SEARCH

SEARCH BY CITATION

Keywords:

  • adolescent;
  • overweight;
  • family dinner;
  • longitudinal;
  • race-ethnicity

Abstract

  1. Top of page
  2. Abstract
  3. Introduction
  4. Research Methods and Procedures
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

Objective: To explore associations between overweight status and the frequency of family dinners (FFD) for adolescents and how those associations differ across race and ethnicity.

Research Methods and Procedures: A sample of 5014 respondents between 12 and 15 years of age from the 1997 wave of the National Longitudinal Survey of Youth 1997 (NLSY97) was used. BMI was calculated using self-reported height and weight; 13.3% of respondents qualified as overweight, 16.4% qualified as at-risk-of-overweight, and 1.9% qualified as underweight. The remainder were normal weight. FFD was defined as the number of times respondents had dinner with their families in a typical week in the past year. Multinomial logistic regression models were estimated separately for non-Hispanic whites vs. blacks and Hispanics for odds of belonging to the other weight categories compared with normal weight. A supplementary longitudinal analysis estimated the odds of change in overweight status between 1997 and 2000.

Results: In 1997, the FFD distribution was as follows: 0, 8.3%; 1 or 2, 7.3%; 3 or 4, 13.4%; 5 or 6, 28.1%; 7, 42%. For whites, higher FFD was associated with reduced odds of being overweight in 1997, reduced odds of becoming overweight, and increased odds of ceasing to be overweight by 2000. No such associations were found for blacks and Hispanics.

Discussion: Reasons for racial and ethnic differences in the relationship between FFD and overweight may include differences in the types and portions of food consumed at family meals. More research is needed to verify this.


Introduction

  1. Top of page
  2. Abstract
  3. Introduction
  4. Research Methods and Procedures
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

Over the last 30 years, the prevalence of overweight and its associated health and psychosocial consequences has increased among all age groups in the United States. Among adolescents, the prevalence of overweight has increased from 6% in 1971–1974 to 16% in 1999–2002 (1). Growing alarm over the obesity epidemic has prompted increasing research into the etiology of obesity and the extent to which both genetics and modifiable social, environmental, familial, and behavioral factors play a role therein.

As overweight among adolescents has shown a rising trend, one phenomenon that has shown a corresponding declining trend is the frequency of children eating dinner with family (2). The concurrence of these trends raises the question of whether there exist associations between adolescent overweight and the frequency of family dinners.

Family meals may have relevance for the prevention and correction of childhood overweight. Regular family meals give parents the scope to provide their adolescent children with nutritious and healthy fare, to monitor and limit children's intake of calorically dense and junk food, and to serve as role models for healthy eating behavior. Regular family meals may help develop healthy eating habits that persist even when the adolescent is not eating with the family. Regular family meals may also provide structure, stability, and improve family communications, thereby protecting adolescents against feelings of isolation, depression, and other psychosocial problems that encourage turning to food as a refuge. Additionally, regular family meals may correlate with other protective factors against adolescent overweight—like better family connectedness—and may, to some extent, serve as a proxy for the beneficial influences of such factors. Finally, there is some possibility of bidirectional effects, in that adolescents who are not overweight may derive more enjoyment from family meals and, therefore, participate more regularly in them.

Extant research has found associations between frequency of family meals and improved dietary quality and eating behavior for adolescents. The frequency of family meals is positively related to healthy food choices and negatively related to the consumption of fried food, trans fat and saturated fat, and soda (3, 4). Adolescents express greater confidence in making healthful food choices when eating with families than otherwise (5), and family meal frequency is negatively related with prevalence of eating disorders among adolescent girls (6). On a related note, frequency of family meals is positively correlated with adolescents being well adjusted (7) and negatively correlated with low self-esteem, depressive symptoms, and suicide ideation (8). However, only one study to date by Taveras et al. (9) has comprehensively investigated the issue of how family dinners relate to adolescent overweight status. The study found that young respondents who report eating dinner with families all or most of the time were less likely to be overweight than counterparts who did so some or none of the time at the baseline, but there was no statistical relationship between family dinners and becoming overweight within 1 year. Although the above study makes a very valuable contribution to the literature, it arguably has a few limitations. The sample was non-representative: respondents were all offspring of registered nurses and were 90% white; overweight was defined as being >85th percentile of age- and sex-specific BMI, whereas it is more conventionally defined as being >95th percentile of age- and sex-specific BMI, with those between the 85th and 95th percentile typically categorized as at-risk-of-overweight (10, 11, 12); and the empirical models use limited controls for family socioeconomic status (SES)1 status, which is an important correlate of overweight in most Western societies (13). The question about regularity of family dinners is somewhat perplexing; respondents were asked to categorize their frequency of family dinner as none, some, most, and all of the time, but the reader was not provided with further guidelines as to what these categories should quantitatively indicate. For example, what is the demarcation between some and most? Or, how might most respondents categorize eating dinner four times a week? This potentially makes interpretation of the results somewhat difficult. Additionally, the longitudinal analysis in the above study considers changes in overweight status that occur within 1 year only. However, it is possible that influences of family dinner on body weight will sometimes take longer to manifest themselves; thus, considering changes in overweight status over a longer period of time is of value. Finally, the above study considered only the odds of becoming overweight (from non-overweight) over time, but it ignored other potential outcomes, such as ceasing to be overweight or remaining overweight, which are also of interest.

