Traditionally research on diet and health has focused on intakes of individual nutrients and/or foods or food groups. However, diet is complex and intake of individual nutrients or foods is not independent of intakes of other nutrients or foods. More recently a number of statistical methods have been used to identify patterns of dietary behaviour. The most popular of these is principal components analysis (PCA)(Reference Michels and Schulze1), a data-driven method which identifies foods that are frequently eaten together by aggregating items based on the degree to which the amounts consumed are correlated with one another. The aim of PCA is to identify groups of food that account for the largest amount of variation in overall diet between individuals. This method has been mainly used in studies of diet in adult populations but has also been used previously to explore dietary patterns in British, Spanish, Australian, Korean and Finnish children(Reference North and Emmett2–Reference Mikkila, Rasanen and Raitakari11).
In 2005, the Food Standards Agency Scotland commissioned the ‘Survey of Sugar Intake among Children in Scotland’(Reference Sheehy, McNeill and Masson12) to provide information on the diet of children in Scotland, with a particular focus on the intake of non-milk extrinsic sugars and sugar-containing foods. The aims of the present study were to identify dietary patterns in school-age children from the survey and investigate associations with socio-economic factors, physical activity and obesity.
Methods
A named sample of 2800 Scottish children aged between 3 and 16 years on 1 May 2006 was drawn from eighty postcode sectors across Scotland using the Department of Work and Pensions Child Benefit records. Only one child per household was selected. After exclusions and an initial opt-out period, a FFQ (see below) was sent to the remaining sample of 2352 children, of whom a further sixty-five opted out and the address was found to be incorrect for forty-two. Completed FFQ were collected by trained fieldworkers when they conducted a face-to-face interview in which information on sociodemographic factors and physical activity levels was collected and the child's height and weight measured. A total of 1700 interviews were carried out between May and September 2006 and 1512 FFQ were returned, giving a combined response rate of 66 %.
The survey used two new versions of the Scottish Collaborative Group FFQ, which has been widely used in epidemiological studies in Scotland (http://www.foodfrequency.org.uk). Version C2 for children aged 3–11 years was designed for completion by a parent or guardian with help from the child and version C3 for young people aged 12–17 years for completion by the young person with help from their parent or guardian. Version C2 lists 140 foods or drinks with a measure defined for each item. Version C3 includes an additional six items covering intake of coffee and alcoholic drinks.
Socio-economic status was assessed using the 2006 Scottish Index of Multiple Deprivation (SIMD) based on postcode(13). Data were analysed according to fifths of the distribution (quintiles). Respondents in the first quintile live in one of the 20 % least deprived areas in Scotland while those in the fifth quintile live in one of the 20 % most deprived areas in Scotland. Data were also analysed by quintiles of equivalised income (a measure of household income adjusted for the number of individuals living in the household) and education level of the main food provider (based on the highest level of recognised qualifications).
The questions on physical activity were the same as those used in the 1998 and 2003 Scottish Health Surveys(Reference A, A and J14–Reference C, K and N15). Information was collected on the number of days and the time spent each day on the four different types of physical activity thought to account for the largest part of children's total physical activity (sport/exercise, active play, walking and housework/gardening) over the previous 7 d. The physical activity questions did not include any activity undertaken as part of the school curriculum. There was no lower time limit for the inclusion of sport/exercise or active play but only episodes of housework/gardening which lasted for at least 15 min and walks of at least 5 min duration were included.
As a measure of physical inactivity, children were asked about the average number of hours spent sitting in front of a television or computer screen (other than at school) on an average week day and on an average weekend day.
BMI was calculated from measured height and weight. Respondents were classified using UK 1990 reference data(Reference Cole, Freeman and Preece16) as neither overweight nor obese (BMI < 85th centile (z-score < 1·04)), overweight (BMI ≥ 85th centile and < 95th centile (z-score ≥ 1·04 and < 1·64)) and obese (BMI ≥ 95th centile (z-score ≥ 1·64)).
