Testing the Longitudinal Structure and Change in Sluggish Cognitive Tempo and Inattentive Behaviors From Early Through Middle Childhood

Abstract

Previous studies have demonstrated that sluggish cognitive tempo (SCT) behaviors are empirically distinct from inattentive (IN) behaviors that are used to define attention-deficit/hyperactivity disorder. However, most studies used cross-sectional designs during middle childhood. Using parent and teacher ratings from the Family Life Project (N = 1,173), we investigated the factor structure, longitudinal measurement invariance, developmental trajectories, and predictors of developmental change in SCT and IN from age 3 years through Grade 5. SCT and IN were dissociable but correlated constructs that exhibited longitudinal invariance for both informants. Mean levels of SCT increased modestly with age, becoming more prominent between age 5 years and first grade, while IN was more stable. Lower parental education was associated with higher parent- and teacher-reported SCT, male sex was associated with higher teacher-reported IN, and African American race was associated with higher teacher-reported IN but lower teacher-reported SCT. These findings support the validity of SCT starting in early childhood.

Sluggish cognitive tempo (SCT) is characterized by excessive daydreaming, inconsistent alertness, slow thinking/slow behavior, mental confusion or “fogginess,” apathy, and a lack of energy. SCT was initially identified as a set of behaviors to potentially characterize a sizeable minority of children with “purely” attention-deficit/hyperactivity disorder (ADHD) inattentive type in the Diagnostic and Statistical Manual of Mental Disorders, Third Edition (DSM-III; American Psychiatric Association [APA], 1980). However, SCT does not appear to indicate a subtype of ADHD (Barkley, 2012, 2013; Becker, Marshall, McBurnett, 2014; Becker et al., 2016; Willcutt et al., 2012). Instead, researchers have increasingly examined SCT as a separate construct in community and clinical samples (Becker, Luebbe, Fite, Stoppelbein, & Greening, 2014; Becker, Burns, Schmitt, Epstein, & Tamm, 2017; Leopold et al., 2016; Servera, Sáez, Burns, & Becker, 2018). These efforts have suggested that SCT may be a construct of transdiagnostic importance (Becker & Willcutt, 2019) or its own mental disorder (Barkley, 2013, 2014).

An initial recurring question was whether SCT behaviors are empirically distinct from the inattentive (IN) behaviors that characterize ADHD. Studies using factor analytic techniques indicate that SCT items load together on a factor that is dissociable from IN items but are also strongly correlated with IN (weighted r = .63; see Becker et al., 2016 for a recent meta-analytic review). However, most of these studies were limited to school-aged children of youth (6-18 years of age; Barkley, 2013; Bauermeister, Barkley, Bauermeister, Martinez, & McBurnett, 2012; Becker et al., 2014; Burns, Becker, Servera, Bernad, & García-Banda, 2017; Lee, Burns, Snell, & McBurnett, 2014; McBurnett et al., 2014; Sáez, Servera, Becker, & Burns, 2018; Willcutt et al., 2014). Thus, examining SCT in early childhood remains an important research priority (Becker et al., 2016).

We are aware of only two studies that tested whether SCT and IN are dissociable constructs in young children (Lee, Burns, & Becker, 2017; Leopold et al., 2016). First, a cross-sectional study of 172 preschool children (ages 4-6 years), demonstrated that maternal ratings of SCT and IN were empirically distinct and differentially associated with functional outcomes (e.g., somatic complaints and sleep problems, respectively; Lee, Burns, & Becker, 2017). Second, a longitudinal study with 489 children (ages 5-15 years) also demonstrated support for the empirical differentiation of parent-rated SCT and IN behaviors (Leopold et al., 2016). The first goal of the current study was to conduct an evaluation of the structure of SCT and IN behaviors in a large community sample of children using multiinformant ratings that spanned early childhood through middle childhood. Consistent with evidence from samples of school-aged youth and the two studies that focused on young children (Lee et al., 2017; Leopold et al., 2016), we hypothesized that SCT behaviors would be correlated with, yet separate from, IN behaviors.

Most research on SCT has used cross-sectional designs (Becker et al., 2016). Longitudinal designs are required to test developmental questions, such as whether a common set of items measure the constructs of SCT and IN in a similar way across time. This is accomplished by testing for longitudinal measurement invariance (Widaman, Ferrer, & Conger, 2010). To date, only one study has examined the longitudinal measurement invariance of SCT and IN (Leopold et al., 2016). The study included six time points spanning from preschool through ninth grade. Parents viewed SCT and IN behaviors as dissociable constructs, and a common set of items provided consistent measurement of both constructs across time (Leopold et al., 2016). The second goal of the current study was to test for longitudinal measurement invariance of parent- and teacher-rated SCT and IN behaviors in a larger sample from age 3 years through fifth grade. We hypothesized that a common set of items obtained across informants and occasions would demonstrate invariant measurement of SCT and IN constructs. We were interested in testing for longitudinal measurement invariance because it is a precondition for making meaningful inferences about the stability and change in SCT and IN behaviors across time.

If longitudinal measurement invariance is established, corollary questions related to the stability and change in IN and SCT behaviors can be addressed. Whereas stability refers to the extent to which individuals maintain similar rank ordering of levels of behavior over time, change refers to the extent to which mean level changes in SCT and IN behaviors are evident across time. The stability of IN behaviors is well-established with high test-retest correlations (r = .78-.82) across periods of less than 1 year, as well over 1 to 5 years (r = .64), demonstrating that the rank order of individuals in the population remains stable across childhood (Willcutt et al., 2012). Much less is known about the stability of SCT behaviors, especially from early into middle childhood. Most studies that have investigated the stability of SCT behaviors have been limited to school-aged children across 1- to 2-year periods (Bernad, Servera, Becker, & Burns, 2016; Bernad, Servera, Grases, Collado, & Burns, 2014; Servera, Bernad, Carrillo, Collado, & Burns, 2016) that have demonstrated that SCT also remains stable with high test–retest correlations (r = .60-.76) from first to third grade. Leopold et al. (2016) found evidence for the overall stability of parent-rated SCT and IN behaviors across preschool to ninth grade. Using linear comparisons, Leopold et al. (2016) also demonstrated that mean levels of IN during this time span remained generally unchanged (d < .17), while there was a modest but significant increase in mean levels of SCT over time, most notably from second to fourth grades and again from fourth to ninth grades (d = .18-.31). This finding is consistent with the large body of literature on ADHD across the life span that has demonstrated that mean levels of IN demonstrate minimal change, particularly during the early school years (e.g., Lahey & Willcutt, 2010). Another objective of the current study is to investigate stability and change in parent and teacher rated IN and SCT behaviors across time. Consistent with the large body of research demonstrating the stability and minimal increases in IN across childhood (e.g., Lahey & Willcutt, 2010; Leopold et al., 2016; Willcutt et al., 2012), we hypothesized that IN behaviors would demonstrate high stability (i.e., individuals maintain similar rank ordering of levels of IN behavior over time) and minimal mean levels changes across childhood. Given the lack of longitudinal evidence examining SCT particularly in early childhood, we did not make a priori hypotheses regarding the stability and mean levels of change in SCT.

