Graham J. Reid PhD, The University of Western Ontario, Westminster Hall, Room 319E, London, ON N6G 2K3, Canada. Tel.: (519) 661-2111x84677; fax: (519) 661-3340; e-mail: email@example.com
The contribution of sleep problems to emotional and behavioral problems among young children within the context of known risk factors for psychopathology was examined. Data on 2- and 3-year-olds, representative of Canadian children without a chronic illness, from three cross-sectional cohorts of the Canadian National Longitudinal Study of Child and Youth were analysed (n = 2996, 2822, and 3050). The person most knowledgeable (PMK), usually the mother, provided information about her child, herself, and her family. Predictors included: child health status and temperament; parenting and PMK depressive symptomatology; family demographics (e.g., marital status, income) and functioning. Child sleep problems included night waking and bedtime resistance. Both internalizing/emotional (i.e., anxiety) and externalizing/behavioral problems (i.e., hyperactivity, aggression) were examined. Adjusting for other known risk factors, child sleep problems accounted for a small, but significant, independent proportion of the variance in internalizing and externalizing problems. Structural equation models examining the pathways linking risk factors to sleep problems and emotional and behavioral problems were a good fit of the data. Results were replicated on two additional cross-sectional samples. The relation between sleep problems and emotional and behavioral problems is independent of other commonly identified risk factors. Among young children, sleep problems are as strong a correlate of child emotional and behavioral problems as PMK depressive symptomatology, a well-established risk factor for child psychopathology. Adverse parenting and PMK symptomatology, along with difficult temperament all contribute to both sleep problems and emotional and behavioral problems. Children’s sleep problems appear to exacerbate emotional and behavioral problems.
Among young children, sleep and psychopathology (i.e., internalizing/emotional, externalizing/behavioral problems) have been found to relate to each other (Gregory and O’Connor, 2002; Lam et al., 2003; Owens-Stively et al., 1997; Paavonen et al., 2002). For example, Richman and colleagues (1982) found 29% of 3 year olds with behavior problems had sleep problems, compared with 11% of children without behavior problems. Similar results have been reported among older children (e.g., El-Sheikh et al., 2007; Goodnight et al., 2007; Paavonen et al., 2002). A major limitation of most previous studies is the failure to examine other factors that may account for the relationship between sleep and psychopathology. For example, parenting practices are well-established correlates and predictors of externalizing problems (e.g., oppositional defiant behavior problems) (Mesman and Koot, 2001). Parenting difficulties have been hypothesized as a key mechanism in the development of problems with young children getting to sleep, and waking in the night (Anders et al., 2000; Pollock, 1994), and one study found lax parenting related to more sleep problems (Owens-Stively et al., 1997). Thus, parenting may be a third variable accounting for the relation between sleep and emotional and behavioral problems. Similarly parental psychological adjustment and family dysfunction have been related to sleep problems (Seifer et al., 1996) and psychopathology (Lavigne et al., 1998; O’Leary et al., 1999) in young children. Other well-established risk factors for the development of psychopathology may also contribute to the relationship between sleep problems and psychopathology (Harland et al., 2002; Pevalin et al., 2003). We examined the contributions of children’s sleep problems to emotional and behavioral problems measured concurrently, after accounting for the effects of: (1) child characteristics (i.e., gender, health status and temperament); (2) demographic factors (i.e., parental age, educational attainment, family income and single parent status); (3) family and parent functioning (i.e., family functioning, parental depressive symptomatology); and (4) parenting. It has been hypothesized that sleep problems during early childhood may have lasting impacts (Dahl, 1996). Thus, the current study’s focus on children age 2-to-3 years old is particularly relevant.