The study did not require the approval of a National Health Service research ethics committee but was carried out in accordance with the research governance and quality-assurance policies of the Scottish Centre for Social Research and the Institute of Applied Health Sciences at the University of Aberdeen (http://www.abdn.ac.uk/iahs/research-governance/governance.shtml). After an initial opt out, consent was implied by completion of the FFQ and written consent obtained for the interview and measurements.
Statistical analysis
In line with standard operating procedures for the Scottish Collaborative Group FFQ, questionnaires with more than ten missed lines were excluded from analysis (n 51 (3 %)). Respondents with extreme values of total energy intake below the 2·5th centile or above the 97·5th centile for age group (3–7, 8–11 or 12–17 years) were also excluded from the analysis (n 70). Analysis of dietary patterns was carried out for school-age children; therefore children aged 3–4 years were excluded. After exclusion of outliers, incomplete questionnaires and 3- to 4-year-old children, 1233 FFQ were available for analysis.
Statistical analysis was carried out using SPSS (version 15.0; SPSS Inc., Chicago, IL, USA)(17). Data were weighted so that the estimates generated from the responding sample more accurately reflected the characteristics of the population of children aged 3–16 years on 1 May 2006 in Scotland. Weights were calculated to take account of selection probabilities and non-response bias and then combined to create a calibration weight which was applied in analyses. Further details can be found in Appendix A of the survey report(Reference Sheehy, McNeill and Masson12).
PCA was carried out separately by age group (5–11 and 12–17 years) and sex. The decision was made to explore the dietary patterns separately by age group due to the questionnaires differing slightly by age with an additional six items in the older children's questionnaire and some differences in portion sizes for certain foods. In addition, dietary patterns were explored separately by sex within each age group due to an a priori assumption that patterns may be different between the sexes.
The weight (g/d) consumed for each item on the FFQ was calculated by multiplying the frequency of consumption by the portion weight. Missing lines were recoded as 0 g/d for children with ≤ ten missing lines.
In order to minimise the number of subjective decisions in determining the dietary patterns and retain the full detail of the diet as assessed by the FFQ we chose to retain the majority of items as they appeared on the FFQ. Food items consumed by < 5 % of the sample were excluded (soya milk for all respondents and some alcoholic beverages and Marmite for 12- to 17-year-olds). PCA was carried out using the weight (g/d) of all remaining food items; 139 food items for 5- to 11-year-olds and 141 items for 12- to 17-year-olds. The number of components was selected primarily by examination of scree plots(Reference Cattell18), but the interpretability of the components was also considered. Where the number of components to choose was unclear from the scree plot, the effect of removing or adding components on the content of the components and therefore ease of interpretation was assessed. Rotation was carried out using Varimax orthogonal rotation(Reference Kline19). Factor loadings < − 0·3 and >0·3 were used to define the patterns. The factor scores for each component for each child were used in subsequent analyses.
Tests of association were carried out using the Complex Samples option in SPSS to take account of sample design and response weighting variables. Associations between factor scores and SIMD quintile, income quintile and education level of the main food provider were assessed using Complex Samples general linear model (ANOVA) for individual associations and Complex Samples general linear model (analysis of covariance) when adjusting for each other, looking at the linear trend across the groups. The associations between factor scores and physical activity and BMI classification were assessed using Complex Samples general linear model (analysis of covariance) adjusting for age at interview and SIMD, again looking at the linear trend across the groups.
Results
PCA identified three dietary patterns in each age and sex group (Table 1 and Appendix 1). The three patterns accounted for 11·6 and 13·0 % of the variance, respectively, in boys and girls aged 5–11 years, and 14·4 and 15·1 % of the variance, respectively, in boys and girls aged 12–17 years.
Dietary patterns (variance explained) for boys and girls by age group
The first component accounted for 5·1–6·2 % of the variance in the separate groups; for girls aged 12–17 years each of the three factors accounted for approximately the same amount of variance (5 %), whereas in the other groups the second and third factors accounted for progressively less of the variance.
