Article Text
PDF
Abstract
Objectives Parents are believed to have a strong influence on children's eating behaviours. However, previous findings on child–parent resemblance in dietary intakes are mixed. We systematically reviewed and meta-analysed the association (correlations) based on published studies.
Methods We searched related studies published since 1980 and found 24 studies meeting inclusion criteria for review and 15 for meta-regression analysis. We compared the associations between parent–child pairs, nutrients, over time and by dietary assessment method.
Results Most studies were based on small samples. Overall, they suggest a moderate or weak association, but findings varied remarkably. Our meta-analysis showed that average Fisher's transformed correlations were 0.20 (95% CI 0.13 to 0.28) for fat (% energy); for energy, 0.21 (0.18 to 0.24). The correlations varied by parent–child pairs, dietary assessment and countries. Food frequency questionnaires or mixed approaches yielded lower correlation than 24-h recalls or food records. Child self-reported intakes showed weaker correlation and better methodology quality showed stronger correlation in fat intake (% energy), which also became weaker over time.
Conclusions Overall, the resemblance is weak, and it varied considerably across studies, nutrients, foods and parent–child pairs.
- Child
- parent
- diet
- resemblance
- association
- review
- meta-analysis
- child health
- international health
- nutrition
Statistics from Altmetric.com
It has been widely believed that parents have a strong influence on children's eating behaviours. Parents are gatekeepers and can serve as role models for their children's health-related behaviours.1 2 Consequently, there is a widespread perception of a strong parent–child association in dietary intakes.3–10 Surprisingly, some studies show that the association is very weak.11–13 This is likely because young people's eating patterns are in fact influenced by many complex factors, and the family environment plays only a partial role.14–16 For example, most children, especially in industrialised countries, consume at least one meal at school. Moreover, eating snack foods is common practice among children and adults, in particular, in industrialised countries.17 18
Furthermore, as children grow older, they become subjected to stronger peer influence and acquire greater autonomy when making food choices. Therefore, it will be of interest to systematically examine and quantify the parent–child association in dietary intakes. To our knowledge, no such effort has been made. Another related question is whether the association has become weaker over time because of many changes in the society. For instance, some of those changes may include growing independence of children, changes in home and social environments, parenting styles, growing proportion of working mothers,19 changes in food supply and distribution 15 20 21 as well as modifications in people's dietary intake.22–25
The present study aimed to systematically review and assess the degree of association and similarity between children's and their parents' dietary intake based on studies published since 1980. Further, we compared the differences in the association between parent–child pairs, countries and over time. We also tested the differences in the association detected across dietary assessment approaches. We hypothesised that studies using food frequency questionnaires (FFQ) were likely to report weaker associations given their better estimation of usual dietary intake but being less quantitative26 27 compared to 24-h recalls or multiple food records. To our knowledge, this is the first such systematic investigation to address these questions. Findings of this study will help enhance our understanding of the factors that may affect children's dietary intake patterns and provide useful insights for developing effective intervention programs to promote healthy eating in young people.
Methods
Literature search strategy
We first searched the PubMed database using related Medical Subject Headings (MESH) key words, including child, adolescent, family, parent, mother, father, diet and food for studies published between 1 January 1980 and 31 September 2009. Then using similar approaches, we conducted separate searches for studies published in Spanish and Chinese using other database that we have access to (see below). We could not examine studies published in other languages.
The titles and abstracts of the related studies were examined on screen first for exclusion and inclusion. Papers that could not be excluded on the basis of the abstracts were obtained in full and reviewed for suitability for inclusion. Only studies that have provided results (eg, correlation coefficients, intraclass correlation coefficients) about the association between children's and their parents' dietary intakes (including energy, nutrients and food groups) were included. Studies that aimed at correlating dietary intake of children with their parents' knowledge and attitudes about dietary intake were excluded. There was no restriction regarding children's age. Then, the full papers that met our selection criteria were carefully reviewed.
