Family Meals and Child Academic and Behavioral Outcomes

Previous Theory and Research

Research from scholars in a variety of different fields has examined the importance of family meals to child and family functioning (Larson, Branscomb, & Wiley, 2006). Although family meals are of relatively short duration -- the average family dinner lasts just 20 minutes (Fiese & Schwartz, 2008) -- previous work has highlighted the unique and powerful role of shared family mealtimes in modeling behavior for children and conveying cultural traditions (Larson et al., 2006), and in providing an opportunity for parents to engage in activities that promote literacy, learning, and healthy behavior (Larson, 2008).

In general, family meals are thought to present an opportunity space to promote the healthy development of children (Larson et al., 2006). As a result, the frequency with which such opportunities are available should be related to child and family wellbeing. According to Fiese, Foley and Spagnola (2006), family meals are comprised of both ritual (symbolic) and routine (activity) elements, each of which may be instrumental to a child’s positive developmental outcomes. Both routine and ritual practices can act as markers for healthy family functioning; respectively, they are indicative of greater organization and a greater sense of belonging and closeness for children (Fiese, Tiomcho, Douglas, Josephs, Poltrock, & Baker, 2002).

A number of previous studies have examined the relation between family meal frequency and adolescent cognitive and social outcomes, frequently finding a positive relation. A separate body of literature has also examined the relation between family meals and child nutrition (Burgess-Champoux, Larson, Neumark-Sztainer, Hannan & Story, 2009), eating behaviors (Neumark-Sztainer, Wall, Story, & Fulkerson, 2004), and child overweight and obesity (Anderson & Whitaker, 2010). However, as the mechanisms relating family meals to these outcomes are likely different than those for academic outcomes and child behavior, we do not consider them here. As described below, this study investigated the effects of family dinner and breakfast frequency on child reading, math, and science scores, and internalizing and externalizing behaviors, each of which has been linked to family meal frequency in previous empirical or theoretical work.

For instance, one study found that family meal frequency was inversely associated with low grade point average among middle and high school girls (Eisenberg et al., 2004). Shared family mealtimes may help promote language development in particular; research has demonstrated that exposure to certain types of talk during mealtimes may foster later reading scores and vocabulary size (Snow & Beals, 2006). These findings are confirmed by a series of unpublished studies by The National Center on Addiction and Substance Abuse at Columbia University (CASA), which have reported a consistent relation between family dinner frequency and teen academic performance, such that children who reported typically eating 5 to 7 family dinners per week were more likely to report receiving mostly As and Bs in school and less likely to report receiving mostly Cs (CASA, 2007, 2009, 2010).

Children and adolescents eating more meals together with their families have also fared better on measures of psychological adjustment and have engaged in fewer risk behaviors. As noted, family meals may provide connection to important family and cultural rituals, which may in turn be beneficial for children’s psychological functioning. In addition, shared family meals may provide increased opportunities for communication and monitoring that may be related to a young person’s opportunity or inclination to engage in risky or harmful behavior (Sen, 2010). For example, one study found that 6^th to 12^th grade students who ate 5 to 7 family dinners per week had significantly lower odds of engaging in a number of high risk behavior patterns such as alcohol, drug, and tobacco use, depression-suicide, violence, antisocial behavior, and school problems when compared to those who typically ate 0–1 dinners (Fulkerson, Story, Mellin, Leffert, Neumark-Sztainer, & French, 2006). Using data from the National Longitudinal Survey of Youth (NLSY), Sen (2010) found that the frequency of family dinners was associated with decreased substance use and running away for adolescent females and decreased alcohol use, physical violence, property destruction, stealing, and running away for adolescent males. One study by Eisenberg and colleagues (2004) found that young people who ate 5 to 7 dinners together with their families were less likely to use drugs and alcohol and had lower depressive symptoms and odds of suicide involvement, while another found evidence of a longitudinal inverse association between FMF and substance use, independent of substance use at first report (Eisenberg et al., 2008). Longitudinal findings from another study also indicated that girls who ate dinner at home every day had lower odds of initiating alcohol use at a follow-up survey (Fisher, Miles, Austin, Camargo, & Colditz, 2007).

