Like parent, like child? Dietary resemblance in families

Study design, setting, and participants

The DAGIS study (Increased Health and Wellbeing in Preschools) investigated energy balance-related behaviors among Finnish preschool children using a cross-sectional design. The DAGIS study protocol has been published previously [31]. Eight municipalities, all of which agreed, were asked to participate in the study in 2015. In order to ensure a wide range of socioeconomic backgrounds in our sample, the selection of the municipalities was based on socioeconomic status indicators (larger variation of educational level, income level and higher Gini coefficient) [32]. From all municipal preschools and private preschools from whom the municipalities purchased education services in the participating municipalities, we randomly selected 169 preschools to be invited to take part in the study. The random selection of preschools was conducted separately for each of the municipalities, and the number of invited preschools was based on power and sample size calculations [31]. Of the invited preschools, 67 (40%) did not wish to participate and 16 (9%) were excluded due to being a 24- h preschool, operating in a language other than Finnish or Swedish, or not having reduced fees for low-income families. Altogether 86 preschools (51% of those invited) provided a written informed consent. From these preschools, all the children from groups with 3–6 year-olds (N = 3592) and their families were invited to participate by an informational letter with a consent sheet attached. The letters were distributed through the preschools and collected from the preschools approximately one week later. The recruitment period took place between August 2015 and February 2016. Children in preschools with a low consent rate (≤30% of the children in each of the preschool groups for 3–6-year-olds consented to participate) were excluded (91 children in 20 preschools) resulting in 892 consented children (25% of those invited) and 864 participating children from 66 preschools (24% of those invited in total; 29% of those invited from the participating 66 preschools). Data collection took place between September 2015 and April 2016. Figure 1 shows the flow of the participating preschools and the participants.

Measurements and variables

Statistical methods

Participants

The participating children for whom we could not calculate the dietary resemblance measure because of missing food consumption data, had younger mothers (Student’s t-test p = 0.029), as well as less educated parents (Chi-Squared test for both mother’s and father’s education p < 0.001). In addition, there were more boys among the excluded children compared to the participants for whom the resemblance measure was calculated (Chi-Squared test p = 0.016). In terms of age of the child or father, number of family meals shared, children living in the same household or respondent, the excluded children did not differ from the children included in the analyses.

Dietary resemblance in families

When we set the dietary resemblance as the outcome variable with one or two observations per child, we found a statistically significant interaction between respondent (the parent providing food consumption data on behalf of the child) and the parent to whom the food consumption of the child is being compared to (p < 0.0001). Figure 2 illustrates this reporter-bias. In univariate models (Table 3), having mother as a respondent was negatively linked with the father-child resemblance, i.e., father-child resemblance was stronger, when food consumption data for the child was reported by the father, and weaker if the mother reported food consumption on behalf of the child. A similar reporter-bias was also detected in mother-child resemblance. In the full models (Table 4), the father-child resemblance was statistically significantly explained only by the respondent: having mother as a respondent was associated with weaker father-child resemblance. The tendency for reporter-bias was also seen in the mother-child resemblance, which was borderline significantly explained by the respondent.

	Father-child resemblance			Mother-child resemblance
	β estimate	95% confidence interval	Significance for difference	β estimate	95% confidence interval	Significance for difference
Child’s gender
Girl	ref.			ref.
Boy	−0.003	− 0.029, 0.022	0.813	0.001	− 0.023, 0.025	0.927
Child’s age
Median or higher	ref.			ref.
Lower than median	0.005	− 0.016, 0.026	0.638	0.011	−0.009, 0.030	0.276
Number of children living in the same household (in addition to the participating child)
Only child	ref.			ref.
One child	0.002	−0.036, 0.040		−0.026	− 0.058, 0.007
Two or more children	0.003	−0.036, 0.043	0.986	−0.016	−0.051, 0.018	0.277
Father’s education
High	ref.			ref.
Middle	−0.039	−0.085, 0.008		−0.030	−0.075, 0.016
Low	−0.045	−0.092, 0.001	0.136	−0.039	− 0.083, 0.005	0.220
Mother’s education
High	ref.			ref.
Middle	−0.019	−0.062, 0.024		−0.015	−0.056, 0.026
Low	−0.050	−0.099, 0.000	0.149	−0.034	− 0.080, 0.012	0.357
Number of family meals as reported by the father	0.014	0.003, 0.024	0.013	0.012	0.001, 0.022	0.032
Number of family meals as reported by the mother	0.018	0.007, 0.030	0.002	0.019	0.008, 0.030	< 0.001
Respondent
Father	ref.			ref.
Mother	−0.068	−0.129, −0.006	0.032	0.066	0.003, 0.129	0.039