A second limitation in the current literature is a lack of models for how risk factors contribute to sleep problems and emotional and behavioral problems. Three studies using structural equation modeling are notable exceptions (Bates et al., 2002; El-Sheikh et al., 2007; Goodnight et al., 2007). Bates et al. (2002) found higher family stress lead to sleep disruption (later and more variable sleep), which in turn contributed to behavior problems at school. Interestingly, better ‘family management’ (parenting and family functioning) related to fewer behavior problems, but not sleep disruption. Reasons for the lack of relationships among parenting, sleep and school adjustment are unclear. Use of an at-risk sample (i.e., low-income families) may have been a factor. Goodnight et al. (2007) found that at age 5 years, temperamentally resistant children had both more sleep and externalizing problems; however, sleep problems were not significantly related to externalizing problems. Among school-age children, El-Sheikh et al. (2007) recently found that marital conflict predicted children’s emotional insecurity, which predicted greater sleep problems and in turn parent- and teacher-rated psychopathology. Gregory et al. (2005) found family disorganization and maternal depression were significant correlates of both sleep and anxiety among 3- and 4-year-old twins. Pathways linking family disorganization and maternal depression were not examined.
We tested a model of proposed linkages relating demographic factors, child characteristics, family and parent functioning and parenting to sleep problems and emotional (internalizing) and behavioral (externalizing) problems (Figs 1 and 2 show the hypothesized and final models). We hypothesized that positive socio-demographic factors (i.e., older parental age, higher educational attainment, adequate family income, and not being a single parent) would be related to less family dysfunction (Landy and Tam, 1998). Family dysfunction was hypothesized to contribute to both poorer parenting and greater parental depressive symptomatology (Belsky et al., 1996; Cummings and Davies, 1994), which we expected to relate to sleep problems and in turn to higher levels of both internalizing and externalizing problems, along with directly relating to internalizing and externalizing problems. Children with poorer health status were expected to have more sleep problems. Finally, we expected children’s temperament would have direct effects on both sleep problems and internalizing and externalizing problems (Rothbart and Bates, 2006). Variables were structured to reflect a shift from distal to proximal effects both temporally and with respect to interactions with the child (Bronfenbrenner, 1986).
Data were from the Canadian National Longitudinal Survey of Children and Youth (NLSCY), a multi-wave national study that gathers information concerning physical, emotional, and social characteristics of children (Statistics Canada., 1997b). The aims of the NLSCY are to provide data on the risk factors and health of Canadian children and youth (from birth to adulthood) permitting an examination of both the prevalence of risk factors, and the impact of risk and protective factors on children’s development. (Further information on each cycle of the NLSCY can be found at http://www.statcan.ca/english/rdc/nlscy_cycle1.htm; http://www.statcan.ca/english/rdc/nlscy_cycle2.htm, etc.)
The NLSCY was developed by an expert advisory group. The survey comprised multiple scales. Most scales are either well-established measures (e.g., the McMaster Family Assessment Device; (Miller et al., 1985) or are short-forms derived from previous measures. Factor analyses conducted on split-half samples from the first wave of NLSCY support the validity of the scales used in the current analyses. The NLSCY sampling reflects the demographic profile of each province. Data were gathered through face-to-face or telephone interviews with the person most knowledgeable (PMK).
Participants and procedure
We analyzed data from children, age 2 or 3 years old, from Cycle 2, conducted in 1996–1997 (Statistics Canada 1999). Analyses were replicated in two independent samples of children age 2 or 3 years old from Cycle 3, conducted in 1998–1999 (Statistics Canada 2001), and Cycle 1, conducted in 1994–1995 (Statistics Canada 1997a)1. As some households had more than one eligible child who participated in the survey, we randomly selected one child from these households in our analysis. We excluded children with chronic illness, as the reasons underlying sleep problems among these children might differ from children without chronic illnesses, and child health problems are independently related to emotional and behavioral problems. Cases who were (a) missing data on one or more of the key outcome variables (i.e., sleep problems, emotional disorder-anxiety, separation anxiety, hyperactivity, and physical aggression) or (b) missing on at least three out of the nine predictors were also deleted.
