Study design and population
We examined trends in family dinner participation among participants of the Growing Up Today Study (GUTS). GUTS consists of two on-going cohort studies of offspring of nurses (participants of the Nurses Health Study II (NHS II) [10]). The first cohort, GUTS1, was established in 1996 and the second cohort, GUTS2, was established in 2004. For the GUTS1 cohort, we contacted mothers in the NHS II who we identified to have an adolescent between the ages of 9–14 years through mailed letters describing the purpose of GUTS and requesting permission to contact their children. In 1996, we mailed questionnaires to 13,261 females and 13,504 males whose mothers had granted consent; 9039 (68 % response rate) females and 7843 (58 % response rate) males returned completed questionnaires, thereby assenting to participate in GUTS1. Follow-up questionnaires were mailed annually in 1997 and 1998. A similar process was followed for recruitment in the GUTS2 cohort; 20,700 women in NHS II who had children aged 9–17 years were contacted and, in 2004, we mailed questionnaires to 8826 females and 8454 males whose mothers granted consent to contact their child. A total of 6002 (68 % response rate) females and 4918 (58 % response rate) males returned completed questionnaires, thereby assenting to participate in GUTS2. Follow-up questionnaires (on-line and mailed paper copies) were sent biannually in 2006 and 2008. This study was approved by the Human Subjects Committee at the Brigham and Women’s Hospital.
For this study, we restricted analysis to participants who were aged 14 or 15 years during the years when frequency of family dinners was assessed (1996–1998 for GUTS1, n = 11,131 observations for 7779 participants; and 2004–2008 for GUTS2, n = 8071 observations for 7987). This age range was selected as it was represented in each year that family dinner was assessed. We excluded observations with missing family dinner data (n = 1041 observations in GUTS1 and 86 observations in GUTS2) from these analyses, resulting in an analytic sample of 10,090 observations for 6889 participants in GUTS1 and 7985 observations for 7908 participants in GUTS2. In the analytic sample, 78 % had one family dinner observation and 22 % had more than one observation and 20 % had at least one sibling in the cohort.
Measures
Family dinner frequency
Family dinner frequency was measured in GUTS1 in 1996, 1997 and 1998 and in GUTS2 in 2004, 2006 and 2008 using the question: “How often do you sit down with other members of your family to eat dinner or supper?” In GUTS1, response categories were, “never,” “some days,” “most days,” and “every day.” In GUTS2, the responses were “never/almost never,” “1–2 times/week,” “3–4 times/week,” and “5 or more times/week.” For GUTS1, we coded the response option “never” as 0 times/week, “some days” as 3 times/week, “most days” as 5 times/week, and “everyday” as 7 times/week. For GUTS2, we coded the response option “never/almost never” as 0.07 times/week, “1–2 times/week” as 1.5 times/week, “3–4 times/week” as 3.5 times/week, and “5 or more times per week” as 6 times/week. To explore whether using different values to represent the response options in GUTS1 would influence our results, we also explored alternative values (never = 0, some days = 1, most days = 3.6, and every day = 7); analyses using these alternative values produced similar results (results not shown).
Household socioeconomic status
As part of the NHS II surveys, mothers reported their annual household income in 2001 and the education level of their partners in 1999.
Analysis
We stratified all analyses by sex as previous research found frequency of family dinners to differ among males and females [8].
To examine secular trends in family dinner among 14- and 15-year-olds in 1996 through 2008, we first calculated mean and standard deviation of the number of family dinners for each year. Then, to model family dinner frequency, we fit logistic models using generalized estimating equations with independence working correlation and empirical variance to account for correlation within individual and between siblings [11, 12]. The outcome was the number of family dinners per week, treated as a binomial random variable with n = 7 trials. We included a covariate indicating GUTS1 (1996–1998 vs. GUTS2 2004–2008), time (1996 coded as 0 to 2008 coded as 12), and the interaction of GUTS1 by time (to test linear trends for GUTS1 and GUTS2). The interaction term was not significant, indicating that the rate of change in family dinner frequency was similar across GUTS1 and GUTS2, so we removed it from the models. We fit models that additionally adjusted for household income and maternal partner’s education level; the effect estimate remained virtually unchanged (results not shown), so we report only estimates from the more parsimonious, unadjusted model. We conducted our analyses using SAS (version 9.3, SAS Institute Inc., Cary NC, USA).
We also explored the 12-year secular trends and rate of change stratified by annual household income (<$75,000/year and ≥$75,000/year) and maternal partner’s education level (less than a college education and a college education or higher) [8]. Results from these stratified models showed that the decrease in family meals and rate of change in family meals were similar across income and maternal partner education levels (results not shown).