Background
Much of our eating behaviors are formed in early childhood and most behaviors are modeled after important caregivers of the child, primarily the parents [
1]. Furthermore, parents shape children's early experiences with food and eating [
2], and can affect children's diet and eating behaviors in numerous ways. For instance: by encouraging them to eat certain foods, by restricting certain foods, or by passively allowing certain foods in the regular diet. Other important parent-related determinants of children's eating behaviors are the physical and emotional environment in which eating behaviors are developed [
3]. Hart, Bishop, and Truby [
4], have stated a need for increased knowledge about parental influence on children's eating behavior. Also Zeinstra [
5] has suggested that further research on child eating behavior should focus on the role of parental strategies in shaping children's food preferences and consumption.
A barrier to this literature has been a lack of validated instruments for quantifying parental feeding behaviors and styles [
6]. Thus, comparability of studies has been a challenge. In a review of 22 studies [
7], only the Child Feeding Questionnaire (CFQ) [
8] was cross-validated in different parental samples and used in multiple settings. Furthermore, most previous measures of parental feeding practices have included just a few feeding practices, such as restrictive feeding and pressure to eat. These practices are aspects of control over child food intake, and are typically measured with the CFQ [
8]. Although controlling feeding practices seem to be widely used by parents in an attempt to secure a well-balanced diet for their children [
7], some studies have proved counterproductive effects of these strategies, as parents who exert too much control over child food intake tend to have children with an increased preference for high-fat foods and higher levels of snack-food intake [
9].
The emphasis on parental control in previous feeding practices measures may have left other important feeding practices rather underexplored. This is especially true for feeding practices that are associated with desirable outcomes in children [
10]. Parental modelling of healthy eating and exposure to healthy foods are examples of feeding practices that may be effective [
11‐
14]. The extent to which parents try to teach their children about nutrition is another aspect not examined in the previous measures of parental feeding practices [
10]. However, more recent research has suggested that additional feeding practices such as these can also be measured in parents and might impact child outcomes [
15]. The Comprehensive Feeding Practices Questionnaire (CFPQ) [
10] is an extension of previous feeding practices measures, and represents a more complete range of feeding practices that may be relevant to child outcomes. It consists of 49 items representing 12 dimensions (subscales), each including 3-8 items. Initial testing of the CFPQ with American parents of 2-to-8-year-olds showed reasonable validity and reliability [
10]. An analysis of nine CFPQ subscales with French parents of 4-to-7-year-olds demonstrated reasonable validity and reliability in this sample as well [
15].
Although the CFPQ appears to be a promising instrument for measuring multiple parental feeding practices, it's important to note that this instrument was designed to measure feeding practices in parents of young children (about 2 to 8 years of age). Zeinstra [
5] suggests that children's cognitive development influences the strategies that parents use to shape the eating behavior of their children. Thus, feeding practices measures developed for parents of young children will not necessarily be valid for parents of older children and adolescents. As far as we know, only one previous study has validated a (pure) feeding practices measure with parents of older children and adolescents. In this study by Kaur et al [
16], a modified version of Birch's CFQ was validated in a multi-ethnic sample of 260 parents of 10-to-19-year-olds (mean age:15 years). The psychometric properties of the modified CFQ were found to be similar to those of the original CFQ. However, consistent with the evolving independence of adolescents, the factor scores for the controlling feeding practices measured by the CFQ decreased with increasing age of the adolescent. That is; controlling feeding strategies seemed to be less used by parents of older children and adolescents than for younger ones. Nevertheless, the fact that parents are considered to be an important social agent impacting upon children's diets, also applies to older children and adolescents [
17].
