The costs of publication of this article were defrayed, in part, by the payment of page charges. This article must, therefore, be hereby marked “advertisement” in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
Objective: The purpose of this study was to examine both cross-sectional and longitudinal associations between frequency of family dinner and overweight status in a large sample of 9- to 14-year-old children.
Research Methods and Procedures: We studied a cohort of 7784 girls and 6647 boys, 9 to 14 years of age at baseline in 1996, participating in the Growing Up Today Study. From annual mailed surveys, we calculated BMI from self-reported height and weight and assessed frequency of family dinner over the previous year. We defined “overweight” as age- and sex-specific BMI >85th percentile. We performed multiple logistic regression analyses; the longitudinal analyses assessed the association of previous year family dinner consumption with 1-year incidence of becoming overweight, using prospective data from 1996 through 1999.
Results: At baseline in 1996, 16% of participants had family dinner “never or some days,” 40% on “most days,” and 44% “every day.” Across these categories, overweight prevalence for girls was 19.4%, 16.6%, and 16.7% and for boys was 24.6%, 23.3%, and 22.7%, respectively. In cross-sectional analyses, adjusting for potential confounders, the odds of being overweight was 0.85 [95% confidence interval (CI): 0.76, 0.96] among children who ate family dinner on “most days” or “every day” compared with those who ate family dinner “never or some days.” In longitudinal multivariate models, the odds ratios between previous year frequency of eating family dinner and 1-year incidence of becoming overweight were 0.95 (95% CI: 0.78, 1.16) and 1.04 (95% CI: 0.85, 1.27) for children who ate family dinner on “most days” and “every day,” respectively, compared with those who ate family dinner “never or some days.”
Discussion: The frequency of eating family dinner was inversely associated with overweight prevalence at baseline but not with likelihood of becoming overweight in longitudinal analyses.
Over the last 30 years, the prevalence of childhood overweight and its associated health and psychosocial consequences has increased dramatically in the United States (1, 2, 3, 4). Whereas the etiology of overweight is almost certainly influenced by genetics, genomic variation alone cannot explain the rapid rise in its prevalence in recent years. Thus, it is critical to study modifiable social and environmental determinants of energy balance and overweight risk from childhood through adolescence.
One trend that has paralleled the rise in overweight in the last two decades has been the decline in frequency of children eating family dinner (5). Eating dinner with family members is associated with healthful dietary intake patterns (6, 7). In the Growing Up Today Study of more than 16, 000 girls and boys 9 to 14 years of age, Gillman et al. (6) found that frequency of eating family dinner was associated with greater consumption of fruits and vegetables, less fried food and soda, less saturated and trans fat, lower glycemic load, and more fiber. Neumark-Sztainer et al. (7) also found a positive association between frequency of family meals and intake of fruits, vegetables, grains, and calcium-rich foods and a negative association with soft drink consumption.
Despite evidence of the healthful properties of family dinner, only one study has related frequency of family dinner to BMI in children (6). In the study of Gillman et al. cited above, age-adjusted BMI was 19.5 kg/m2 among children who ate dinner with other members of their family “never or some days.” 19.2 kg/m2 among those who said “most days,” and 19.0 kg/m2 among those who said “every day.” These estimates, however, were cross-sectional, and they were not adjusted for potential confounders.
The purpose of this study was to expand the previous work of Gillman et al. and examine both cross-sectional and longitudinal associations between the frequency of family dinner and overweight status in a large sample of children who were 9 to 14 years old at baseline.
Research Methods and Procedures
The ongoing Growing Up Today Study, established in the fall of 1996, consists of 16, 882 children, residing in 50 states, who are offspring of Nurses’ Health Study II participants. The study was designed to assess prospectively determinants of adolescent dietary intake, physical activity, and inactivity. We collected all data by means of annual mailed self-administered questionnaires.
