Abstract
This study examined associations between profiles of family routines in early adolescence and profiles of health behaviors during young adulthood. In a sample of 4565 individuals, latent transition analysis indicated individuals in a family characterized by low involvement in adolescence were most likely, across classes, to demonstrate a profile characterized by substance use in young adulthood. The high-involvement class during adolescence was least likely to be a substance user but was relatively likely to be in the poor diet and exercise class during young adulthood. Results highlight the utility of examining complex family influences on health using person-centered methods.
The importance of parents during adolescence has been demonstrated across a variety of domains (Aquilino and Supple, 2001; Baumrind, 1991; Steinberg, 2001). Studies have frequently demonstrated parenting style to be predictive of adolescent achievement and risk behaviors (Calafat et al., 2014; Steinberg et al., 1992), while others have shown constructs like parental knowledge, control, and monitoring are associated with outcomes (Cleveland et al., 2012; Fletcher et al., 2004; Kerr et al., 2012). Related work has gone into determining the concurrent and predictive impacts of parent–child or family engagement in routine activities on youth behaviors.
Family routines have been conceptualized in the literature as observable prosocial behaviors that occur with predictable regularity over time involving more than one family member (Jensen et al., 1983). The potential impact of these types of family activities on youth have been examined for decades (Fiese et al., 2002; Skeer and Ballard, 2013) as routines can be thought of as measurable indices of the more general construct of parental involvement, providing concrete targets for intervention to improve youth outcomes. For example, meals shared by family members have been examined in a variety of contexts, with results frequently (though not always) indicating an association between greater frequency of family meals and more optimal adolescent behavioral outcomes, including reduced alcohol use, tobacco use, marijuana use, and disorder eating behaviors, as well as improved school performance (Fulkerson et al., 2006; Hoffmann and Warnick, 2013; Musick and Meier, 2012; Neumark-Sztainer et al., 2003; Sen, 2010; White and Halliwell, 2010). Family religious routines have also been linked with youth outcomes, such that greater religious behaviors by adolescents are associated with less substance use (Ulmer et al., 2012) and greater physical activity (Pfeiffer et al., 2011). Furthermore, parental attendance at religious services has been linked with decreased likelihood of adolescent substance use (Farmer and Brown, 2013). Routine family activities aimed at shared fun have also been linked to better family functioning (Roggman et al., 1994; Zabriskie and McCormick, 2001) and better adolescent reports of well-being (Offer, 2013). Similarly, involvement in outdoor recreation or sports with parents has been found to promote positive health behaviors and increased physical activity in youth (Lam and McHale, 2015). Even less enjoyable family routines, like chores for children and adolescents, have been related with greater physical health (Francavilla and Lyon, 2003) and psychological well-being (Telzer and Fuligni, 2009).
One limitation present in much of the work on family routines has been either a focus on a single family routine (e.g. family meals and family fun activities) or on a composite of family routines (e.g. Family Routines Inventory) (Jensen et al., 1983), providing limited information on the patterns of routines observed within families, as well as the potential associations between these patterns and youth health outcomes. Individual routines like family meals do not exist in a vacuum, independent of other activities, while summing or averaging routines’ risks inaccurately equating different patterns of activities (e.g. 0 days a week with family fun activities and 7 days a week with family meals do not necessarily equal 4 days with family fun and 3 days with family meals). Furthermore, youth health behavior outcomes in these studies have primarily been examined separately, thus providing limited information on the general patterns of health behaviors engaged across domains (e.g. substance use and nutrition) and their associations with family routines in early adolescence. Finally, the majority of the work on family routines examines their associations with concurrent youth health outcomes or outcomes at a relatively short follow up (e.g. 1–2 years). By limiting the timeframe with which the impact of routines is measured (e.g. routines and outcomes both measured during adolescence), researchers may be missing the true lasting effects, whether sustained or extinguished, of these family activities. This study sought to mitigate these gaps by examining the association between profiles of family routines in early adolescence (i.e. 12–14 years) and profiles of health behaviors 10 years later during young adulthood (i.e. 22–24 years). We hypothesized that profiles characterized by frequent and diverse family routines will be associated with the most optimal profiles of health behaviors during young adulthood.
