task, fully mediated the link between language ability and parent reported attention problems across ages 30, 36, and 42 months (Petersen, Bates, & Staples, 2015). This study showed language development to be important in self-regulation development and attentional control, but left much of the variance in these constructs unexplained. The present study builds on this previous work, using additional data from the same cohort, in two critical ways. First, sustained attention measures were tested along with language and caregiver ratings of attention problems to examine connections between these constructs. Second, toddlers’ sleep was considered as an additional child factor that might be involved in self-regulation development and attentional control.

Sleep.

Studies have examined associations between sleep problems and attention problems (Yoon, Jain, & Shapiro, 2012), but this literature is, so far, inconclusive about the nature of the association. Previous research has shown associations between sleep problems and attention problems longitudinally from infancy to early childhood (Sadeh et al., 2015) and from early childhood to adolescence (O’Callaghan et al., 2010). However, previous studies have not examined toddlerhood in particular and have not used repeated measures of each variable, so they could not ask whether early attention problems predict later sleep problems.

Studies examining associations between sleep deficits and laboratory measures of sustained attention have been similarly inconclusive. Some research has shown that sleep deficits have no significant effect on sustained attention (Astill, Van der Heijden, Van IJzendoorn, & Van Someren, 2012), while other studies have shown that sleep deficits result in worse sustained attention (Lim & Dinges, 2010). For toddlers, one cross-sectional study from our research group has shown that irregular sleep was associated with less efficient attentional processing as reflected by longer latencies of P3 EEG components in response to a target stimulus, and, in turn, longer target P3 latencies were associated with poorer sustained attention observed during free play (Hoyniak, Petersen, McQuillan, Staples, & Bates, 2015). However, this study used only one attention measure and had a small sample of toddlers that were not followed longitudinally. Longitudinal studies with repeated measures of toddlers’ sleep and attention will clarify the direction of effects – whether toddlers with poorer sleep show poorer attention and/or whether toddlers with poorer attention experience poorer sleep. If the former is true, sleep may be an especially important target for future prevention efforts.

Toddlerhood provides a crucial period in which to examine links between sleep and attention because toddlers experience substantial development in sustained attention (Ruff & Rothbart, 2001) as well as decreases in daytime sleep (Staples, Bates, & Petersen, 2015), night wakings (Byars, Yolton, Rausch, Lanphear, & Beebe, 2012), and the total amount of sleep (Galland et al., 2012). No studies to date have used repeated measures to examine change in actigraphic measures of sleep and observed measures of attention across toddlerhood. Previous related studies have relied upon parent reports of children’s sleep or on experimental sleep restrictions (for a review, see Lundahl, Kidwell, Van Dyk, & Nelson, 2015), but actigraphic measures of naturally occurring sleep deficits are more accurate than caregiver reports of sleep (Dayyat, Spruyt, Molfese, & Gozal, 2011) and may be more generalizable and representative of what children actually experience in their day-to-day lives.

The Current Study

The present study had three aims. The first was to examine the links and non-links between measures of attention across situations (unstructured and structured lab tasks), informants (parents and babysitters/daycare staff), and time (30, 36, and 42 months of age). We expected that toddlers who sustained attention in the lab during free play and a sorting task would be less likely to be rated by their parents and babysitters as toddlers who “can’t pay attention for long”, “quickly shift from one activity to another”, and “wander away”. Previous research has confirmed this notion, showing that the longer children orient to toys during free play at age three, the lower maternal ratings of hyperactivity are at age six (Campbell, Breaux, Ewing, & Szumowski, 1986). However, in cross-sectional research with toddlers by our research group, sustained attention in the lab was not linked with parent ratings of attentional control, but was linked with secondary caregiver ratings (Acar, Frohn, Prokasky, Molfese, & Bates, 2018). Given these mixed findings in the extant literature, we were not sure how attention measures would inter-relate across situations, sources, and time in the present study. Nevertheless, we expected to find cross-time reliability in each of the attention measures, and we expected that attention measures might be more strongly inter-related at later stages of toddlerhood, based on what has been seen in prior research (Petersen et al., under review; Acar et al., 2018).

The second aim was to examine the independent and interactive effects of language and sleep on attention. We expected that toddlers with poorer language may be at risk for poorer attention, and we expected that toddlers who often experience poor or insufficient sleep may have difficulties sustaining attention compared to their more well-rested peers. We also expected a significant interaction between language and sleep, such that toddlers with less advanced language skills who are at risk for poor attentional control may be especially affected when they experience poor or insufficient sleep. While examining the independent and interactive effects of language and sleep on attention, we also examined the role of socioeconomic differences as a proxy for potential differences toddlers may experience in environmental stressors, household resources, and caregiving. Accounting for socioeconomic differences allowed us to test whether language and sleep contribute to attentional control beyond the caregiving environment.

The third aim was to test bidirectional links between sleep and attention using cross-lagged panel models. This also allowed us to determine whether sustained attention, as measured in the laboratory, mediates the expected link between poor quality and quantity of sleep and parent reports of attention problems. Cross-lagged analyses with toddlers have rarely been undertaken. The present study was therefore novel in its use of cross-lagged analyses to clarify relations among actigraphic sleep and observed attention across toddlerhood. Given limited and mixed prior findings about the effects of sleep duration, variability, fragmentation, or timing on functioning, we considered all of these aspects, as well as daytime naps, and did not have specific hypotheses about which aspect of sleep would be most important for toddlers’ attention.

Method

Participants

Participants in this study included 314 toddlers recruited at 30 months of age (M = 30.12 months, SD = 0.06, 54% male) and assessed at 30, 36, and 42 months. Families were recruited in several ways, including through a database search of county birth records, through community outreach efforts (e.g., Head Start and the Housing Authority), and through public advertising (e.g., postcards and flyers). Compensation was provided, and transportation offered. The Institutional Review Board at Indiana University approved the Toddler Development Study (protocol # 0811000120), and parents provided informed consent. Sample sizes and rates of missingness by measure at each age are listed in Table 1. The participants were part of a larger, longitudinal project on the development of self-regulation.

Table 1.