This article revisits the relationship between adolescent body weight and frequency of family dinner (FFD) using a large, nationally representative sample of adolescents that includes an oversampling of blacks and Hispanics. FFD is defined as the actual number of days in a typical week respondents dined with their families. Weight status definitions correspond to those provided by the Centers for Disease Control and Prevention. The empirical models control extensively for demographic, SES, and other characteristics. Cross-sectional models estimate the relationship between FFD and overweight status in the first survey year, and the longitudinal analysis studies the relationship between FFD and change in overweight status over a 3-year period. The analyses were done for the full sample and also repeated separately for non-Hispanic whites vs. blacks and Hispanics.

Research Methods and Procedures

  1. Top of page
  2. Abstract
  3. Introduction
  4. Research Methods and Procedures
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

Data and Methods

The data were drawn from the National Longitudinal Survey of Youth, 1997 (NLSY97). This annual survey, initiated in 1997, consisted of 6748 youth respondents who were a representative sample of the U.S. population who were 12 to 16 years of age on December 31, 1996. It was coupled with a supplemental oversample of 2236 black and Hispanic adolescents representative of the U.S. population of blacks and Hispanics of that same age range. The youth respondents were interviewed in each subsequent year since 1997, with a low attrition rate. In the first survey year only, a parent or parent figure was interviewed to provide supplementary information on the respondent's family background, parental education, and family income.

In surveys from 1997 to 2000, those youth respondents ≤14 years of age as of December 31, 1996, who were living with at least one parent/parent figure/legal guardian as of the survey date were asked about their household environment in a typical week. A total of 5325 respondents answered these questions in 1997. The respondents reported the number of days that their family ate dinner together in a typical week in the past year, as well as the number of days in a typical week that they participated in religious activities together and did something fun together.

All NLSY97 respondents were asked their height and weight in each survey year. I used this information to calculate their BMI. While self-reporting of height and weight by respondents leads to concerns about accuracy, a study based on National Health and Nutrition Examination Study III data (14) on adolescents 12 to 16 years of age found a high degree of correlation between self-reported and measured weight and height and that the use of self-reported weight and height resulted in the correct classification of weight status in 94% of children. Moreover, as long as inaccuracies in self-reported height and weight are not systematically correlated with reported FFD, it should not bias the estimates of interest.

Cross-sectional Analysis

The cross-sectional empirical model uses data from the 1997 survey year. This was the year when the largest number of respondents responded to questions about FFD and the only year when household income and parental education information was obtained by interviewing a parent. Respondents are assigned to one of four mutually exclusive and exhaustive categories based on how their BMI compared with the sex-specific BMI-for-age charts that are readily available from the Centers for Disease Control and Prevention's website. The weight categories follow the Centers for Disease Control and Prevention's prescription and are defined as follows: overweight includes respondents with BMI at or above the 95th percentile of age- and sex-specific BMI distribution, at-risk-of-overweight includes those with BMI between the 85th and 95th percentile, underweight includes those with BMI at or below the 5th percentile, and the remainder are categorized as normal-weight.

The empirical method used is that of multinomial logistic regression. This is a widely used method that is an extension of the standard binary logistic model and is designed to handle cases where the dependent variable can belong to more than two categories rather than just being dichotomous, as in the case of the standard binary logistic model (15). They are sometimes also referred to as polychotomous or polytomous logistic regression models. One of the categories is chosen to be the reference category, and the regression model simultaneously estimates the odds of belonging to each of the other categories relative to the reference category. I used normal-weight as the reference category or category 1. The other categories are underweight (category 2), at-risk-of-overweight (category 3), and overweight (category 4). Therefore, the ith respondent's odds of belonging to any other weight category j relative to being of normal weight is

  • image

The log-odds of belonging to any weight category j relative to that of normal weight is derived as

  • image

where FFD is by a vector of four binary variables denoting the number of days respondents dined with their families in a typical week in the past year: 1 or 2 days, 3 or 4 days, 5 or 6 days, or 7 days, with the basis for comparison being 0 or no days. Xi refers to the vector of additional control variables, which are described in greater detail below. The models are estimated using the statistical software STATA (StataCorp LP., College Station, TX).

Two model specifications are considered. For the first specification (Model A), Xi includes selected demographic characteristics and measures of physical development for the respondent. These are age in 1997 measured in actual years, dichotomous controls for sex (1 if male) and race-ethnicity (1 if black, and 1 if Hispanic origin), height in 1997 measured in inches, and a dichotomous control for attainment of puberty (1 if attained puberty by 1997). For female respondents, the NLSY97 defines attaining puberty as having started menarche; for male respondents, the NLSY97 defines puberty as initiation or completion of phenomena such as voice change, appearance of body hair, etc.).