Methods
Participants
Data for this study come from the baseline assessment (12–14 years) and the 10-year follow up (ages 22–24 years; transition to adulthood) of the National Longitudinal Study of Youth 1997 (NLSY97). The baseline sample consisted of 5419 participants (52% men). With regard to race/ethnicity, 2823 participants were non-Hispanic White (52%), 1388 were non-Hispanic Black (26%), 1157 were Hispanic (21%), and 51 were non-Hispanic, mixed race (1%). The mean baseline age was 13.33 years (standard deviation (SD) = 0.95). The majority of youth were raised by Roman Catholic (n = 1412, 26%), Baptist (n = 1083, 20%), or non-religious parents (n = 372; 7%). A total of 4565 participants (84%) were surveyed at the 10-year follow up, representing a high degree of retention.
Measures
Measures at baseline
Family routines
Family routines were measured at baseline using the Index of Family Routines. This index consisted of four items measured from 0 = No days per week to 7 = All seven days per week. The specific items were as follows: In a typical week, how many days from 0 to 7—(a) do you eat dinner with your family? (b) does housework get done when it is supposed to, for example, cleaning up after dinner, doing dishes, or taking out the trash? (c) do you do something fun as a family such as play a game, go to a sporting event, go swimming, and so forth? and (d) do you do something religious as a family such as go to church, pray, or read the scriptures together?
Theoretically, meaningful cut-points were employed to break the distributions into five bins: 0 days, 1–2 days, 3–4 days, 5–6 days, and 7 days.
Covariates
Participant sex was coded 1 = male and 2 = female, and race/ethnicity was recoded to be 1 = non-Hispanic White (termed White herein) and 0 = other. Birth year, 1982–1984, was employed to represent participant age. Youth at baseline reported on their use of alcohol, cigarettes, and marijuana at some point during their life (yes = 1, no = 0).
Measures at follow up
Health behaviors
A representative, though not exhaustive, set of distinct health behaviors were examined. Participants reported on their past year alcohol, cigarette, and marijuana use (yes = 1, no = 0). Fruit and vegetable intakes were individually measured on a seven-point scale from 1 = do not typically eat to 7 = four or more times per day. These values were binned into four substantively meaningful categories: 0 days/week (e.g. never), 1–3 days/week (e.g. rarely/sometimes), 4–7 days/week (e.g. sometimes/often), two or more times/day (e.g. very often). Exercise was operationalized as the number of days in the past week that a participant reported engaging in exercise that lasted 30 minutes or longer. These responses were binned into 0 days/week, 1–2 days/week, and 3–7 days/week. Finally, daily sleep was self-reported from 0 to 10 hours or more hours per night. These responses were binned into 1–6 hours (e.g. <Centers for Disease Control and Prevention (CDC) recommendation; Hirshkowitz et al., 2015) 7–8 hours, and 9 or more hours per night.
Substance use
At the 10-year follow up, participants reported on their use of alcohol, cigarettes, and marijuana in the past year (yes = 1, no = 0). No data were available on the use of other specific substances in the NLSY97.
Plan of analysis
Analyses were split into three phases, examining (1) profiles of family routines in early adolescence, (2) profiles of health behaviors during the transition to adulthood, and (3) associations between the two sets of profiles.
Phase 1
A series of latent class analyses were performed using the set of family routines variables collected at baseline, with classes added one at a time until the best fitting solution was found. Latent class models seek to describe latent, or unobserved, subgroups of the population using a set of categorical manifest indicators. Statistically, the minimum Akaike Information Criteria (AIC; Akaike, 1987) and Bayesian Information Criteria (BIC; Schwarz, 1978) values across solutions were examined, in addition to the adjusted likelihood ratio test (aLRT; Lo et al., 2001) were used to evaluate fit of the models. Substantively, considerations included class sizes (i.e. profiles smaller than 5% are often unstable), distinguishability of profiles, and model interpretability (Lubke and Muthén, 2005). Once the best fitting model was identified, we further described class differences by examining demographic and baseline substance use covariates using the auxiliary function in Mplus.
Phase 2
The analytic plan for examining classes of health behaviors, collected at the 10-year follow up, was identical to Phase 1.