Rates of Missingness for all Measures

Construct	Measure	Age (months)	N	% Missing (out of 314)
Sustained Attention	Free Play Composite	30	295a	6
	Free Play Composite	36	247	21
	Free Play Composite	42	247	21
Sustained Attention	Token Sort Time Sorting	30	296a	6
	Token Sort Time Sorting	36	251	20
	Token Sort Time Sorting	42	252b	20
Attention Problems (Primary Caregiver Report)	CBCL	30	314	0
	CBCL	36	261	17
	CBCL	42	247	21
Attention Problems (Secondary Caregiver Report)	CBCL	30	139	56
	CBCL	36	129	59
	CBCL	42	141	55
Language	DAS Verbal Ability	30	307c	2
	DAS Verbal Ability	36	263	16
	DAS Verbal Ability	42	255	19
Sleep	Actigraphy	30	294d	6
	Actigraphy	36	242	23
	Actigraphy	42	223	29

Note. Secondary caregivers include regular babysitters or daycare staff, and these reports were only used in the analyses for Aims 1 and 2.

^aSix percent of the sample did not provide sustained attention data for either the free play task or the token sort task. Missingness was either due to the family withdrawing from the study before the lab visit (n = 4), child fussiness (n = 8), or camera equipment failure (n = 7 for free play; n = 6 for token sort).

^bThree families began the study at 36 months and were retained at 42 months, and retention efforts were more pronounced at this final wave of data collection with some families missing their 36 month visit but returning for their final visit, resulting in a higher n for Token Sort at 42 months than at 36 months.

^cTwo percent of the sample did not provide DAS data at 30 months either because the child refused to play (n = 6) or because the DAS was not administered for a nonverbal child (n = 1).

^dFor the key study variables, the highest rates of missingness across ages were observed with actigraphy data. At 30 months, 6% of the sample did not provide actigraphy data either due to losing/refusing to wear the actigraph (n = 7) or to actigraph error (n = 13). At 36 months, 23% of the sample did not provide actigraphy data due to losing/refusing to wear the actigraph (n = 5), actigraph error (n = 16), or the family not participating (n = 51). At 42 months, 29% of the sample did not provide actigraphy data due to losing/refusing to wear the actigraph (n = 8), actigraph error (n = 25), or the family not participating (n = 58).

The sample was 80% White, 5% Latinx, 3% Black, 2% Mixed Race, 3% Other, and 7% Unknown, Not Reported, or Missing, reflecting the general racial breakdown of the Midwestern city in which these data were collected (6 – 8% Black and Latinx according to recent US Census Data). Ninety-seven percent of the primary caregivers were mothers, most of whom were married (89% married or remarried, 11% single/never married, divorced, or separated). Other primary caregivers were fathers (2%) or grandmothers (1%). Primary caregivers ranged in age from 19 to 53 (M = 32.85, SD = 5.17 years). In 35% of the families, the target child was the only child, in 58% there were one or two other children in the home, and in 7% there were four or more children. Family socioeconomic status (SES) was calculated using the Hollingshead Four Factor Index (Hollingshead, 1975), which takes into account the parents’ educational attainment and the prestige of their occupation (based on US Census codes), with both parents’ education and occupation scores equally informing estimates when both parents were employed, but with only one parent’s score informing estimates in families with a single parent or a non-employed parent. SES estimates ranged from 13 to 66, with M = 47.59 (SD = 14.13), suggesting that the sample was predominantly middle class, although not uniformly so.

Procedure

At each assessment, a trained experimenter visited the family’s home and administered a standardized assessment of the toddler’s language ability in a quiet area of the home. Parents were asked to remain silent without giving any help or feedback to the toddler. The experimenter also gave the toddler an actigraph to wear over the next one to two weeks and provided instructions to the parent about actigraph use. The toddler’s parent was asked to complete daily sleep diaries and several questionnaires, concerning demographics and child behavior. One week after the home visit, the toddler visited the lab with his or her parent to complete a battery of tasks, including measures of sustained attention. Next, the toddler’s secondary caregiver, if identified by the parent, was asked to complete questionnaires about the toddler’s behavior.

Measures

Sustained attention during free play in the lab.

Toddlers were seated at a table with a standardized set of developmentally appropriate toys to play with independently for five minutes while their parent completed other research procedures in a nearby room. A trained experimenter sat on a couch across from the toddler and limited her interactions with the toddler by telling the toddler that she needed to complete paperwork. The free play task was video recorded and later independently coded by trained research assistants using the coding protocol outlined by Ruff and Capozzoli (2003). Coders used ELAN (Brugman, Russel, & Nijmegen, 2004) to create time-stamped annotations for each moment of either casual, settled, or focused attention. Casual attention was coded anytime the toddler looked at the toys. Settled attention was coded if the toddler looked at and touched the same toy. Focused attention was coded when the toddler looked at and touched a toy, while also leaning in towards the toy and displaying a serious facial expression, little to no talking, and effortful, goal-directed, fine-motor movements (Ruff et al., 1998). These levels of attention built upon each other, such that settled attention could only occur when casual attention occurred, and focused attention could only occur when settled attention occurred. The proportion of time the toddler spent in each attentional level during the total free play task was computed (ranging from 0-1). The three proportions for the three levels of attention were then weighted, with the proportion of casual attention multiplied by 1, settled attention multiplied by 2, and focused attention multiplied by 3. These scores were then averaged, forming a weighted free play composite with a possible range of 0-2. For example, if a toddler spent 80% of the five-minute task at casual, 60% at settled, and 20% at focused, the weighted scores to be averaged would be .8, 1.2, and .6, respectively, with a composite score of .87. Twenty percent of the videos were coded by two raters, and the coders were reliable with each other (ICC = .83 for the weighted composite).

Sustained attention during the token sort task in the lab.

Later in the lab visit, the token sort task was administered (Goldsmith, Reilly, Lemery, Longley, & Prescott, 1999). The experimenter presented the toddler with a board on which were fastened one large container and three smaller ones. The large container held a mix of blue, white, and red tokens. The smaller containers were marked, each with a different color, to indicate how the tokens should be sorted. The experimenter asked the toddler to sort the tokens from the large container into the smaller containers based on color. Once the experimenter left the room, the toddler was given three minutes to independently sort the tokens into their respective color containers. Video recordings of the task were later coded using ELAN for time spent sorting tokens. The total amount of time spent sorting ranged from 0-180 seconds. Twenty percent of videos were coded by two raters, who were reliable with each other (ICC = .99).

Attention problems.