Limitations of Previous Work

Although previous work has consistently suggested a beneficial link between family meal frequency and child outcomes, there is some reason to approach these findings with caution. For one, previous studies have been inconsistent in their operationalization of family meal frequency. In their review of the literature on family meals, Fiese and Schwartz (2008) identify inconsistent measurement and operationalization of frequency of family mealtimes as one of the chief shortcomings of previous research and also suggest that there is little agreement regarding a ““critical” number of meals that seems to be essential for healthy outcomes” (p.5). This lack of consensus is most clear in the body of research from the Project Eat Surveys, the dataset from which the majority of previous work on family meal frequency and child outcomes has been published. In the Project Eat survey, adolescent respondents were asked, “During the past 7 days, how many times did all or most of your family eat a meal together?” with response choices of 0, 1–2, 3–4, 5–6, 7, or greater than 7. Based on this question, researchers have variously identified meal frequency using midpoints (e.g. 0, 1.5, 3.5, etc.) (Neumark-Sztainer, Story, Ackard, Moe, & Perry, 2000), categories (e.g. 0, 1–2, >=3) (Fulkerson, Neumark-Sztainer, Hannan, & Story, 2008), or using cut-points to determine “regular” family meals such as 5 or more (Burgess-Champoux et al., 2009; Eisenberg et al., 2008) or 3 or more (Eisenberg, Neumark-Sztainer, & Feldman, 2009; Feldman, Eisenberg, Neumark-Sztainer, & Story, 2007). Mostly as a consequence of the structure of survey instruments, studies have tended not to explicitly acknowledge the possible differences between the effects of the frequency of all meals and the frequency of just dinner or breakfast separately. The single previous study to use the ECLS-K to investigate the effect of family meals frequency (Gable, Chang, & Krull, 2007) used a summary measure (range 0–14) of the number of breakfasts and dinners typically eaten together.

Although previous research with the ECLS-K dataset has investigated the associations between FMF and child obesity in children (Gable et al., 2007), the majority of research on academic and behavioral outcomes has focused on adolescents. This is largely a limitation of existing information, as many datasets containing information on FMF (Project Eat, the NLSY, AddHealth) are focused on adolescents explicitly. However, family meals may differ in their importance as children develop. Data from previous research (Fulkerson, Neumark-Sztainer, & Story, 2006; Fulkerson, Story, Mellin, Leffert, Neumark-Sztainer, & French, 2006) suggest that family meals become less common as children age, and a recent meta-analysis indicates that age may have a moderating effect on the relation between family meals and child nutritional outcomes (Hammons & Fiese, 2011). Variation in meal frequency may be more meaningful for younger children, as differences might be indicative of departures from more typical behavior. Alternatively, as children age and expand their range of potentially risky or otherwise detrimental activities, frequent family meals might be distinguished by their ability to protect children against risk and promote healthy development. To our knowledge, no study has systematically investigated the differential impacts of FMF by child age.

Previous research on FMF and adolescent outcomes has also relied primarily on cross-sectional data or has typically been limited to estimating the effect of family meal frequency by including a limited number of controls, which suggests that the findings of earlier studies may be biased as a consequence of endogeneity. The endogeneity problem (Duncan, Magnuson, & Ludwig, 2004) has been extensively addressed in previous work in human development (Berger, Bruch, Johnson, James, & Rubin, 2009; National Institute of Child Health and Human Development Early Child Care Research Network [NICHD ECCRN] & Duncan, 2003) and so will not be reviewed in full here. In short, endodeneity occurs most frequently when unobserved characteristics are correlated with both an independent and dependent variable of interest, a problem commonly referred to as omitted variables bias. In regression analyses of the type used in most previous research on family meals, endogeneity can bias coefficients leading to inaccurate estimates of the effect of family meals on children. Concerns about the endogeneity of FMF are expressed by Neumark-Sztainer (2008), who noted “A question that often emerges is whether the family meal is truly contributing to these better outcomes or if the family meal is only a proxy measure for important confounders such as familial relationships” (p.11). Indeed, while previous work has demonstrated a significant relation between family meals and adolescent controls after controlling for measures of familial relationships like family connectedness (Eisenberg et al., 2004), there may be many other factors not controlled for in previous work that could likewise confound estimates.