So far, most studies of parental influence on child eating behavior have focused on young children. In the present study, we have focused on feeding practices in parents of children on the onset of adolescence (10-12-year-olds). Adolescence is the period from about the age of eleven to the late teen years, and represents a transitional stage from childhood to adulthood. It is characterized by the elaboration of identity, and it is a time of growing independence when individuals want to make their own decisions including what and when to eat [
18,
19]. This stage is typically a time of gradual shift from parental to peer influence [
20]. Thus, during adolescence parental influence over food choice may be displaced by the effects of advertising and peer pressure [
21], and the age at which these changes set in appears to be diminishing [
22]. However, the eagerness of adolescents to take over responsibility for food choice is not necessarily matched with their ability to make healthy food decisions. Adolescents have a reputation for unhealthy food choices [
23,
24], and interventions directed towards this group of the population have had mixed success [
25]. Furthermore, research has found that adolescents understand at an abstract level the (un)healthiness of foods, but have limited concern about future health [
26]. Therefore, the influence of parents should be assessed at all stages of this "hand-over-of-control" period to assist in the development of concurrent parental and peer group intervention programs [
27]. The rationale for focusing on 10-to-12-year-olds in the present study is that children this age are still highly influenced by parents. Accordingly, it might be easier to implement intervention programs involving parents among individuals within this age range than among older ones.
Given the lack of validated instruments measuring feeding practices that might be relevant for parents of older children and adolescents, we aimed to test the validity of the CFPQ with Norwegian parents of 10-to-12-year-olds to check if it is a suitable tool for measuring feeding practices in this part of the population. We believe that development and validation of broad feeding practices measures such as the CFPQ is of great importance for applied research aiming to develop interventions to improve children's and adolescents' diets, whether it is for public health purposes or for clinical purposes.
Discussion
The aim of the present study was to test the validity of a slightly adapted version of the CFPQ with Norwegian parents of 10 to12 year old children. Analyses of both the individual subscales and a unified 42 item version of the instrument suggested reasonable validity of the CFPQ in our sample.
The initial scale analyses included assessment of subscale dimensionality (convergent validity) and internal consistency. Reasonable convergent validity and internal consistency was found for most scales. However, there were indications of some problems within the following four subscales: Child control, Environment, Modelling, and Teaching nutrition. We found some very low communality items within the Child control, Modelling and Teaching nutrition subscales, and the Environment subscale showed a two-factor solution, thus indicating some problems with the convergent validity of these scales. Moreover, the low alphas found in three of these scales may be questioned (Child control = 0.55, Teaching nutrition = 0.44, Environment = 0.57). Some low alphas were also found by Musher-Eizenman & Holub (2007) (e.g. Encourage balance and variety = 0.58 for American mothers) and Musher-Eizenman et al [
15] (e.g. Teaching nutrition = 0.54 and 0.56 for French mothers and fathers respectively). However, it is important to note that all CFPQ subscales have few items. According to Cortina [
42], it is well known that the number of items has an effect on alpha, especially at low levels of average item inter-correlation. That is, if a scale has enough items (e.g. more than 20), then it can have an alpha of ≥ 0.70 even when the correlation among items are very small [
42]. Thus, lower values of alpha can be expected from shorter scales like the subscales of the CFPQ. Developing survey instruments always involves a trade-off between internal consistency (using multiple items) and practicality. The CFPQ is an instrument aiming to tap many different aspects of feeding practices. Using only a few items in each subscale makes it less tiresome, and therefore more applicable. However, one may question if the brief subscales of the CFPQ sufficiently captures the different aspects of feeding practices.
Initial testing of discriminant validity by running correlation analyses between the CFPQ subscales revealed some substantial correlations, but these were not large enough to compromise the discriminant validity of the scales [
40]. The correlation (r = 0.56, p < 0.01) between the Restriction for weight control and Restriction for health subscales could be expected, as these scales represent conceptually close constructs. Musher-Eizenman and Holub [
10] indicated that parents may not spontaneously differentiate between restriction motivated by weight and by health, suggesting that parents who limit or restrict child food intake for weight control reasons may also be doing so for health reasons (or vice versa). Yet, Musher-Eizenman and Holub [
43] was the first to articulate the distinction between restriction for health reasons and restriction for weight control reasons. They argue in favor of the distinction between different types of restrictive feeding, as there may be many different motivations behind the restriction, including child health outcomes, child weight loss or maintenance, to teach the child healthy eating habits for the future, or for religious or ethical beliefs. In their directions for future research, they suggest further exploration of the effect of different restrictive feeding practices on child eating, weight and health outcomes. The correlation (r = 0.52, p < 0.01) between the Encourage balance and variety and Teaching nutrition subscales was also expected, as these scales both deal with explicit nutrition communication with the child. The relation and discrimination between these scales are further discussed below.
