Cumulative risk and process evaluation
Stolk, M.N.
Citation
Stolk, M. N. (2007, March 8). Parenting intervention and the caregiving environment. Cumulative risk and process evaluation. Retrieved from https://hdl.handle.net/1887/11404
Version: Corrected Publisher’s Version
License: Licence agreement concerning inclusion of doctoral thesis in the Institutional Repository of the University of Leiden Downloaded from: https://hdl.handle.net/1887/11404
Note: To cite this publication please use the final published version (if applicable).
Parent and family characteristics in relation to
preschoolers’ externalizing problems:
single or cumulative risk?
Mirjam N. Stolk, Judi Mesman, Jantien van Zeijl, Lenneke R. A. Alink,
Marian J. Bakermans-Kranenburg, Marinus H. van IJzendoorn, Femmie Juffer, and Hans M. Koot Manuscript submitted for publication
Abstract
The objective of this study was to examine single and cumulative parent and family risk in relation to early childhood externalizing problems. The sample consisted of 651 1-year-old, 604 2-year-old, and 614 3-year-old Dutch children and their parents.
Externalizing problems were reported by both parents; parent and family characteristics by the mother. The 1- to 2-year follow-up sample consisted of 724 of these children.
Both cross-sectional and longitudinal results supported the cumulative risk model. The effect of cumulative parent and family risk on child externalizing problems exceeded that of single risk factors only in the longitudinal analyses. The inclusion of the number of parent and family risk factors, rather than information about single specific risk factors may enhance screening processes to identify families with young children at risk for the development of externalizing problems.
Introduction
Externalizing behaviors are quite common before the age of four years and can, to a certain extent, be regarded as age-appropriate (Campbell, 1995, 2002). While more than fifty percent of children between 18 months and 36 months old show externalizing behaviors such as hitting, bullying, disobedience, angry moods, or temper tantrums (Achenbach &
Rescorla, 2000; Koot & Verhulst, 1991), not all children develop externalizing problems nor show persistent externalizing behavior (Keenan & Wakschlag, 2004). About 7 to 15 percent of preschool children develop externalizing problems, based on criteria such as duration and severity (Campbell, 2002).
Influences of parent and family characteristics on the risk for externalizing problems during the preschool years have been shown in several studies (Campbell, 1995), but never from a cumulative risk perspective. The studies that did concern cumulative risks were aimed at children in middle childhood and adolescence. These studies were based on different models or systems arranging influences on child development according to their ‘proximity’ to the child. Examples are, Bronfenbrenner’s (1979) model of different micro-level systems of influences on child development (individual, microsystem, mesosystem, exosystem, and macrosystem), and Sameroff’s transactional model, describing the effects of bidirectional influences between parents and children and factors in their environment (i.e., the social and economic context; Sameroff & Chandler, 1975; Sameroff & Fiese, 2000). The cumulative risk factors in previous studies included influences from several or all of these levels (e.g., Atzaba-Poria, Pike, & Deater-Deckard, 2004). For early childhood prevention purposes, information about (cumulative) risk at the level of parent and family characteristics is of particular importance, because this may point to important indicators of whether a family is at risk for inadequate parenting and subsequent child externalizing problems (Scarr & Deater-Deckard, 1997). The present study explores the relation between both single and cumulative parent and family risk factors and externalizing problems in 1-, 2-, and 3-year-old children. The predictive value of (cumulative) parental and family risk over a 12- to 24-month follow-up period is also investigated.
Parent and family risk factors for child externalizing problems
Research has shown that externalizing problems at an early age are developmentally significant (e.g., Campbell, 2002; Denham, Workman, Cole, Weissbrod, Kendziora,
& Zahn-Waxler, 2000). The development of externalizing problems has been shown to be multifactorial and transactional (e.g., Olson, Bates, Sandy, & Lanthier, 2000).
Externalizing problems have been demonstrated to result from dynamic interactions between child characteristics and characteristics of the caregiving environment (Bronfenbrenner, 1979; Sameroff, 1995). In the first four years of life, parental influence on child development through caregiving is especially important, when there are fewer additional socialization influences like peers and teachers as compared to later childhood (Maccoby, 2000; Rothbaum & Weisz, 1994). The extensive developmental changes and
the significance of family life during the preschool years may set a child on a course of adaptation or maladaptation (Campbell, 1995; Sroufe, Carlson, Levy, & Egeland, 1999).
Findings from the longitudinal Minnesota High Risk Study on low-income families suggest that in the first 4 years of life factors that place a child on an antisocial pathway may best be found in parent-child interactions in the microsystem, instead of individual factors (Aguilar, Sroufe, Egeland, & Carlson, 2000; Sroufe, Egeland, Carlson, & Collins, 2005).