From Cycle 2, 161 cases were deleted when there were two children per family and 602 children with chronic illness were excluded. Cases (n = 131) who were (a) missing data on one or more of the key outcome variables (i.e., sleep problems, emotional disorder-anxiety, separation anxiety, hyperactivity, and physical aggression), or (b) missing on at least three out of the nine predictors were deleted. One case missing a normalized weight value was also deleted, resulting in a final weighted sample size of 2996. Cases excluded due to missing data were more likely to (a) be from single parent families; (b) have inadequate family income; (c) greater family dysfunction; and lower levels of (d) sleep problems, (e) emotional problems, and (f) separation anxiety.
From Cycle 3, 48 cases were deleted when there were two children per family. Children with chronic illness (n = 581) and cases (n = 131) missing data for the key outcome variables or multiple predictors were deleted. The final weighted sample size was 2822. Excluded cases were more likely to (a) be from single parent families; (b) have poorer child health status; (c) have inadequate family income; (d) greater family dysfunction; (e) less rational parenting; and (f) lower levels of physical aggression and oppositional behavior.
From Cycle 1, 169 cases were deleted when there were two children per family. Children with chronic illness (n = 272) and cases (n = 347) missing data for the key outcome variables or multiple predictors were deleted. The final weighted sample size was 3050. Excluded cases were more likely to (a) be from single parent families; (b) have poorer child health status; and (c) have inadequate family income.
When some data were missing on a variable, mean substitution was used. NLSCY cross-sectional normalized weights were used to derive population estimates representative of the Canadian population, while maintaining the original sample size. Weighting is used when studies employ complex sampling procedures. It corrects for minor variations between the actual sample and the target population with respect to the sampling framework (the NLSCY sampling aimed to select households with children to represent all geographic areas of all provinces). Table 1 presents the sample characteristics after the normalized weight was applied.
Table 1. Demographic characteristics of study samples
n (%) or mean (±SD)
n (%) or mean (±SD)
n (%) or mean (±SD)
Percentages values are given in parenthesis. PMK, person most knowledgeable. Statistics Canada limits the disclosure of data for cell sizes < 5; some variables were regrouped to accommodate this restriction.
*Due to small cell sizes, for Cycle 1 on the relationship of PMK to child variable, other ‘unrelated female’ were included in this group.
Relationship to child
Step, adoptive, or foster mother, or unrelated female
Biological, adoptive, or step father
Other related adult (e.g., grandmother, aunt)
Less than high school
High school graduate
Beyond high school
College or university degree (including trade)
Persons in household including children
Single parent status
One parent only or does not live with a parent
Less than 15 000
15 000–29 999
30 000–49 999
50 000–79 999
More than 80 000
Table 2 presents the subscales and items that comprised the emotional/internalizing and behavioral/externalizing problems scales for 2- and 3-year-olds in the NLSCY. The subscale items were based on items measuring the same types of problems from the Ontario Child Health Study Scales (Boyle et al., 1993), Montreal Longitudinal Survey (Tremblay et al., 1994) and the Child Behavior Checklist (Achenbach, 1992).
Table 2. Summary of predictor variables and measures of internalizing and externalizing problems
10 items based on the Infant Characteristics Questionnaire (Bates et al., 1979) related to child’s emotional and behavioral tendencies. Rated on 7-point Likert-scale (α = 0.80, Cycle 2; α = 0.82, Cycle 3; α = 0.79, Cycle 1).
Average response across 10 items; higher scores reflect more negative emotional (e.g., gets upset easily, more negative response to new situations and people), and less adaptable and more persistent/unstoppable behavioral (e.g., persists when told not to play with object) tendencies.
1, 15–24 years; 2, 25–29 years; 3, 30–34 years; 4, 35–39 years; 5, over 40 years
Household income adequacy
Statistics Canada guidelines based on household income and number of persons in the household. (e.g., lowest = income ≤$10 000 and household size 1–4 persons or income ≤$15 000 and household size ≥ 5).