Our literature search resulted in a total of 24 papers3–13 28–40 finally included in the review and 15 in our meta-analysis (which reported correlations for intakes of total energy and fat). Most of these papers were obtained through initial screening, while some were identified by using the “Related Article” feature on Pubmed and reference lists in the selected articles. In addition, some studies identified in the course of reading or brought to our attention by colleagues and experts consulted were included.
Note that our search of PubMed using different combinations of the related MESH key words yielded 2590 related abstracts, but only 23 studies met our inclusion criteria. Our search of Embase using the related key words identified 455 related abstracts, but only one met our inclusion criteria for systematic review; our search of Índice Médico Español identified 15 related abstracts, but none met our inclusion criteria. These are two comprehensive databases of biomedical research published in Spanish. We searched the Chinese National Knowledge Infrastructure, the most comprehensive database of biomedical research published in China, for papers published in Chinese, identified 344 related abstracts, but none met our inclusion criteria.
Data extraction
Using a standardised data extraction form, we extracted and tabulated the related data. Information extracted included first author's name, study publication year, country of data collection, sample characteristics (eg, age, sex, ethnicity/race of the participants, sample size, socioeconomic status (SES)), type of parent–child pairs, methods of dietary assessment, main research findings, whether and what confounders and covariates were controlled for and main strengths and limitations. Some studies did not provide all the related details.
Most studies reported correlation coefficients and computed Pearson or Spearman rank correlation coefficients. Our meta-regression analysis (simply termed “meta-analysis” in our study) was conducted based on the reported correlation coefficients (used as “data points”) and those that could be approximately interpreted as correlation coefficients such as intra-class correlation coefficients. Each of these correlation coefficients were reported for a specific parent–child pair, and data points were classified according to sex of parent, child or both for each of these coefficients. The 15 studies provided3–12 28 31 33 36 117 data points on energy and fat intakes and for different types of parent–child pairs (ie, parent–child, parent–son, parent–daughter, father–son (FS), father–daughter (FD), mother–son (MS), mother–daughter (MD)). Note that we selected energy because it helps reflect overall diet and affects people's risk of obesity, which has become a global epidemic. We selected fat intake, both as absolute (in grams) and relative intake (ie, % of energy derived from fat), to reflex dietary composition and quality, and fat intake affects the risk of a number of chronic diseases such as cardiovascular disease. The other nutrients were not included also because of their small number of data points.
Assessment of methodological quality of each study
This was assessed systematically (see appendix A). Each study was assigned seven subscores (from one worst to three best for each category) for different components (sample size, age range of children, dietary assessment, nutrients and foods reported, types of parent–child dyads, adjustment for potential confounders and representativeness of sample) and a total score. The total score ranged between 0 and 21.
The scores for each study were assessed by two coauthors (raters) in a blinded manner, and the agreement between their scores was tested thereafter. When discrepancies were found, a third rater scored the study considering the other two raters' assessment. Next, they had a discussion to reach a consensus. The correlation between the initial scores assigned by the first two raters was 0.88, and their average difference was 0.6. The correlations between the total scores assigned by the third rater and one of the first two raters were 0.94 and 0.95, and their average differences were 0.1 and 0.4, respectively. Based on the median of the final total scores, the studies were considered as good versus less good quality studies. This binary variable was added in our meta-analysis to assess its impact on the correlations.
Statistical analysis
We conducted a set of analyses. The mutually exclusive specific categories of parent–child pair (eg, mother–daughter) were considered in some, while in others another more general classification, termed “non-exclusive”, was used (eg, mother–child includes mother–child, mother–son and mother–daughter data points). We first estimated the arithmetic mean, standard deviation and range of the correlation coefficients by the non-exclusive categories of parent–child pairs. Correlations of 0.10 to <0.30 are considered to be weak; r=0.30 to <0.50 as moderate; while r≥0.50 indicates a strong association.