Several parent and family characteristics have been investigated as sources of risk in relation to externalizing problems, including low educational level, young motherhood, low well-being, poor physical health, lack of social support, daily hassles, and (perceived) marital discord. Although these parental characteristics are not directly related to or aimed at child development, they may influence parenting, and through parenting increase the risk of child behavior problems. Maternal educational level as a social and cultural indicator has been demonstrated to influence parenting, mediated by maternal depressive symptoms (Jackson, Brooks-Gunn, Huang, & Glassman, 1998). Low educational level predicted elevated levels of depressive symptoms, which negatively influenced parenting and indirectly child outcome. In turn, the quality of parenting was directly associated with children’s behavior problems, for parents with higher education tend to pursue values and attitudes that promote child development (Anselmi, Piccinini, Barros, & Lopes, 2004). Maternal age has been linked to externalizing problems and ineffective parenting (e.g., Brooks-Gunn & Chase-Lansdale, 1995). Teenage motherhood, but also motherhood before age 25 years is significantly related to an increase in reported externalizing problems and inadequate parenting strategies such as inconsistent and harsh discipline (Coley & Chase-Lansdale, 1998; Trautmann-Villalba, Gerhold, Laucht, & Schmidt, 2004). Maternal general well-being or psychological health, mostly defined as maternal stress and depressive symptoms, has been found to be related to maladaptive parenting, in particular for mothers of children with externalizing problems (Baker & Heller, 1996). In addition, parental physical health problems have been reported to predict externalizing problems in school-aged children (Stein & Newcomb, 1994) and adolescents (Forehand, Biggar, & Kotchick, 1998). In several studies, parental lack of social support has been shown to be associated with both inadequate, aversive parenting styles (e.g., Koeske & Koeske, 1990; Turner & Avison, 1985) and child externalizing problems (e.g., Greenberg, Lengua, Coie, & Pinderhughes, 1999). Further, parents of children with high levels of externalizing problem scores reported frequent general daily hassles, indicating an association between daily hassles and externalizing problems (Belsky, Woodworth, & Crnic, 1996). In addition, high levels of daily stress foster negative parenting outcomes (see Crnic & Acevedo, 1995, for a review). Marital discord was reported to have both a direct effect on the child through exposure to overt parental conflict (Katz & Gottman, 1995) and an indirect effect through inadequate parenting as a result of parental conflict. Parents experiencing high marital discord show more negative parental behaviors, such as inconsistent discipline, emotional unavailability, or unresponsive behavior towards the child, behaviors that are associated with externalizing child outcomes (e.g., Reid & Crisafulli, 1990).
Considering their associations with externalizing behavior problems in the child, these parent and family characteristics may be of particular relevance for screening purposes and interventions to prevent or reduce externalizing behaviors in young children.
However, some questions remain about the nature of the association between parent and family risk factors and child externalizing problems. One of these questions concerns the issue of single versus cumulative risk. In other words: does the association between parent and family risk and child externalizing problems arise from the effect of specific single risk factors, or is it simply the aggregate number of risk factors that predicts child maladaptation?
Cumulative parent and family risk and externalizing problems
According to the cumulative risk hypothesis, not any particular array of risk factors, but the number of risk factors is of relevance in the development of behavior problems (Rutter, 1979; Sameroff, Seifer, Baldwin, & Baldwin, 1993). It is suggested that risk factors act in a cumulative way in the sense that the presence of an increasing number of risk factors is related to an increasing probability of negative outcomes (Seifer, Sameroff, Baldwin,
& Baldwin, 1992). Several studies investigating the effects of cumulative risk focused on intellectual development and maladjustment (e.g., Guttman, Sameroff, & Cole, 2003;
Jessor, Van den Bos, Vanderrijn, Costa, & Turbin, 1995; Sameroff et al., 1993). Studies on externalizing problems were mostly conducted in middle childhood or adolescence.
In one study cumulative effects on behavior problems in 3-year-olds were investigated (Liaw & Brooks-Gunn, 1994). Behavior problems did not change as the number of risk factors increased. However, the authors did not differentiate between internalizing and externalizing problems, precluding conclusions on the outcome-specificity of the results.
Deater-Deckard and colleagues (Deater-Deckard, Dodge, Bates, & Petit, 1998) predicted externalizing behaviors from age 5 to 10 years, based on the cumulative influence of multiple risk factors from four different domains (child characteristics, sociocultural, parenting, and peer experiences). Their cumulative sociocultural risk factor included mainly parent and family characteristics, such as low socioeconomic status, life events, maternal age, and lack of social support. This cumulative parent and family risk factor was significantly correlated with externalizing problems (r = .24). However, as the authors pointed out, this correlation was not higher than the correlations between some of the single risk factors and externalizing problems (e.g., socioeconomic status: r = - .25; life events: r = .23). Atzaba- Poria and colleagues (Atzaba-Poria et al., 2004) found similar results. In a sample of 7- to 10-year-old middle class children a cumulative risk factor, including parent and family risk factors such as lack of social support, problems in the marital relationship, and job spillover, was related to externalizing problems. The cumulative risk factor was significantly related to child externalizing problems (r = .24), but not more strongly than some of the single risk factors (lack of social support: r = -.28; job spillover:
r = .30). Shaw and Emery (1988) investigated the association between parental stressors, including parental conflict, depressive symptoms, overcrowding, and family income, on internalizing and externalizing problems in 5- to 10-year-olds. They concluded that scores on externalizing problems increased as the number of risk factors increased,
supporting an additive, cumulative model of risk factors. Because this study did not present a comparison between the effects of single and cumulative risk factors, it is left undecided whether the predictive value of the cumulative factor exceeds that of single risk factors. Several other studies in (young) adolescent samples provided support for the cumulative model of the association between parent and family risk and child externalizing problems (Forehand et al., 1998; Williams, Anderson, McGee, & Silva, 1990). However, none of these studies provides comparisons of the influence of single versus cumulative risk. Finally, Jones, Forehand, and Brody (2002) also studied family risk in relation to externalizing problems in 6- and 7-year-old African American children in single-mother families, including maternal depressive symptoms, parenting, inadequate income, and community risks. Their cumulative factor was not significantly related to externalizing problems.