The components were assigned labels based on the foods with high loadings within that component. For boys and girls aged 5–11 years and boys aged 12–17 years the first component was characterised by high intakes of ‘healthier’ foods such as fruit and vegetables; however, for girls aged 12–17 years the first component was characterised by high intakes of more ‘unhealthy’ foods such as puddings.
For most of the groups at least one of the other two patterns identified was a more ‘unhealthy’ ‘snacks’ or ‘puddings’ pattern. For 5- to 11-year-old boys a third ‘fish and sauce’ pattern was identified and for 12- to 17-year-old boys a third pattern loading highly for ‘starchy food and drinks’ was identified.
For boys and girls aged 5–11 years and boys aged 12–17 years fruit and vegetables loaded highly within the same component (component 1), although for 12- to 17-year-old boys there was a predominance of vegetables; however, for girls aged 12–17 years fruit and vegetables loaded highly in separate components (components 2 and 3, respectively).
Association between dietary patterns and socio-economic indicators
Table 2 shows the relationship between the dietary components and SIMD by age and sex group.
(Mean values and 95 % confidence intervals)
* P value for linear association.
In boys and girls aged 5–11 years the ‘fruit and vegetables’ component was associated with lower levels of deprivation and the ‘snacks’ component was associated with higher levels of deprivation. In girls aged 5–11 years the ‘puddings’ component was also associated with higher levels of deprivation. In boys and girls aged 12–17 years the ‘vegetables’ component was associated with lower levels of deprivation and the ‘puddings’ component was associated with higher levels of deprivation.
An almost identical relationship was found between the dietary patterns and quintiles of household income (data not shown), with income positively associated with the ‘fruit and vegetables’ component in 5- to 11-year-olds (P for linear trend 0·008 boys and 0·014 girls) and the ‘vegetables’ component in 12- to 17-year-olds (P for linear trend < 0·001 boys and 0·008 girls) and inversely associated with the ‘puddings’ component in girls of both ages and 12- to 17-year-old boys (P for linear trend 0·002 5- to 11-year-old girls, < 0·001 12- to 17-year-old girls and 0·001 12- to 17-year-old boys) and the ‘snacks’ component in 5- to 11-year-old boys (P for linear trend < 0·001).
Again a similar relationship was found with education level of the main food provider (Table 3). The ‘fruit and vegetables’ pattern in 5- to 11-year-old girls and the ‘vegetables’ pattern in 12- to 17-year-olds were most likely to be followed by children whose main food provider was educated to degree level. In contrast, the ‘snacks’ and ‘puddings’ patterns were most likely to be followed by children whose main food provider had no qualification.
(Mean values and 95 % confidence intervals)
* P value for linear association.
The ‘fish and sauce’ component in 5- to 11-year-old boys, the ‘starchy food and drinks’ component in 12- to 17-year-old boys and the ‘fruit’ component in 12- to 17-year-old girls showed no significant association with any of the socio-economic indicators.
In order to assess any independent associations, all three socio-economic indicators were entered into the same model (data not shown). In younger children aged 5–11 years the education level of the main food provider seemed to be the socio-economic indicator most associated with dietary patterns after adjustment for the other indicators. It remained significantly related to the ‘fruit and vegetables’ patterns in boys and girls (P for linear trend 0·001 boys and 0·013 girls) and the ‘snacks’ and ‘puddings’ patterns in girls (P for linear trend 0·020 and 0·027, respectively). However, household income was significantly related to the ‘snacks’ pattern in boys (P for linear trend 0·025) and SIMD to the ‘snacks’ pattern in girls (P for linear trend 0·012). In older children aged 12–17 years, however, household income was the only indicator which remained significant after adjustment for the other indicators, with the ‘vegetables’ component in 12- to 17-year-old boys (P for linear trend 0·003) and the ‘puddings’ component in 12- to 17-year-old boys and girls (P for linear trend 0·024 boys and 0·041 girls) remaining significantly associated with income.