In the second specification (Model B), Xi includes the controls in Model A and additionally includes controls for SES and a proxy for family connectedness. SES controls consist of dichotomous measures of family composition (1 if the respondent lives with both own parents), maternal education (1 if the respondent's mother is a college graduate), poverty status (1 if family income was below poverty level), and a computer in the household (1 if there is a computer in household). Family connectedness is proxied by the respondent's report of the number of days in a typical week in the past year that the family participated in fun activities together. All these factors may potentially influence the odds of adolescents being overweight and may also affect the frequency of family dinners. For example, families that are intact, economically better off, and well connected may be more likely to dine together. Thus, omitting these controls from the regression model may bias the estimates of the relationship between FFD and weight status. Arguably, however, causality may also run in the other direction: families that have a history of dining together may be more stable, thus increasing the likelihood of both parents being present and/or of economic well being. FFD may also promote family connectedness and thereby lead to more family fun activities. Therefore, the prudent approach is to estimate the models both without (Model A) and with (Model B) these controls and to compare the results from the two models to see whether the relationship between FFD and adolescent weight is robust to the inclusion of these additional control variables.

After excluding the respondents with missing information (i.e., those who refused to answer or answered don't know) on height, weight, FFD, and other key variables, the final sample consists of 5014 respondents between 12 and 15 years of age at the time of the 1997 interview. Table 1 presents the descriptive statistics for the full sample and by FFD. The means are not weighted to account for the oversampling of blacks and Hispanics in the NLSY97 and, thus, do not represent population means.

Table 1.  Descriptive statistics from the 1997 survey year by self-reported FFD in a typical week
 FFD = 0 (N = 418) (8.3%)FFD = 1 or 2 (N = 375) (7.3%)FFD = 3 or 4 (N = 682) (13.4%)FFD = 5 or 6 (N = 1433) (28.1%)FFD = 7 (N = 2106) (42.0%)Full sample N = 5014
  1. FFD, frequency of family dinners; SD, standard deviation. The means are not adjusted for the oversampling of minorities in National Longitudinal Survey of Youth 1997 and, hence, not representative of the population.

  2. FFD is the number of times respondents reported dining with their families in a typical week in the past year. The percentages in the column heading indicate percent of full sample. BMI is calculated using respondents’ self-reported height and weight. The weight categories follow the Centers for Disease Control and Prevention guidelines.

Mean BMI in 1997 (SD)22.16 (4.40)21.71 (4.43)21.72 (4.00)21.09 (4.02)21.34 (4.27)21.42 (4.20)
Percent at-risk-of-overweight17.215.717.115.516.716.4
Percent overweight17.413.014.611.113.513.3
Percent underweight1.61.60.82.12.21.9
Percent male44.846.948.953.755.452.5
Percent black44.931.229.019.424.125.7
Percent Hispanic24.920.021.018.220.621.2
Percent living with both own parents41.848.250.357.054.052.9
Percent in poverty26.814.417.113.217.016.4
Percent with computer in household40.452.050.758.949.551.8
Percent whose mother has college degree8.119.716.120.615.917.0
Percent attained puberty58.454.054.948.649.851.2
Mean height (in) (SD)63.89 (3.99)64.16 (4.20)64.03 (3.87)63.88 (3.97)63.47 (4.17)63.75 (4.06)
Mean age in 1997 (years) (SD)13.46 (0.95)13.56 (1.02)13.50 (0.92)13.37 (0.96)13.22 (0.92)13.33 (0.95)
Mean days family does fun activities together (SD)1.61 (2.22)1.57 (1.64)2.17 (1.69)2.65 (1.81)3.32 (2.13)2.71 (2.07)

Estimation results for Model A and Model B are presented in Table 2. Table 3 shows models estimated separately for non-Hispanic whites (sample size, 2736) and for blacks and Hispanics (sample size, 2278). The separate estimations are primarily motivated by the literature about differences in self-perceptions of overweight among racial and ethnic groups (16, 17), as well as differences in attitudes toward overweight (18, 19), which suggest that blacks, and to a lesser extent Hispanics, are less likely to perceive themselves as overweight and less likely to associate overweight with unattractiveness than whites. This arguably may lead to less motivation for preparing low-calorie food for family meals among blacks and Hispanics compared with their white counterparts, and, thus, the relationship between FFD and body weight may differ for non-Hispanic white respondents vs. others.2 Likelihood ratio tests also support estimating the models separately.

Table 2.  Results from multinomial logistic regression for weight status in 1997
  Model AModel B
 UnderweightAt-risk-of-overweightOverweightUnderweightAt-risk-of-overweightOverweight
  • OR, odds ratio; CI, confidence interval; FFD, frequency of family dinners. The results are from multivariate regression estimates. The reference category in the multinomial logistic model is those with normal weight. N = 5014.

  • *

    p < 0.05.

  • p < 0.10 but not p < 0.05.