When running factor analysis on the unified 42-item version of the CFPQ, a 10-factor solution, largely corresponding with the original instrument developed by Musher-Eizenman and Holub [
10], was found to be conceptually sound in our sample. However, there were some small, but noteworthy differences in factor structure between our solution and the one suggested by Musher-Eizenman and Holub [
10]: In our solution the items on the Encourage balance and variety and Teaching nutrition subscales clustered together to form one factor. Moreover, the four items on the Environment subscale split into two different factors reflecting availability of healthy and unhealthy foods respectively. Also worth noting, is that the following items did not load onto their assigned scales: item 10 on the Child control subscale, item 41 on the Modeling subscale, and item 42 on the Restriction for weight subscale.
If we take a closer look at the problematic scales and items revealed by our analyses, it may seem as if some of the items are not conceptualized in an adequate way. Starting with the low-communality item 10 ("Do you allow this child to leave the table when s/he is full, even if your family is not done eating?") on the Child control subscale, this item might reflect a breach of meal-related social norms rather than child control over what and when to eat. In other words, leaving the table when full before the rest of the family is done eating, might reflect a breach of good table manners, an ideal learned through family meals in most Western cultures [
44]. Thus, item 10 did not seem feasible as a measure of child control over food intake in our sample.
Moving on to the Environment subscale, the items of this scale split into two different factors with items 12 ("Most of the food I keep in the house is healthy") and 20 ("A variety of healthy foods are available to my child at each meal served at home") loading onto one factor reflecting to what extent healthy foods are available in the home environment, and items 14 ("I keep a lot of snack food in my house") and 34 ("I keep a lot of sweets in my house") loading onto a second factor reflecting to what extent unhealthy foods are available in the home environment. Both correlation analysis between the two "new" factors (r = 0.21, p < 0.01) and face validity supports this distinction. Thus, the items on the original Environment subscale seem to specify two different kinds of behavior (having healthy foods available in the home vs. having unhealthy foods available in the home) in our sample.
Initial analysis of the Modelling subscale revealed one low-communality item that needed further investigation (item 41 "I model healthy eating for my child by eating healthy myself"). This item is distinct from the other items on the Modelling scale, as it seems to reflect a general, "passive" form of modelling (...eating healthy myself...), while the remaining three items involve specific, "active" components (...eating healthy foods
in front of my child..., ...
show enthusiasm about eating healthy foods..., ...
show how much I enjoy eating healthy foods...). This distinction was supported when running factor analysis on the unified 42 item CFPQ; item 41 did not load onto the Modelling factor, but onto the factor reflecting availability of healthy foods in the home environment. One obvious assumption for healthy eating is availability of healthy foods. Thus, if parents practice healthy eating, healthy foods are most likely available in the home. Collectively, studies do suggest that readily available and easily accessible healthy foods within the home are likely to enhance healthy eating among families [
45]. To sum up on this; healthy eating practices among parents might be more related to the availability of healthy foods in the home environment than to "active" modeling of healthy eating.
The initial scale analyses revealed that the Teaching nutrition subscale had one very low communality item (item 39 "I tell my child what to eat and what not to eat without explanation") and two high communality items (item 23 "I discuss with my child why it's important to eat healthy foods", and item 29 "I discuss with my child the nutritional value of foods"). Item 39 seems to reflect an authoritarian interaction with the child, while a more democratic, authoritative mode of interaction is reflected by items 23 and 29. In the feeding domain, authoritarian practices include parental control and indisputable instructions on what to eat [
46,
47], while authoritative practices include using discussion, negotiations, and reasoning for desirable eating behavior [
48]. In light of this, item 39 seems to reflect a different type of parental food-related behavior than items 23 and 29, which might explain its lack of communality with the latter two. A suboptimal performance of the Teaching about nutrition scale was also found by Musher-Eizenman and co-workers [
15].
