Children's Sleep and Externalizing Problems: A Day-to-day Multilevel Modeling Approach

Abstract

Background: Sleep problems and externalizing problems tend to be positively associated, but the direction of this association is unclear.

Method: Day-to-day associations between sleep and behavior were examined in children (N = 22) ages 3-8 with clinical levels of externalizing problems. These children were enrolled in Parent Management Training and behavioral sleep intervention. During assessments before and after treatment, children wore actigraphs for seven days and parents concurrently completed sleep diaries and daily tallies of noncompliance, aggression, and tantrums. Multilevel modeling was used to account for the nested structure of the data, at the day-to-day level (level 1), within assessment points (level 2), and within children (level 3).

Results: Late sleep timing and fragmentation were predictive of next-day noncompliance and tantrums, respectively. There were fewer associations for a given day's behavior predicting that night's sleep, although children who showed more aggression and noncompliance at baseline tended to have later bedtimes and sleep onset times compared to other children.

Trial registration: ClinicalTrials.gov NCT02783560.

Figure 1.

Intervention overview diagram. Squares represent the treatment phases. Circles represent the study procedures. Intake, or the initial assessment, occurred over two weeks, with one session happening each week, and eligible families were invited to consent to the study at the first intake session. Consented families received their baseline measures at the second intake session to complete at home in the week before the first treatment session, goal setting. Eligible, consented families were also randomly assigned to either the sleep first or the mealtime first condition. All families received the same treatment, with pertinent sleep and mealtime components. Randomization determined the order of intervention (i.e., mealtime first or sleep first) but not intervention content. The number of treatment sessions varied depending on how quickly families moved through the treatment material. The entire treatment protocol was expected to last for 14 weeks at a minimum, including one week for goal setting, a minimum of two weeks for sleep (or mealtime), a minimum of 8 weeks for PMT, a minimum of two weeks for mealtime (or sleep), and a final “graduation” session.

Figure 2.

Diagram of sample recruitment and retention.

Figure 3.

Sample Sleep Train. Sleep Trains were available in four to eight cars, depending on how many steps were in the child’s bedtime routine. After selecting a train, the family was invited to select cars that corresponded to the steps of their child’s bedtime routine. These steps were then attached to the train cars with Velcro. Six sample steps (bath, teeth brushing, potty, pajamas, prayer, and story time) are also depicted above. Gray-scaled pictures are presented here, but full-color copies were used in clinic.

Figure 4.

Linear regression lines of actigraphic sleep across days of data collection. Days 1–7 correspond to the Baseline measurement point, before treatment began. Days 8–14 correspond to the Final measurement point, after treatment was complete. Gray lines represent individual trajectories; overlaid black lines represent the average trajectory for the whole sample. SleepMin refers to the actigraphic variable indexing true sleep minutes. SleepTim refers to the actigraphic variable indexing the timing of sleep onset, using 24 hour time. NightWake refers to the actigraphic variable indexing the number of wakings that lasted five or more minutes. SOL refers to the number of minutes between parent-reported bedtime and actigraphically determined sleep onset time. The unconditional growth model, which incorporated the effect of time, significantly improved model fit for each sleep index, except for sleep minutes (χ²(1) = 12.2, 13.7, 3.9, p < .05 for sleep timing, night wakings, and sleep onset latency, respectively).

Figure 5.

Linear regression lines of parent-reported sleep across days of data collection. Days 1–7 correspond to the Baseline measurement point, before treatment began. Days 8–14 correspond to the Final measurement point, after treatment was complete. Gray lines represent individual trajectories; overlaid black lines represent the average trajectory for the whole sample. In addition to significant between-child differences in both intercepts and slopes, parent-reported bedtimes significantly declined (moved earlier) across days of data collection (beta = −132.15, p = .03 for the effect of time in the unconditional growth model, which had significant improvement in model fit over the unconditional means model (χ²(1) = 14.0, p = .00).

Figure 6.

Linear regression lines of parent-reported behavior problems across days of data collection. Days 1–7 correspond to the Baseline measurement point, before treatment began. Days 8–14 correspond to the Final measurement point, after treatment was complete. Gray lines represent individual trajectories; overlaid black lines represent the average trajectory for the whole sample. NC refers to noncompliance. Agg refers to aggression. In addition to significant between-child differences in both intercepts and slopes, each parent-reported behavior problem significantly declined across days of data collection (beta = −.30, −.13, −.12, p = .00 for the effect of time in the unconditional growth models for noncompliance, aggression, and tantrums, respectively, each of which had significant improvement in model fit over the unconditional means models (χ²(1) = 103.4, 46.6, and 80.8, p = .00 for each model).