1, less than secondary; 2, secondary school; 3, beyond high school; 4, college or university graduate
PMK depressive symptomatology
12-item short-form of the Center for Epidemiologic Studies – depression scale (Radloff, 1977). Ratings of mood over the previous 2 weeks using a 4-point scale of 0 (rarely or none of the time) to 3 (most or all of the time) (Cycle 2 and 3, α = 0.82; Cycle 1, α = 0.81).
Total scores range from 0 to 36; higher scores reflect more depressive symptoms.
12-item version of the McMaster Family Assessment Device (Miller et al., 1985) measuring problem solving, communication, roles, affective involvement, affective responsiveness and behavioral control among family members. Ratings used a 4-point scale: 0 = strongly agree to 3 = strongly disagree (Cycle 2, α = 0.90; Cycle 3, α = 0.91; Cycle 1, α = 0.89).
Total family dysfunction scale; higher scores reflect greater family dysfunction.
(i) Seems to be unhappy, sad or depressed. (ii) Is not as happy as other children. (iii) Is too fearful or nervous. (iv) Is worried. (v) Is nervous, high strung or tense. (vi) Has trouble enjoying him/herself.
(b) 4-items from the separation anxiety scale (α = 0.59, Cycle 2; α = 0.56, Cycle 3; α = 0.67, Cycle 1); one item on child’s refusal to sleep alone was excluded
(i) Cries a lot. (ii) Clings to adults or is too dependent. (iii) Constantly seeks help. (iv) Gets upset when separated from parents.
(i) Can’t sit still, is restless/hyperactive. (ii) Is distractible, has trouble sticking to any activity. (iii) Fidgets. (iv) Can’t concentrate for long time. (v) Is impulsive, acts without thinking. (vi) Cannot settle to anything for more than a few moments. (vii) Is inattentive.
(i) Is defiant. (ii) Gets into many fights. (iii) Punishment doesn’t change his/her behaviour. (iv) Has temper tantrums or hot temper. (v) Has difficulty awaiting turn in games or groups. (vi) When another child accidentally hurts him/her (such as by bumping into him/her), assumes that the other child meant to do it, and then reacts with anger and fighting. (vii) Has angry moods. (viii) Kicks, bites, hits other children.
Four sleep problems were rated using a 5-point scale (1 = almost never; 5 = almost all of the time). Problems getting to sleep included (a) trouble falling asleep (‘When you put him/her to bed, how often does he/she have trouble falling asleep?’), (b) particular and long bedtime routine (‘Does he/she have a particular and long routine, more than 30 min, to go to bed [rocking, songs, nursery rhymes, etc.] that he/she cannot go to sleep without?’); problems staying asleep included (c) waking up several times at night (‘Does ... wake up several times during his/her sleep?’), and (d) ‘Does ... have a restless sleep?’. An overall sleep problems score was computed by averaging the four items; higher scores reflect more sleep problems (α = 0.62, Cycle 2; α = 0.63, Cycle 3; α = 0.54, Cycle 1).
We first conducted bivariate correlations between sleep problems and internalizing and externalizing problems. Then, hierarchical regressions tested the incremental validity of sleep problems in predicting internalizing and externalizing problems, above and beyond other risk factors (Babyak, 2004). Predictors were entered in five steps using forced entry. (1) child characteristic variables (i.e., gender, health status and temperament); (2) family demographics (i.e., PMK age, PMK education level, income adequacy and single parent status); (3) family and parent functioning (i.e., PMK depressive symptomatology, family dysfunction); (4) parenting; and (5) sleep problems. To evaluate effect sizes [see (NICHD Early Child Care Research Network 2006)] we examined: (a) the absolute effect sizes, based on Cohen (1988) (small, r ≤ 0.19; medium, 0.19 < r < 0.39, and large = r ≥ 0.40); and (b) the relative effect sizes, based on the strength of the relation of sleep problems with emotional and behavioral problems versus established predictors of childhood psychopathology using partial correlations (McCartney and Rosenthal, 2000).