 OR(95% CI)OR(95% CI)OR(95% CI)OR(95% CI)OR(95% CI)OR(95% CI)
FFD 1-20.89(0.29 to 2.69)0.89(0.60 to 1.30)0.77(0.51 to 1.15)0.94(0.31 to 2.88)0.93(0.63 to 1.37)0.81(0.54 to 1.22)
FFD 3-40.51(0.17 to 1.53)1.00(0.72 to 1.39)0.89(0.63 to 1.26)0.51(0.17 to 1.55)1.02(0.73 to 1.43)0.90(0.63 to 1.27)
FFD 5-61.09(0.47 to 2.55)0.85(0.62 to 1.15)0.66*(0.48 to 0.91)1.07(0.46 to 2.53)0.87(0.64 to 1.19)0.68*(0.49 to 0.95)
FFD 71.15(0.51 to 2.61)0.93(0.70 to 1.25)0.79(0.59 to 1.06)1.06(0.46 to 2.44)0.93(0.69 to 1.25)0.78(0.57 to 1.05)
Male1.20(0.75 to 1.91)1.66*(1.38 to 1.99)2.02*(1.65 to 2.48)1.20(0.75 to 1.90)1.65*(1.37 to 1.99)2.01*(1.64 to 2.47)
Black1.21(0.75 to 1.94)1.24*(1.04 to 1.51)2.09*(1.72 to 2.54)1.01(0.60 to 1.70)1.12(0.92 to 1.36)1.85*(1.50 to 2.28)
Hispanic0.70(0.38 to 1.28)1.40*(1.15 to 1.51)1.83*(1.48 to 2.27)0.61(0.32 to 1.16)1.21(0.98 to 1.48)1.59*(1.27 to 1.99)
Age (in 1997)0.81(0.67 to 1.08)0.87*(1.15 to 1.70)0.82(0.83 to 1.02)0.86(0.92 to 1.02)0.87*(0.80 to 0.96)0.92(0.83 to 1.01)
Attained puberty0.52*(0.32 to 0.83)1.22*(0.79 to 0.95)1.10(0.90 to 1.35)0.52*(0.32 to 0.83)1.23*(1.02 to 1.47)1.08(0.89 to 1.33)
Height0.98(0.92 to 1.02)1.03*(1.01 to 1.47)1.01(0.98 to 1.03)0.97(0.92 to 1.02)1.03*(1.01 to 1.06)1.02(0.98 to 1.03)
Lives with own parents      0.89(0.57 to 1.37)1.09(0.92 to 1.29)1.01(0.84 to 1.20)
In poverty      1.83*(1.07 to 3.10)1.12(0.90 to 1.38)1.11(0.89 to 1.40)
Mother is college graduate      0.95(0.54 to 1.66)0.89(0.65 to 1.03)0.70*(0.53 to 0.93)
Computer in household      1.11(0.70 to 1.73)0.73*(0.62 to 0.87)0.82*(0.67 to 0.98)
Days of family fun      1.07(0.97 to 1.18)1.01(0.96 to 1.04)1.02(0.97 to 1.06)
Log likelihood−4398.8     −4360.9     
Table 3.  Results from multinomial logistic regression for weight status in 1997 by race
  Model AModel B
  • OR, odds ratio; CI, confidence interval; FFD, frequency of family dinner. The results are from multivariate regression estimates. The reference category in the multinomial logistic model is those with normal weight. N = 2736 for non-Hispanic whites; N = 2278 for blacks and Hispanics.

  • *

    p < 0.05.

  • p < 0.10 but not p < 0.05.

 UnderweightAt-risk-of-overweightOverweightUnderweightAt-risk-of-overweightOverweight
 OR(95% CI)OR(95% CI)OR(95% CI)OR(95% CI)OR(95% CI)OR(95% CI)
Non-Hispanic whites            
 FFD 1-20.86(0.18 to 3.96)0.99(0.52 to 1.87)0.61(0.26 to 1.02)0.86(0.18 to 4.01)1.05(0.55 to 1.99)0.56(0.28 to 1.13)
 FFD 3-40.58(0.13 to 2.47)0.91(0.51 to 1.63)0.54*(0.30 to 0.98)0.59(0.13 to 2.50)0.97(0.54 to 1.75)0.59(0.32 to 1.08)
 FFD 5-60.80(0.23 to 2.75)0.73(0.43 to 1.24)0.34*(0.19 to 0.59)0.77(0.22 to 2.68)0.80(0.46 to 1.37)0.39*(0.22 to 0.67)
 FFD 70.75(0.22 to 2.56)0.85(0.51 to 1.44)0.51*(0.30 to 0.85)0.68(0.19 to 2.37)0.90(0.52 to 1.54)0.55*(0.32 to 0.94)
 Log likelihood2716.6     2149.6     
Blacks and Hispanics            
 FFD 1-20.70(0.12–3.88)0.79(0.47 to 1.33)0.88(0.53 to 1.45)0.82(0.14 to 4.62)0.82(0.48 to 1.38)0.91(0.55 to 1.51)
 FFD 3-40.21(0.02 to 1.92)1.10(0.72 to 1.67)1.10(0.73 to 1.69)0.32(0.02 to 2.02)1.10(0.72 to 1.67)1.08(0.71 to 1.65)
 FFD 5-61.24(0.38 to 4.05)0.97(0.66 to 1.43)0.93(0.63 to 1.37)1.36(0.41 to 4.49)0.94(0.64 to 1.41)0.92(0.62 to 1.36)
 FFD 71.46(0.49 to 4.34)1.02(0.71 to 1.45)0.92(0.64 to 1.32)1.45(0.47 to 4.44)0.96(0.69 to 1.40)0.88(0.60 to 1.27)
 Log likelihood2274.2     2168.3     