It is important to clarify that previous research has not typically claimed a direct link between FMF and child outcomes. Typically, studies have concluded by suggesting the need for additional research to identify the mechanisms by which family meals might affect outcomes (Eisenberg et al., 2004; Eisenberg et al., 2008) or by indicating steps like public education campaigns to raise awareness about the value of family meals (Eisenberg et al., 2004; Sen, 2010). Although an investigation of potential pathways by which shared mealtimes might affect family functioning or child well-being is a potentially important task for research, this study attempts to rigorously investigate the more primary question of whether FMF has a main “effect” on child outcomes.

Generally, the sort of endogeneity of greatest concern to the study of family meal frequency is selection bias, whereby children “select” into both family meal frequency and an outcome of interest as a consequence of unobserved characteristics of the children or their families. For example, parents who recognize the value of shared family mealtimes may also engage in other activities that they believe are good for children, such as promoting extracurricular activities, reading to children, or pursuing high-quality schooling, each of which might affect children’s school performance. In such a case, the frequency that a family eats meals together may act as a marker for other such activities, but may not actually be causally related to child outcomes.

Similarly, the tendency of a family to eat meals together may reflect congruence between parents’ work schedules and children’s routines, and thus may be emblematic of greater overall possibilities for communication and monitoring, which could account for the relation between family meal frequency and child behavioral and academic outcomes. Finally, greater closeness among family members and with parents in particular might increase the frequency of shared family meals, but could also foster an environment that promotes healthy outcomes for children (Boyer, 2006).

Many studies reviewed for this paper relied upon cross-sectional survey data. The handful of studies using longitudinal data to examine the links between family meal frequency and the outcomes of interest to this paper have tended to employ more sophisticated methods, but have also been limited. For instance, while Eisenberg et al. (2008) included controls for a baseline measure of their dependent variable, they additionally controlled for only family connectedness, child race, and socioeconomic status. Using repeated observations on adolescents from the National Longitudinal Survey of Youth, Sen (2010) included a large number of controls and a measure of the dependent variable measured at a future time to address concerns of reverse causality and selection bias, but did not employ other panel data methods that might have better addressed possible bias.

A reliance on cross-sectional data is additionally potentially troublesome because of concerns about reverse causality (Larson, Branscomb, & Wiley, 2006). This is especially true for studies examining internalizing or externalizing behaviors. That is, although it is possible that increased FMF promotes positive child outcomes, it is equally plausible that children are less likely to engage in any sort of shared family activity (including family meals) because they are engaged in risky behavior or have psychological distress.

This study extended previous research of the effects of FMF in a number of ways. First, using a large panel sample of children we examined the relation between FMF and child outcomes for all sample children (ages five to 15) and then separately for younger and older children, whereas previous analyses have exclusively examined adolescents. Second, to address issues regarding the operationalization of family meals, we utilized the relatively comprehensive information on family meal frequency in the ELCS-K dataset to examine the separate effect of the number of shared breakfasts and shared dinners and to explore whether the operationalization of FMF affects its relation to child outcomes. In addition, we took advantage of the longitudinal nature of our data source to reduce concerns regarding endogeneity. We utilized lagged regression models, which examined the effect of family meal frequency on child outcomes at a later time point, thereby creating a temporal set up that precludes the possibility of reverse causality. In addition, to more clearly estimate the causal effect of FMF, in all models we included a large number of controls, which addressed the selection bias associated with observed characteristics of children, families, and schools that might be correlated with both FMF and child outcomes. Last, we utilized panel data methods ---lagged dependent variables (LDV) models and child fixed effects models --- each of which improves upon previous analyses by controlling for unobserved factors that might confound the effect of family meal frequency on child outcomes.