Issues in cumulative risk research
A number of studies have shown significant associations between cumulative parent and family risk and child externalizing behaviors. Nevertheless, it seems that the cumulative model is not necessarily a better predictor of child adaptation than some of the single risk factors. The usefulness of the cumulative model may thus be questioned. As it is clearly more complicated and less efficient to assess a wide array of risk factors rather than just one or two, cumulative effects should exceed single risk effects to warrant the assessment of many versus few risk factors.
As a second issue concerning cumulative risk, it can be noted that most studies on the effects of cumulative risk were based on continuous outcome factors, whereas for screening purposes a categorical approach to both risk and outcome may be more informative. A disadvantage of a categorical approach is the decrease in variance and statistical power.
On the other hand, the focus on extreme groups may enhance the predictive power of risk factors. Most importantly however, in prevention practice, professionals need to make decisions based on group membership (i.e., ‘Is this family at risk or not?’). Research results based on continuous risk factors do not provide clear directions for this type of decision making. For both the risk and the outcome factors, a categorical approach may provide a clearer contribution to decision making in screening processes.
Finally, none of the studies examining cumulative parent and family risk factors in relation to externalizing problems included preschool children, whereas the development of externalizing problems before the age of four years has been demonstrated to be of particular developmental interest (Campbell, 2002). Several authors have recommended more research in different developmental periods to understand change and continuity in risk influences (Atzaba-Poria et al., 2004; Deater-Deckard et al., 1998) and to examine the etiology of behavior problems.
Present study
The aims of the present study were to (1) investigate whether single and cumulative parent and family risk factors are related to externalizing problems in early childhood, both cross- sectionally and longitudinally, (2) compare the contribution of single versus cumulative parent and family risk factors to the prediction of early childhood externalizing problems, and (3) explore the feasibility of a categorical approach to associations between parent and family risk factors and early childhood externalizing problems.
Based on studies conducted so far, we expected to find that both single and cumulative parent and family risk factors would predict externalizing problems in early childhood.
In addition, we expected that cumulative parent and family risk would be a stronger predictor of early childhood externalizing problems than some of the single family risk factors or the combined effect of single risk factors. Finally, we hypothesized that a categorical approach to parent and family risk and child externalizing problems would yield similar results compared to an approach based on continuous data, with the advantage of increased applicability in screening processes.
Method
Participants and procedure
Participants were recruited in the Dutch SCRIPT study (Screening and Intervention of Problem behavior in Toddlerhood). The study investigates the effectiveness of an early intervention program aimed at reducing externalizing problems in 1- to 3-year-old children by enhancing parental sensitivity and adequate discipline strategies.
It consists of a screening phase in a large general population sample and a randomized case-control intervention phase in a selected subsample of children with high levels of externalizing problems. In addition, the study included a 1- to 2-year follow-up in the general population screening sample. The data for the present paper are derived from the general population screening phase and the follow-up.
During the screening phase (Time 1) names and addresses of children aged 10 to 15 months (1-year-olds), 22 to 27 months (2-year-olds), and 33 to 40 months (3-year-olds) were drawn from municipal registers from several cities and towns in the western region of the Netherlands between May 2001 and December 2002. Because the screening phase of the SCRIPT study was designed to provide participants for the intervention study, sample homogeneity in terms of cultural background (Dutch) was important for statistical reasons (confounding) and practical reasons (cultural/language difficulties in home visits). Therefore, children who had both a non-Dutch surname and a non-Dutch first name were not included in the target sample. In addition, if they or one of their siblings had already participated in the present study or other research projects at the Centre for Child and Family Studies, they were excluded from the sample. Parents of 4,615 eligible
children were asked to complete several questionnaires on the child and family, sent by mail. Both caregivers were requested to complete the questionnaires independently and to return them by mail. For 2,408 children, primary caregivers completed and returned the questionnaires (response rate 52%). For 2,105 of these children, data were also obtained from the second caregiver, response rate 46%. For the present paper, only those children were included for whom the primary parent was the mother (biological or otherwise) and the second caregiver (if present) was the father (biological or otherwise). A final selection was based on the presence of full data on all the measures of interest.
The selection for the present paper resulted in a sample of 1,869 children: 651 1-year-old children, M = 11.70 months, SD = 1.02, 604 2-year-olds, M = 23.77 months, SD = 0.99, and 614 3-year-olds, M = 35.73 months, SD = 1.20. This sample consisted of 950 boys and 919 girls, of whom 52% was first-born, 61% had one or more siblings, and 40%
received more than 20 hours of daycare per week. Virtually all children, (99%), were living with both biological parents. Mean age of the mothers was 34 years, SD = 4.18, range = 20 - 48. In 53% of the sample, mothers had received high education (Bachelor’s or Master’s degree).