Association between dietary patterns and physical activity and inactivity
Time spent in physical activity was positively related to the ‘snacks’ component for boys and girls aged 5–11 years, i.e. scores for these patterns were higher in children who spent longer in physical activity and also positively related to the ‘fruit and vegetables’ component for girls aged 5–11 years (Table 4).
(Mean values and 95 % confidence intervals)
* P value for linear association.
Table 5 shows the relationship between the dietary components and time spent sitting at a screen. In all age and sex groups, components loading highly for fruit and/or vegetables were inversely associated with time spent in front of a screen. In girls aged 5–11 years the ‘puddings’ and ‘snacks’ components were positively associated with more time spent in front of a screen.
(Mean values and 95 % confidence intervals)
* P value for linear association.
Association between dietary patterns and overweight and obesity
The only significant association between BMI group and dietary patterns was found in boys aged 5–11 years, in whom there was a significant linear association between ‘snacks’ and BMI group, with the lowest factor score in obese children, i.e. obese children ate less of the snack foods associated with this component (P for linear trend 0·047). There was also a significant linear association between ‘fish and sauce’ and BMI group, with the highest factor score in obese children (P for linear trend 0·023).
Discussion
For most of the age and sex groups, one ‘healthy’ dietary pattern characterised by high intakes of foods such as fruit and vegetables was identified and this was usually the first pattern explaining the greatest variance. At least one of the other two patterns identified was a more ‘unhealthy’ ‘snacks’ or ‘puddings’ pattern consisting mainly of high-fat and/or high-sugar energy-dense foods. For younger boys a third ‘fish and sauce’ pattern was identified and for older boys a third ‘starchy food and drinks’ pattern was identified.
Older girls (aged 12–17 years) differed from the other groups with respect to the dietary patterns identified in that the first component was a more ‘unhealthy’ pattern loading highly for puddings, although the percentage variance explained did not differ much from the second and third components. Each of the three factors in older girls accounted for approximately the same amount of variance (5 %), whereas for the other groups the first component accounted for 5·2–6·2 % of the variance, with the other components accounting for less. Also for older girls fruit and vegetables loaded highly in separate components, whereas for the other groups fruit and vegetables loaded highly within the same component, although for 12- to 17-year-old boys there was a predominance of vegetables.
The three dietary patterns identified for each age and sex group only accounted for 12–15 % of the variance; however, the low variance explained is not unique to the present study, as dietary patterns derived using PCA generally tend to account for only a small amount of the total variance of diet in a dataset(Reference Michels and Schulze1). Therefore, results should be interpreted with caution, as other patterns exist within the data which may individually explain less of the variance but jointly account for a substantial proportion of the variation between individuals. The variance explained in the present study was very slightly lower than that of three factors identified at 4, 7 and 9 years of age in the ALSPAC Study children (17·7 % at 4 years, 18·1 % at 7 years and 19·2 % at 9 years)(Reference Northstone and Emmett4) and three factors identified in an Australian study of 12- to 18-year-olds which accounted for 21·7 % of the variance(Reference McNaughton, Ball and Mishra7). However, this is most probably due to the larger number of food items used in the PCA in the present study, as it has been shown that with more items the lower the proportion of variance explained; however, prior grouping of foods may attenuate relationships between derived patterns and outcome variables(Reference McCann, Marshall and Brasure20). As we chose not to pre-group the foods before analysis there were approximately 140 input variables, whereas the Australian study had eighty-six items(Reference McNaughton, Ball and Mishra7) and the ALSPAC Study approximately forty(Reference Northstone and Emmett4).