Primary (parent) and secondary (babysitter/daycare provider) caregivers completed the Child Behavior Checklist (CBCL 1 ½ - 5; Achenbach & Rescorla, 2000) at all ages. The attention problems subscale (α = .92) comprised of 5 items, which were rated from 0 = not true, 1 = somewhat or sometimes true, to 2 = very true or often true, was used in the present study to reflect problems with inattentiveness (e.g., “can’t concentrate, can’t pay attention for long”) and hyperactivity (e.g., “can’t sit still, restless, or hyperactive”). Higher scores indicate more attention problems.

Sleep.

Sleep was assessed using a MicroMini Motionlogger actigraph (Ambulatory Monitoring, Inc., Ardsley, NY), a watch-like accelerometer that monitors minute-by-minute motor activity to determine sleep and wake patterns. Toddlers wore actigraphs on their non-dominant wrists continuously for up to two weeks. There were 9.58 days of data on average across all participants and ages (M = 9.62, SD = 3.60 days at 30 months; M = 9.61, SD = 3.76 days at 36 months; M = 9.52, SD = 4.07 days at 42 months), following the accepted recommendation to collect at least five days of actigraphy data for reliable estimates (Acebo et al., 1999). Variability in the number of days that the toddler wore the actigraph was due to child noncompliance with wearing the device, family scheduling preferences for the lab visit (as actigraphs were returned to the research team during this visit), and actigraph failure. The toddler’s primary caregiver completed daily sleep diaries that were used to mark sleep and wake times, and any times the toddler did not wear the actigraph. Actigraph data were processed in the AW-2 software package (Motionlogger Analysis Software Package Action W-2 software, version 2.6.92, Ambulatory Monitoring, Inc) using the Sadeh algorithm, an algorithm that has been validated for use in early childhood (Sadeh, Sharkey, & Carskadon, 1994).

Based on previous principal components analysis (Staples et al., 2019), the large number of AW-2 actigraph variables were summarized into four composite indexes. Composite indexes were formed by standardizing the actigraphy variables (based on sleep data across all assessments) and averaging them. These four composites—sleep duration, sleep timing, sleep variability, and sleep activity—represent broad dimensions of actigraphy that are often examined in the child sleep literature (Meltzer et al., 2012). The sleep duration composite is composed of the mean of z-scored actigraph variables including average time the toddler spent in bed each night, the time the toddler spent in bed after sleep onset, and the time the toddler spent asleep each night, excluding night awakenings. The sleep variability composite is composed of the mean of z-scored actigraph variables including the night-to-night standard deviations of: time of sleep onset, duration of time spent in bed at night, time spent asleep at night not including night awakenings, and the midpoint of sleep (halfway between actigraphically determined true sleep onset and wake time). The sleep activity composite is composed of the mean of z-scored actigraph variables including the average time awake after sleep onset, the average minute-by-minute activity level, the average number of night awakenings lasting at least five minutes, the average duration of the longest wake episode after sleep onset, and the average percent of active epochs after sleep onset. The sleep timing composite is composed of the mean of z-scored variables including the average time of midsleep, average time of sleep onset, and the average bedtime reported on the sleep diary. The specific actigraphy variables included in each composite are also listed in Supplementary Table S1. A separate actigraphy variable that was not in any composite was nap sleep minutes, which indexes the number of minutes scored as sleep during the period between the diary-reported nap time and the first epoch after a period of sleep for which the activity count reached 50 and remained above that threshold until the next sleep interval. The average total nap minutes per day across the two weeks of data collection was used in all analyses. We also accounted for total sleep minutes, a sum of nighttime and daytime sleep.

Language ability measured in the home.

The Differential Abilities Scale (DAS; Elliott, 1990) – Early Years Battery was used to assess language at each age, with higher raw scores indicating better performance. Both receptive (Verbal Comprehension) and expressive language (Naming Vocabulary) were included. Cronbach’s alpha was .85 on average across ages.

Analysis Plan

To accomplish Aim 1, which concerned links and non-links among measures of attention across situations, informants, and time, bivariate correlations were computed across sustained attention measures, caregiver reports of attention problems, and ages. For all bivariate correlations, we also corrected for multiple tests using Bonferroni alpha corrections.

To accomplish Aim 2, which concerned the independent and interactive effects of language ability and sleep on attention, bivariate correlations were computed between language, sleep, and attention. We then tested potential interactions between aspects of sleep and language predicting attention in toddlerhood. For each interaction, a hierarchical regression equation was estimated predicting attention with language and the proposed sleep index (as step 1), followed by entry of the two-way interaction between language and the proposed sleep index. Given the expected overlap between SES and our constructs of interests, we controlled for SES in additional sensitivity analyses with partial correlations and multiple regression equations.

To accomplish Aim 3, we used cross-lagged panel models to test bidirectional links across time between sleep and sustained attention, and to determine whether sustained attention mediates the expected link between poor and insufficient sleep and parent-reported attention problems. To maximize the amount of data used in these models, we only used parent reports of attention problems since secondary caregiver reports were only available for about half of the sample (see Table 1). Panel models are useful for examining between person differences, or change relative to others, and can be used to document whether there is a lead-lag relationship among two or more constructs. All path analysis models were fit using Mplus 8.4 (Muthen & Muthen, 2017). Mplus implements full information maximum likelihood estimation, which is a robust estimation method when data are missing at random. All of the longitudinal mediation models used maximum likelihood estimation with bootstrapping. We followed the recommendations of Kenny, Kaniskan, and McCoach (2015) to estimate additional paths to check for significantly improved model fit. We used likelihood ratio tests from Satorra–Bentler scaled χ² statistics for non-normal outcomes to compare proposed models to models with additional paths estimated (Satorra & Bentler, 1994). To determine if the proposed model fit the data well, the proposed “simpler” model was tested against a saturated “full” model with no degrees of freedom and perfect fit. The proposed simpler models included autoregressive paths for each construct across time, concurrent covariances between the three constructs, and cross-lagged regression paths (see Supplementary Figure S1). The full model estimated the additional covariance paths necessary for a saturated model, with all variances, covariances/regressions, and means freely estimated. If the full model fit better than the proposed simpler model, it suggested that the simpler model sacrificed accuracy for parsimony, and that additional paths (resulting in a saturated model) were necessary to account for the covariance structure of the data. The simpler model was selected for parsimony unless the full model had better fit.

Results

Descriptives for all analysis variables across ages are provided in Table 2.

Table 2.