Details of initial recruitment are available elsewhere (8). The baseline 1996 sample included 8843 girls and 7650 boys, age 9 to 14 years. In the fall of 1997, 1998, and 1999, we mailed participants follow-up questionnaires to update all information. Response rates to at least one of these follow-up surveys were 94% in girls and 89% for boys. For the current analyses, we excluded 107 girls and 72 boys with conditions that affected their growth (i.e., juvenile rheumatoid arthritis, inflammatory bowel disease, and cerebral palsy) and 256 girls and 98 boys who reported binging and/or purging behavior in 1996. Binging was defined as eating a very large amount of food at least once monthly and feeling out of control and unable to stop during the eating episodes. We defined purging as at least monthly use of laxatives or vomiting to control weight. We further excluded 334 girls and 493 boys who had missing or implausible physical activity hours and 362 girls and 340 boys who were missing BMI data, had an outlying BMI, (9), or were missing family dinner data. Thus, our final sample size was 7784 girls and 6647 boys. There were no substantial differences in exposure or outcome between excluded participants and the final study sample. The Human Subjects Committees at the Harvard School of Public Health and Brigham and Women's Hospital approved this study.
Children self-reported their height and weight each year. Our questionnaire provided specific instructions for measuring height and weight but suggested asking someone to help. Previous studies reported high validity for self-reported heights and weights in preadolescents (10) and adolescents (11). BMI was computed from the heights and weights reported by children each year from 1996 to 1999 using the following formula: BMI = weight/height2 (kilograms per meters squared). For the purposes of this study, we defined overweight as a BMI >85th percentile for age and sex based on the Centers for Disease Control and Prevention standards (12).
Intake of Family Dinner
On each questionnaire from 1996 through 1998, we asked the question, “How often do you sit down with other members of your family to eat dinner or supper?” Response categories to this question, which formed our exposure variable, were “never, some days, most days, and every day.” Because of the very low proportion (1%) of subjects answering “never,” we grouped these subjects with those reporting “some days” for the analyses.
Physical Activity and Inactivity
We developed a physical activity questionnaire specifically for youth that asked the participants to recall the typical amount of time spent, within each season over the past year, in various activities and team sports (13). We computed each child's typical hours spent per week in total physical activity for each season and over the entire year. Estimates of total physical activity that exceeded 40 h/wk were excluded as implausible. For watching television and videos and for time spent playing video/computer games, children reported their typical hours per day for each activity, reporting separate amounts for weekdays and weekends. Totals exceeding 80 h/wk were excluded as implausible.
To estimate energy intake, we used a self-administered semiquantitative food frequency questionnaire, designed by members of our group specifically for older children and adolescents (14). This food frequency questionnaire for youth has been shown to be valid and reproducible (15). It included questions regarding the usual frequency of intake of 132 specific food items over the past year. Using nutrient composition databases, food portion sizes, and each child's reported intake of each food, we estimated total energy intake (kilocalories per day).
At baseline, children reported their race/ethnic group by marking all that applied (six options). We assigned each child to a race/ethnic group following U.S. Census definitions, except that we kept Asians as a separate group rather than pooling them with “Other.” Each year, children reported their Tanner maturation stage in five categories of pubic hair development using a validated self-rating measure (16), and girls reported whether/when their menstrual periods began. We computed each child's age from his or her birth date and the date when the questionnaire was returned to us.
We used logistic regression models with estimation by generalized estimating equations (GEEs)1 (17) to examine 1) the cross-sectional relationship between prevalence of overweight and frequency of family dinner (in categories) and 2) the longitudinal relationship between incidence of becoming overweight in the past year and frequency of family dinner in the previous year. For cross-sectional analyses, we used GEEs to account for correlations among the small numbers of siblings in the cohort. The cross-sectional cohort included 14, 486 participants from 11, 403 families. For longitudinal analyses, we adjusted for clustered, non-independent observations among participants (participants could contribute one to three observations). The longitudinal analysis included 10, 420 participants with 22, 249 observations. All models were fit separately for boys and girls; we combined results and adjusted for sex if sex-specific estimates were similar. We also assessed the relationship between overweight and family dinner consumption within two baseline age strata: 9 to 12 years and ≥13 years.