No longitudinal study has examined specific demographic characteristics as predictors of level and/or rates of change in SCT and IN in a sample of youth starting in early childhood. Thus, it is unknown whether certain children are at heightened risk for the development of SCT or IN behaviors (e.g., demographic characteristics, family factors). Evidence from primarily cross-sectional samples of school-aged youth suggest that SCT and IN have different and unique associations with key demographic and family characteristics including race/ethnicity, sex, parental education, and family income (Barkley, 2013; Garner, Marceaux, Mrug, Patterson, & Hodgens, 2010). Based on limited previous research with samples of older children (Becker et al., 2016), we tentatively hypothesized that socioeconomic variables (lower parent education and family income) and male sex would be associated with both IN and SCT, and African American race would be unassociated with SCT.

In sum, this study tested the factor structure, longitudinal measurement invariance, stability and change, and predictors of change in parent-rated (four waves from age 3 to first grade) and teacher-rated (six waves from pre-K to fifth grade) SCT and IN behaviors in a community sample of 1,173 children. Building on previous studies, we hypothesized that SCT and IN behaviors would be dissociable but correlated constructs across early and middle childhood and that SCT and IN items would exhibit longitudinal measurement invariance. In addition, we hypothesized that IN would remain stable across development and that children would exhibit meaningful individual differences in level and increases in IN and SCT. Given the absence of directly relevant previous studies, we adopted a largely exploratory approach for determining the functional form of change in SCT and IN as well as which demographic and/or socioeconomic factors were associated with that change.

Method

Participants and Procedures

The Family Life Project was designed to study young children and their families who lived in two of the four major geographical areas of the United States, east of the Mississippi River, with high poverty rates (Dill, 2001). Specifically, three counties were in Eastern North Carolina and three counties in Central Pennsylvania (PA) were selected to be indicative of the Black South and Appalachia, respectively. The Family Life Project adopted a developmental epidemiological design in which complex sampling procedures were employed to recruit a representative sample of 1,292 children whose families resided in one of the six counties at the time of the child’s birth. Low-income families in both states and, in NC, African American families were oversampled; however, using weighted analyses, all inferences generalize back to the six-county study area as if participants were selected using simple random sampling. The primary caregiver completed the ratings in the present study, and biological mothers accounted for 96% to 99% of the primary caregivers at each age. Detailed information on the study design and sampling plan were detailed elsewhere (Vernon-Feagans et al., 2013).

Of the 1,292 children whose families were enrolled in the main study, 1,173 were included in this current study based on having parent- and/or teacher-rated IN behavior data from at least one of the 10 available waves. Participating children and families (N = 1173) did not differ from nonparticipating children and families (N = 119) with respect to recruitment state (40% vs. 43% from Pennsylvania, p = .53), recruitment into the poor strata (78% vs. 75% poor, p = .45), child race (43% vs. 36% African American, p = .14), child sex (50% vs. 55% male, p = .29), or primary caregiver education (80% vs. 78% had high school degree, p = .56).

Following hospital screening, participants who were selected and agreed to participate were formally enrolled into the study by way of completion of a home visit when the target child was approximately 2 months old. Participating families were invited to participate in seven additional home visits when their child was 6, 15, 24, 36, 48, and 60 months old, and when their child was in first grade. At each visit, parents and children completed a variety of standardized tasks, observational procedures, interviews, and questionnaires. Teacher ratings were collected in the spring of the preschool, kindergarten through third grade, and fifth grade years.

Measures

Inattention and Sluggish Cognitive Tempo Behaviors.

Following precedent (e.g., Pelham, Evans, Gnagy, & Greenslade, 1992), the nine DSM-IV (APA, 1994) inattentive behaviors were rated by parents and teachers using a 4-point Likert-type scale (0 = not at all, 1 = just a little, 2 = pretty much, 3 = very much). Four additional items that represented SCT behaviors were also included (i.e., daydreams; is sluggish, slow moving or lacks energy; is apathetic or unmotivated; seems in a “fog”). The selection of SCT items was inspired by the literature at the time in which data collection decisions were made (Carlson & Mann, 2002; McBurnett, Pfiffner, & Frick, 2001; Todd, Rasmussen, Wood, Levy, & Hay, 2004). The current study focuses on 13 items (i.e., the nine DSM ADHD IN items, plus the four SCT items) that were completed by parents (at the age 3-year, 4-year, 5-year, and first-grade home visits) and/or teachers (at the kindergarten, first, second, third, and fifth grade school assessments). The confounding of informant and child age is a consequence of testing our questions in a study that was not designed specifically to investigate questions related to ADHD/SCT onset or persistence. Alphas for the nine inattention items (ranges .891-.955) and four SCT items (ranges .715-.884) are provided for each wave and informant in Table 1. For the parent ratings, of the 1,173 sampled children, 10 (0.9%) did not have a parent rating at any time point, 48 (4.1%) had a rating at 1 time point, 49 (4.2%) had a rating at 2 time points, 96 (8.2%) had a rating at 3 time points, and 970 (82.7%) had a parent rating at all time points. For teacher ratings, 85 (7.2%) had a rating at 1 time point, 25 (2.1%) had a rating at 2 time points, 41 (3.5%) had a rating at 3 time points, 83 (7.1%) had a rating at 4 time points, 208 (17.7%) had a rating at 5 time points, 366 (31.2%) had a rating at 6 time points, and 365 (31.1%) had a teacher rating at all time points.

Table 1.