The dietary patterns found by the ALSPAC Study at age 4, 7 and 9 years of age were ‘processed’ (originally named ‘junk’), ‘health conscious’ and ‘traditional’(Reference Northstone and Emmett4). At age 3 years a fourth ‘snacks’ pattern was found(Reference North and Emmett2). The foods that made up the ‘processed’ pattern are similar to those in the ‘snacks’ and ‘puddings’ patterns in the present study. The ‘traditional’ pattern contained meat, poultry, potato and vegetables whilst the ‘health conscious’ pattern loaded highly for fruit, vegetables, fish, pasta and rice at ages 3, 4 and 7 and inversely with meats at 9 years. The three dietary patterns identified in the Australian study of 12- to 18-year-olds were a ‘fruit, salad, cereals and fish’ pattern, a ‘high fat and sugar’ pattern and a ‘vegetables’ pattern(Reference McNaughton, Ball and Mishra7). The results from these two studies and the present one suggest that intake of vegetables may be highly correlated with the intake of fruit in younger children but not necessarily in older children.
The present study found significant relationships between dietary patterns and socio-economic indicators. In general, ‘healthier’ dietary patterns (‘fruit and vegetables’) were associated with lower levels of deprivation, higher incomes and higher education levels of the main food provider whereas more ‘unhealthy’ dietary patterns (‘snacks’ and ‘puddings’) showed the opposite association. It should be noted, however, that although there were clear associations between dietary patterns and socio-economic status, the survey did not find as clear an association between socio-economic status and nutrient intakes, with a small increase in non-milk extrinsic sugars intake with increasing deprivation measured by SIMD and no difference in total fat or SFA intakes(Reference Sheehy, McNeill and Masson12) which is also consistent with data from other surveys(Reference Gregory, Lowe and Bates21).
In older girls, where fruit and vegetables loaded highly in separate components, socio-economic factors were related to ‘vegetables’ but not ‘fruit’, suggesting that intake of vegetables may be more influenced by socio-economic factors than intake of fruit. Very few other studies have investigated fruit and vegetables separately and most other studies of dietary patterns have found that fruit and vegetables aggregate together in the same component, similar to most of the subgroups in the present study. However, the results from the main analysis of the survey looking at individual food groups showed a stronger socio-economic gradient for vegetables than fruit in these children(Reference Sheehy, McNeill and Masson12).
Other studies have found similar associations between dietary patterns in children and socio-economic factors. The ALSPAC Study(Reference North and Emmett2–Reference Northstone and Emmett3) found that their ‘junk’ (renamed ‘processed’) pattern was related to socially disadvantaged households (council housing, financial difficulties, lower levels of maternal education), whereas their ‘healthy’ component was associated with higher maternal education levels. A Spanish study of 2-to 24-year-olds found similar associations with maternal education levels(Reference Aranceta, Perez-Rodrigo and Ribas6). Those children whose mother had a lower education level were more likely to have a ‘snacky’ pattern and less likely to have a ‘healthy’ pattern. However, the Australian study of 12- to 18-year-olds failed to find an association between the three dietary patterns identified in their sample and socio-economic indicators(Reference McNaughton, Ball and Mishra7).
In younger children education level of the main food provider seemed to be the socio-economic indicator most associated with dietary patterns after adjustment for the other indicators; however, in older children household income was the only indicator which remained significant after adjustment for the other socio-economic indicators. It is conceivable that as children age and become more independent and make more of their own food choices the amount of money available to them may become more of an issue and perhaps their own education level becomes more important than that of a parent. Most other studies have considered only one age group, focusing either on younger or older children and often considering different socio-economic indicators separately; therefore it is difficult to compare these findings. However, as highlighted earlier, the ALSPAC Study(Reference North and Emmett2–Reference Northstone and Emmett3) found that maternal education levels in particular were associated with dietary patterns in children up to 7 years of age and the enKid Study found that maternal education levels were associated with dietary patterns in 2- to 24-year-olds in Spain(Reference Aranceta, Perez-Rodrigo and Ribas6).