Descriptive Statistics

Construct	Variable/Measure	Age	M	SD	Lower	Upper
Sustained Attention	Free Play Composite	30	.83	.20	.07	1.32
	Free Play Composite	36	.90	.20	.18	1.67
	Free Play Composite	42	.95	.17	.20	1.51
	Token Sort Time Sorting (s)	30	37.37	43.38	0	180
	Token Sort Time Sorting (s)	36	67.05	59.79	0	180
	Token Sort Time Sorting (s)	42	94.69	63.41	0	180
Attention Problems (Primary Caregiver Report)	CBCL	30	2.58	1.78	0	8
	CBCL	36	2.47	1.66	0	8
	CBCL	42	2.15	1.65	0	7
Attention Problems (Secondary Caregiver Report)	CBCL	30	2.04	1.88	0	8
	CBCL	36	1.86	1.82	0	8
	CBCL	42	1.74	1.90	0	9
Language	DAS Verbal Ability	30	77.72	19.21	19.00	118.50
	DAS Verbal Ability	36	94.17	17.80	33.50	139.50
	DAS Verbal Ability	42	106.85	16.95	43.50	154.00
Sleep	Sleep Duration (z)	30	.06	.87	−2.97	2.80
	Sleep Duration (z)	36	.12	.94	−2.82	2.51
	Sleep Duration (z)	42	.12	.88	−3.07	2.01
	Sleep Activity (z)	30	.17	.94	−1.92	4.31
	Sleep Activity (z)	36	−.04	.80	−2.08	2.20
	Sleep Activity (z)	42	−.10	.81	−1.79	2.89
	Sleep Timing (z)	30	−.01	.93	−2.08	3.17
	Sleep Timing (z)	36	.07	1.02	−2.28	3.79
	Sleep Timing (z)	42	−.01	1.01	−2.22	4.73
	Sleep Variability (z)	30	.05	.90	−1.40	6.81
	Sleep Variability (z)	36	−.02	.76	−1.37	3.64
	Sleep Variability (z)	42	.00	.81	−1.40	3.53
	Total Nap Minutes	30	90.53	31.64	0	299.00
	Total Nap Minutes	36	85.04	33.89	0	193.00
	Total Nap Minutes	42	79.46	32.91	0	231.00

Note. “Lower” is the lower bound of the 95% confidence interval; “Upper” is the upper bound of the 95% confidence interval around the mean.

Aim 1. Measurement of Attention across Situations, Informants, and Time

Bivariate correlations across laboratory measures of sustained attention, caregiver reports of attention problems, and ages are provided in Table 3. Within each attention measure, cross-time correlations were statistically significant (.18 ≤ r ≤ .66; for scatter plots, see Supplementary Figure S2), indicating that each attention measure showed rank-order stability across time. We also tested whether these associations remained statistically significant using Bonferroni alpha corrections for multiple tests (p < .00076). All of the cross-time, within-measure correlations remained significant after Bonferroni corrections with two exceptions – free play sustained attention at 30 and 36 months were no longer significantly associated with free play sustained attention at 42 months. However, the lagged association between free play sustained attention at 30 and 36 months remained significant (r = .29, p = .00001). Most remarkable in the pattern of cross-time, within measure correlations was that each measure was correlated most clearly with its counterpart at the other ages and much less with the other attention indexes—a clear pattern of differential linkage across time. In effect, each measure was validated as a stable trait, even with Bonferroni corrections. The between measure correlations were much less prominent than the within measure correlations, and none of the between measure correlations remained significant with Bonferroni corrections for multiple tests.

Table 3.

Bivariate Correlations between Attention Measures at 30, 36, and 42 months of age

	1	2	3	4	5	6	7	8	9	10	11	12
1. Free Play Composite 30	1.00
2. Time Sorting 30	.04	1.00
3. Attention Problems P 30	.04	.00	1.00
4. Attention Problems S 30	.09	−.22*	.22 *	1.00
5. Free Play Composite 36	.29 **	.04	−.02	.12	1.00
6. Time Sorting 36	.02	.28 **	.00	−.02	.01	1.00
7. Attention Problems P 36	.01	−.01	.53 **	.16	.06	−.01	1.00
8. Attention Problems S 36	.08	.00	.04	.61 **	.14	−.01	.06	1.00
9. Free Play Composite 42	.17 *	.07	.01	.00	.18 **	.13	−.04	.01	1.00
10. Time Sorting 42	.06	.23 **	−.08	−.21 *	−.06	.44 **	−.05	−.12	.15 *	1.00
11. Attention Problems P 42	−.06	.04	.48 **	.16	.08	−.01	.66 **	.10	−.09	.00	1.00
12. Attention Problems S 42	.09	−.17	.18*	.43 **	−.05	−.14	.17	.45 **	−.18*	−.12	.12	1.00

Note. Attention Problems P represents primary caregiver reports of inattention and hyperactivity on the CBCL. Attention Problems S represents secondary caregiver reports of inattention and hyperactivity on the CBCL. N range = 124 - 314. Significant associations are bolded.

^**Correlation is significant at the 0.01 level (2-tailed).

^*Correlation is significant at the 0.05 level (2-tailed). Associations that remained statistically significant after Bonferroni correction for multiple tests are underlined.

The lab-based measures of sustained attention were only weakly associated with each other at 42 months (r = .15, p = .02), an association which did not hold with corrections for multiple tests. However, for the sake of reducing the number of analyses for our more complex questions, we chose to test the usefulness of a condensed summary of children’s attention characteristics across these measures. Based on the weak correlations among the attention indexes at 42 months (absolute r value range =.15 – .18, p range = 0.02 – 0.04; see scatter plots in Supplementary Figure S3), we tested a confirmatory factor analysis (CFA) model to test whether a higher-order construct reflecting attention would be justified. We set the mean and variance of the latent factor to zero and one, respectively, which allowed the factor loadings to vary freely. The CFA model converged only at age 42 months. The factor loadings with the latent attention construct were β = .47 for free play, β = .31 for token sort, and β = −.40 for secondary-caregiver reports of attention problems. Based on these results, we standardized (i.e., z-scored) sustained attention scores in free play and token sort, and the inverse of attention problems reported by secondary caregivers, and averaged these z-scores together to form a 42-month attention composite. This attention composite was tested in subsequent correlations and interactions.