Bivariate Correlations, Means, and Standard Deviations Between Inattention and Sluggish Cognitive Tempo.

	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16	17	18	19	20
1. PR-IN-Age 3	—	.633**	.554**	.471**	.521**	.344**	.320**	.260**	.295**	.254**	.213**	.232**	.220**	.277**	.133**	.166**	.110**	.151**	.183**	.214**
2. PR-IN-Age 4		—	.666**	.524**	.394**	.48**	.372**	.343**	.293**	.309**	.293**	.309**	.274**	.302**	.131**	.206**	.148**	.192**	.212**	.239**
3. PR-IN-Age 5			—	.528**	.365**	.382**	.592**	.314**	.292**	.269**	.226**	.270**	.246**	.273**	.160**	.175**	.083*	.175**	.170**	.255**
4. PR-IN-1st				—	.259**	.268**	.337**	.603**	.367**	.392**	.381**	.407**	.334**	.360**	.163**	.294**	.258**	.271**	.229**	.310**
5. PR-SCT-Age 3					—	.462**	.414**	.277**	.158**	.130**	.125**	.136**	.120**	.142**	.105**	.088**	.075*	.107**	.122**	.148**
6. PR-SCT-Age 4						—	.452**	.337**	.130**	.106**	.123**	.118**	.149**	.131**	.133**	.081*	.098**	.098**	.134**	.133**
7. PR-SCT-Age 5							—	.414**	.163**	.133**	.067*	.158**	.183**	.187**	.118**	.101**	0.025	.115**	.142**	.191**
8. PR-SCT-1st								—	.216**	.233**	.202**	.214**	.184**	.202**	.134**	.228**	.217**	.180**	.183**	.218**
9. TR-IN-PreK									—	.520**	.422**	.411**	.380**	.370**	.471**	.347**	.222**	.216**	.186**	.283**
10. TR-IN-K										—	.542**	.534**	.429**	.415**	.278**	.689**	.342**	.359**	.276**	.339**
11. TR-IN-1st											—	.664**	.531**	.463**	.255**	.360**	.700**	.483**	.382**	.374**
12. TR-IN-2nd												—	.620**	.459**	.214**	.319**	.400**	.720**	.407**	.358**
13. TR-IN-3rd													—	.499**	.203**	.221**	.323**	.449**	.725**	.392**
14. TR-IN-5th														—	.153**	.255**	.277**	.340**	.357**	.794**
15. TR-SCT-PreK															—	.300**	.303**	.259**	.154**	.249**
16. TR-SCT-K																—	.335**	.307**	.195**	.289**
17. TR-SCT-1st																	—	.449**	.344**	.323**
18. TR-SCT-2nd																		—	.412**	.348**
19. TR-SCT-3rd																			—	.326**
20. TR-SCT-5th																				—
Ms	.920	.827	.733	.732	.316	.282	.308	.364	.567	.736	.785	.808	.808	.759	.378	.470	.510	.552	.558	.582
SDs	.609	.586	.585	.619	.481	.398	.434	.477	.635	.806	.845	.830	.790	.787	.550	.681	.723	.711	.705	.744
Alphas	.891	.812	.895	.715	.905	.751	.907	.783	.933	.850	.952	.872	.955	.882	.954	.878	.946	.866	.951	.884
N	1096	1062	1069	1087	816	984	929	928	848	703	1096	1062	1066	1086	816	984	929	928	848	703

Note. Mean values for behavior severity items range from 0 to 3. Boldfaced values represent bivariate associations across inattention and sluggish cognitive tempo behavior severity within time. PR = parent rated; TR = teacher rated; IN = inattention behaviors; SCT = sluggish cognitive tempo; PreK = preschool; K = kindergarten; M = mean; SD = standard deviation. 1st-5th = grade years.

^*p < .05.

^**p < .01.

Risk Factors.

Biological parental history of ADHD, low parental educational attainment, and household poverty were examined as potential predictors of SCT and IN.

Biological parental history of ADHD.

A single item was asked to establish whether either biological parent of the target child had a childhood history of ADHD (i.e., “Has a doctor or other medical professional ever told you [him/her] that you [she/he] have [has] Attention Deficit Disorder?”). When the respondent was the biological mother of the target child, she answered the question about herself and the child’s biological father. Likewise, when the respondent was the biological father of the target child, he answered the question about himself and the child’s biological mother. When the respondent was not a biological parent of the target child, she/he answered the question with reference to the child’s biological parents. Mother and father history were combined into a single index to accommodate the low base rates of these variables when considered separately.

Low educational attainment.

As part of the 2-month interview, primary caregivers self-reported their educational attainment. A dummy variable was used to index those who had not completed a high school degree or general education diploma at the time of study recruitment.

Household poverty.

Following the approach of Hanson, McLanahan, and Thomson (1997), household income was tallied for anyone who resided in the household, not just those related by blood, marriage, or adoption. Individuals were considered co-residents if they spent three or more nights per week in the household. Using this information, the total annual household income was divided by the yearly federal poverty threshold for a family of that size and composition to create an income/needs ratio. Given the highly stable nature of household income in this sample (Vernon-Feagans et al., 2013), the average income/needs ratio across the 6-, 15-, 24-, and 36-month assessments was used to index household poverty.

Analytic Strategy

This study tested four research questions. The first question tested whether IN and SCT behaviors were best represented by one versus two factors across multiple parent and teacher assessments that spanned age 3 years through Grade 5. Confirmatory factor analysis (CFA) models were fit to IN and SCT items using a robust weighted least squares estimator, and likelihood ratio tests (LRTs) were used to determine which factor structure best represents the observed data.

The second question tested whether IN and SCT items exhibited longitudinal measurement invariance, separately within informant. Given the categorical (ordinal) nature of items, we followed the approach described by Liu et al. (2017), which included estimating and comparing models that imposed configural (i.e., same pattern of factor loadings across time), metric (i.e., equal factor loadings across time), and scalar (i.e., equal factor loadings and thresholds) invariance. The fit of these increasingly restrictive models was assessed using both LRTs and comparative fit index (CFI) difference tests. The LRTs are reported because they represent standard practice. However, given the multiple assessments and large N in the present study and evidence that chi-square LRTs have excessive statistical power in large samples (MacCallum, 1990), we emphasized fit statistics to inform model comparisons. Specifically, models in which CFI differences were ≤.01 (Chen, 2007; Cheung & Rensvold, 2002) were considered to provide an equally good fit to the data. The results of longitudinal CFA models also provided estimates of the stability of SCT and IN behaviors across time.