N = 584–632 for father-child pairs and 627–717 for mother-child pairs

	Father-child resemblance			Mother-child resemblance
	β estimate	95% confidence interval	Significance for difference	β estimate	95% confidence interval	Significance for difference
Child’s gender
Girl	ref.			ref.
Boy	0.003	−0.025, 0.030	0.857	0.005	−0.022, 0.032	0.719
Child’s age
Median or higher	ref.			ref.
Lower than median	−0.003	−0.026, 0.020	0.800	0.006	−0.016, 0.028	0.588
Number of children living in the same household (in addition to the participating child)
Only child	ref.			ref.
One child	0.002	−0.037, 0.040		−0.024	− 0.061, 0.013
Two or more children	0.004	−0.037, 0.045	0.977	−0.016	−0.055, 0.023	0.418
Father’s education
High	ref.			ref.
Middle	−0.025	−0.078, 0.028		−0.024	−0.075, 0.027
Low	−0.021	−0.079, 0.037	0.648	−0.036	− 0.092, 0.020	0.448
Mother’s education
High	ref.			ref.
Middle	−0.005	−0.055, 0.045		0.004	−0.044, 0.052
Low	−0.030	−0.090, 0.030	0.567	−0.020	− 0.078, 0.039	0.657
Number of family meals as reported by the father	0.003	−0.014, 0.020	0.695	−0.002	−0.018, 0.014	0.819
Number of family meals as reported by the mother	0.015	−0.003, 0.034	0.109	0.022	0.004, 0.040	0.015
Respondent
Father	ref.			ref.
Mother	−0.090	−0.162, − 0.019	0.013	0.062	−0.007, 0.132	0.078

N = 574 for father-child pairs and 577 for mother-child pairs

In the univariate models (Table 3), family meals as reported by both parents separately were positively associated with father-child and mother-child resemblance. In the full models (Table 4), family meals as reported by mother was positively associated with the mother-child resemblance, whereas father-child resemblance was not explained by the number of family meals. Sociodemographic factors were not associated with parent-child dietary resemblance.

Key results

Our study results demonstrate that the diets of the children and their parents were largely similar, as illustrated by the moderate resemblance. Mother-child resemblance seemed to be stronger compared to father-child resemblance, but we also found a significant parent-respondent interaction indicating that the diet of the child resembled more the diet of the parent providing responses for the child compared to the other parent. After taking the reporter-bias into account, the resemblance no longer differed between father-child and mother-child pairs. To our knowledge, this is the first study to investigate and report such an effect (i.e., stronger resemblance with the parent who reported the child’s diet), which can possibly explain some of the previous contradictory results on dietary similarities.

Interpretation of the findings

The present study is the first dietary resemblance study to provide evidence of reporter-bias: the parent providing responses on behalf of the child might artificially strengthen the responding parent-child dietary resemblance. In former studies, mothers have traditionally provided the food intake information on behalf of their children [38]. Hence, it is unclear, whether the stronger mother-child correlations reflect the truth or if they simply reflect reporting bias and confounding, as discussed by Oliveria et al. and Shrivastava et al. [10, 14]. This view was also supported by a review of Wang et al., which detected weaker correlations in studies using child self-reported intake compared to intakes reported by parents [25]. However, mothers may be more willing to report the child’s food consumption because they are, in fact, more aware of the child’s diet. For example, in a study by Blissett et al., mothers reported greater feeding responsibility than fathers: 63% of mothers were mostly responsible for their child’s feeding, whereas the corresponding proportion among fathers was only 4% [39]. In future studies, child self-report of food consumption should be preferred whenever possible: it has been shown, that children aged 8–11 years can most accurately report their energy intake using an FFQ when compared to their parents [40]. If, however, child self-report is not possible due to the age of the participants, researchers should report clearly who provided the food consumption information and adjust for this confounding factor in the subsequent analyses. Since different family members have distinctive insights into child behavior, as discussed by Morgan et al., the two caregivers could both report the child’s diet [41].

In our study, the parent-child dietary resemblance on a ‘whole-diet’ level was relatively strong. Earlier studies have investigated dietary resemblance by calculating parent-child correlations for the consumption of different food groups. Thus, comparison between studies is challenging. Beydoun & Wang investigated similarity in dietary quality and reported a parent-child correlation of 0.26 in HEI_n [12]. However, HEI_n is designed to measure dietary quality and includes the intake of 12 food groups, whereas our approach was broader and we included a much wider range of foods, both healthy and unhealthy, covering the ‘whole-diet’ of the participants and their parents. Interestingly, a recently published paper suggested that parent-child similarity may be stronger in healthy food consumption than in unhealthy food consumption [27]. However, our data did not support this conclusion: no apparent pattern in the differences of the rankings was detected. Furthermore, our stronger resemblance can partly be related to the age of the participants: according to a review by Wang et al., younger children (< 10 years of age) have stronger correlations with their parents than older children [25]. Our results support this view: participants were preschool-aged children and thus probably more dependent on their parents’ food choices than older children are. Interestingly, studies using foods or food groups have reported slightly higher correlations than studies reporting similarities in dietary quality (HEI): in an Australian study, unadjusted father-child correlations ranged from 0.31 (fruit juice) to 0.54 (cookies) [20], whereas a study from the U.S. reported adjusted mother-child correlations ranging from 0.34 (snacks) to 0.48 (vegetables) [21]. We applied a statistical method innovatively to food frequency data in order to compare parent-child resemblance over the ‘whole-diet’ instead of comparing the use of single foods or food groups.