Aim 2. Independent and Interactive Effects of Language and Sleep

Table 4 presents the bivariate correlations between language ability and attention across ages and measures, including the attention composite at 42 months. Language and attention were significantly associated at each age across measures of attention, but here only the associations that remained significant after Bonferroni corrections for multiple tests are highlighted (p < .0013). Language at 30 months was significantly predictive of sustained attention in free play at 36 months (r = .24, p = .0002). Language at 36 months was significantly associated with sustained attention in the free play task at 36 (r = .21, p = .0011) and 42 months (r = .24, p = .0003) and with sustained attention observed during token sort at 42 months (r = .27, p < .0001). Language at 42 months was significantly, concurrently associated with sustained attention observed during token sort (r = .23, p = .0002). The summary attention composite at 42 months was significantly associated with language at 36 months (r = .35, p < .0001) and at 42 months (r = .30, p < .0001); scatter plots for these associations are in Supplementary Figure S4.

Table 4.

Bivariate Correlations between Language and Attention at ages 30, 36, and 42 months

	Lang30	Lang36	Lang 42
1. Free Play Composite 30	.19 **	.18**	.13*
2. Free Play Composite 36	.24 **	.21 **	.18**
3. Free Play Composite 42	.16*	.24 **	.19 **
4. Time Sorting 30	.10	.10	.03
5. Time Sorting 36	.12	.19 **	.17*
6. Time Sorting 42	.12	.27 **	.23 **
7. Attention Problems Primary Caregiver 30	−.16**	−.14*	−.13*
8. Attention Problems Primary Caregiver 36	−.06	−.16*	−.14*
9. Attention Problems Primary Caregiver 42	.00	−.08	.00
10. Attention Problems Secondary Caregiver 30	.03	−.03	.03
11. Attention Problems Secondary Caregiver 36	.05	.04	.06
12. Attention Problems Secondary Caregiver 42	−.13	−.22*	−.27**
13. Attention Composite 42	.19 **	.35 **	.30 **

Note. Nrange = 109-294. Attention composite is a z-scored composite reflecting higher scores on the free play task, longer time sorting, and fewer attention problems reported by the secondary caregiver. Concurrent, significant associations are bolded.

^**Correlation is significant at the 0.01 level (2-tailed).

^*Correlation is significant at the 0.05 level (2-tailed). Associations that remained statistically significant after Bonferroni correction for multiple tests are underlined.

There were far fewer significant associations between attention and sleep. The 42-month attention composite was used in these correlations to reduce the number of tests, and Bonferroni alpha corrections for multiple tests were also conducted (see Supplementary Table S2). The only sleep index that was significantly related to the 42-month attention composite was nap duration (r = −.14, −.19, and −.18, p = .04, .01, and .03 at 30, 36, and 42 months, respectively). None of these remained significant after Bonferroni corrections for multiple tests (p < .003).

Next, we examined potential interactions between language and sleep in predicting the attention composite at 42 months. Of the six interactions tested (with language at 42 months and sleep duration, timing, variability, activity, nap minutes, and total sleep minutes at 42 months), only the interaction between language and nighttime sleep duration was statistically significant (see Figure 1). As expected, language ability had a significant main effect (β = .20, p < .01); better language ability was associated with better attention scores. The two-way interaction effect with language ability by sleep duration was also significant (β = −.10, p < .01). To interpret this interaction, post-hoc probing using estimation of simple slopes at the mean, and at 1 SD above and below the sample mean, of language ability was conducted. Nighttime sleep duration was most strongly predictive of higher attention scores at lower levels of language ability 1 SD below the mean: β = .22, p < .05). In contrast, the association between sleep duration and attention was nonsignificant at high levels of language ability (1 SD above the mean: β = .05, p .53). As shown in Figure 1, toddlers with poorer performance on the language assessment were especially likely to have lower attention scores when they experienced short sleep durations. Toddlers with better language tended to have better attention, regardless of sleep duration.

Figure 1.

Standardized slopes between nighttime sleep duration and attention as a function of language ability. Low Language Ability = 1 standard deviation below the mean, and High Language Ability = 1 standard deviation above the mean. The full equation was significant, F (3, 206) = 6.45, p < .01

To examine the replicability of this interaction effect, the same regression equation predicting the 42-month attention composite was fit at 30 and 36 months using sleep and language measured at those ages. The same pattern emerged at 30 and 36 months, but the interaction term was only trending towards statistical significance at 30 months (β = −.10, p < .10) and was not significant at 36 months (β = .00, p = .85). At 30 months, toddlers with lower language abilities again showed a positive association between sleep duration and attention (β = .14, p = .02 at 1 SD below the mean on language ability).

SES as a covariate.

SES was also considered as a covariate in the Aim 2 analyses. First, we examined the bivariate correlations between SES, attention, language, and sleep (see Table 5). Only nap minutes and sleep duration were tested here as these were the only sleep indexes significantly associated with attention in the Aim Two analyses. SES was significantly, positively associated with language ability at each age (.20 ≤ r ≤ .29, .000 ≤ p ≤ .002) and with sleep duration at 30 months (r = .17, p = .004). Toddlers from higher socioeconomic backgrounds tended to have more advanced language across ages and longer sleep durations at 30 months. These associations remained significant after Bonferroni corrections (p <.005).

Table 5.

Bivariate Correlations between SES, Attention, Language, and Sleep

	Hollingshead SES
1. Attention Composite 42	.11
2. Language 30	.29 **
3. Language 36	.21 **
4. Language 42	.20 **
5. Total Nap Minutes 30	.04
6. Total Nap Minutes 36	.02
7. Total Nap Minutes 42	.10
8. Sleep Duration 30	.17 **
9. Sleep Duration 36	−.04
10. Sleep Duration 42	.05

Note. N range = 166-292. Attention is a z-scored composite at 42 months, reflecting higher scores on the free play task, longer time sorting, and fewer attention problems reported by the secondary caregiver. Nap minutes and sleep duration were the only sleep indexes tested here because these were the only sleep indexes significantly associated with attention in the Aim Two analyses. Significant associations are bolded.

^**Correlation is significant at the 0.01 level (2-tailed).

^*Correlation is significant at the 0.05 level (2-tailed). Note that 2-tailed tests overestimate the risk of a spurious result to the extent that the direction of the correlation is expected in advance; based on the literature, positive correlations were expected for variables 1-4 and 8-10. Associations that remained statistically significant after Bonferroni correction for multiple tests are underlined.

Of the key findings from Aim 2, language ability at 42 months was still significantly associated with the attention composite at 42 months when SES was statistically controlled in a partial correlation (r = .32, p < 0.01), and nap sleep minutes at 42 months was still inversely related to attention at 42 months when SES was controlled in a partial correlation (r = −.26, p < 0.05). Lastly, the interaction between language and sleep duration in predicting the attention composite at 42 months also remained significant when controlling for SES (β = −.10, p < .05).