Longitudinal Analysis

In a supplementary longitudinal analysis, I studied how change in overweight status between the baseline survey year of 1997 and survey year 2000 associated with the reported FFD for the interim years. The 2000 survey is the last one that includes questions about FFD.

This analysis uses the subsample of respondents who qualified to answer questions on family environment in 1997, were not lost to attrition by year 2000, did not have missing information of height, weight, or FFD in 1997 or 2000, and were <18 years of age at time of the 2000 survey.3 The final sample contains 2089 non-Hispanic white respondents and 1685 black or Hispanic respondents. BMI for the 2000 was calculated based on the respondent's reported height and weight at that survey year, and respondents were assigned to weight categories using the Centers for Disease Control and Prevention's age- and sex-specific BMI charts and the same cut-offs as in the cross-sectional analysis. However, the focus now was specifically on longitudinal changes (or lack of change) in overweight status. This section extends on the longitudinal model of Taveras et al. (9) (which only considered moving into overweight vs. not) by considering moving into overweight, as well as ceasing to be overweight, and remaining overweight. Once again, a multinomial logistic regression framework was used, with the dependent variable belonging to one of four mutually exclusive and exhaustive categories. This time, the reference category used is not being overweight either in 1997 or in 2000, and the three other categories were moving into overweight (not overweight in 1997 survey, overweight in 2000 survey), remaining overweight (overweight in 1997 survey, overweight in 2000 survey), and ceasing to be overweight (overweight in 1997 survey, not overweight in 2000 survey).4

For the longitudinal analysis, I took the most generalized approach toward measuring FFD in the interim years by including separate vectors of binary variables for FFD for each of 1998, 1999, and 2000. The basis for comparison was dinner on zero or no days in a typical week for all years. Although this approach may pose some multicollinearity problems, it is preferred over a priori imposing functional form restrictions on the FFD variables, as well as over a priori leaving out FFD from one of the years, and potentially creating a situation for omitted variable bias.5 The other control variables are respondent's sex (1 if male); age, height, and family composition (1 if the respondent lives with both own parents) in 2000; poverty status and maternal education from 19976; and, finally, a vector of binary variables for FFD from the baseline year of 1997.7 The analysis was again done separately for non-Hispanic whites and blacks and Hispanics. Table 4 presents these results.

Table 4.  Results from multinomial logistic regressions for change in overweight status between 1997 and 2000 by race
 Non-Hispanic WhitesBlacks and Hispanics
  • OR, odds ratio; CI, confidence interval; FFD, frequency of family dinner. Additional controls are included for sex, age, height, and family structure in 2000, poverty status and maternal education in 1997, and a vector of binary controls for FFD in 1997. The reference category is not overweight in 1997 or 2000. N = 2089 for non-Hispanic whites; 1685 for blacks and Hispanics.

  • *

    p < 0.05.

  • p < 0.10 but not p < 0.05.