Phase 3
Using latent transition analysis (LTA; Lanza and Collins, 2008), the final phase examined the association between the two sets of latent classes described in Phases 1 and 2. This model provided probabilities of being a member in each of the health behavior classes at the 10-year follow-up period, given membership in the family routines’ profiles at baseline.
All models were performed in Mplus 7.0 using full information maximum likelihood estimators robust to non-normality and missing data over time. This study was approved by the Institutional Review Board of the College at Brockport.
Results
Family routines’ preliminary analyses
We first examined the overall pattern of responses to each of the four family routines examined (presented in the Overall column of Table 1). In general, family meals were very common, with 70 percent of the sample reporting family dinners occurred on 5–7 days per week. Housework was similarly common, with 81 percent reporting work on 5–7 days per week. Family fun occurred at less frequent rates, with the majority of participants reporting family fun on 1–4 days per week (62%). Religious activity was the least common of the routines, with 34 percent reporting no weekly activities and 45 percent reporting religious activities on 1–2 days per week.
Overall and conditional probabilities of family routine latent class indicators.
Family routines’ latent class analysis
Results indicated that a four-class solution fits the data best. The minimum value of the BIC was observed for the four-class solution (BIC3 class = 55,623, BIC4 class = 55,565 and BIC5 class = 55,574), and the aLRT indicated the four-class model fits better than the three-class model (p = 0.005). Although AIC values continue to decline with the inclusion of a fifth class and the aLRT indicated that the five-class model fits better than the four-class model (p = 0.008), there were serious convergence problems at the five-class level. Given these results and the desire for model parsimony, the four-class model was retained.
The first and largest class, representing 58 percent of the sample, was labeled the Average class, as the conditional probabilities for this class closely resembled the overall pattern of probabilities for the sample. The second class, representing 23 percent of the sample, was labeled the Low Involvement class, as this group was characterized by infrequent family fun and religious practice and below average frequency of family meals. The third class and smallest class, representing 6 percent of the sample, was labeled the Extremely Low Involvement class, as this group was characterized by very high probabilities of 0 days of family meals, family fun, or religion in a typical week (ps ≥ 0.80) and a relatively high probability of no housework, as well. The fourth class, representing 14 percent of the sample, was labeled the High Involvement class, as this group was characterized by very high probabilities of daily family meals and housework (ps = 0.80), high probabilities of family fun at least 5–6 days a week (p = 0.76), and the highest probability of daily religious activities (p = 0.28; four times higher than closest other class).
Predicting family routines’ latent class membership
We then sought to provide more thorough descriptions of the classes by examining differences on demographic and baseline substance use characteristics. The Average class served as the reference group for these analyses. Results indicated the Average class was significantly younger than the High Involvement class and older than the Low Involvement class (ps < 0.01). There was a greater proportion of White individuals in the Average class (56%) than in the Extremely Low Involvement (34%) and High Involvement (43%) classes (ps < 0.001). Baseline use of each substance was higher in the Low Involvement class (alcohol = 42%, cigarettes = 41% and marijuana = 19%) than the Average class (alcohol = 29%, cigarettes = 29% and marijuana = 10%; ps < 0.001). Alcohol use was less common in the High Involvement class (20%) than the Average class (p < 0.001), and marijuana use was much more common in the Extremely Low Involvement class (20%) than in the Average class (p < 0.001).
Health behaviors’ preliminary analyses
Overall patterns of young adult health behaviors are presented in the Overall column of Table 2. Alcohol was used relatively frequently in the past year, whereas use of cigarettes or marijuana was more moderate and mild, respectively. Consumptions of fruits and vegetables were moderate, with the majority (77%) reporting consumption 1–7 days a week (e.g. 43% eat fruit 1–3 days/week and 34% eat fruit 4–7 days/week). Exercise was fairly common, with most (48%) reporting these activities on 3–7 days, while most participants reported sleeping at least 7 hours per night.
Overall and conditional probabilities of health behavior profile indicators.
Health behaviors’ latent class analysis
Results indicated that a five-class solution fits the data best. The minimum value of the BIC was observed for the five-class solution (BIC4 class = 53,930, BIC5 class = 53,923 and BIC6 class = 53,926). Although AIC values continue to decline with the inclusion of a sixth class, there were convergence and interpretation problems at the six-class level. As with the family routines’ models, these results and the desire for model parsimony led us to retain the five-class model.