Next, we tested the differences in correlation coefficients between nutrients and within each parent–child pair using analysis of variance (ANOVA) and pairwise t tests. Statistical significance of the correlation coefficients was considered as well; the proportion significant across nutrients and within parent–child pairs was compared using χ2 test.
Moreover, using the Fisher's z transformation, we converted Pearson's or Spearman's correlation coefficients to z's to obtain approximate normality and then calculated a mean transformed correlation weighted by the sample sizes in the studies,41 with its 95% CI, taking into account heterogeneity between data points, and thus using random effect estimates. Subsequent in-depth analyses, however, focused mainly on Fisher's z transformed correlation coefficients.
Considering the small sample size and the expected heterogeneity of these studies (eg, large variations in people's dietary intakes, study design and dietary assessment approaches), we conducted meta-regression analyses with random effects and added a limited number of predictors, using Fisher's z transformed correlation with their corresponding standard errors as the outcome variable. Three models were conducted: model 1 included study-level characteristics (such as publication year, study country and methods of dietary assessment, overall study quality) as the predictors; models 2 and 3 forced parent–child dyads to be additional predictors. Models 2 and 3 were reduced, given the limited sample size, by backward elimination of study-level predictors with p<0.10.
Taking the average of Fisher's z transformed correlations and their corresponding average standard errors for each study, we tested homogeneity of the effect size between studies using Q test, which indicated heterogeneity across studies included in our meta-analysis (Q=81.93, df=14; p<0.001). Finally, we assessed publication bias through Begg's funnel plot by plotting Fisher's transformed correlation values against their standard errors and by the Begg's adjusted rank correlation test.42 43 Our analysis indicated no publication bias (appendix B). All analysis was performed using STATA release 11,44 particularly using the meta, metabias and metareg commands for the meta-analyses among others. Statistical significance was set at p<0.05.
Results
Findings of systematic review
Table 1 presents the detailed characteristics and main findings of the 24 studies we identified that reported parent–child associations in dietary intakes. The studies differed considerably in their design, study samples, settings and findings and used various methods such as 24-h recalls, dietary records, FFQ or a mixed approach for dietary assessment. Very few of them are based on large, national samples, and over half were conducted in the USA.
Main characteristics and findings of relevant studies that examined the association between children's and their parents' dietary intakes†
Overall, they suggest moderate to weak correlation coefficients, though findings varied remarkably across studies and nutrients. For example, in a recent study, based on nationally representative US data with diet being assessed using two 24-h recalls in the mid-1990s, we examined a set of dietary intake measures including an overall dietary quality index score (the Healthy Eating Index), which measured the overall dietary pattern and quality. This is the largest available study (included 2692 child–parent pairs) that assessed the familiar association in dietary intakes. The parent–child correlations were weak or moderate (correlations were ranged 0.20–0.33 for key dietary measures such as the dietary quality score and total energy intake) for various parent–child pairs, but the resemblance was stronger in some groups for some intake measures (eg, 0.56 for total fat intake as grams in the “other ethnic group”).28 In addition, the resemblance was stronger for younger children (<10 years of age) than their older counterparts (>10 years of age) in terms of overall dietary quality. In another Chicago-based study, we assessed the resemblance in 121 pairs of urban low-income African-American adolescents and their mothers.33 Their dietary intakes were assessed using two similar FFQ developed by Harvard University, one for adults and one for children. None of the mother–son correlations for nutrients and food groups were greater than 0.20. Mother–daughter pairs had stronger correlations (0.26 for energy and 0.30 for fat).
The second largest study was conducted among 1077 households in The Netherlands, based on 2-day dietary records collected in the 1987 and 1992 national surveys.11 The study also showed weak to moderate parent–child correlations, although there were considerable variations in the correlations, ranged from 0.55 for mother–daughter pairs' cholesterol intake to 0.09 for mother–son pairs' energy intake.