Structural equation modeling
Structural equation modeling (SEM) can be used to determine the degree to which a hypothesized model fits observed data. An estimate of the fit of a structural model (i.e., relationships among constructs) is examined by testing whether the pattern of the hypothesized relations among latent constructs is consistent with an observed covariance structure. Fit indices indicate the degree of model fit; in cases where fit is less than optimal, model re-specification may be employed to increase fit. The interested reader might wish to consult Kline (2005) for a primer.
The hypothesized models (see Figs 1 and 2) were tested using the AMOS 5.0 program. In estimating the latent constructs, either item parcels or individual items were used as observed indicators (Hau and Marsh, 2006). Modification indices were examined and some paths in the hypothesized models were altered. Fit indices recommend by Hu et al. (1992) were used, with values falling above (or below) the cut-offs as reflective of a good fit to the data: adjusted goodness-of-fit (AGFI; >0.90); Tucker–Lewis index (TLI; ≥0.95); comparative fit index (CFI; ≥0.95); and root mean squared error of approximation (RMSEA; <0.06). Full results are presented from analyses of Cycle 2 data. We repeated all analyses on data from Cycle 3 to confirm the findings and examine the stability of the parameter estimates (La Sorte, 2007); given some differences in the results between Cycles 2 and 3, analyses were replicated on data from Cycle 1. Results related to sleep problems from Cycle 2 are highlighted. Findings that replicated across all three cycles are emphasized.
Incremental validity of sleep problems
Children’s sleep problems were correlated with both internalizing (r = 0.25, P < 0.001; r = 0.22, Cycle 3; r = 0.21, Cycle 1) and externalizing problems (r = 0.24, P < 0.001; r = 0.19, Cycle 3; r = 0.21, Cycle 1). After controlling for other risk factors, child sleep problems accounted for a small, but significant, proportion of the variance (1.6%) in internalizing problems (see Table 3). (Results were replicated in analyses of Cycles 3 [β = 0.117, ΔR2 = 0.013] and 1 [β = 0.120, ΔR2 = 0.014]. Full results of the regressions for all three cycles are reported in the online Supporting information.) Inspection of the partial correlations revealed that temperament was a modest predictor and PMK depressive symptomatology, parenting, and child sleep problems were small predictors of internalizing problems. A similar pattern was found with Cycles 3 and 1 (child sleep problems, pr = 0.13, both cycles). In terms of relative effect size, child sleep problems had effects similar to PMK depressive symptomatology and parenting.
Table 3. Hierarchical regression analyses predicting internalizing and externalizing problems from Cycle 2 data
PMK, person most knowledgeable; pr, partial correlations.
†Standardized regression coefficients and partial correlations are estimates from the final regression model.
*P < 0.05, **P < 0.01 and ***P < 0.001.
Step 1: Child characteristics
Child health status
Child difficult temperament
Step 2: Household demographics
PMK education level
Single parent status
Step 3: Family and parent functioning
Step 4: Parenting
Step 5: Child sleep problems
Child sleep problems
Child sleep problems accounted for a very small (0.4%) but significant proportion of the variance in externalizing problems, after controlling other risk factors (see Table 3). (Results were replicated in analyses of Cycles 3 [β = 0.034, ΔR2 = 0.001] and 1 [β = 0.099, ΔR2 = 0.009].) Temperament and parenting were moderately strong predictors of externalizing problems while the rest of the predictors were small. The effect size for child sleep problems was similar to PMK depressive symptomatology. Results with Cycles 3 and 1 were similar (child sleep problems, pr = 0.04, Cycle 3; pr = 0.12, Cycle 1).