The first class, representing 12 percent of the sample, was labeled the Healthy Eating and Exercise class, as these individuals demonstrate frequent fruit and vegetable intake and a high likelihood of exercising 3–7 days per week. The second class, representing 24 percent of the sample, was labeled the Substance Users class, as these individuals reported the highest probabilities of reporting the use of each substance. Specifically, rates of past year cigarette and marijuana use were more than twice those of all other classes. The third class, representing 8 percent of the sample, was labeled the Abstainers and Good Health class, as these individuals were very unlikely to report past year substance use, reported relatively frequent fruit and vegetable intake, and the highest probability of sleeping 9 or more hours a night. The fourth class, representing 28 percent of the sample, was labeled the Average Health class, as these individuals reported values very similar to the overall sample. The fifth class, representing 28 percent of the class, was labeled the Poor Diet and Exercise class, as these individuals demonstrated relatively high probabilities of infrequent fruit and vegetable intake, as well as the highest likelihood of no exercise per week.
Predicting health behaviors’ latent class membership
We then sought to provide more thorough descriptions of the health behaviors’ classes by examining differences on demographic characteristics. The Average Health class served as the reference group for these analyses. Results indicated there was a greater proportion of females in the Abstainers and Good Health Class (59%) than in the Average Health class (50%; p = 0.006) and a lower proportion of females in the Substance Users class than the Average Health class (43%; p = 0.005). There was a much greater proportion of White individuals in the Average class (54%) than in the Poor Diet and Exercise (42%) and Abstainers and Good Health classes (41%; p < 0.001). There were no differences across birth year.
LTA linking the two sets of classes
We then concluded by relating the two sets of latent classes using LTA. The primary output from this analysis, the transition matrix, is presented in Table 3. Results indicated that individuals in the Average Involvement class during adolescence demonstrated the highest probability, across classes, of being in the Healthy Eating and Exercise class during the transition to adulthood. Individuals in the Low Involvement class demonstrated the highest probability of membership in the Substance Users class across classes. Individuals in the Extremely Low Involvement class demonstrated the highest probability of membership in the Poor Diet and Exercise class and the lowest probability of membership in the Healthy Eating and Exercise class. Members of the High Involvement class in early adolescence had the highest probability of membership in the Abstainers and Good Health class during the transition to adulthood across all classes, though they also had a relatively high probability of membership in the Poor Diet and Exercise class.
Transition matrix associating two sets of latent classes.
Values represent the probability of membership in the outcome classes, given membership in the family routines’ classes.
Discussion
This study sought to relate profiles of family routines during early adolescence with profiles of health behaviors during young adulthood. The study utilized a large, national sample of youth prospectively examined over a 10-year period, and profiles were derived and related using categorical latent variable models. Four profiles of routines and five profiles of health behaviors were observed, with distinct associations between profiles identified.
The four family routines’ profiles followed a linear pattern (i.e. Extremely Low, Low, Average, and High), with groups distinguished largely by the quantity of routines observed across contexts, though the High Involvement profile was relatively distinct from the other profiles with regard to routine religious activities. Given this linear pattern of profiles, it was interesting to see that the Low Involvement profile actually presents the poorest pattern of adolescent substance use outcomes rather than the Extremely Low Involvement profile. It is possible that during adolescence, Low Involvement manifests as inconsistent family routines, whereas there is more consistency in the Extremely Low Involvement profile. Consistency in parenting has been linked with a variety of adolescent outcomes (Chassin et al., 2002; Simons and Conger, 2007). There could also be a lower quantity of routines but higher quality in the Extremely Low profile than in the Low profile. This would support the work by Windlin and Kuntsche (2012) showing that quality of shared family activities was more predictive of outcomes like substance use and violence than the quantity of these activities.