Statistically significant differences between age groups were found for maternal educational level: mothers of 1-year-olds had a higher educational level than mothers of 3-year-olds: F(2, 1866) = 4.55, p < .05. In addition, older children had siblings more often than younger children, overall χ²(2, N = 1,868) = 118.84, p < .01. There were no differences between the three age-groups regarding (1) gender of the child, overall χ²(2, N = 1,869) = 0.70, p = .70, (2) living situation, overall χ²(2, N = 1,869) = 0.01, p = .99, (3) nationality of child, overall χ²(2, N = 1,865) = 0.10, p = .95, (4) nationality of mother, overall χ²(2, N = 1,865) = 0.83, p = .66, and (5) nationality of father, overall χ²(2, N = 1,857) = 0.36, p = .84.
Follow-up data (Time 2) were obtained approximately 1 to 3 years after the screening phase, a range of 9 to 40 months in this study. Questionnaires were sent to all families of which primary caregivers responded at Time 1. The follow-up sample for the present paper was selected using the following criteria: (1) we only included children for whom the follow-up interval was 12 to 36 months, M = 21.70, SD = 5.74, n = 1446, to avoid large differences in follow-up intervals; (2) because data were collected as part of an intervention study, a subsample of children (n = 78) had received an intervention between the screening phase and follow-up. These children were removed from the follow-up sample to avoid interference of potential intervention effects; and (3) longitudinal data on the stability of problem behaviors can be influenced by informant-effects if data from only one informant are used. We therefore selected only those children for whom data were available from both parents at both times of assessment (n = 724).
This selection resulted in a follow-up sample of 724 children, 281 from the originally 1-year-olds, 225 from the 2-year-olds, and 218 from the originaly 3-year-olds. Statistically significant demographic differences between this follow-up sample and the families
that were not selected in the follow-up sample were found for maternal educational level. Mothers in the follow-up sample had received higher education than mothers in the non-follow-up sample, F(1, 1,867) = 11.68, p < .01. In addition, in the follow-up sample there were more originally 1-year-olds and fewer originally 2- and 3-year-olds, F(1, 1,867) = 11.02, p < .01. There were no differences for presence of siblings, χ²(1, N = 1,869) = 0.88, p = .35, and for living situation, χ²(1, N = 1,869) = 2.79, p = .10.
Measures
Externalizing behaviors
The Child Behavior Checklist for 1½ to 5-year-old children (CBCL/1½-5; Achenbach
& Rescorla, 2000) was used to measure externalizing behaviors and was obtained from both mothers and fathers. The Dutch translation of the previous version of the CBCL/1½-5 (the CBCL/2-3) has been proven valid and reliable in a Dutch sample of 2- and 3-year-old children (Koot, Van den Oord, Verhulst, & Boomsma, 1997). The CBCL/1½-5 has not been previously used for children under 18 months old. Recently the applicability of the factor structure as found for 2-/3-year olds by Koot et al. (1997) was tested for 12-month-old children and the applicability of the Externalizing Problems syndrome confirmed (for details see Mesman et al., 2003; Mesman et al., 2003; Van Zeijl et al., 2006b). The present paper focuses on the Externalizing Problems syndrome, consisting of the subsyndromes Oppositional, Aggressive and Overactive. The internal consistency (Cronbach’s alpha) in this study on the Externalizing Problems syndrome was .91 for both mothers and fathers at Time 1 and Time 2. Mother-father agreement on Externalizing Problems, Time 1: r (1,868) = .58, Time 2: r (723) = .52, was comparable to that found in other studies (see meta-analysis by Duhig, Renk, Epstein, & Phares, 2000).
For the present study, mother and father reports of externalizing problems were averaged to measure externalizing problems more validly and protect against informant effects (Van der Valk, Van den Oord, Verhulst, and Boomsma, 2003).
Age of mother at birth of the target child and educational level
The age of mother at the birth of the target child was calculated by diminishing the age of mother by the age of the child. Maternal educational level was measured by a single item question using a 5-point scale (1 = Elementary school to 5 = Master’s degree).
Well-being
The Cantrill ladder (Cantrill, 1965) was used to assess general well-being. This self- anchoring single item indicator was scored on a scale ranging from 0 (very poor) to 10 (very good). The mothers were asked to place themselves on a ladder-figure based on how they had felt in the past month. The Cantrill Ladder has been reported to have good validity, stability, and reasonable reliability (Atkinson, 1981; Palmore & Kivett, 1977). Scores on the Cantrill ladder tend to be skewed, with most responses falling in the positive categories (Diener & Diener, 1996; Veenhoven, 1993). In the present study, the mean score for general well-being was 7.31 (SD = 1.44). For our analyses this factor was reversed to indicate low well-being.
Physical health
Physical health of mothers was measured using a one-item indicator of how mothers’
health was at time being with a 5-point Likert scale ranging from 1 (very well) to 5 (very bad). For the present study this indicator was reversed to indicate poor physical health.