The third question tested mean level change in SCT and IN behaviors across time. Initially, a series of three unconditional growth models were fit to SCT and IN behaviors separately for parent- and teacher-rated behavior (i.e., 2 behaviors × 2 informants × 3 models = 12 models). These models—which implied no change (intercept only), linear change (intercept and linear slope), or nonlinear change (intercept and freed-loading slope)—were used to establish the optimal functional form of change. For the freed loading model, the first wave was fixed to 0 and the last time point was fixed to 1, with all intermediate waves freely estimated to accommodate potential nonlinear change (see, e.g., McArdle, 2009). Parent models were parameterized such that the age 3 assessment was the intercept. Teacher models were parameterized such that the preschool assessment was the intercept. Chi-square difference tests, model fit statistics, and parameter estimates were used to determine the best fitting model for each combination of behavior (IN, SCT) and informant (parents, teachers). Building on the results of the best fitting univariate growth models, a series of multivariate growth curve models were estimated. The multivariate growth models provided a comprehensive representation of change in parent and teacher rated IN and SCT behaviors, including how growth parameters were related to each other.

The fourth question tested whether risk factors predicted individual differences in level and/or rates of change in parent and teacher rated SCT and IN behaviors. This question was addressed by extending the unconditional multivariate growth curve model to include risk factors as time invariant predictors of growth parameters.

Except for descriptive statistics, which were computed using SPSS, all models were estimated in Mplus Version 8.2 (Muthén & Muthén, 1998-2018) and accommodated the complex sampling design (stratification and individual probability weights). Full-information maximum likelihood also accommodated missing data (i.e., all available data from parent and teacher symptom ratings for each child were included) that is considered a statistical best practice (Schafer & Graham, 2002). For all CFA models, items were treated as ordered categorical manifest variables using a robust weighted least-squares estimator (WLSMV). For growth models, IN and SCT mean scores were treated as continuous manifest variables using a robust full information maximum likelihood estimator (MLR). All likelihood ratio tests that were used for model comparisons made appropriate adjustments given the estimators (Satorra & Bentler, 2001). Model fit was based on the likelihood ratio chi-square test, as well as the CFI, the Tucker-Lewis Index (TLI), and the root mean square error of approximation (RMSEA), with values of CFI and TLI > .95 and RMSEA < .05 being indicative of good fit (Hu & Bentler, 1999).

Results

To provide an initial understanding of the observed data, parent- and teacher-rated SCT and IN items were combined into mean scores. Descriptive statistics and bivariate correlations are summarized in Table 1. In general, mean ratings of IN were consistently higher than ratings of SCT for both parent and teacher informants. No prominent changes in mean ratings of either IN or SCT were evident for parents or teachers. Mean levels of IN were minimal for both parents and teachers (ds < .10). Mean level increases in SCT were minimal for parents (age 3 years to first grade, d = .10) and moderately increased for teachers (preK to fifth grade, d = .32). Within measurement occasions (denoted in boldface in Table 1), IN and SCT behaviors were moderately to strongly correlated for both parents (rs = .48-.60) and teachers (rs = .4-7.79). Across measurement occasions, IN appeared more stable (i.e., individuals seemed to retain their relative position across time) than SCT (rs = .47-.67 vs. rs = .28-46 for parents; rs = .37-.66 vs. rs = .15-.45 for teachers).

Are SCT and IN Dissociable Constructs in Early and Middle Childhood?

The first research question was to test whether the two-factor or one-factor model provides the best fit for the data. A synopsis of model fit is provided in Table 2. The chi-square difference tests (all ps < .001) and global fit statistics indicate that the two-factor model consistently fit better than the one-factor model, and the two-factor model distinguishing IN from SCT was deemed to provide the best fit for the observed data at all waves from age 3 through fifth grade (RMSEAs ranged from .06 to .08 and CFIs ranged from .98 to .99). The estimated latent correlations between factors in the two-factor model were moderately high and fairly consistent over time (φs = .65 to .75 and φs = .63 to .87 for parent and teacher ratings, respectively). Standardized factor loadings across waves and informants are presented in the Supplemental Materials (Table S1).¹

Table 2.

Confirmatory Factor Models of ADHD Inattentive Behaviors and Sluggish Cognitive Tempo Behaviors From Age 3 Years to Fifth Grade per Parent and Teacher Reports.

Informant	Wave	Factors	χ2	df	Δχ2	CFI	TLI	RMSEA [90% CI]
Parent	Age 3	2	346.126***	64	—	.975	.970	.063 [057, .070]
		1	1037.701***	65	182.090***	.915	.898	.117 [.111, .123]
	Age 4	2	300.420***	64	—	.980	.975	.059 [.052, .066]
		1	923.630***	65	155.365***	.927	.912	.112[.105, .118]
	Age 5	2	420.506***	64	—	.976	.971	.072 [.066, .079]
		1	938.336***	65	146.434***	.942	.931	.112 [.106, .119]
	1st Grade	2	308.949***	64	—	.983	.979	.059 [.053, .066]
		1	780.629***	65	129.921***	.950	.940	.101 [.094, .107]
Teacher	PreK	2	292.079***	64	—	.987	.984	.066 [.059, .074]
		1	1147.166***	65	167.564***	.938	.926	.143 [.136, .150]
	K	2	470.780***	64	—	.990	.987	.080 [.074, .087]
		1	974.867***	65	136.443***	.977	.972	.119 [.113, .126]
	1st Grade	2	392.359***	64	—	.993	.991	.074 [.067, .081]
		1	828.273***	65	120.858***	.983	.980	.112 [.106, .119]
	2nd Grade	2	402.300***	64	—	.992	.990	.075 [.069, .083]
		1	862.166***	65	125.153***	.981	.977	.115 [.108, .122]
	3rd Grade	2	384.593***	64	—	.989	.986	.077 [.070, .084]
		1	661.664***	65	103.372***	.979	.975	.104 [.097, .111]
	5th Grade	2	332.911***	64	—	.989	.987	.077 [.069, .086]
		1	524.800***	65	71.448***	.981	.978	.100 [.092, .108]

Note. PreK = preschool; K = kindergarten; 1st-5th = grade years; ADHD = attention-deficit/hyperactivity disorder; CFI = comparative fit index; TLI = Tucker–Lewis index; RMSEA = root mean square error of approximation; CI = confidence interval.