Structural equation models
The correlation matrix for all observed indictors included in the SEM analyses is reported in the online Supporting information. The hypothesized model predicting internalizing problems yielded an inadmissible solution (see Table 4). Modification indices suggested relaxing the path linking temperament and parenting, which is logical given reciprocal relationships between child characteristics and parent behavior. The alternative model (M2) had a moderately good fit; modification indices suggested the family demographics – PMK depressive symptomatology path should be estimated. This was theoretically justifiable as income inadequacy and being a single parent has been shown to be associated with elevated depression symptoms (Cairney et al., 2006; Wade and Cairney, 1997). The alternative model (M3) was a good fit to the data, and the chi-square difference test supported model M32. Modification indices suggested more paths could be relaxed, but these were less justifiable on theoretical or logical grounds. The fit of M3 model tested against data from Cycles 3 and 1, fit adequately. Fig. 1 shows the path coefficients for all three cycles.
Table 4. Fit indices and model comparisons in the prediction of internalizing and externalizing problems
M1, hypothesized model; M2, alternative model with a regression path added from difficult temperament to negative parenting; M3, alternative model with regression paths added from difficult temperament to negative parenting and from positive family demographics to PMK depressive symptoms, for Cycle 2; M3. B, replication of Model M3 with data from Cycle 3; M3. C, replication of Model M3 with data from Cycle 1; AGFI, adjusted goodness-of-fit index; TLI, Tucker–Lewis index; CFI, comparative fit index; RMSEA, root mean square error of approximation. – = not applicable; *P < 0.001.
M2 versus M3
M1 versus M2
M2 versus M3
The SEM models predicting externalizing problems are presented in the bottom panel of Table 4. The hypothesized model (M1) failed to yield a satisfactory fit. Similar models to internalizing were tested. The final model M3 was the best fit and when tested against data from Cycles 3 and 1, fit adequately. Fig. 2 shows the path coefficients for all three cycles.
Common to models for both internalizing and externalizing problems, temperament was a stronger predictor of sleep problems than the other predictors. PMK depressive symptomatology was predicted by family dysfunction and demographics. Parenting was predicted predominantly by temperament and, to a lesser extent, by PMK depressive symptomatology; the association between family dysfunction and parenting was weaker and less consistent across cycles.
The externalizing and internalizing models differed in important ways. PMK depressive symptomatology and internalizing had a moderately strong relation, whereas the PMK depressive symptomatology – externalizing relation was much weaker. Parenting and externalizing problems were more strongly associated than parenting and internalizing. Sleep problems was a stronger and more consistent predictor of internalizing than externalizing.
Children age 2- or 3-years-old who had problems getting to sleep and staying asleep had higher levels of both internalizing and externalizing problems based on bivariate correlations. The relation between these sleep problems and both emotional and behavioral problems was not due to co-occurring relationships with known risk factors for psychopathology. These findings were replicated in two large independent samples that were representative of the population of Canadian children age 2–3 years, excluding those with chronic health problems. Sleep problems accounted for a small proportion of the variance in predicting internalizing and a very small proportion of the variance in predicting externalizing problems. Nevertheless in the regression analyses, sleep problems were as strong a correlate of internalizing and externalizing problems as some well-established risk factors, including parenting and maternal depressive symptomatology, and the relationship between sleep problems and internalizing and externalizing problems was independent of relations with these risk factors. These results are consistent with some, but not all other studies examining cross-sectional relations between sleep problems and psychopathology among young children. Some studies have found sleep problems to be related to internalizing problems (Gregory et al., 2005), externalizing problems (Lavigne et al., 1999; Richman et al., 1982), or both (Gregory et al., 2004; Lam et al., 2003). However, in some cases sleep problems have been related to externalizing but not internalizing (Lavigne et al., 1999) or neither (Stoleru et al., 1997) while Goodnight et al. (2007) reported non-significant correlations between sleep problems and externalizing but did not examine internalizing. Reasons for inconsistency across studies may include differences in the measurement of sleep problems (e.g., sleep duration versus bedtime resistance) or populations (e.g., community versus clinical samples).