When running factor analysis on the unified 42 item version of the CFPQ the items on the Teaching nutrition and Encourage balance and variety subscales loaded highly onto the same factor, indicating a lack of discriminant validity between the scales. A certain overlap between these measures could be expected since they both deal with explicit nutrition communication with the child. Although the bivariate correlation between them (r = 0.52, p < 0.01) was not high enough to suggest a complete conceptual overlap, factor analysis did not support the discriminant validity of the two scales in our sample.
While item 42 ("I often put my child on a diet to control his/her weight") on the Restriction for weight subscale performed well in the initial analysis on the individual subscales, factor analysis of the unified version of the CFPQ showed that item 42 did not load onto its assigned scale, but loaded (together with item 41) onto the factor reflecting availability of healthy foods in the home environment. Like for healthy eating, (successful) dieting also requires availability of healthy foods in the home. Thus, item 42 might be more related to the availability of healthy foods in the home environment than to restriction for weight control reasons. If we see this in light of the general recommendation about not focusing on dieting in front of children and adolescents [
49], the unsubstantial loading of item 42 on its assigned scale was not totally unexpected. Furthermore, the CFPQ was first developed and tested in the US [
10], where child overweight and obesity is substantially more prevalent than in Norway [
50,
51]. Thus, one might speculate if this item is more appropriate in an American than in a Norwegian setting.
Regarding nomological validity, significant correlations between CFPQ subscales in our 10-factor solution and related attitude scales supported theoretically expected relations. For instance; parents who were concerned about their child being or becoming overweight reported more restrictive feeding practices of both types, whereas parents who were concerned about their child being or becoming underweight reported more pressure to eat. Furthermore, parents feeling responsible for child eating reported less child control over feeding interactions, a healthier home environment, more modelling, monitoring, encouragement and teaching about nutrition, and more restriction of both types.
To sum up: our findings largely supported the validity of a slightly adapted, 42 item version of the CFPQ with parents of 10 to12 year old children in a Norwegian setting. Although some subscales and items seemed problematic as a result of our statistical scale analyses, face validity indicated that most of these items still were relevant for measuring feeding practices in parents of 10-to-12-year-olds. Furthermore, it is important to note that our findings are sample specific, and thus cannot be used as a sole foundation for changing the original CFPQ subscales. The CFPQ has previously been validated with parents of younger children, and in other cultural settings (USA; Musher-Eizenman & Holub, 2007, and France; Musher-Eizenman et al., 2009). Thus, some differences when it comes to factor structure and other validity measures between these studies and the present one are not unexpected.
Strengths and limitations
Among the strengths of this study is its large sample size. According to Guadagnoli and Velicer [
52], a sufficiently large sample size is one of the most important factors for determining a stable factor structure. Pett, Lackey and Sullivan [
40] recommend that there be at least 10 to 15 subjects per item, preferably aiming for a sample size of 500 or more. We more than satisfy these recommendations with our sample of 963 respondents. Furthermore, most previous validation studies on feeding practices measures have focused on parents of young children and on rather parsimonious instruments largely tapping aspects of parental control over child eating behavior. Thus, the present study extends the current literature by validating a multi-dimensional feeding practices instrument with parents of older children. We believe this is a relevant contribution, as valid instruments are needed to assess a wider range of feeding practices in diverse groups of parents, including parents of older children and adolescents [
6].
A few limitations of this study need comments. The findings are limited to Norwegian parents of pre-adolescent children. Furthermore, four items were excluded from the Norwegian version of the CFPQ, and only 10 out of 12 subscales were validated in the present study (thus, a reduced version of the CFPQ was tested). Second, the study sample was a cluster sample drawn from a confined geographic area (two municipalities in the South-Western part of Norway). However, as Norway is a rather homogeneous country [
53], we believe the results are likely to be generalized to other areas in Norway.
Competing interests
The authors declare that they have no competing interests.
Authors' contributions
ELM designed the study, collected and analyzed the data, and drafted the manuscript. NCØ and TØ supervised the study and contributed to the analyses and writing of the article. All of the authors read and approved the final manuscript.