 Moving into overweight (3.8% of sample)Remaining overweight (6.2% of sample)Ceasing to be overweight (3.7% of sample)Moving into overweight (5.9% of sample)Remaining overweight (10.6% of sample)Ceasing to be overweight (6.6% of sample)
 OR(95% CI)OR(95% CI)OR(95% CI)OR(95% CI)OR(95% CI)OR(95% CI)
1998 FFD 1-21.20(0.30 to 4.73)1.00(0.37 to 2.66)1.56(0.41 to 5.95)1.88(0.36 to 2.37)0.96(0.49 to 1.86)0.23*(0.07 to 0.64)
1998 FFD 3-41.61(0.47 to 5.52)0.80(0.32 to 2.00)1.68(0.46 to 3.12)1.52(0.34 to 2.14)0.72(0.38 to 1.36)0.64(0.31 to 1.28)
1998 FFD 5-61.83(0.54 to 6.11)1.00(0.37 to 2.16)1.87(0.51 to 3.70)1.83(0.37 to 2.39)0.92(0.49 to 1.70)0.50*(0.23 to 1.00)
1998 FFD 71.79(0.52 to 6.13)1.10(0.32 to 2.00)2.74*(1.32 to 6.99)1.42(0.36 to 2.51)0.98(0.53 to 1.78)0.75(0.37 to 1.48)
1999 FFD 1-20.39(0.14 to 1.11)0.72(0.41 to 2.46)1.17(0.37 to 3.65)0.93(0.20 to 1.20)0.62(0.30 to 1.26)0.85(0.35 to 2.07)
1999 FFD 3-40.34*(0.11 to 0.98)0.83(0.44 to 2.76)0.79(0.24 to 2.60)0.86(0.47 to 2.35)0.78(0.40 to 1.50)1.17(0.51 to 2.68)
1999 FFD 5-60.20*(0.06 to 0.59)0.79(0.27 to 1.89)0.84(0.13 to 1.71)0.94(0.47 to 2.44)0.67(0.33 to 1.35)0.89(0.36 to 2.16)
1999 FFD 70.38(0.12 to 1.07)0.54(0.18 to 1.51)0.47(0.39 to 2.06)0.96(0.34 to 2.12)0.65(0.31 to 1.33)1.54(0.63 to 3.73)
2000 FFD 1-20.98(0.25 to 2.67)1.53(0.67 to 3.46)0.90(0.39 to 2.06)0.49(0.20 to 1.20)1.30(0.65 to 2.59)1.11(0.54 to 2.27)
2000 FFD 3-41.08(0.40 to 2.90)1.47(0.65 to 2.32)0.41(0.16 to 1.15)1.05(0.47 to 2.35)2.24*(1.14 to 4.37)0.72(0.33 to 1.54)
2000 FFD 5-60.94(0.33 to 2.63)1.30(0.55 to 3.01)0.66(0.26 to 1.63)1.06(0.46 to 2.44)1.65(0.80 to 3.37)0.74(0.33 to 1.61)
2000 FFD 72.30(0.79 to 5.62)1.62(0.63 to 4.11)0.67(0.24 to 1.88)0.84(0.33 to 2.12)1.54(0.72 to 3.30)0.42*(0.17 to 0.99)
Log likelihood−1037.32     −1206.18     

Results

  1. Top of page
  2. Abstract
  3. Introduction
  4. Research Methods and Procedures
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

A little more than 8% of the final sample reported never having dinner with family on a typical week in the past year, and approximately 20.7% reported having dinner four or fewer times. Approximately 42% of the sample reported having dinner with their families every day in a typical week. Table 1 shows the associations between FFD and the variables when other factors are not controlled for. Mean BMI and percentage overweight were higher among respondents reporting FFD of zero than any of the other groups. Poverty rate was also higher and maternal education was lower among this group than any of the other groups. Respondents reporting FFD of four or less had proportionally larger shares of girls, blacks, those not living with both of their own parents, and those without a computer in the household than respondents reporting FFD of five or more.

Results using the pooled sample (Table 2) showed that FFD was not significantly related to the odds of underweight or at-risk-of-overweight relative to normal weight. However, FFD showed some associations with the odds of being overweight relative to normal weight. Specifically, compared with those reporting FFD of 0, those reporting FFD of 5 or 6, on average, had 32% to 34% lower odds of overweight [odds ratio (OR) 0.66; 95% confidence interval (CI), 0.48 to 0.91 in Model A; OR, 0.68; 95% CI, 0.49 to 0.95 in Model B]. Those reporting FFD of 7 had 21% to 22% odds of overweight on average, but this association was only weakly significant, with a p value of 0.10 in both models.

Among the control variables in Model A, being male, Hispanic, and black were associated with increased odds, and age was associated with reduced odds of risk-of-overweight and overweight relative to normal weight. Attaining puberty was associated with increased odds of risk of overweight and underweight but not of overweight. Among the additional controls included in Model B, maternal college education and a computer in the household were associated with reduced odds of overweight and risk-of-overweight, and poverty was associated with increased odds of underweight. Readers are invited to inspect for themselves the magnitudes of these results in Table 2.

Table 3 presents estimation results separately by race-ethnicity. For economy of space, only the estimates pertaining to FFD are shown, with estimates for the other controls available from the author on request. It is now seen that the negative associations between FFD and the odds of being overweight (relative to normal weight) are confined to non-Hispanic whites only. In Model A, compared with those reporting FFD of 0, the odds of overweight are, on average, 46% lower for those reporting FFD of 3 or 4 (OR, 0.54; 95% CI, 0.30 to 0.98), 66% lower for those reporting FFD of 5 or 6 (OR, 0.34; 95% CI, 0.19 to 0.59), and 49% lower for those reporting FFD of 7 (OR, 0.51; 95% CI, 0.30 to 0.85). Although the odds of overweight are, on average, 39% for those reporting FFD of just 1 or 2 compared with those reporting FFD of 0, these results are only weakly significant (p = 0.09). In Model B, the results remain quite similar. Compared with those reporting FFD of 0, those with FFD of 5 or 6 and 7, respectively, had 61% and 45% lower odds of being overweight (OR, 0.39; 95% CI, 0.22 to 0.67 and OR, 0.55; 95% CI, 0.32 to 0.94, respectively). Those with FFD of 3 or 4 also had 41% lower odds, on average, of being overweight, but now the relationship is found to be only weakly significant (p = 0.08). Results for FFD of 1 or 2 were no longer significant, even at the 10% level.