Method

Results

Discussion

As discussed at the outset, previous research on FMF and child outcomes has been limited in a number of different ways. Using extremely rich data on a nationally representative sample of children who entered kindergarten in fall 1998 and were followed through eighth grade and applying more rigorous methods than have typically been used in previous studies on this topic, we found little evidence for beneficial effects of frequent family meals. In our preferred models, effects of family meals on child academic and behavioral outcomes were either small or effectively zero. Our results did not vary when we employed different operationalizations of our main independent nor did they vary by child age.

Of course, our study is not without limitations. As with all non-experimental research, we are not able to be certain that we have eliminated all possible confounders and estimated true causal effects. We have included an extensive set of confounding variables to ensure that we have controlled, to the extent possible, for factors associated with the selection into family meals. However, doing so raises the potential concern that we are also controlling for factors that are causal antecedents to family meals – factors such as family status, parental employment, and so on. As such, our analyses cannot address the question of whether family meals might influence child outcomes as part of a more complex set of processes, whereby they act as mediators between such causal antecedents and child outcomes. Although this seems unlikely, given that we fail to find significant effects of family meals on child outcomes, a firmer answer would require a fuller set of analyses that begin with family predictors and then explore the role of family meals as a mediating variable. Such analyses were beyond the scope of this paper but would be worth exploring in future research.

Although this study contributed to the existing literature by investigating whether the impacts of FMF differed by child age, an important caveat to our results is that we examined children through eighth grade only (when children were on average 13.6 years old). Many of the prior studies finding strong positive associations between frequent family meals and child outcomes focused on older adolescents, for whom family meals may be more protective. Relatedly, because our sample is relatively young, we were not able to examine all the outcomes that have been studied in prior research. Most notable is the lack of information on substance use, which other studies have found to be related to family meal frequency. By analyzing the effects of FMF on a younger sample of children, this paper can be understood as an extension rather than a repudiation of previous work.

Nevertheless, our results are generally at odds with prior studies of adolescents. We note that in our pooled models with controls and to a lesser extent in our LDV models (both in Table 2), there were numerous significant associations between FMF and child outcomes. Thus, when we implemented modeling strategies most similar to those of previous work, we found that FMF was indeed associated with child academic and behavioral outcomes, if not always in the way predicted by previous work. This indicates that an important reason for the discrepancy between our results and other studies’ findings was not the age of the children studied, but rather the methods we applied.

Our study found differences between the measures of FMF in the ECLS-K and those found in other studies. There are a couple of likely causes that underscored these differences. For one, both our own data and data from previous studies (Fulkerson, Neumark-Sztainer, & Story, 2006, Fulkerson, Story, Mellin, Leffert, Neumark-Sztainer, & French, 2006) confirmed that that the frequency of family meals is lower among older children. As a result, it is not surprising that the parents of children in the ECLS-K dataset (the majority of whom are 14 or younger) reported eating meals together as a family more frequently. In addition, differences between the ECLS-K and other datasets were not so pronounced when comparing response categories indicating more frequent meals. For instance, when the data from 8^th grade respondents in ECLS-K were compared to adolescent respondents from the NLSY (Sen, 2010) and CASA (2010), results were similar, where approximately 60% of respondents reported eating dinner 5 or more times per week, compared to 70% in the ECLS-K (Table 1).

Second, there appear to be differences between youth and parental report about family frequency and about the family meal environment in general (van Assema, Glanz, Martens, & Brug, 2007). Since the ECLS-K used only data from parental reports, we would expect to find higher reported rates of family meals than in prior studies that relied on child reports. One study published on a subset of Project EAT data (Fulkerson, Neumark-Szatiner, & Story, 2006) found that parents were only half as likely (6.8% vs. 13.7%) as adolescents to report never eating together but more than twice as likely (22% vs. 9.9%) to report eating seven meals together as a family in the previous week; parents were only slightly less likely to select any of the other responses, including the choice of more than seven meals. The authors of this study suggested that parents may report more frequent family meals because they may count meals that do not include children; this is a valid concern for the questions in the ECLS-K, which did not specifically ask parents to report on meals involving the study child.