Amongst school-age children and adolescents, experimentally manipulated reductions in sleep result in increased attentional problems and changes in neurocognitive functioning (Fallone et al., 2005; Sadeh et al., 2003); similar studies have not been conducted with preschool-age children but the relative effects of sleep problems compared to other risk factors supports the importance of sleep problems as a factor contributing to emotional and behavioral problems, particularly internalizing problems, in young children. Our finding that sleep problems relate to internalizing and externalizing problems after accounting for other potential ‘third variables’ adds confidence that sleep has a unique effect on young children’s emotional and behavioral functioning.
A model of how child, parent, and family factors contribute to sleep problems and internalizing and externalizing problems among young children was proposed and showed a reasonable fit to the data, particularly given the complexity of the model and replication in two independent samples. The pathway linking sleep problems and internalizing was stronger and more consistent than for externalizing. Among school-age children, a similar pattern was found between children’s sleep quality and duration and teacher ratings of psychopathology, while these sleep parameters were equally predictive of parent ratings of internalizing and externalizing (El-Sheikh et al., 2007).
Difficult temperament contributed to both sleep problems and internalizing and externalizing problems. Previous studies have found temperament to be concurrently related to night waking (Atkinson et al., 1995; Goodnight et al., 2007; Sadeh et al., 1994) and temperament in infancy predicted toddlers’ night waking (Morrell and Steele, 2005). Children who are emotionally negative and persistent in their behavior in general would be expected to have problems settling either because of resistance to go to bed, or difficulties with quieting and self-soothing needed to fall asleep. Children with more difficult temperaments also tended to elicit more negative parenting practices from parents which contributed to sleep problems. This is consistent with the notion that temperament plays a role in children’s responses to their environment, and how responsive they are to parenting strategies (Rothbart and Bates, 2006).
Previous studies have found relations between families’ sociodemographic characteristics (e.g., occupation, education; Rona et al., 1998; Scott, 1990), parenting (Bates et al., 2002; Owens-Stively et al., 1997), family dysfunction and parental psychopathology (Bates et al., 2002; El-Sheikh et al., 2007; Gregory et al., 2005; Seifer et al., 1996) and sleep problems in young children. Our model suggests how these factors may relate to children’s sleep problems. Positive socio-demographic characteristics (i.e., two parents, higher incomes, older parents, higher parental education) related to less family dysfunction and PMK depressive symptomatology; family dysfunction and PMK depressive symptomatology both related to negative parenting which in turn related to more sleep problems. Gregory et al. (2005) found maternal depressive symptoms and family dysfunction contributed to anxiety and sleep problems among 3- and 4-year-old twins and speculated on the role parenting might play in these relations. We found family dysfunction had its effect most directly on PMK depressive symptomatology, and to a lesser extent on parenting, which then influenced both children’s sleep problems and both internalizing and externalizing problems. Maternal mood disturbance during infancy has been found to predict the development of toddlers’ sleep problems (O’Connor et al., 2007). In the present cross-sectional study, however, it is possible the PMK depressive symptomatology may be the result, rather than the cause of children’s sleep problems.
The present study looked at risk factors for sleep and emotional and behavioral problems. There are other variables that may have more proximal influences on children’s sleep, and could explain why and how the risk factors we examined operate. For example, a positive bedtime routine (Adams and Rickert, 1989) that minimizes arousal (Dahl, 1996), combined with an appropriate bedtime, and optimal room environment (e.g., dark without excessive noise) (Spruyt et al., 2005) would create conditions that facilitate children’s sleep. Future studies need to examine how distal risk factors such as family dysfunction relate to these more proximal determinants of children’s sleep (Luthar, 1993). Greater family dysfunction may increase children’s stress in general, while parents with depressed mood may be unable to provide a consistent bedtime routine, both of which would impair a smooth transition to sleep. Ideally, studies examining parent and family characteristics and both psychological and physiological individual difference factors among children would lead to a more complete understanding of the relation between sleep and emotional and behavioral problems.