Table 2 summarises the characteristics of the studies included in our review and meta-analysis. Only the 15 studies that reported parent–child correlations in energy and fat intakes were included in our meta-analysis. They provided a total of 117 data points (ie, correlation values; 45 for energy, 45 for total fat and 27 for fat as % energy).
Summary of the characteristics of included studies and data points (ie, correlation coefficients)
Most characteristics of these data points were significantly different across nutrients considered. For instance, while approximately 60% of the energy data points were from US studies, it was 80% and 37% for fat and fat (% of energy), respectively. In addition, fat (% energy) was more likely than the other two to be assessed using FFQ (44% of data points) compared to energy (27%) and fat (18%). However, study overall quality score and self-report of diet by child (yes vs no) did not differ significantly by nutrient data points.
We grouped parent–child pairs using a mutually exclusive and a non-exclusive method, respectively. For the mutually exclusive method, it is important to note that none of the data points considered parent–son or parent–daughter correlations. In the case of total energy intake, the majority of data points consisted of correlations between mothers and their children or fathers and their children, irrespective of children's sex. The same was true for fat intake. However, in the case of fat (% of energy), the distribution of data points by mutually exclusive parent–child pair was more even, with the highest proportion being for mother–daughter correlations (26%).
Meta-analyses and meta-regression analyses for energy and fat intakes
Our meta-analyses showed that, overall, parent–child correlations (not Fisher transformed) for fat intake in grams and per cent energy were similar, 0.20 (0.19) versus 0.19 (0.15); for energy intake, it was 0.17 (0.14). This pattern was not replicated, however, across non-exclusive parent–child pairs. In fact, it was only consistent in the case of parent–son, father–son and mother–son correlations. Analysis of variance and pairwise t tests indicated that mean correlations did not differ significantly between the three intake measures considered within each parent–child pair (p>0.05), which may be due to the small sample size (table 3). Mean correlations for the three intake measures across parent–child pairs are shown in figure 1a,b. These mean correlations ranged between 0.08 and 0.28, which is considered a weak to moderate correlation.
Meta-analysis: comparison of parent–child correlations of energy and fat intakes, by type of parent–child pairs
Mean correlation coefficients of dietary intakes and proportion (%) of significant correlations (p<0.05), by parent–child relationship type. (A) and (B) p, parent; c, child; m, mother; f, father; s, son; d, daughter. The differences across the pairs were not statistically significant based on one-way ANOVA and pairwise t tests, p>0.05. (C) None of the negative correlations were statistically significant. Chi-square tests of significance of correlation by nutrient type were not significant across pairs.
Furthermore, we tested whether the proportion of significant correlations (ie, p<0.05 for null hypothesis that correlation=0) differed between nutrients across parent–child pairs (figure 1c). We only considered pairs that had enough power to conduct a χ2 test. While in most parent–child pairs, energy intake had the highest proportion, the differences in proportions between nutrients were only statistically significant in parent–daughter dyads (for energy=82%, for fat=75%, for fat as % of energy=18%; p=0.005).
Means of Fisher's transformed r with their 95% CI were estimated using random effects models, taking into account the associated standard error. In general, estimated Fisher's transformed correlations became stronger, and most were significantly greater than 0 (see 95% CI, table 3).
Table 4 presents the results examining study and data point characteristics in relation to the magnitude of correlations (Fisher's z transformed). In models 1A through 1C, only these predictors were considered, and the full model is presented. In the case of energy intake (model 1A), dietary assessment was the only significant predictor with FFQ revealing weaker correlation by −0.307 compared to multiple 24-h recalls or records. In contrast, in model 1B (fat intake), the only significant predictor was study country; non-European countries showed a significantly higher correlation compared to the USA (p<0.001). In model 1C, correlation in fat (% energy) was significantly lower over time (β=−0.032 per decade; p<0.001), higher based on FFQ compared to 24-h recalls or records (β=0.473; p=0.013), lower for child self-reported intake (β=−0.365; p=0.046) and higher with better study quality (β=0.278; <0.001).