For whites, there was no statistical association between FFD and being underweight or risk-of-overweight relative to normal weight. For blacks and Hispanics, there was no statistical association between FFD and the odds of belonging to any of the other weight categories relative to normal weight.

As part of robustness checks, the multinomial logistic models were re-estimated after the reference category was redefined to include normal weight and underweight. Thus, the new model showed how FFD related to the odds of being risk of overweight or of being overweight relative to being either normal weight or underweight. Results remained virtually identical to those described above and are available from the author on request.

The results for the longitudinal analysis (Table 4) show that, for non-Hispanic white respondents, there are negative associations between FFD in 1999 and moving into overweight by 2000. Compared with those with FFD of 0, those with FFD of 3 or 4 and FFD of 5 or 6 in 1999 had, on average, 66% (OR, 0.34; 95% CI, 0.11 to 0.98) and 80% (OR, 0.20; 95% CI, 0.06 to 0.59) lower odds of moving into overweight. Those with FFD of 7 in 1999 had, on average, 62% lower odds of moving into overweight, but this result is only weakly significant (p = 0.08). Additionally, compared with those reporting FFD of 0, those reporting FFD of 7 in 1998 showed extremely large (174%) increases in the odds of ceasing to be overweight by 2000 (OR, 2.74; 95% CI, 1.32 to 6.99). No similar relationships were found for blacks and Hispanics. In fact, for these groups, higher FFD occasionally seems to have some disadvantageous associations with overweight status. Specifically, compared with those reporting FFD of 0, those reporting FFD of 5 or 6 in 1998 seemed to have ∼50% lower odds of ceasing to be overweight (OR, 0.50; 95% CI, 0.23 to 1.00), and those reporting FFD of 7 in 2000 had ∼58% lower odds of ceasing to be overweight (OR, 0.42; 95% CI, 0.17 to 0.99). Those reporting FFD of 3 or 4 (compared with FFD of 0) in 2000 also had very substantially increased odds of remaining overweight in 2000 (OR, 2.24; 95% CI, 1.14 to 4.37).

Discussion

  1. Top of page
  2. Abstract
  3. Introduction
  4. Research Methods and Procedures
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

Eating meals with the family relates to numerous positive outcome associations for adolescents, including improved academic achievement, reduced drug use (1, 8), reduced eating disorders (6), lower risk of premature sexual activity, and lower risk of depression and suicide ideation (8). This article contributes to the literature by exploring how frequency of family meals relates to adolescent overweight.

Cross-sectional results show that higher frequency of dining with family is associated with reductions in the odds of being overweight (compared with normal weight) for non-Hispanic white adolescents, but no associations were found for black and Hispanic adolescents. There are no statistical relationships between frequency of family dinners and the odds of being at risk-of-overweight or being underweight for either racial group. Notably, for white adolescents, the relationship between family meals and the odds of overweight do not follow a monotonic pattern. Specifically, those who report dining with their families 5 or 6 times in a typical week have ∼66% lower odds of being overweight than those who report never dining with families, but those who report dining all 7 days in a typical week have only 49% lower odds.

Longitudinal analyses for changes in overweight status between 1997 and 2000 also showed some correlations between frequency of family dinners, reduced odds of becoming overweight, and increased odds of ceasing to be overweight for non-Hispanic white adolescents. Specifically, more frequent family dinners in 1999 were statistically associated with lower odds of becoming overweight by 2000, and more frequent family dinners in 1998 were statistically associated with higher odds of ceasing to be overweight by 2000. It should be noted that this analysis only reveals the cumulative change in overweight status between the 1997 and 2000 surveys and does not inform on precisely when within that period the change in overweight status occurred, or, indeed, whether multiple changes occurred within that period. Thus, the associations between FFD of 1998 and 1999 and observed change in overweight status by 2000 may partly derive from weight changes occurring before 2000, but they may also partly derive from healthy eating habits acquired through regular family meals in the earlier years, whose benefits appear a year or two later. No similar results were found for black and Hispanic adolescents. In fact, for these groups, it seems that higher frequency of family dining was occasionally associated with increased odds of remaining overweight and reduced odds of ceasing to be overweight by the 2000 survey.

This study has several advantages. It is able to use a nationally representative longitudinal dataset on adolescents; the oversampling of blacks and Hispanics in the data permits separate analyses by race/ethnicity. The questions regarding frequency of dinners with family are well defined and allow for easy interpretation of results. Extensive controls are available for demographic, physical, and familial characteristics. The longitudinal nature of the data permits inspection of change in overweight status over a 3-year period, which has not been attempted before.

This study also has certain limitations. These primarily include the self-reported nature of the information in height and weight, as well as the lack of information on participation in organized sports and physical activity, and the lack of information on actual types of food consumed either at family dinners or otherwise.8

Additionally, this study did not investigate whether the relationship between FFD and adolescent weight varies by sex and age. That empirical analysis was judged to be beyond the scope of this study but should be studied in future work.