Social support
The existence of different sources of social support, as well as mothers’ satisfaction with that support, were measured using a social support questionnaire, based on the Social Support Scale (Van den Boom, 1988; Westgeest, 1985). Mothers were asked to indicate whether or not they received social support concerning 15 different sources of support (e.g., partner, extended family, community). Subsequently, mothers rated their satisfaction on a 5-point Likert scale ranging from 0 (not satisfied) to 4 (very satisfied). In the present study, the internal consistency (Cronbach’s alpha) for satisfaction of support was .78. For our analyses only satisfaction with support was used and this scale was reversed to indicate lack of satisfaction with social support.
Daily hassles
Typical everyday life events that can be perceived as stressful were assessed by the General Daily Hassles Questionnaire (Kanner, Coyne, Schaefer, & Lazarus, 1981).
The 25-item questionnaire was translated into Dutch, and was completed by the mother. The intensity of the hassles was rated on a 5-point Likert scale, ranged 0 (no hassle) to 4 (big hassle). The questionnaire concerns the areas of work, health, family, friends, the environment, practical considerations, and occurrence by chance (e.g., difficulties parents may experience in making plans, doing groceries, and fixing dinner). Cronbach’s alpha for internal consistency in the present study was .87.
Marital discord
Difficulties in the marital relationship as reported by the mother were measured by a subscale of the Family Problems Questionnaire: marital discord (Koot, 1997). This 5-item subscale, rated on a 3-point Likert scale ranged 0 (not true), 1 (somewhat true), and 2 (true or often true), includes items like “I worry about my relationship with my partner”. Reliability of the subscale was indicated by Koot (1997) as sufficient, Cronbach’s alpha = .75. In the present study, the internal consistency (Cronbach’s alpha) of the subscale was .66.
Categorical factors
All factors were dichotomized to reflect low versus high externalizing problems and low versus high risk states. The externalizing problems factor was dichotomized based on the borderline range cutoff (82.7th percentile; Achenbach, 1992) for each age group.
Maternal educational level was dichotomized so that low educational level reflected higher risk. Mothers who had not completed high school were considered at risk (cf., Gerard & Buehler, 2004; Gutmann et al., 2003). Maternal age at birth of the target child was dichotomized so that ages 25 years or younger reflected higher risk. This cutoff was
also used in other studies that investigated young motherhood as a risk factor (Orlebeke, Knol, Boomsma, & Verhulst, 1998; Trautmann-Villalba et al., 2004). We did not use teenage motherhood as a criterion, because this is very uncommon in the Netherlands, where the average maternal age at birth of the first child is very high as compared to other European countries (M = 29 years; United Nations Economic Commission for Europe, 2002). For maternal physical health, those mothers who indicated that their health was very bad to mediocre were classified as high risk. For the other parent and family risk factors no logical or previously defined cutoffs were available. Based on similar studies, we decided to use a cutoff at the 75th percentile (e.g., Liaw & Brooks-Gunn, 1994; Gerard
& Buehler, 2004).
Results
Descriptive statistics and preliminary analyses
Table 2.1 shows the descriptive statistics for all parent and family risk factors and child externalizing problems as collected during the screening phase (Time 1).
Statistical significant relations in the screening sample (N = 1,869) were found among several dichotomized factors (see Method section): (1) age mother and low education, χ²(1, N = 1,869) = 45.75, p < .00, (2) low well-being and, respectively, poor health, χ²(1, N = 1,869) = 189.11, p < .00, lack of satisfaction with support, χ²(1, N = 1,869) = 152.32, p < .00, daily hassles, χ²(1, N = 1,869) = 193.3, p < .00, and marital discord, χ²(1, N = 1,869) = 78.10, p < .00, (3) poor health and, respectively, lack of support, χ²(1, N = 1,869) = 31.42, p < .00, daily hassles, χ²(1, N = 1,869) = 63.02, p < .00, and marital discord, χ²(1, N = 1,869) = 12.73, p < .00, (4) lack of satisfaction with support and, respectively, daily hassles, χ²(1, N = 1,869) = 140.01, p < .00, and marital discord, χ²(1, N = 1,869) = 181.02, p < .00, and (5) daily hassles and marital discord, χ²(1, N = 1,869) = 128.98, p < .00. No significant relations were found among (1) low education and, respectively, low well-being, p = .88, poor health, p = .48, lack of satisfaction with support, p = .32, daily hassles, p = .19, and marital discord, p = .78, and (2) age mother and, respectively, low well-being, p = .40, poor health, p = .17, lack of satisfaction with support, p = .13, daily hassles, p = .24, and marital discord, p = .50.
The follow-up sample consisted of a subsample from the screening sample, including 724 children with complete data on all measures of interest. We found no significant correlation between age and Time 2 externalizing problems, and the cutoff scores were virtually identical for the three original age groups. There was no significant association between the follow-up interval and the mean of externalizing scores at Time 2, F(2, 19) = .71, p = .49. In addition, no associations were found between the follow- up interval and the age of the children with t(279) = .13, p = .19 for the 1-year-olds, t(223) = -.48, p = .63 for the 2-year-olds, and t(216) = -.68, p = .50 for the 3-year-old children. In addition, logistic regression showed no relation between the follow-up interval
and externalizing problems at Time 2, n = 724, Wald = 1.55, p = .21, and no significant interaction effect of the follow-up interval on the association between Time 1 and Time 2 externalizing problems, n = 724, Wald = 2.17, p = .14. Therefore, a single cutoff score at the 82.7th percentile was used for all age groups, for the dichotomization of the mean externalizing problems scores at Time 2.