Meta-analysis: tests of the differences in the Fisher's transformed correlation coefficients, by country, publication year, dietary assessment approach, whether child dietary intake was self-reported, study quality, and type of parent–child pairs*
Furthermore, we tested whether the correlations differed by the types of parent–child pairs, in particular, the influence of parental and child sex, but the small sample sizes limited our statistical power. None of the differences were significant (p>0.05). In model 2, the type of parent–child pair was entered as an additional predictor and forced to be retained. The other predictors from model 1 were only retained if their associated p values in the full model was <0.10. In model 3, a different variant of the parent–child pair variable was introduced, focussing on the sex of the child.
Discussion
In the present study, we systematically reviewed and analysed the relevant studies published since 1980. In order to quantify the associations, we calculated the average correlation coefficients and the variations across intake variables and child–parent pairs. We also compared the differences in the associations across study population groups, over time, as well as dietary intake assessments. Our research showed that only a relatively small number of (n=24) previous studies have examined the child–parent association in dietary intakes. Most of them are based on small samples, and about half are conducted in the USA. Only two of them, a recent study we conducted in the USA28 and another one from The Netherlands,11 are based on national data. Very limited studies compared the differences between different parent–child pairs or tested the resemblance in overall dietary intake pattern.28
Nevertheless, these studies revealed that, although the reported degree of association and similarity varied considerably across studies, nutrients and foods, overall, the association is weak. For example, our meta-analysis shows that on average, the mean correlation coefficient was only 0.17 (SD=0.14; range, −0.24–0.39) for energy intake and 0.19 (SD=0.15; range, −0.04–0.44) for fat intake (% energy). We suspect these weak correlation coefficients probably reflect the fact that the parental and family influence on young people's dietary intake is not as strong as many people have speculated. In addition, it is also possible that the difficulty to assess children's and their parents' intakes accurately using comparable assessment methods (ie, measurement) have weakened the observed association.
Our review and meta-analysis also reveal several other interesting differences in the correlations. First, the differences in the association are noticeable across nutrient intake variables. For instance, the correlation seemed to be slightly stronger for fat intake than for energy intake. This may be due to the potentially greater similarity in dietary composition and the possibility that many parents in some societies (eg, Western societies) may want to control for their total energy intake due to concerns of weight gain and obesity.
Second, our meta-analysis provided some evidence supporting our hypothesis that the association has become weaker over time as indicated by the effect of publication year for fat intake (as % of energy, lowered by 0.032 per year). In the case of energy intake and fat intake (as grams), we cannot rule out the possibility that the lack of significant finding may be due to small sample sizes. In addition, as a marker for parental influence on children's eating behaviours, fat intake as per cent of energy may be better than energy intake. Family members could share many food items, but parents might restrict total caloric intake due to concerns of weight gain.
Third, our meta-analysis findings suggest parent–child pairs in the USA seem to have weaker association in intakes of energy and total fat compared to other non-European countries. We suspect these differences are contributed by differences in food environment (eg, food supply and availabilities) and parenting styles between the USA and other countries. For example, US children may have more opportunities to eat away from home including school meals and to eat snack foods without the presence of their parents (thus, weaker association). We suspect the parent–child similarity in dietary intakes would be stronger in at least some developing countries as children and their parents are more likely to eat more meals at home (or children were more likely to eat food brought from home in schools), less likely to eat snack foods and eat the same kinds of foods compared to industrialised countries. For example, a study in Brazil showed that the average correlation between consumption of food groups by mother–child pairs by maternal education levels was 0.49 for ≤4 years and 0.41 for >4 years, respectively.40 Another study in Mexico also showed a good correlation between parental and child food preferences.39 However, the limited available studies in other countries, especially developing countries, do not allow us to fully explore the between-population differences in our meta-analysis.