^*p < .05.

^**p < .01.

^***p < .001.

Do SCT and IN Items Exhibit Longitudinal Measurement Invariance?

The second research question was to establish whether the two-factor CFA fit the data equally well across time (parents: age 3 years to first grade; teachers: preschool to fifth grade). Three increasingly restrictive longitudinal CFA models (configural, metric, scalar) were fit to each combination of informant and behavior (see Table 3).

Table 3.

Longitudinal Invariance Testing Within Rater Across Time (N = 1,173).

	Inattention							SCT
	χ2	df	Δχ2	CFI	ΔCFI	TLI	RMSEA	χ2	df	Δχ2	CFI	ΔCFI	TLI	RMSEA
Parent
Configural	519.494***	294	—	.991	—	.989	.026	83.827***	39	—	.995	—	.992	.032
Metric	564.602***	310	52.027***	.990	.001	.989	.027	93.001***	45	9.967ns	.995	.000	.992	.030
Scalar	743.629***	344	211.883***	.984	.006	.984	.032	110.613***	59	19.075ns	.994	.001	.994	.027
Teacher
Configural	549.776***	294	—	.996	—	.995	.029	87.504***	39	—	.997	—	.994	.034
Metric	576.653***	310	34.171**	.996	.000	.995	.028	91.093***	45	3.591ns	.997	.000	.995	.031
Scalar	616.904***	344	45.388ns	.996	.000	.996	.027	109.293***	59	20.012ns	.997	.000	.996	.028

Note. ns = not significant; CFI = comparative fit index; TLI = Tucker–Lewis index; RMSEA = root mean square error of approximation; SCT = sluggish cognitive tempo. Configural invariance assumes the same pattern of factor loadings across times. Metric invariance assumes equal factor loadings across time. Scalar invariance assumes equal factor loadings and thresholds across time.

^*p < .05.

^**p < .01.

^***p < .001.

Parent IN.

The baseline (configural) model fit the data well. Although imposing equality constraints on factor loadings across time (i.e., metric invariance) resulted in significantly worse model fit (p < .001), the model fit indices were largely unchanged (ΔCFI = .001) and differences in factor loadings were negligible across configural and metric models. Similarly, although the imposition of constraints on item thresholds across time (i.e., scalar invariance) resulted in statistically worse fit (p < .001), model fit indices (ΔCFI = .006) and comparisons of parameter estimates indicated no appreciable change.

Teacher IN.

The configural model fit the data well. Although the metric invariance model resulted in significantly worse model fit (p = .005), the model fit indices were largely unchanged (ΔCFI < .001) and examination of the factor loadings did not reveal any noteworthy differences. The scalar invariance model did not significantly differ from the metric model (p = .092), the fit indices were unchanged (ΔCFI < .001), and parameter estimates were comparable.

Parent SCT.

The configural model fit the data well. The metric invariant model did not differ from the configural model (p = .126; ΔCFI < .001). Moreover, the scalar invariant model did not differ from the metric model (p = .162; ΔCFI = .001).

Teacher SCT.

The configural model fit the data well. The metric invariant model did not differ from the configural model (p = .732; ΔCFI < .001), and the scalar invariant model did not differ from the metric model (p = .130; ΔCFI = .001)

What Is the Stability of SCT and IN Across Time?

The third research question asked how stable parent and teacher rated SCT and IN behaviors were across time. In contrast to descriptive statistics that were presented in Table 1, here we report latent correlations of each construct across time, using results from the scalar invariant models that were established above. The estimated latent correlations within parent-rated IN demonstrated that IN behaviors were highly stable from age 3 years to Grade 1 (φs = .55-.64). Similarly, teacher-rated IN was highly stable from pre-K to Grade 5 (φs = .62-.74). SCT behaviors were moderately to highly stable for both parent-rated SCT (φs = .37-.57) and teacher-rated SCT (φs = .47-.62).

What Are the Patterns of Individual Change in SCT and IN?

Preliminary unconditional univariate growth models were estimated to determine the best functional form of change (i.e., no change, linear change, nonlinear/freed-loading change) in IN and SCT behaviors across time. As is summarized in Table 4, the nonlinear models consistently provided the best fit to both parent and teacher rated IN and SCT behaviors. This was evident from global model fit statistics, as well as formal model comparisons. Thus, we used the freed-loading parameterization to represent both parent- and teacher-rated IN and SCT trajectories in subsequent multivariate growth curve models.

Table 4.

Evaluating the Functional Form of Unconditional Univariate Growth Models of ADHD Inattentive Behaviors and Sluggish Cognitive Tempo Behaviors Within Rater Across Time.

	Model fit indices
Model	χ2	df	Δχ2	CFI	TLI	RMSEA
Inattention
Parent
Intercept only	105.934	8	—	.903	.927	.103
Intercept + Linear slope	25.391	5	78.952***	.980	.976	.059
Intercept + Freed-loading slope	14.319	3	10.817**	.989	.978	.057
Teacher
Intercept only	225.856	19	—	.815	.854	.100
Intercept + Linear slope	139.598	16	84.961***	.890	.897	.084
Intercept + Freed-loading slope	40.292	12	68.198***	.975	.968	.047
SCT
Parent
Intercept only	29.024	8	—	.911	.934	.048
Intercept + Linear slope	16.079	5	12.152**	.953	.944	.044
Intercept + Freed-loading slope	2.495 (ns)	3	24.012***	1.000	1.004	.000
Teacher
Intercept only	139.706	19	—	.778	.825	.076
Intercept + Linear slope	61.090	16	74.331***	.917	.922	.051
Intercept + Freed-loading slope	33.693	12	21.070***	.960	.950	.041

Note. ADHD = attention-deficit/hyperactivity disorder; CFI = comparative fit index; TLI = Tucker–Lewis index; RMSEA = root mean square error of approximation; SCT = sluggish cognitive tempo. Parent models included age 3, 4, 5 years and first grade ratings; teacher models included preK, K, 1st-, 2nd-, 3rd-, and 5th-grade ratings.

^*p < .05.