Identification of risk factors and computing a cumulative risk factor
We investigated whether seven parent and family risk factors were associated with externalizing problems and therefore be fit to compute a cumulative risk factor. To simultaneously investigate the feasibility of a categorical approach, both externalizing problems and the parent and family risk factors were dichotomized (see Method section). To ensure the identification of robust associations between risk factors and externalizing problems, the initial analyses were performed twice: once on each half of the sample (first half n = 934; second half n = 935). For all dichotomized factors χ²-values were computed. Results showed that all parent and family risk factors were significantly related to externalizing problems in both halves of the sample, first half range χ²(1, n = 934) = 4.95 to 24.21, p < .05, and second half range χ²(1, n = 935) = 5.68 to 20.00, p < .05. All parent and family factors were then included in the cumulative risk factor. The cumulative risk factor was computed by summing the dichotomized single risk factors. Table 2.2 depicts the distribution of the presence of different numbers of risk factors as measured during the screening phase, both in the screening sample (Time 1) and in the follow-up sample (Time 2). The table shows that in over half of the families, one or no risk factors were present, and that very few families experienced five or more risk factors. There were no families with all seven risk factors present.
Due to small sample sizes in the groups with five and six risks the cumulative risk variable was recoded into four categories (or cutoff points), with the highest category representing the presence of four or more risks.
Single and cumulative parent and family risk related to externalizing problems To investigate the association between the number of risk factors and child externalizing problems, we computed the percentages of children with externalizing problem scores above the 82.7th percentile at Time 1 and Time 2 as a function of the number of risk factors at Time 1. The results are shown in Figure 2.1.
The figure indicates that with every added risk factor there was an increase in the number of children showing high externalizing problems. Logistic regression showed that this increase was significant, both at Time 1, Wald = 68.80, p < .01, with an odds ratio of 1.41 and a 95% confidence interval (95% CI) of 1.30 - 1.53, and at
Wald = 21.50, p < .01, with an odds ratio of 1.38, 95% CI = 1.20 - 1.58. The more parent and family risk factors were present, the more children exhibited high externalizing problems.
To investigate associations between single and cumulative parent and family risk factors on the one hand and externalizing problems at Time 1 and Time 2 on the other, logistic regressions were performed. First, cross-sectional analyses on a categorical level were performed by computing initial and unique contributions to the prediction of externalizing problems for all single risk factors (logistic regressions). For the cumulative risk factor, logistic regression analyses were carried out for all potential cutoff points (i.e., 1 or more risks, 2 or more risks, etc.). The cross-sectional results are shown in Table 2.3. All single risk factors were significantly related to externalizing problems on an initial level with daily hassles showing the strongest association to concurrent externalizing problems, odds ratio = 2.29, 95% CI = 1.79 - 2.94, Wald = 42,68, p < .01. For the cumulative factors all cutoffs yielded significant initial results in relation to externalizing problems. The presence of four or more risk factors showed the highest odds ratio, 3.17, with a reasonable confidence interval (95%) of 2.34 - 4.49, Wald = 42.81, p < .01.
For the single risk factors, the unique contribution to the prediction of externalizing problems on a categorical level was investigated as well. Low well-being, lack of satisfaction with social support, maternal physical health problems, and marital discord failed to reach significance after controlling for the other risk factors, leaving low educational level, low maternal age, and daily hassles as significant unique predictors. The strongest initial and unique contribution of the single risks was made by daily hassles.
Second, the predictive values of single and cumulative risk factors at a categorical level in the longitudinal sample (Time 2) were tested. In the regression models for the single risk factors, all risk factors were entered simultaneously. For the cumulative risk factor we performed logistic regressions four times, once for each cutoff point. Results are shown in Table 2.4. In addition, results are shown for regression models with and without correcting for Time 1 externalizing problems.
The analyses showed that Time 1 externalizing problems was the strongest predictor for externalizing problems at Time 2, Wald = 78.13, p < .00. Time 1 externalizing problems showed an odds ratio of 8.75, with a 95% CI of 5.41 - 14.14. In the logistic regressions with all single risk factors included (Table 2.4) only low well-being of mother remained a significant predictor of Time 2 externalizing problems in the longitudinal analyses.
However, after correcting for Time 1 levels of externalizing problems none of the single risk factors remained significant. In contrast, all cumulative risk factors (one for each cutoff point) were significant predictors of Time 2 externalizing problems before correcting for Time 1 externalizing problems. After correction, only the cumulative factors referring to one or more, three or more, and four or more risks remained significantly related to Time 2 externalizing problems, all three with similar test statistics.
A critical number of risk factors?
To investigate which cutoff point in the number of Time 1 risk factors is most strongly related to child externalizing problems at Time 1 and Time 2, additional logistic regressions were performed, using the ‘simple’ contrast method. In this contrast, every
category of the predictor factor is compared to a reference category, in our study the presence of zero risks. By using this method, we can assess which number of risk factors produces the highest increase in the risk for externalizing problems compared to families with no risk factors. The results for the cross-sectional analyses are shown in Figure 2.2 (N = 1,896), and for the longitudinal analyses in Figure 2.3 (n = 724).