Aim 3: Cross-lagged Panel Models of Sleep, Sustained Attention, and Attention Problems

Cross-lagged models were used to test bidirectional links between sleep and sustained attention and whether sustained attention, in either the free play or the token sort task, mediated the expected effect of poor and insufficient sleep on later parent-reported attention problems. We first examined whether there was systematic missingness for actigraphy data and sustained attention data as a function of SES or child sex at the 30 month assessment. There were no significant differences between toddlers with and without actigraphy data in terms of their SES (t (312) = −0.12, p = .90), or their sex (t (312) = −0.58, p = .57). There were also no significant differences between toddlers with and without sustained attention data measured in the lab at 30 months in terms of SES (t (312) = −0.46, p = .65). The difference between toddlers with and without sustained attention data measured in the lab at 30 months in terms of child sex was trending towards significance (t (312) = 1.94, p = .06). Toddlers whose sustained attention was observed in the lab were, to a slight degree, more often girls than boys. Next, we examined whether toddlers who were lost to attrition at 36 and 42 months differed from toddlers who participated at these waves. Toddlers who were lost at 36 months compared to toddlers who completed the 36 month assessment did not differ in terms of SES (t (312) = 0.34, p = .73), sex (t (312) = 1.00, p = .32), sustained attention (t (293) = −1.19, p = .24), language (t (305) = −0.80, p = .43), or sleep (t (292) = −1.03, p = .30). Additionally, toddlers who were lost at 42 months compared to toddlers who completed the 42 month assessment did not differ in terms of sex (t (312) = 1.00, p = .32), sustained attention (t (293) = .85, p = .40), or language (t (305) = −1.45, p = .15). There were slight differences between toddlers with and without data at 42 months in terms of their SES (t (312) = −1.73, p = .09) and sleep (t (292) = −1.78, p = .08), but these differences were only trending towards significance. Overall, there was little to no evidence of systematic missingness. Thus, we did not include covariates in these models to account for missingness. To maximize data used in these cross-lag models, we only used parent reports of attention problems since secondary caregiver reports were only available for about half of the sample (see Table 1).

To determine whether sustained attention mediated the effect of sleep on later parent-reported attention problems, we used the longitudinal mediation model recommended by Cole and Maxwell (2003). The proposed model examined whether (a) each index of problematic sleep predicted later sustained attention controlling for prior sustained attention, (b) sustained attention predicted later attention problems controlling for prior levels of attention problems, (c) problematic sleep predicted later attention problems controlling for prior levels of sustained attention and attention problems, and (d) whether the effect of problematic sleep on later attention problems was mediated by sustained attention. Indirect effects were tested by bootstrapping 95% confidence intervals from 1,000 bootstrap samples, as recommended for tests of mediation with small to moderate sample sizes (Shrout & Bolger, 2002)

Sustained attention in free play.

Using the template depicted in Supplementary Figure S1, we tested five separate longitudinal mediation models using sustained attention in free play with five sleep indexes (duration, activity, timing, variability, and nap minutes). We did not find the expected mediation effect on attention problem outcomes, but we did find other effects that are worth considering. Sustained attention in free play was significantly associated with sleep variability (see Figure 2). The full model did not result in significant improvement in fit, so the simpler model was retained (χ²[6] = 3.38, p = .24 for sleep variability). Free play sustained attention was not significantly associated with the other sleep indexes across time.

Figure 2.

Longitudinal mediation model for sleep variability, sustained attention in free play, and parent-reported attention problems. Estimates represent standardized regression coefficients. Nonsignificant paths are not displayed. CFI = .99, TLI = .95, RMSEA = .04. *p < .05, **p < .01, ***p < .001.

Figure 2 shows that more variable sleep at 30 months was concurrently associated with less sustained attention (r = .11). The figure also shows bidirectional effects across time, with more variable sleep predicting less growth in sustained attention from 30 to 36 months (β = −.17), and more sustained attention at 30 months predicting a reduction in night-to-night sleep variability at 36 months relative to 30 months (β = −.16). Scatter plots for these associations are presented in Supplementary Figure S5. By 42 months, sustained attention observed during free play was also inversely associated with parent-reported attention problems (r = −.11).

Sustained attention in token sort.

We again used the template depicted in Supplementary Figure S1 to test separate models for sustained attention observed during token sort with the five sleep indexes. As with the free play measure, the token sort measure of sustained attention was not found to mediate associations between sleep and parent-reported attention problems. Time spent sorting was significantly, negatively associated with nap minutes (see Figure 3) and with sleep activity (see Figure 4), but not with any of the other sleep indexes. The full models did not result in significant improvement in fit; therefore, the simpler models were retained (χ²[6] = 10.44, p = .11 for nap minutes; χ²[6] = 8.55, p = .20 for sleep activity).

Figure 3.

Longitudinal mediation model for total nap minutes, sustained attention in token sort, and parent-reported attention problems. Estimates represent standardized regression coefficients. Nonsignificant paths are not displayed. CFI = .99, TLI = .90, RMSEA = .06. *p < .05, **p < .01, ***p < .001.

Figure 4.

Longitudinal mediation model for sleep activity, sustained attention in token sort, and parent-reported attention problems. Estimates represent standardized regression coefficients. Nonsignificant paths are not displayed. CFI = .99, TLI = .97, RMSEA = .03 ^p < .10, *p < .05, **p < .01, ***p < .001.

Figure 3 shows that toddlers who napped more at 36 months tended to concurrently have less sustained attention during the token sort task (r = −.20). We also found that toddlers who napped more at 36 months showed less growth in sustained attention at 42 months (β = −.17), The scatter plot for this association is presented in Supplementary Figure S5. Some toddlers who nap more may go to bed later, which could, in turn, affect attentional control. Thus, we conducted a sensitivity analysis using multiple regression to control for lateness of bedtime (see Table 6). The number of nap minutes at 36 months was still inversely associated with time spent sorting tokens at 42 months, even with statistical controls for sleep timing (β = − 0.21, p < .01). Lastly, Figure 4 shows that toddlers who spent more time sorting in the structured lab task later had less active sleep (β = −.15). The scatter plot for this association is presented in Supplementary Figure S5.

Table 6.