^**p < .01.

^***p < .001.

An unconditional multivariate (IN, SCT) multi-informant (parent, teacher) growth curve model fit the data well, χ² (170) = 443.46, p < .001, CFI = .96, TLI = .95, RMSEA = .03, confidence interval [CI; .03, .04]. Supplemental Figure S1 depicts model-implied mean-level changes in parent and teacher-rated SCT and IN behaviors from age 3 years through Grade 5.

For parent-rated IN, the mean (μ = .881, p < .001) and variance (φ = .220, p < .001) of the intercept term were statistically significant that indicated individual differences in IN behaviors at age 3. The mean of the slope term was significant (μ = −.169, p < .001) but the variance was not (φ = .039, p = .054), which indicated that all children exhibited modestly decreasing levels of parent-rated IN behaviors from age 3 years to Grade 1. The pattern of estimated freed loadings (see Supplemental Materials, Table S2) indicated that 54% of the decrease in parent-rated IN occurred between age 3 and 4 years.

For parent-rated SCT, the mean (μ = .276, p < .001) and variance (φ = .080, p < .001) of the intercept term were also statistically significant that indicated individual differences in SCT behaviors at age 3 years. The mean of the slope term was significant (μ = .083, p < .001) but the variance was not (φ = .129, p = .148), which indicated that all children exhibited modestly increasing levels of parent-rated SCT behaviors from age 3 years to Grade 1. However, a closer inspection of the freed loading estimates (Supplemental Table S2) indicated a complex pattern of change, such that parents reported modestly decreasing levels of SCT behaviors from age 3 through 5 years followed by an overall increase at Grade 1.

For teacher-rated IN, the mean (μ = .538, p < .001) and variance (φ = .299, p < .001) of the intercept term were statistically significant that indicated individual differences in IN behaviors at the preK assessment. The mean of the slope term was significant (μ = .154, p < .001), but the variance was not (φ = .180, p = .110) that indicated that all children exhibited modestly increasing level of teacher-rated IN behaviors from preK through Grade 5. The pattern of estimated freed loadings (see Supplemental Table S2) indicated that that 70% of the increase occurred between preschool and kindergarten assessments.

For teacher-rated SCT, the mean (μ = .378 p < .001) and variance (φ = .118, p < .001) of the intercept parameter were statistically significant that indicated individual differences in SCT behaviors at the Pre-K assessment. The mean of the slope term was significant (μ = 218, p < .001), but the variance was not (φ = .039, p = .363) that indicated that all children exhibited modestly increasing level of teacher-rated SCT behaviors from preK through Grade 5. The pattern of estimated freed loadings (see Supplemental Table S2) indicated that 40% of the increase in teacher-rated SCT occurred between preschool and kindergarten assessments.

Because of the lack of significant variation in slope terms, questions about whether patterns of change across informants and behaviors were not testable. However, initial levels of parent and teacher-rated IN and SCT were related to each other. Specifically, initial level of parent-rated SCT was significantly associated with teacher-rated SCT (φ = .42, p < .001) and as well as significantly correlated with parent- and teacher-rated IN (φ = .73, p < .001 and φ = .38, p <.001, respectively). Higher initial levels of parent-rated IN were also associated with higher initial levels of teacher-rated SCT (φ = .56, p < .001). Furthermore, higher levels of teacher-rated SCT were highly associated with higher levels of parent- (φ = .47, p <.001) and teacher-rated IN (φ =.83, p <.001).

What Demographic Predictors Are Associated With Growth Parameters?

Next, the growth curve model was extended to include associations between key demographic predictors. Specifically, the intercept parameters for IN and SCT were regressed on sex, race, parental education level, family income status, and parental ADHD (see Table 5). This conditional model fit the data well, χ² (266) = 606.571, p < .001, CFI = .958, TLI = .948, RMSEA = .033 [90% CI: .030, .037]. Parent education level exerted significant direct effects on both parent-rated (β = −.12, p = .02) and teacher-rated (β = −.23, p < .001) SCT, such that higher parent education was associated with lower SCT but not parent- or teacher-rated IN. Children who were African American were more likely to have higher levels of teacher-rated IN (β = .34, p < .001), but lower levels of teacher-rated SCT (β = −.20, p = .009); there was no association for race with parent-rated IN or SCT. Male children were more likely to have higher levels of teacher- but not parent-rated IN (β = .38, p < .001); there was no association of sex with SCT for either rater. Family income and parental ADHD did not have an effect on either SCT or IN. These results are consistent with bivariate correlations (Supplemental Table S3) and regressions (Supplemental Table S4) of family and demographic characteristics with parent- and teacher-rated behavior at age 3 through fifth grade.

Table 5.

Conditional Multivariate Model With Demographic Characteristics Predicting Intercepts.

	Model parameter (intercept)
	PR IN	PR SCT	TR IN	TR SCT
Demographic predictor	β (SE)	β (SE)	β (SE)	β (SE)
Male	−.025 (.069)	−.051 (.064)	.383 (.076)***	−.078 (.078)
African American	−.031 (.066)	.111 (.066)	.342 (.076)***	−.202 (.077)**
Parental Education	−.041 (.069)	−.124 (.056)*	−.073 (.062)	−.230 (.062)***
Income-to-Needs Ratio	−.053 (.065)	−.104 (.061)	−.089 (.070)	−.119 (.071)
Parental ADHD	−.014 (.073)	.110 (.080)	.099 (.109)	.038 (.115)

Note. ADHD = attention-deficit/hyperactivity disorder; SCT = sluggish cognitive tempo behavior severity; β = partial standardized regression coefficient; SE = standard error; PR = parent rated; TR = teacher rated; IN = inattention behavior severity; CFI = comparative fit index; TLI = Tucker-Lewis Index; RMSEA = root mean square error of approximation. Intercept parameters for IN and SCT were regressed on the demographic predictors. Income-to-needs ratio ranges from 0 to 15. Overall model fit: χ² (266) = 606.571, p < .001, CFI = .958, TLI = .948, RMSEA = .033 [90% CI: .030, .037].