In the cross-sectional analyses, we considered two groups of confounders that may be related to both frequency of family dinner and overweight prevalence: socio-demographic and physical factors (including age, Tanner maturation stage, baseline height, race/ethnicity, and girls’ menstrual status) and total hours of physical activity and inactivity. To examine the effects of these factors on the relationship between overweight prevalence and frequency of eating family dinner, we sequentially entered these covariates into regression models of our outcome and main predictor. We then included total energy intake in a further model as a hypothesized intermediary in the pathway between frequency of family dinner and overweight prevalence.
To study the effects of family dinner frequency during the year of incident overweight, we related the past year frequency of eating family dinner reported in 1996 to incident overweight from 1996 to 1997, frequency reported in 1997 to incident overweight from 1997 to 1998, and frequency reported in 1998 to incident overweight from 1998 to 1999 in multivariate longitudinal analyses. Because each child could contribute up to three observations, the assumption of independent observations required by ordinary regression models was not met, so we used logistic regression models with estimation by GEEs. Participants who were already overweight in the past year were excluded from the analyses of subsequent year overweight incidence, i.e., 1996 was “baseline” for the analyses of 1996 to 1997, 1997 was “baseline” for 1997 to 1998, and 1998 was baseline for 1998 to 1999. A participant who was overweight in 1996 would not be included in the 1996 to 1997 analyses, but this participant could enter the analyses in 1997 to 1998 if he or she was not overweight in 1997 (went from overweight in 1996 to not overweight in 1997). All models were adjusted for age and race/ethnicity, and to account for increases in BMI that typically occur during growth and maturation, we included height growth during the same year, menstrual history, Tanner stage, non-linear age trends, and prior BMI z-score to adjust for nearness to the cut-point. Models also adjusted for activity and inactivity during the year of incident overweight. Finally, we again included total energy intake in further models as a hypothesized intermediary in the pathway between frequency of family dinner and incident overweight. We used SAS Version 8.0 (SAS Institute Inc., Cary, NC) for all statistical analyses.
Baseline characteristics of study participants are summarized in Table 1. Mean age-adjusted BMI in girls was 18.4 ± 3.1 (SD) kg/m2 in 9 to 12 year olds and 20.2 ± 3.0 kg/m2 in the 13 to 14 year olds, corresponding to percentiles of 52.2% and 53.3% in U.S. national data, respectively. In boys, the mean age-adjusted BMI was 19.2 ± 3.4 kg/m2 in 9 to 12 year olds and 19.3 ± 3.3 kg/m2 in the 13 to 14 year olds, corresponding to percentiles of 62.5% and 45.3% in U.S national data, respectively. The cohort was 93% non-Hispanic white, reflecting the race/ethnicity representation of their mothers, who are participants in the Nurses’ Health Study II.
Table 1. . Baseline (1996) values for 14, 431 Growing Up Today Study participants*
Total hours per week watching television and videos and time spent playing video/computer games.
Tanner stage of pubic hair development.
NA, not applicable.
9 to 12
18.4 ± 3.1
13.5 ± 8.1
25.0 ± 14.1
2042 ± 634
20.2 ± 3.0
17.3 ± 9.1
25.4 ± 15.1
2038 ± 626
9 to 12
19.2 ± 3.4
16.5 ± 8.8
30.8 ± 15.9
2285 ± 719
19.3 ± 3.3
19.1 ± 9.3
33.1 ± 16.6
2322 ± 732
Intake of Family Dinner
At baseline in 1996, ∼17.4% of girls and 15.3% of boys ate dinner with members of their family “never or some days,” 39.7% of girls and 40.1% of boys ate dinner with members of their family on “most days,” and 42.8% of girls and 44.6% of boys ate dinner with members of their family “every day.”
We observed an age-related trend of reduced frequency of family dinner for all participants from 1996 to 1998. In 1998, only 34.0% of girls and 38.4% of boys were eating family dinner every day. The observed decline in family dinner frequency was greater among older boys and girls (data not shown).