Sensitivity Analysis: Hierarchical Regression Predicting Time Sorting at 42 Months

	Time Sorting (42 months)
Variable	B	SE	β	p
Sleep Timing (36 months)	.93	.01	.01	.855
Nap Minutes (36 months)	−.39	.14	−.21	.005
R 2	.04
F for change in R2	8.05
p	.005

Discussion

The present study measured individual differences in toddlers’ attention across situations, informants, and time, examined the influence of relevant child factors, specifically language ability and sleep, in predicting attention, and examined bidirectional links between sleep and sustained attention across a year of toddlerhood.

Attention Tasks and Ratings of Attention Problems

The first aim of the present study was to examine toddlers’ attention across situations (i.e., sustained attention during free play and a more structured token sorting task), informants (primary and secondary caregiver reports of attention problems), and time. Notably, although the two laboratory measures of sustained attention showed clear cross-time continuity, they were only weakly associated with each other at 42 months, and not associated with each other at 30 or 36 months. Additionally, at 42 months, lower levels of sustained attention during free play were weakly associated with higher levels of attention problems as reported by secondary caregivers. Although these associations were weak and did not hold after corrections for multiple tests, a CFA did support a latent construct of the three measures at 42 months. For the sake of reducing the number of tests, a composite of these measures was used in several additional analyses.

Previous research with children has similarly shown minimal cross-situational correlations for sustained attention performance (Ruff et al., 1998). We found that toddlers’ sustained attention varies with task demands and with toddlers’ abilities to meet those demands, particularly at younger ages. The associations between the lab and secondary caregiver-reported attention measures at 42 months, although modest, may suggest that sustained attention becomes more coherent with age, as evidenced by the increase in cross-situational stability in performance. This corresponds with Acar et al. (2018) who used similar measures with toddlers and found that a single attention factor, with consistent factor structure and scale, only emerged after 30 months. This cross-measure linkage at 42 months is also consistent with data from this same cohort of toddlers showing that several measures of inhibitory control are inter-related by 42 months (Petersen et al., under review).

Neither of the sustained attention performance measures was associated with parent reported attention problems at any age, except for a small association between free play sustained attention and parent reported attention problems shown in the cross-lag models at 42 months. Our assessment battery included parent- and secondary caregiver-report measures of attention-relevant, functional behavior in everyday life as a way of capturing naturally occurring behavior outside of the lab to complement lab measures of sustained attention. We expected that toddlers who sustained attention in the lab during free play and the sorting task would be less likely to be rated by their parents and babysitters as children who “can’t pay attention for long”, “quickly shift from one activity to another”, and “wander away”. However, our findings of a general lack of association between sustained attention and parent reported attention problems is consistent with previous findings. Acar and colleagues (2018) found that parent ratings of attentional focusing on a temperament questionnaire were not related to secondary caregiver ratings or to lab attentional performance. Notably, we saw a lack of agreement between primary and secondary caregiver reports of attention problems at each age, despite prior literature showing that parents and teachers tend to agree to a modest level on ratings of behavior problems (Kolko & Kazdin, 1993). Child adjustment may be distinct in different settings, perhaps especially so in toddlerhood.

Independent and Interactive Effects of Language and Sleep

We highlight three findings concerning how language and sleep predict child attention. First, toddlers with more advanced language tended to have better sustained attention in both free play and token sort. Since language was linked with attention across measures and ages in this study, language may be a promising, early target for assessment and, ultimately, intervention to maximize cognitive development and prevent the downstream consequences of poor attentional control. Of course, the potential bidirectional nature of language-attention associations must also be considered. Just as language may play a self-regulatory, causal role in the development of attentional control, so too can attentional control facilitate language development.

Second, we found very few bivariate associations between sleep and the attention composite at 42 months. Notably, the duration of naps at each age was inversely linked with attention; toddlers who napped more tended to have worse attention. These associations between naps and attention did not hold with alpha corrections for multiple tests, but this finding was replicated in subsequent structural equation models. The nap finding is discussed in more detail below, but first, we consider the lack of main effects for nighttime sleep on attention.

Prior research on associations between sleep and sustained attention has been inconclusive. Some studies have found that poor and insufficient sleep were not significantly associated with sustained attention (Astill et al., 2012). Other studies have shown that sleep deficits were associated (inversely) with sustained attention (e.g., Hoyniak et al., 2015). The present study longitudinally followed a large sample of toddlers with multiple lab measures of sustained attention, functional attention traits in normal life, and actigraphic measures of sleep, yet there was a general lack of findings of concurrent sleep and attention relations. Nevertheless, the pattern of findings does not rule out possible sleep-attention relations—small correlations can sometimes obscure higher correlations in some meaningful subgroups. The current findings, while not supporting concurrent links between sleep and attention across the whole sample, did yield an interaction between sleep and language in accounting for attention problems.

The third key finding, then, showed that, despite the general lack of concurrent linkage between sleep and attention, nighttime sleep duration mattered more clearly for toddlers with lower language development than for toddlers with higher language development. This finding is notable because it is consistent with the interpretation posed by Durmer and Dinges (2005) that advanced cognitive resources may help individuals compensate when sleep deprived, using their faster processing speed, greater memory capacity, and larger vocabularies to function at high levels even when sleep deprived (Durmer & Dinges, 2005). In the context of this study, toddlers with advanced language abilities may be able to compensate and attend well even when sleep deprived, but toddlers with poorer language abilities appear to lack this compensatory resource and are more negatively affected by shorter sleep. Nighttime sleep duration has been studied extensively with other age groups, and this finding is consistent with prior research and theory, but even so, it should be interpreted with caution, given that it was the only one of six possible interactions to reach statistical significance. If the pattern were to hold in future research, it could imply an interesting approach to prevent attention problems in early childhood, by focusing on the sleep of toddlers with lagging language development in particular.

SES as a Covariate

While examining the independent and interactive effects of language and sleep on attention, we also examined the role of socioeconomic differences as a proxy for potential differences toddlers may experience in environmental stressors, household resources, and caregiving. The caregiving environment alone has been demonstrated to significantly contribute to developing sustained attention. Accounting for socioeconomic differences allowed us to test whether the child factors of language and sleep also contributed to attentional control beyond the broad SES index of the caregiving environment. SES was significantly, positively associated with language ability at each age, meaning that toddlers from higher socioeconomic backgrounds tended to have more advanced language across ages, which is consistent with prior research (Fernald et al., 2013). Higher SES was also associated with longer 30-month sleep durations and with higher 42-month attention scores, again converging with prior research (El-Sheikh et al., 2013; Russell et al., 2016). However, the contributions of language and sleep predicting attention remained when socioeconomic differences were controlled.