Discussion

This is the first longitudinal investigation of both parent and teacher ratings SCT and IN behaviors across 10 time points in early and middle childhood. Specifically, this is the first study to examine multi-informant longitudinal measurement invariance and the functional form of developmental change and associations between SCT and IN across age 3 years through fifth grade. We demonstrated that SCT and IN are unique but correlated constructs that exhibited longitudinal measurement invariance for parents and teachers across time. Across development both parents and teachers viewed individual differences in SCT and IN behaviors as highly stable. Mean levels of SCT became more prominent between age 5 years and first grade and increased modestly with age, while mean rates of change in IN were smaller. Lower parental education was associated with higher parent- and teacher-reported SCT, male sex was associated with higher teacher-reported IN, and African American race was associated with higher teacher-reported IN but lower teacher-reported SCT. Overall, our findings add to a growing literature supporting the validity of SCT from IN across early childhood into middle childhood.

Dimensionality of SCT and IN From Early to Middle Childhood

As hypothesized, SCT and IN were strongly correlated although dissociable constructs across early and middle childhood. Confirmatory factor analyses strongly supported a two-factor model starting at age 3 years and this continued to provide the best fit for the data through fifth grade. Consistent with evidence from samples of school-aged youth, and the two other studies of preschool children (Lee et al., 2017; Leopold et al., 2016), SCT and IN behaviors were separate yet moderately to highly correlated and these associations were fairly consistent over time. The separation of SCT and IN has been repeatedly shown in cross-sectional studies (Becker et al., 2014; McBurnett et al., 2014; Willcutt et al., 2014) and in a small number of longitudinal studies (Leopold et al., 2016; Servera et al., 2016), and our study indicates that this separation occurs early in childhood. However, the discriminant validity between SCT and IN factors using teacher ratings was weak at some time points (correlations >.80; Brown, 2015; Molina, Servera, & Burns, 2019), which points to the need for additional studies examining the discriminant and convergent validity of SCT and IN behaviors when teacher ratings are used in early childhood.

Measurement Equivalence and Stability of SCT and IN

The SCT and IN constructs exhibited strong longitudinal measurement invariance across informants, which permitted consideration of mean-level changes in SCT and IN behaviors across time. Furthermore, the finding that loadings and thresholds of were statistically equivalent across measurement occasions provides further support for the construct validity of SCT as assessed by the four-item measured used in the present study. These findings are consistent with the Leopold et al. (2016) study, which demonstrated longitudinal measurement invariance of parent-rated SCT and IN in preschool through ninth grade. Longitudinal invariance is an important prerequisite for future research endeavors that aim to understand developmental pathways to SCT, mechanisms linking SCT to functional impairment, and potential moderators that may mitigate or exacerbate the risks of SCT for subsequent impairment. Consistent with the large body of research supporting the stability of IN in childhood, the present study found high latent correlations across time, within rater provided strong support for the stability of IN from preK to Grade 5. Similarly, SCT behaviors were highly stable, providing further evidence of stability in the extent to which individuals maintain similar rank ordering of levels of behavior over time.

Mean Level Change in Developmental Trajectories of SCT and IN Behaviors

Our results partially supported our hypotheses about the mean level change in SCT and IN in early development. As hypothesized, mean levels of change in SCT tended to increase gradually with age as rated by teachers, although somewhat earlier than anticipated. That is, the proportion of total change in parent-rated SCT was the greatest between preschool and kindergarten, rather than increasing after second grade as demonstrated by Leopold et al. (2016). However, in contrast to Leopold et al. (2016), in the present study, parent ratings of SCT demonstrated no significant changes across waves. As expected, and in line with previous studies of children with ADHD (e.g., Lahey & Willcutt, 2010; Willcutt et al., 2012), mean levels of change in IN behaviors were minimal, with the greatest proportion of change occurring early in development across both parent and teacher ratings. This is perhaps not particularly surprising and is likely related to the increase of structure and academic, socioemotional, and behavioral demands in kindergarten as compared with preschool. Thus, with increasing demands associated with starting kindergarten, the impact of a sluggish, daydreamy, and/or inattentive presentation becomes more evident. This is also consistent with a recent study that demonstrated that SCT and IN contained more trait than occasion-specific variance starting in kindergarten (Burns, Becker, Geiser, Leopold, & Willcutt, 2019). In the present study, this pattern was observed for teacher ratings, and not parent ratings, of SCT. It has been argued that teachers may be better reporters of SCT, and better able to distinguish between SCT and inattentive behaviors given the classroom context and ability to compare with other children, particularly in young children (Garner et al., 2010; McBurnett et al., 2001; Tamm, Brenner, Bamberger, & Becker, 2018).

Demographic Predictors of Growth Parameters

Our findings add to a growing body of research suggesting sociodemographic characteristics may be an important correlate of SCT (Barkley, 2013; Becker, 2014; Marshall, Evans, Eiraldi, Becker, & Power, 2014). Specifically, lower parental education was associated with higher SCT, although this association was not observed for income. Parental education, in particular, appears important to the expression of SCT, particularly as it predicted mean initial levels of SCT for both parent and teacher ratings. Maternal education has previously been shown to be a general risk factor for child psychopathology, and in particular, internalizing disorders (Velez, Johnson, & Cohen, 1989). However, in the present sample, family income did not predict SCT, which was somewhat surprising as parental education and income are often highly correlated. Although highly inter-correlated, specific components of socioeconomic status (e.g., parent education, income, employment status, free or reduced lunch status) may independently influence child outcomes through unique underlying mechanisms (Duncan & Magnuson, 2012). For example, more educated parents may be more likely to focus on promoting children’s cognitive development, and more educated parents may themselves be more organized in their daily routines allowing them to more effectively achieve parenting goals (Duncan & Magnuson, 2012). Of note, twin studies have reported nonshared environmental factors influence the presence of SCT more than the presence of ADHD symptomatology (Moruzzi, Rijsdijk, & Battaglia, 2014). Further research is warranted to explore the nuances of the association between socioeconomic variables, family-level factors, and SCT.