It was surprising that parenting did not relate more strongly to sleep problems, given the treatment literature showing changes in parenting effects children’s sleep (Kuhn and Elliott, 2003). We examined general parenting behaviors, not how parents manage children’s bedtime resistance or night waking. Among infants, the later parents shifted from active comforting (e.g., cuddling) to strategies encouraging child autonomy related to the persistency of child sleep problems (Morrell and Cortina-Borja, 2002). We need to better understand how sleep-specific parenting strategies relate to sleep problems among preschool-age children, as well as the factors that relate to parents’ strategy choice (Morrell and Steele, 2005).
Similar analyses as in the present study need to be conducted using samples of slightly older preschool-age children (4–5 year olds), school-age children, and adolescents. Due to limitations in the data, we could not examine other risk and protective factors for internalizing and externalizing problems and sleep problems such as stressful life events (Bates et al., 2002), poor marital adjustment (El-Sheikh et al., 2007), or parents’ social support (El-Sheikh et al., 2006). The prospective relation between sleep problems among young children and later internalizing and externalizing problems in light of known risk factors also needs to be examined. Experimental studies are needed to determine causal relations between risk factors and children’s sleep problems. We modeled factors predicting children’s sleep problems and emotional and behavioral problems. Other studies have examined these relations in the reverse. Namely, the effects of children’s sleep on parental adjustment (Meltzer and Mindell, 2007). Examining the bi-directional relationship between these variables is complex, but would be particularly informative.
All data are based on parent report. This approach is common in large community-based studies (O’Connor et al., 2007). However, as parents rated both their children’s sleep problems and internalizing and externalizing problems, shared method variance and/or shared bias in parental ratings may have affected the results. Use of observational measures with smaller samples could help disentangle the effects of parental perception of child characteristics such as temperament and parental depressive symptoms and parenting.
Although parents’ ratings of children’s sleep problems are highly related with objective measures such as actigraphs (Aronen et al., 2000; Tikotzky and Sadeh, 2001), there is a need to examine the generalizability of this study’s findings using measures of sleep problems other than parent ratings. The NLSCY does not include information on co-sleeping (either in the same bed or room); co-sleeping arrangements could affect both the nature of children’s sleep problems and parents’ awareness of their children’s sleep difficulties. Similarly, there is a need to examine other aspects of sleep, such as sleep duration, in relation to internalizing and externalizing problems among young children. A related issue is the clinical significance of sleep problems among young children. The measures used in the NLSCY do not have established clinical cut-offs, and as such we modeled the variables using continuous scores. We found that the magnitude of the relation between sleep problems and internalizing and externalizing problems was similar to other risk factors, also modeled as continuous variables. However, there may be threshold effects for these common problems of bedtime resistance and nightwaking, whereby lower intensity or frequency of these problems has virtually no effect on children’s behavioral or emotional functioning but more intense and frequent problems could have marked effects. The present study did not examine the relative contribution of specific sleep disorders (e.g., sleep apnea), versus other risk factors, in relation to children’s emotional and behavioral problems. This is an important issue for future research.
Finally, some of the scales used in our analyses of the NLSCY are short-forms of well-established measures. Although the psychometrics of the original measures (e.g., the Child Behavior Checklist) are known, the properties of the short-forms used in the NLSCY do not have extensive documentation.
In the off-chance that there were difficulties with the first administration of the NLSCY, we choose to conduct our first set of analyses on Cycle 2 and then replicate the data on Cycle 3. When inconsistencies in the results between Cycles 2 and 3 were found, analyses were repeated on data from Cycle 1.
Due to our large sample size, the chi-square difference test will produce significant results even when differences between nested models are trivial. To ensure that our decision inclusion of additional paths was not a statistical artifact, we considered an improvement in fit indices of 0.005 or greater for at least two of the fit indices, and if the path coefficient for the relaxed parameter was 0.20 or greater to be substantial (Byrne and Goffin, 1993).
While the research and analysis are based on data from Statistics Canada, the opinions expressed do not represent the views of Statistics Canada.