Sleep, Sustained Attention, and Attention Problems in Cross-Lagged Panel Models

The cross-lagged panel analyses yielded three significant associations between sleep and sustained attention across time, but given the lack of cross-lag associations between the attention tasks and the parent-rated attention problems, the proposed mediation effects were not found. The first significant association between a sleep variable and a sustained attention variable involved an inverse, reciprocal relation over time between free play sustained attention and sleep variability. Toddlers who had less variable sleep schedules at 30 months experienced more growth from 30 to 36 months in sustained attention compared to their peers, and toddlers who had higher levels of sustained attention at 30 months also had less growth in sleep variability compared to their less attentive peers. The first link is consistent with previous cross-sectional research showing that more variable sleep is associated with less sustained attention (Hoyniak et al., 2015) and more behavior problems (Bates, Viken, Alexander, Beyers, & Stockton, 2002). This result underscores the importance of consistent timing and duration of sleep from night to night, which has been demonstrated in previous research showing the benefits of consistent bedtime routines (Mindell & Williamson, 2018). The reciprocal link in the pattern, with more sustained attention predicting less sleep variability is consistent with another recent study using cross-lagged panel model analyses that found that greater parent-reported attention regulation at age 2-3 predicted fewer parent-reported sleep problems at age 4-5 (Williams, Berthelson, Walker, & Nicholson, 2017). Toddlers with greater attentional regulation may benefit more from their parents’ involvement at bedtime or may be generally more positively involved in the bedtime routine, which would help to make routines more efficacious, which, in turn, could make it easier for a family to be consistent from night-to-night in sleep timing and duration. Indeed, recent research has shown that warm, mutually enjoyable bedtime routines contribute to more regular sleep patterns during toddlerhood (Hoyniak et al., 2020).

The second significant association between sleep and sustained attention in the cross-lagged panel models was an inverse association between nap duration and time spent sorting tokens. Toddlers who napped more showed worse sustained attention during the token sort task, which is consistent with the weak bivariate correlations between nap minutes and the attention composite found in relation to Aim 2. One might have expected that toddlers who nap more would be better rested and more adept at regulating and sustaining their attention in various contexts. It is important to note, though, that toddlerhood is characterized by substantial change in sleep patterns, with decreasing daytime sleep and increasingly consolidated nighttime sleep (Lushington, Pamula, Martin, & Kennedy, 2013). Toddlers who nap more than their peers may have developmentally immature sleep habits and this might be associated with other forms of developmental immaturity, such as poorer sustained attention abilities both in the lab and as observed by secondary caregivers. Indeed, a recent meta-analysis showed that children who cease napping earlier in development tend to be more advanced on other neurocognitive and language assessments (Staton et al., 2020).

The third significant association between sleep and sustained attention in the cross-lagged panel models involved sleep activity and sustained attention in the token sort task. Toddlers who spent less time sorting tokens at 36 months had more sleep activity 6 months later compared to their peers. Toddlers who are less able to sustain attention in a structured sorting task may share a similar connection between hyperactivity and nighttime activity as observed in prior research (Konofal, Lecendreux, Bouvard, & Mouren-Simeoni, 2001), but this association requires replication and further probing in future research. Additionally, toddlers in this study who spent less time sorting tokens may not be inherently hyperactive. Thus, this theorized connection between sustained attention performance and hyperactivity also requires further probing.

Limitations and Future Directions

The present study adds to our understanding of attentional control in toddlerhood. It suggests that attentional control is a multi-faceted and multiply determined developmental construct. Several limitations of the study could inform directions for future research. The sample included a relatively small number of low SES families, which may limit the generalizability of these findings. For example, prior work has shown that children from lower SES backgrounds tend to sleep significantly less overall, have later bedtimes, and show more variability in their sleep schedules (El-Sheikh et al., 2013; Hoyniak et al., 2019), potentially because these children tend to experience poorer sleep environments and poorer sleep hygiene (Jones & Ball, 2014) and are more likely to experience chronic health problems that can also interfere with sleep (Chen & Miller, 2013). Although the key findings from this study cannot be attributed to SES, because they remained statistically significant when socioeconomic differences were controlled, a sample with more lower-SES families might show a different pattern of associations involving sustained attention. Moreover, recent research has shown that children from lower SES backgrounds, who tend to be at risk for poorer sustained attention (Tomalski & Johnson, 2010), can benefit from training to improve attention (Ballieux et al., 2016) and thus would be important to include and target in future research and clinical efforts.

Other measurement procedures or analyses could be used in future research to replicate and extend the present findings. First, secondary caregivers, such as teachers, daycare staff, or babysitters, provide useful information about children’s attention problems that may surpass the informational utility of parent ratings, as has been demonstrated in other research (e.g., McQuillan et al., 2018). Secondary caregivers can evaluate child performance in more structured, academic contexts and in comparison to other age mates, providing a more comprehensive description of child strengths and weaknesses. In the present study, secondary caregiver reports were only provided for about half of the sample, a limitation that could be addressed in future research. Second, although the present study used two measures of sustained attention, sustained attention could be measured in other ways, too, such as with measures of sustained attention in the home or with lab-based Continuous Performance Tests, which have been previously linked with sleep deficits (Lim & Dinges, 2010). Third, among the key study variables, the highest rates of missingness across ages were observed with actigraphy data. Some of this missingness was due to toddlers losing or refusing to wear the device. We offered to help families find lost devices and provided small, daily gifts for parents to consider rewarding their toddler for wearing the actigraph each day. However, this missingness is important to highlight, especially because the toddlers most likely to lose or refuse to wear the device may be those with the most sleep problems and attention problems.

Conclusion

Using multiple measures of attention, actigraphy, and a standardized language assessment, we found that attention measures were stable across time but only weakly linked with each other, and mainly at 42 months. We also found that attention and language were linked and that sleep duration was positively associated with attention scores at 42 months among toddlers with poorer language abilities, even when SES was controlled. Lastly, we found that more variable sleep and longer naps were associated with less growth in sustained attention. These findings advance our understanding of attention development, and point to potential targets (i.e., language and sleep interventions) for preventing the negative impact of poor sustained attention on social and academic functioning.