Similarly, the frequency with which parental respondents to the ECLS-K chose “7” breakfasts or dinners might reflect the social desirability of these choices, which may be a stronger source of bias for parents than for adolescent respondents in other surveys and thus may represent error in these variables. If the higher reporting by parents is indeed indicative of over-reporting either as a consequence of the social desirability of reporting more meals together, or because parents in the ECLS-K reported having meals together but when a child was not present, then the coefficients for FMF reported in Tables 2 and 3 would be biased towards zero and would thus underestimate the true effect of FMF on child academic and behavioral outcomes. This type of bias may be unique to the ECLS-K dataset among other large, nationally-representative studies as it is the only such study to use parents’ reports. However, we note that this sort of bias would be present across all our models, and was not strong enough to eliminate the significant associations between FMF and child outcomes in our pooled models and LDV models. Once again, this bolsters our confidence that discrepancies between our findings and those of previous analyses were not a consequence of differences in the manner by which data on family meal frequency were collected. Moreover, because children in the ECLS-K were younger on average than those in previous analyses, it is less likely that parents would report having family meals that did not include them. Future research using datasets with adolescent or child report of FMF could provide useful evidence in this regard.

A strength of the current study was our examination of several possible operationalizations of family meal frequency. Overall, our results were robust to different specifications of family meals, increasing our confidence that the lack of beneficial effects of family meals is not due to our choice of specification of the family meals variables.

In sum, despite differences between our study and previous analyses, our results suggest that the findings of previous work regarding FMF and adolescent outcomes should be viewed with some caution. When compared to those from the pooled and LDV models, the coefficients for breakfast and dinner frequency in our FE models were substantially attenuated and close to zero in many cases. These differences suggest that invariant and unmeasured characteristics of children or their families related both to FMF and child outcomes may bias estimates from pooled or LDV models. For instance, frequent family meals may be an investment by families who consistently engage in other activities and provide other supports that promote positive outcomes for children. Fixed effects analyses are not without their weaknesses, however. It may be the case that unobserved, time-varying characteristics may account for both changes in FMF and child outcomes, in which case fixed effects model results would be biased, although these factors would have to account for both increased family meal frequency and poorer outcomes for children. With longitudinal data at their disposal, researchers should continue to take advantage of more robust tests of the causal relation between FMF and both child and adolescent outcomes, and future analyses might attempt to replicate the results of previous studies using the sort of methods included in this paper.

Because this study was the first to find no association between FMF and child outcomes and differs from previous investigations, it may be premature to make conclusions about the value of family meal frequency. Previous research and theory have both suggested that shared family meals are a valuable and important developmental context for children, and their worth may go beyond the ability of surveys to detect or measure or may affect different outcomes than those assessed here. Indeed, even if family meals act as a proxy for other behaviors that are related to positive outcomes for children and adolescents, future research that investigates the processes underlying such behaviors could prove informative for policymakers, practitioners, and the public.

Further, quantity of shared meals may not be the most relevant measure of the worth of family meals. Indeed, some contemporary research focuses on the quality of interactions that occur at the family dinner or breakfast table (Fiese & Schwartz, 2008). Our data do not afford us the ability to distinguish the quality of the mealtime environment, and so we cannot draw conclusions about the value of family meals in this regard.

Last, research on the quality of the home environment (Bradley, Corwyn, McAdoo, & Garcia Coll, 2001) indicates that there is significant variation in mealtime frequency between racial and ethnic groups and within groups by family socioeconomic status. While our analyses controlled for race and ethnicity and socioeconomic status, our findings do not rule out the possibility that the importance of the mealtime environment varies across different groups, which might be the case if different processes were at work during mealtimes Thus, there are several ways in which future research might more comprehensively investigate the effects of both quality and quantity of family meals on child and adolescent outcomes, by for example, examining the moderating role of factors such as race or ethnicity, single parent household-status, and family socioeconomic status and utilizing data sources that contain richer detail on the routine and ritual aspects of shared mealtimes.

Ultimately, we cannot conclude that shared family times over breakfast or dinner should be ignored or that families should not attempt to cultivate these types of collective experiences. Rather, we suggest that the magnitude of the effect of FMF may be less than suggested by previous work. Future analyses should be undertaken in such a way as to acknowledge and address possible sources of endogeneity that can bias the results of regression analyses.