In this sample, we did not detect statistically significant associations between parental educational level and parent-child dietary resemblance. Regardless of the socio-economic background, families appear to share a fairly uniform diet. Although many of the studies investigating parent-child similarities in dietary intake have adjusted for parental education, to our knowledge, only one has reported analyses stratified by parental education. A study by Walsh et al. suggested that the relationships between fathers’ and children’s consumption of savory snacks and take-away foods was moderated by father’s education (university-educated vs. non-university-educated) [42]. In an adult population, the educational level of the husband (male spouse) was associated with dietary resemblance in food choices between spouses in unadjusted models, but the association disappeared after adjustment for having children [43]. In terms of other socioeconomic status markers, a stronger parent-child correlation in the consumption of artificially sweetened soft drinks in families with sufficient income (≥300 on household poverty income ratio) has been reported, although no such interaction was detected regarding other food groups [12]. Additionally, Wang et al. found a stronger mother-child correlation in energy intake among working mothers compared to unemployed mothers in a sample of African-American families [44]. Thus, our study is one of the first studies to examine the association between parental education and parent-child dietary resemblance.

The more a family eats together, the more there are opportunities for the child to assimilate health behaviors, such as healthy or unhealthy food intake. In this sample of preschool-aged children, the frequency of family meals as reported by mothers was positively associated with the parent-child dietary resemblance, whereas father-reported frequency of family meals was not. This finding could be explained by fathers’ and mothers’ different interpretations of the question (‘How often your family usually gets together for at least one meal?’). In a recent European study, an opposite result was obtained: concerning the ‘sweet and fat’ dietary pattern, the odds of being allocated to the same dietary pattern with the father was higher with more meals shared, whereas no such association was detected in the mother-child similarities [24]. Oliveria et al. reported non-significant differences in parent-child correlations between parents who ate at home more frequently and those eating at home less frequently, except for mother-child cholesterol intake [10]. In our study, parents reported having a family meal on average almost six days per week. Not many studies have reported family meal frequencies in this age group: in an Australian study, 77% of the children shared dinner (the most frequently shared family meal) with at least one parent on at least five evenings per week [45], whereas in a sample of 4-year-olds, 57% of the mothers reported that at least a part of the family ate the evening meal together on six or seven evenings per week [46].

Strengths and limitations of the present study

In our study, only 9% of the parents reporting the child’s food consumption were fathers. These fathers were more educated compared to the fathers who did not report their child’s food consumption, and they might also have some other distinctive characteristics. We also had a fairly low participation rate: less than 30% of the children from the participating preschools took part in the study, which was notably less than anticipated. In addition, our sample was generally highly educated, which might have affected our results. Furthermore, the sample for whom the dietary resemblance measure could be calculated was more highly educated compared to the excluded participants. Thus, data was not missing completely at random, which may have biased the results and weakened the contrast and power of the corresponding tests on education. In addition, our results are from a cross-sectional study making causality hard to judge: it would be logical to reason that the child’s food consumption is influenced by the parents’ food consumption and not the other way around. However, parenting also is reacting to child characteristics, and thus, the child can also influence the food consumption of the parents [47].

The FFQ used among the children was designed to measure food intake outside preschool hours, since the parents would not have been able to estimate the consumption of foods eaten at preschool. As for the parents, they also reported foods eaten outside the home (for example at work during the days). Measuring food consumption of the children only outside preschool hours tends to make the parent-child dietary resemblance stronger than it may actually be. However, we believe that the extent of the possibly overestimated resemblance is common to all participating children, as children in preschools within the same municipality are offered same meals provided by a single catering company. Therefore, this gap in the assessment of daily food consumption probably has little effect on the ranking of the food items and the subsequent analyses. Another concern is the validity of the FFQs, which is yet unknown. However, we are currently conducting a validation study against 3-day food records. To our knowledge, no validated, culturally appropriate FFQs with an emphasis on fruit and vegetable as well as sugar-enriched food intake among preschoolers were available. Since we looked at the dietary resemblance over the whole FFQ instead of focusing on one or two food items, it is unlikely that the effect of the possible misreporting is substantial. Additionally, we used Spearman rank correlation coefficient in the calculation of dietary resemblance measures and thus, the absolute food consumption frequencies were not of importance as long as the misreporting did not interfere with the within-individual ranking of the food items. A strength of the study is that we used the same FFQ to measure food consumption both among the children and their parents, albeit the parents reported their food consumption during the whole week, whereas the children’s food consumption during preschool hours was not reported.

We did not adjust for home food availability in the present study. It has been suggested that food availability in the home modifies the associations between parental and child food intake. For example, in families where the availability of fruit and vegetables is high, parental fruit and vegetable consumption has been shown to be more strongly associated with the child’s fruit and vegetable consumption compared to the families with low fruit and vegetable availability [48, 49]. Furthermore, in a recent study by Hebestreit et al., the chance of the child sharing a ‘sweet and fat’ dietary pattern with both the mother and the father were higher where soft drinks were available during meals [24]. However, since we did not focus on certain food items, but were interested in similarities in the ‘whole-diet’, adjusting for home food availability would have been challenging and could also have led to overadjusting and misinterpretation.