In addition, the use of different dietary assessment methods affected findings regarding parent–child correlation in dietary intake. Studies using FFQ yielded lower correlation in energy intake than those using 24-h recalls or food records. This may be due to the fact that the latter may be conducted for the same days for parents and children and thus yielded stronger correlations, while FFQ assesses frequency of a wider range of foods over a longer period of time (eg, usually over the past year).26 27 However, when examining fat as per cent of energy (indicating dietary composition), FFQ revealed a stronger parent–child correlation.
The present study has several strengths including its inclusion of related studies published over the past three decades since 1980 that we could find and application of several statistical analysis approaches including regression analysis to compare differences between several factors that we hypothesised might affect the child–parent association in dietary intakes, and test of publication bias. In addition, we assessed the methodology quality of studies being reviewed and in our meta-analysis tested its influence on the observed parent–child correlation in intakes.
On the other hand, as a meta-analysis, our study was limited by the small number of data points provided by available studies. Only 15 studies were included, which provided a relatively small number of data points. This limited the statistical power and our models to test the differences across sample characteristics. Second, over half of the studies were from the USA, and only three were conducted in developing countries. Third, another possible good way to characterise the resemblance would be based on food groups instead of nutrients. However, the limited available results did not allow us to conduct such meta-analysis. In addition, we could only search related studies published in limited languages and data sources.
Our findings have a number of implications. First, they provide useful insight to guide future research. More future studies are needed to study the parent–child resemblance in diet, the differences in the association between population groups, the determinants, and related secular trends. It is desirable to carry out such studies based on nationally representative data with sound dietary assessment. Application of more sophisticated statistical analysis approaches such as multilevel models than simple correlations in order to account for the influence of potential confounders on the results is desirable.
Second, the weak associations we found challenge the widely held belief that children's and their parents' dietary intakes are alike. Although we also believe that parents have an important influence on their children's dietary intake, especially at young ages (eg, Beydoun and Wang28) we suggest that this influence should not be over-stated or interpreted. More attention should be given to the influence of the other players in children's eating patterns such as that of schools, local food environment and peer influence, government's guidelines and policies that regulate school meals, the broader food environment that is influenced by food production, distribution and advertisement. A growing number of studies show that people's dietary intakes are affected by the complex interactions of a large number of variables at different levels.14 16 45
Third, more research is needed in developing countries and societies that are under more remarked social and nutrition transitions. Finally, with the many changes that many societies have been experiencing, we suspect parental influence on children's dietary intake is likely to continue to diminish. To help young people to develop lifelong healthy eating habits, parents and families should not be the sole primary focus; more vigorous and population-based approaches are needed. In addition, parents need to be better empowered to assist their children to eat a healthy diet.
What is already known on this subject
Parents are believed to have a strong influence on children's eating behaviours as role models and gatekeepers. However, previous findings on child–parent resemblance in dietary intakes are mixed. Some provided evidence to support the association, while others did not.
What does this study adds
It provides a systematic review of 24 related studies published since 1980 and an analysis of the association (correlations) based on 15 published studies. Overall, this systematic analysis showed a moderate or weak association (average correlations were approximately 0.2), and the related published findings varied remarkably across studies, nutrients, parent–child pairs and by some other study characteristics including dietary assessment methods.
Acknowledgments
We would also like to thank Drs. Xinxue Liu and Ting Wang from Peking University Health Science Center in China and Ms Silvia Bel Serrat for their assistance in searching related studies published in Chinese and Spanish.
References
Supplementary materials
Web Only Data jech.2009.095901
Files in this Data Supplement:
Footnotes
Funding The study was supported in part by the Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD, 1R03HD058077-01A1, R03HD058077-01A1S1), the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK, R01DK81335-01A1), and the National Institute on Aging Intramural Research Program.
Competing interests None.
Provenance and peer review Not commissioned; externally peer reviewed.