At baseline, overweight prevalence was 19.4% in girls and 24.6% in boys who ate dinner with members of their family “never or some days,” 16.6% in girls and 23.3% in boys who ate dinner with members of their family on “most days,” and 16.7% in girls and 22.7% in boys who ate family dinner “every day” (Figure 1).
Table 2 shows the multivariate adjusted odds ratios (ORs) of overweight prevalence at baseline associated with family dinner frequency. We observed a 15% reduction in odds of adolescent overweight for boys and girls combined [OR, 0.85; 95% confidence interval (CI), 0.76, 0.96] after adjusting for age, sex, race/ethnicity, Tanner stage, menarche (girls), and height. Further adjustment for physical activity and inactivity did not change the estimates. In subgroup analyses by age group, the observed protective effect of family dinner on adolescent overweight was apparently limited to girls and younger boys (ages 9 to 12 years); however, the estimates were not statistically significant. Energy adjustment minimally attenuated the estimates by ∼1.5% to 7% (data not shown). The fit of the final cross-sectional model was assessed as adequate based on Hosmer-Lemeshow tests and c-statistics. The Hosmer-Lemeshow p value for the overall cross-sectional logistic model was 0.59, which was substantially >0.05. The area under the receiver operating characteristic curve (c-statistic) for the overall cross-sectional logistic model was 0.66.
Table 2. . Multivariate adjusted* ORs of overweight† prevalence at baseline associated with family dinner frequency at 9 to 14 years of age
Estimates from logistic regression models with GEE estimation adjusting for correlations among siblings of the same sex. Adjusted for age, race/ethnicity, Tanner stage, menarche, height, physical activity, inactivity, and (in overall analyses) sex.
Overweight is defined as a BMI >85th percentile for age and sex.
The p value for test of significance of the sex-by-age interaction term was 0.001. Therefore, we stratified the models and present the results by sex and by age group.
Prevalence of overweight
OR (95% CI)
Never or some days
0.85 (0.76, 0.96)
0.84 (0.68, 1.03)
0.78 (0.57, 1.06)
0.83 (0.69, 1.01)
1.12 (0.74, 1.68)
0.85 (0.76, 0.96)
0.84 (0.68, 1.02)
0.84 (0.62, 1.15)
0.83 (0.68, 1.00)
1.20 (0.80, 1.79)
Boys in our cohort were more likely to become overweight than girls (Table 3). In longitudinal multivariate models, we found the relationship between previous year frequency of eating family dinner and 1-year incidence of becoming overweight to be essentially null in combined analyses of boys and girls (Table 3). The ORs between previous year frequency of eating family dinner and 1-year incidence of becoming overweight were 0.95 (95% CI, 0.78, 1.16) and 1.04 (95% CI, 0.85, 1.27) for children who ate family dinner on “most days” and “every day,” respectively, compared with those who ate family dinner “never or some days.” The estimates of the impact of family meal frequency were minimally changed by inclusion of prior BMI z-score to the model.
Table 3. . Multivariate adjusted* OR of 1-year incidence of becoming overweight† associated with past year frequency of eating family dinner
9 to 12 years of age
≥13 years of age
9 to 12 years of age
≥13 years of age
Pooled longitudinal data from the 1996, 1997, 1998, and 1999 Growing Up Today Surveys.
Estimates from logistic regression models with GEE estimation adjusting for clustered, nonindependent observations among participants. Adjusted for age, age2, race/ethnicity, baseline and follow-up menstrual status (girls) and Tanner stage, baseline height, annual change in height, BMI z-score, physical activity, inactivity, and (in overall analyses) sex.
Overweight is defined as a BMI >85th percentile for age and sex.