(see p 35)
In the cross-sectional analyses all cutoffs of the cumulative risk factor values, all of which were compared to the presence of zero risks, were significantly related to externalizing problems at p < .05. The strongest cross-sectional increase in odds ratio’s could be seen from three to four risks, i.e., from 2.67, 95% CI = 1.76 - 4.05, Wald = 21.49, to 4.02, 95% CI = 2.52 - 6.41, Wald = 34.24, p < .01. The longitudinal results showed a similar pattern. Again, using the simple contrast method, all cutoffs were compared to the presence of zero risks. The presence of one, three and four risks remained significant predictors of externalizing problems at p < .05, after correcting for Time 1 externalizing problems. The presence of three to four risks showed the steepest increase in odds ratio’s, from 2.23, 95% CI = 1.12 - 4.47, Wald = 3.36, at three risks, to 3.36, 95% CI = 1.38 - 8.17, Wald = 5.17, at four risks.
Finally, all analyses were repeated using a continuous approach of cumulative risk.
Results yielded similar results for all analyses. Therefore, only results from the cate- gorical approach were reported.
Discussion
Our results showed that, using a categorical approach to risk, both single and cumulative parent and family risk factors were related to externalizing problems in 1-, 2-, and 3-year-old children. With every added risk factor, the number of children with high levels of externalizing problems increased. In the cross-sectional analyses, the cumulative parent and family risk factor was not more strongly related to child externalizing problems than some of the single risk factors. In contrast, in the longitudinal analyses cumulative parent and family risk was a stronger predictor of externalizing problems than any of the single risk factors. These results support the feasibility of the categorical cumulative model for both the risk factors and the outcome measure.
Parent and family risk and externalizing problems
As in other studies on cumulative risk (Deater-Deckard et al., 1998; Gerard & Buehler, 1999; Shaw & Emery, 1988), we found more children with high externalizing problems as the number of risk factors increased, in both the cross-sectional sample and the longitudinal sample. The presence of several risks yielded negative effects on problem behavior. These results demonstrate the effect of the presence of multiple risk factors on the development of externalizing problems even as early as in the preschool period.
As hypothesized, significant concurrent relations with externalizing problems were found for all single risk factors as well as for the cumulative risk factor. These results are consistent with other studies (Atzaba-Poria et al., 2004; Campbell, 1995;
Deater-Deckard et al., 1998). Because several previous studies did not find cumulative risk to be a better predictor of child externalizing problems than single risk factors (Deater-Deckard et al., 1998), or did not report such comparisons (Atzaba-Poria et al., 2004; Forehand et al., 1998; Shaw & Emery, 1988; Williams et al., 1990), we were particularly interested in comparing single and cumulative risk in terms of their associations with externalizing problems. As expected, cross-sectional analyses showed that cumulative parent and family risk was a stronger predictor of child externalizing problems than all of the single risk factors, except one. Notably, the experience of higher levels of general daily hassles was strongly related to child externalizing problems.
However, three of the four cutoffs of the cumulative risk factor yielded a stronger relation to high concurrent externalizing problems, suggesting that it is not the type but the number of risk factors that matters.
Our longitudinal results confirmed these findings. Longitudinal predictions over a 1- to 2-year follow-up period confirmed findings from previous studies that early childhood problem behavior is the strongest predictor of future child problem behavior (e.g., Mesman & Koot, 2001). After accounting for this stability of problem behavior, we found that none of the single risk factors added to the prediction of later externalizing problems, while all except one of the cumulative factors did show additive predictive value. The presence of one or more, three or more, or four or more parent and family risks remained roughly equally strong predictors of later externalizing problems.
Risk threshold for high externalizing problems?
Additional analyses showed that the presence of four risks yielded the strongest relation with later externalizing problems, both in the cross-sectional sample and in the longitudinal sample. These results are in concordance with Rutter (1979), who showed that the presence of two risks yielded a fourfold increase in the likelihood of mental disorder, while four risk factors yielded a tenfold increase. Results obtained by Sameroff, Bartko, Baldwin, Baldwin, and Seifer (1998) however, revealed a cutoff of eight or more risks to increase the likelihood of poor outcome, while Greenberg, Speltz, DeKlyen, and Jones (2001) pointed at an increase when three or more risks were present. In the present study, the differences between the predictive values of each of the cutoffs were relatively small, indicating no substantial evidence for a particular threshold beyond which the child outcomes increased substantially. Indeed, these results are more suggestive of an additive model of risk (Appleyard, Egeland, Van Dulmen, & Sroufe 2005; Gerard & Buehler, 1999; Jones et al., 2002).
Remarkably, in the present study cumulative risk was not a particularly stronger predictor for externalizing problems than single risk when measured cross-sectionally, whereas when measured longitudinally cumulative risk is a better predictor for externalizing
problems than single risk. This finding suggests that in the long run it may not be the nature but the number of risks that is predictive of the development of externalizing problems. The different cumulative risk factors provided unique statistical prediction of externalizing problems, even when initial measures of externalizing problems were accounted for, highlighting the importance of the measurement of multiple risks.