Consistent with the literature, teacher-rated IN was associated with male sex (APA, 2013). However, in contrast to our hypotheses, male sex was not associated with SCT. The failure to find sex differences in SCT is consistent with multiple studies that do not report sex differences for SCT (Barkley, 2012; Becker & Langberg, 2013; Carlson & Mann, 2002; Jarrett, Rapport, Rondon, & Becker, 2014; Marshall et al., 2014; Servera et al., 2018). Notably, participants in these studies were all drawn from community or school-based samples with two exceptions (Marshall et al., 2014). Thus, sample characteristics may be important to understanding sex differences, though meta-analytic results found only modest (though statistically significant) support for an association between male sex and SCT (Becker et al., 2016). We did, however, observe that male sex was associated with IN. This is consistent with prevalence data indicated that ADHD is more frequent in males, typically a 3:1 ratio (APA, 2013). Many studies of SCT have been conducted in studies of individuals diagnosed with ADHD (e.g., Becker & Langberg, 2013; Garner et al., 2017; McBurnett et al., 2014) that may have biased the results of studies reporting sex differences in SCT.

In contrast to expectations, we found African American race to be associated with lower teacher-rated SCT, however African American race was positively associated with teacher-rated IN that is consistent with prior literature (Miller, Nigg, & Miller, 2009). The finding that African American race was associated with SCT stands in contrast to the majority of studies that have demonstrated no association between race/ethnicity and SCT (Becker et al., 2016), and the one study that showed non-White children had higher rates of SCT than White children (Becker et al., 2013). The divergence in findings may be related to informant differences; the majority of studies reporting that no race differences included parent ratings of SCT. Like those studies, our findings do not show race related differences for parent-rated SCT but instead show African American race to be associated with lower teacher-rated SCT. There has been a call for research to examine informant differences given that SCT may operate differently in the home versus school contexts (Becker et al., 2016), and this may extend that hypothesis to also include that SCT may operate differently for African American youth in the school setting. Alternatively, there may be biases in teacher expectations for behavior of African American than White children (Gershenson & Papageorge, 2018). Consistent with the current findings, there is literature showing that African American students may be rated as having more ADHD behaviors than White students, even after accounting for socioeconomic status (Miller et al., 2009). However, the interpretation of such findings is difficult given the small pool of empirical work in this domain, as well as the lack of inclusion of potential explanatory variables (e.g., the race of raters, racial identity and context, socioeconomic status, neighborhood context). Future research examining this complex issue is needed to better understand possible links between race and SCT behaviors.

Strengths and Limitations

Strengths of the current study include the use of a large, developmental epidemiological sample enriched for low income and African American families. This is the first study to demonstrate that SCT and IN can be distinguished in youth as young as 3 years old, earlier in development than any previous study. The same measures of SCT and IN were obtained across age 3 through fifth grade across both informants, which helped simplify interpretation. Despite these strengths, the current study also has several important limitations.

The primary limitation of this study is that the SCT measure is comprised of only four items. There is significant variability in the SCT literature with the number of items used to assess SCT (ranging from two to up to 44 items, with some items being more related to IN than others; Becker et al., 2016). The items used to assess SCT in the present study may not capture the full breadth of SCT symptomology (Becker et al., 2016). For instance, our particular measure of SCT may be lacking important factors of the SCT including appearing tired or lethargic, getting lost in one’s thoughts, being easily confused, and seeming to be in a world of one’s own. Decisions about item selection were made by the Family Life Project investigators in 2005, far earlier than current recommendations for assessing SCT existed (Becker et al., 2016). Nevertheless, as shown Supplemental Table S5, the four items used in the present study map onto items supported by meta-analytic findings as consistently loading on an SCT factor rather than an IN factor (Becker et al., 2016). However, in contrast to previous examinations of SCT items, the “daydreams” item did not consistently demonstrate convergent and discriminant validity (see Note 1 and Supplemental Table S6). Given the strong historical and empirical precedent for daydreaming to be a core item of the SCT construct (Becker et al., 2016), we retained this as an SCT item in our study. However, this may have contributed to SCT and IN demonstrating weak discriminant validity for teacher ratings in our study. Furthermore, although the “unmotivated or apathetic” item used in the present study was supported in the meta-analysis, more recent studies have called into question the use of a motivation/apathy item for assessing SCT as distinct from ADHD-IN (Becker et al., 2017; Sáez et al., 2018). It is interesting to consider whether the mixed evidence for the validity of the motivation/apathy item may be an artifact of the specific content of the item. The wording for this item in the recent studies that failed to find convergent validity and discriminant validity was “lacks motivation to complete tasks” (Becker et al., 2017; Sáez et al., 2018), whereas the present study used more general wording not specific to task-related motivation (i.e., “unmotivated or apathetic”). It is possible that excluding the “completes tasks” language in the motivation/apathy item provides better distinction from IN, especially given that sustained attention to tasks is a core symptom of ADHD and prior evidence indicates that SCT items assessing task effort and persistence have not shown discriminant validity from IN (Becker et al., 2016). Additional studies will be needed to determine the extent to which motivation/apathy is best conceptualized as an aspect of SCT. Despite these considerations, we have added confidence in our SCT items given they included the slowed behavior/thoughts, inconsistent alertness, and daydreaming aspects of SCT, and our findings largely replicate previous research (Becker et al., 2016). Nonetheless, additional studies with well-validated measures of SCT are needed to replicate and extend the present findings.

An additional limitation is that the waves at which the children were assessed were not the same for parents and teachers. The lack of overlapping assessments (except first grade) make it somewhat difficult to directly compare growth curve results. However, recent 1- and 2-year longitudinal studies of SCT and ADHD behaviors using parent and teacher report stability in SCT and IN that closely match those found in the present study, supporting the generalizability of our findings (Bernad et al., 2014, 2016). Nonetheless, future research that examines both parent and teacher ratings at the same time points across development would be a useful extension of the current results by better examining the degree of developmental change that is due to these constructs themselves relative to potential rater effects.

Conclusions

It is important to establish when in development SCT behaviors emerge and increase as this knowledge may inform the optimal period for assessing and intervening before behaviors increase. Using longitudinal parent and teacher ratings of SCT and IN behaviors spanning 10 time points in early and middle childhood, we demonstrate that the SCT construct can be reliably measured starting at age 3 years, that SCT is empirically distinct from IN across early childhood, and that SCT increases gradually across development. Furthermore, we demonstrate that SCT is differentially associated with sociodemographic factors, namely parent education level. This work is also essential for setting the groundwork for future studies examining etiological pathways (e.g., prenatal and perinatal factors, parenting, temperament) and developmental pathways and consequences (e.g., learning disorders). These findings support the assessment of SCT beginning in early childhood and future work should explore whether SCT is related to school readiness, preacademic skills, and early social-emotional functioning.