Average yearly incidence of becoming overweight
OR (95% CI)
Never or some days
0.95 (0.78, 1.16)
1.14 (0.76, 1.70)
0.93 (0.61, 1.44)
0.87 (0.57, 1.30)
0.86 (0.60, 1.23)
1.04 (0.85, 1.27)
1.35 (0.91, 1.98)
0.85 (0.53, 1.35)
0.87 (0.58, 1.32)
0.87 (0.60, 1.27)
Subgroup analyses did not reveal any definite associations within age- and sex-specific groups, but with relatively low incidence rates of overweight, the CIs were wider than those in the cross-sectional analyses.
In this study, we found, in cross-sectional analyses, that children who ate dinner with other members of their family at least most days were less likely to be overweight at baseline. After adjusting for several potential confounders, the reduction in prevalence of being overweight was 15%. However, we found no longitudinal association between likelihood of becoming overweight and frequency of family dinner.
Our cross-sectional findings extend and amplify results previously reported from this same cohort, in which we simply showed lower BMI among adolescents who ate family dinner more frequently (6). In this study, we adjusted for several potential confounders, including sociodemographic factors and physical activity and inactivity levels, and we quantified the inverse association between family dinner frequency and overweight prevalence. Multivariate adjustment for energy intake minimally attenuated our OR estimates, suggesting that total energy intake could partially explain the associations between family dinner frequency and overweight prevalence.
There are several plausible reasons why the longitudinal results were null in the face of significant cross-sectional results. One is that the critical time period for family influences on children's eating patterns may be earlier in childhood; therefore, the cross-sectional results may be the product of a mechanism that occurred earlier in life. Studies by Birch and colleagues (18, 19) suggest that children's eating patterns are influenced by their parents and their family environment as early as infancy. Food choices and eating patterns of adolescents may be more a function of individual, environmental, and societal influences, including influences from peers, schools, mass media, marketing, and cultural norms (20). Second, as in our study, the Coronary Artery Risk Development In Young Adults study, which has examined the longitudinal relationship between diet and weight change among black and white young adults, did not consistently find that dietary intake predicted weight change over a 7-year follow-up period (21). Another possible explanation is reverse causality; adolescents who were more overweight could have been skipping family dinner as a way to diet, thereby biasing our longitudinal results toward the null. The wider CIs around the longitudinal estimates suggest that our sample size may not have been adequate enough to detect an effect of family dinner on overweight incidence. Finally, there is the possibility that there is truly no association between family dinner frequency and overweight incidence.
The reduced frequency of eating dinner as a family as children age is a trend supported by other studies in the literature (22). Surveys of families that include younger children indicate that 41% to 46% of them eat dinner together every day (23, 24) compared with 27% of families with 12 to 17 year olds (5). Previous studies have also shown declines in diet quality as children progress from elementary school to middle school and high school (7, 25, 26). The observed changes in family dinner frequency may contribute to the trends in diet quality, suggesting a need to look for ways to increase family dinner frequency or to find alternative strategies to ensure diet quality among older adolescents.
A major strength of our analysis was the longitudinal design, which allowed us to study changes over time in family dinner frequency and in overweight incidence, while accounting for growth and maturation. BMI typically goes up from year to year among children in this age range, and we took these changes into account. There are also several limitations to our study. The first was the necessity of collecting data on youth by self-report on mailed questionnaires, but the resulting measurement error should be random. Second, although the subjects in this study come from all 50 states and several U.S. territories, generalizing the results may be limited because the subjects are sons and daughters of registered nurses, and the cohort is >90% white.
The benefits of eating family dinner seem to include improved diet quality, reduced high-risk adolescent behaviors such as tobacco, alcohol, and marijuana use, and improved school and psychological performance (27, 28, 29). Our data, however, do not strongly suggest that prevention of overweight is an additional benefit of increasing the frequency of adolescents eating dinner with family members.
This study was funded by NIH Grants HL68041, DK46834, and DK59570 and Boston Obesity Nutrition Research Center Grant DK46200). E.M.T. is supported, in part, by the Minority Medical Faculty Development Program of the Robert Wood Johnson Foundation. We thank the participants of the Growing Up Today Study for their contribution to this study.