Clinical implications
Evidence for the stability of externalizing problems, especially when they emerge at an early age (Olson et al., 2000) underscores the importance of identifying parent and family factors that contribute to the development of externalizing problems. As Sroufe and colleagues (1999) state, early experience often plays an important role in the initiation of pathology. Investigating parent and family characteristics in the preschool period may thus be a first step in detecting children with a high risk for problem behavior, and thereby identifying families who may benefit from early interventions.
A categorical approach to assessing risk and outcome is especially useful in screening for externalizing problems, because this approach facilitates the search for concrete decision criteria. The goal in screening processes is to determine whether or not a particular child is at risk for externalizing problems. These decisions are inherently based on categorical criteria for both risk and outcomes. However, previous studies regarding cumulative risk for externalizing problems have most often employed a continuous approach to risk and outcome, making translation of the findings to a categorically based decision process difficult.
The present study revealed that a categorical approach to risk and child outcome yielded significant results similar to those found in studies based on continuous data (Atzaba-Poria et al., 2004; Deater-Deckard et al., 1998). More importantly, we showed that cumulative parent and family risk is a longitudinal predictor of child externalizing problems even after controlling for previous problem behavior and that cumulative effects in that case exceed single risk effects. These findings imply that early screening of young children at risk for persistent externalizing problems may be most effective if aimed at assessing multiple parent and family characteristics and focusing on the number rather than the exact nature of parent and family risk factors. In this study, together with high levels of externalizing problems the number of risks constituted a strong predictor of the presence of problems across a 12-/36- months interval. For example, the odds of showing high levels of problem behavior after this interval when earlier high levels of problems and four or more risk factors were present amounted to 18.7 (8.75 * 2.14).
Strengths and limitations
Some limitations should be noted. Firstly, the generalizability of the results is limited by selective sample attrition. The modest response rates in our study may be due to the size of the questionnaires. However, sample sizes still provided sufficient statistical
power. Secondly, families with high education were overrepresented compared to low educated families, with education as an indicator of socioeconomic status. Research has demonstrated that especially families living under low socioeconomic status are more vulnerable to risks (Garmezy, 1991) and have a higher incidence of behavior problems (Qi & Kaiser, 2003). Hence it is likely that we have underestimated some effects because of restriction of range and did not find associations that may be seen in families living in more extreme adverse social conditions. Thirdly, the predictors were parental self-reports. Therefore it cannot be ruled out that informant bias has influenced our findings, although, if present, it would have been mitigated by the second caregiver’s ratings of problem behavior (also included in our analyses).
Several strengths may also be mentioned. First of all, studies on cumulative risk have not yet focused on early childhood externalizing problems, whereas children’s development before the age of 4 years in relation to later adjustment problems appears to be particularly important (Campbell, 2002). Our study has demonstrated that different parent and family risk factors and the accumulation of risk are already present in early childhood, and from that time on may influence parenting and (through parenting) child behavior. Secondly, where most studies on cumulative risk used cross-sectional data only (e.g., Atzaba-Poria et al., 2004), our study also included longitudinal data. This is essential to draw conclusions about the direction of effects. Because parent and family risk factors predicted later externalizing problems over and above the predictive effect of earlier externalizing problems, it is likely that the parent and family risk factors indeed preceeded and therefore potentially influenced child externalizing problems rather than the other way around. Thirdly, to measure externalizing problems more validly and to prevent any informant effects, we used the averaged score of mother and father reports on externalizing problems, as is recommended by Van der Valk et al. (2003). Fourth, we focused on one element of Bronfenbrenner’s ecological system, namely parental influence on child behavior. Whereas accumulation of risks in different ecological domains have been documented in middle childhood and adolescence (Sameroff et al., 1993), parent and family risk factors are of particular relevance in early development (Sroufe, et al., 2005). Because parent and family characteristics may be assessed when children are very young or not even born yet, this domain of risk factors is especially salient for prevention efforts. In addition, assessing characteristics in a single domain, namely characteristics of the mother, enhances efficiency in screening processes.
Conclusion and future directions
This study provides support for a cumulative model of parent and family risk in relation to child externalizing problems in early childhood. The categorical approach to risk identification and child problems allows for practical implications regarding the early detection of families and children at risk. On a methodological level, future studies should include multiple informants and multiple methods on both risk factors and outcome measures, to enhance external validity. In addition, investigations with
longer follow-up intervals may provide more insight in the longitudinal prediction of externalizing problems, and in stability and change of risk factors. Further, more research is needed to explore the relations between parent and family characteristics, parenting, and externalizing problems. Although a few studies have provided empirical evidence for the mediator model that is widely assumed to underlie the association between these three factors (i.e., parent and family risk leads to inadequate parenting, which leads to child externalizing problems; e.g., Brook, Zheng, Whiteman, & Brook, 2001), this model has not been extensively tested in longitudinal studies with young children. Future studies testing such a model should include a cumulative risk approach to further explore the usefulness of this approach within the broader mediator model explaining the indirect influence of parent and family risk on child externalizing problems.