The five health behavior profiles demonstrated a greater degree of non-linearity, with young adults reporting differential combinations of risk and health behaviors. For example, nearly one-quarter of participants were distinguished by a pattern of elevated substance use, while another quarter was best distinguished by their pattern of poor dietary intake and limited exercise. Clear patterns of reduced health risk were exemplified by roughly 20 percent of participants (i.e. Healthy Eating and Exercise, and Abstainers and Good Health). These findings highlight the potential salience and importance of examining health behaviors as a multi-faceted construct. Previous research has taken a similar approach to modeling health behaviors during childhood/adolescence (Connell et al., 2009; Laska et al., 2009). Several of the classes observed by Laska et al. (2009) among college students correspond to those observed in this study of young adults. For example, the “Higher Risk” profile they describe corresponds well to the Substance Users profile observed here with regard to alcohol and tobacco use, exercise, and diet, though their study did not identify a class specifically exemplifying abstinence and good health behaviors across both males and females.
When profiles during early adolescence and the transition to adulthood were related, an interesting pattern of associations were observed. For example, a family environment characterized by the highest level of family routines (all relatively prosocial/positive routines) was not shown to be consistently related to the most optimal health behavior outcome profiles. Specifically, although the High Involvement class was most likely to be abstainer and least likely to be a Substance User, these individuals were also at an elevated risk for membership in the Poor Diet and Exercise class. It is possible this association is partially induced by the fact that individuals in the Poor Diet and Exercise also demonstrate relatively low levels of substance use. It is also possible that for some youth, High Involvement families might be perceived as intrusive or hovering, leading to a negative pattern of young adult health behaviors as a reaction to a perceived need for autonomy (Schiffrin et al., 2014). Conversely, although the Extremely Low Involvement class was at the greatest risk for membership in the Poor Diet and Exercise class, they were also relatively unlikely to be a Substance User. This may be due to the association between very low parental involvement/parental neglect and parental substance use (Collins et al., 2003), leading these youth to strongly avoid alcohol and other substances. This study’s use of person-centered analysis to demonstrate these distinctly non-linear sets of associations highlights the importance of considering parenting/familial effects in a domain-specific fashion, such that one cannot assume that greater family routines and involvement are necessarily predictive of better outcomes across the range of young adult health behaviors.
The results of the LTA performed also lend support to the notion of “Good Enough” parenting (Hoghughi and Speight, 1998; Ramaekers and Suissa, 2012), which implies that moderate parental/familial efforts are sufficient to produce generally positive outcomes. Our study supports this perspective in that the Average Involvement class during early adolescence was associated to a generally favorable range of outcomes. Specifically, this class demonstrated mildly elevated probabilities of membership in the Healthy Eating and Exercise and the Average Health classes, with a mildly decreased probability of membership in the Poor Diet and Exercise class. This perspective is further supported in this study by the above-discussed elevated risk of Poor Diet and Exercise class membership for youth raised in High Involvement families, potentially implying involvement beyond average levels could be perceived as intrusive. Similar findings have been observed linking relatively average levels of parenting behaviors with positive substance use outcomes (Abar, 2012).
There were several limitations to this study. First, while generally representative of latent construct modeled, the items available in the NLSY97 describing family routines were relatively rough, somewhat lacking in specificity and potentially over-valuing the impact of specific routines measures (e.g. family meals and family religious activities). Future person-centered work should seek to examine profiles using a more diverse and/or specific set of family routines. Second, the family routines examined did not take into account valence of the interactions (e.g. are family meals pleasant/enjoyable?). Subsequent research may benefit from efforts to examine valence distributions within profiles of family routines. Third, the potential impact of family socioeconomic status (SES) on family routines and/or moderating influence on the association between family routines and health outcomes were not examined in this study. Future research should seek to examine the role SES plays in predicting family routines’ profile membership and profile associations with young adult health outcomes. Finally, despite efforts to represent the construct of health behaviors using a diverse set of outcomes, there are a variety of domains that could be explored in additional profile research including health activities like preventive health screenings, dental behaviors, and/or sexual behaviors. More granular measures of some of the measured used (e.g. past year heavy episodic drinking rather than any use) might also be of substantive value.
Conclusion
This study demonstrated an association between profiles of family routines engaged during early adolescence and profiles of health behaviors 10 years later. Results presented imply there is considerable utility to the examination of complex family influences on substance use and health behaviors using a person-centered, latent variable approach. Associations between family routines and health outcomes are oftentimes non-linear, such that more nuanced analyses than traditional, variable-centered methods might continue to provide unique insight into young adult health behaviors.
Footnotes
Declaration of conflicting interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.