Two aspects of the findings of this study particularly warrant further analysis. The first aspect is why higher frequencies of family dinners do not correlate to reduced odds of overweight for blacks and Hispanics, although they do for non-Hispanic whites. Possible reasons may include racial and ethnic differences in the caloric content of food and/or differences in portion sizes of food typically consumed during family dinners. Meals with the family do not automatically preclude consumption of fast food, takeout food, or other unhealthy fare at those meals, and this might occur more frequently for blacks and Hispanics than whites. For instance, among participants in the Coronary Artery Risk Development in Young Adults study, black participants had a higher frequency of visits to fast food places than whites (20), although we do not know whether these visits occurred as a family. Among a sample of adolescents in Minnesota, respondents who frequented fast food restaurants were more likely to be non-white than white and were also more likely to have unhealthy food available at home (21). It also has been argued that blacks and Hispanics are less likely than whites to reside in areas where fresh produce is available for purchase in supermarkets (22). All this provides grounds to speculate that perhaps black and Hispanic adolescents consume more fast food and processed food at family meals than do their white counterparts. Moreover, the Coronary Artery Risk Development in Young Adults study (19) found that changes in fast food frequency over 15 years were associated with changes in body weight for white people but much less so for black people, which suggests that what black people are consuming when not consuming fast food also has higher caloric content than what their white counterparts are consuming. Thus, more research is called for to verify the types of food and portion sizes consumed at family meals by black and Hispanic adolescents vs. white adolescents.

The second aspect, although arguably less important, is why the relationship between frequency of family dinners and the odds of overweight among non-Hispanic whites is non-monotonic. Specifically, why are the odds of overweight lower for respondents reporting FFD of five or six in a typical week than those reporting FFD of seven. Again, the answer may relate to the frequency of consuming fast food, takeout food, or processed food at those family dinners, and it may be speculated that white families who dine together every day during a typical week have a greater propensity for consuming such food during a family meal than families who dine together on most, but not all, days. Another speculation is that frequency of family meals may correlate with levels of physical activity, whereby respondents who report somewhat lower FFD than seven may participate in organized sports or other physically demanding activities in lieu of time for family meals. These activities may more than compensate for any deleterious effects that the few missed family meals might have on overweight status. Finally, it may be speculated that FFD has a non-linear relationship with family income; families where the parent(s) have particularly demanding but lucrative careers may not be able to dine together on all nights in a typical week, but the higher parental income may increase the ability to purchase healthful foods and provide opportunities for physical activity for the children. Higher parental income may also correlate to greater awareness of weight issues and greater social pressure to avoid being overweight. Further research is required to verify whether the above speculations have any merit.

In conclusion, this study shows that the relationship between family meals and adolescent overweight are complex and cautions against assuming that more frequent family meals are a panacea for adolescent overweight problems. Particularly, while dining more frequently with families per se may provide some protection against overweight for non-Hispanic white adolescents, there is no evidence in this study to support that it does so for their black and Hispanic counterparts.

Footnotes
  • 1

    Nonstandard abbreviations: SES, socioeconomic status; FFD, frequent family dinner; NLSY97, National Longitudinal Survey of Youth 1997; OR, odds ratio; CI, confidence interval.

  • 2

    The NLSY97 does not provide detailed information on race-ethnicity other than black and Hispanic. A very small fraction of respondents classify themselves as “other”. They are included with non-Hispanic whites and are omitted from the sample as part of specification tests. Their presence or absence does not noticeably impact results.

  • 3

    Most 18-year olds had left home for college or work in the 2000 survey. A few were married.

  • 4

    Because there are 4 weight categories in 1997, conceivably, one could have a 16-category model, with moving into, remaining in, ceasing to be, and never being for each of overweight, at-risk-of-overweight, underweight, and normal weight. However, a 16-category multinomial logistic model is impractical to estimate because of statistical power issues; it is also not necessary because the main issue of interest is adolescent overweight. Therefore, respondents who were not overweight in either 1997 or 2000 are treated as the reference category, even if some of them moved in the other categories—for example, went from being underweight to normal-weight.

  • 5

    The correlation coefficient between FFD of 1998 and 1999 is 0.52, FFD of 1999 and 2000 is 0.57, and FFD of 1998 and 2000 is 0.43. The significance level is better than 1% in all cases. Thus, arbitrarily excluding any one year's FFD can bias the estimates of the other years’ FFD.

  • 6

    Maternal education is not updated after the 1997 survey, nor is poverty status, unless respondents move out and form their own households.

  • 7

    A reviewer correctly pointed out that including the 1997 FFD allows us to interpret results as the relationship between change in overweight status from the baseline year and change in FFD from the baseline year.

  • 8

    The NLSY97 includes one question on days of exercise and one question on days of consumption of some fruits or vegetables, and they are only asked in the first survey year to the subsample 12 to 13 years of age.

References

  1. Top of page
  2. Abstract
  3. Introduction
  4. Research Methods and Procedures
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References