The present study found significant linear relationships between dietary patterns and physical activity and inactivity, with associations being more consistent with inactivity. ‘Healthier’ dietary patterns were consistently associated with less time spent in front of a screen for all age and sex groups whereas more ‘unhealthy’ dietary patterns were associated with more time spent in front of a screen for girls aged 5–11 years. However, time spent in physical activity was positively related to the ‘snacks’ component in 5- to 11-year-olds of both sexes and the ‘fruit and vegetables’ component in 5- to 11-year-old girls.
Few studies have examined associations between dietary patterns identified by PCA and physical activity. The Spanish study of 2- to 24-year-olds(Reference Aranceta, Perez-Rodrigo and Ribas6) found that their ‘snacky’ pattern was positively associated with time spent watching television in all ages and their ‘healthy’ pattern inversely associated with time spent watching television in 2- to 13-year-olds. Similar to the present study, the Australian study of adolescents(Reference McNaughton, Ball and Mishra7) failed to find a consistent association between dietary patterns and physical activity in 15- to 18-year-olds, with only their ‘high fat and sugar’ pattern being associated with higher levels of physical activity. In general, other studies have concentrated on the relationship between diet and inactivity (usually measured by television viewing), with few looking at the relationship with physical activity. More ‘unhealthy’ diets have been found to be associated with greater time spent watching television(Reference Muller, Koertzinger and Mast22–Reference Miller, Taveras and Rifas-Shiman25).
There was little association between dietary patterns and overweight and obesity. The only significant relationships were in boys aged 5–11 years, with the lowest intake of ‘snacks’ and the highest intake of ‘fish and sauce’ in obese children. The few other studies that have looked at dietary patterns identified by PCA in relation to obesity in children have also failed to find significant associations(Reference Reilly, Armstrong and Dorosty5–Reference McNaughton, Ball and Mishra7) apart from one which found a positive association between an ‘animal foods’ pattern and overweight in Korean preschool children(Reference Shin, Oh and Park9). Another study utilising cluster analysis also failed to find a significant association between dietary patterns and BMI in German children(Reference Alexy, Sichert-Hellert and Kersting26). The lack of an association between diet and BMI could be due to the fact that the data are cross-sectional. The results could be influenced by dieting and/or under-reporting, particularly in the overweight and obese children, but as no information was collected on dieting behaviour and no adjustment of the data has been made for possible under-reporting, the results should be interpreted with caution. However, in general there is a lack of consistent association between diet and BMI in children in the literature(Reference Rodriguez and Moreno27–Reference Newby28).
There are several strengths and limitations of PCA which are discussed in detail by other authors(Reference Michels and Schulze1, Reference Newby and Tucker29). One of its main limitations is that it involves several subjective decisions, in particular choosing the number of components to retain and naming them and, also, grouping of foods if researchers choose to do so. As mentioned earlier, in order to minimise the number of subjective decisions we chose not to pre-group the foods before analysis. However, PCA requires a large sample size due to the fact that the analysis is based on a correlation matrix of the variables, and correlations usually need a large sample size before they stabilize. It has been suggested that a sample size of 200 is fair and 300 is good(Reference Comrey and Lee30), but it has also been suggested that the ratio of sample size to the number of variables is important, with anything from two to ten subjects per variable being recommended(Reference Costello and Osborne31). The sample sizes for the various age and sex groups in the present study ranged from 250 to 381, meaning that they were fair to good by the first definition; however, as there were approximately 140 items on the FFQ this would need 280–1400 subjects to fulfill the second criteria for sample size. The minimum number of 280 was achieved in the younger age groups but not quite in the older children. Pre-grouping the foods may have meant that some of these new groups would have contributed more to components than individual items; however, it would not be clear which of the foods within these new groupings were most important.
The main advantage of PCA is that it combines information across the diet based on food intakes, taking into account the complexity of the diet and revealing underlying food consumption patterns, meaning that it may be more relevant to dietary choices than analyses based on individual foods and/or nutrients.
There are two main approaches to dietary patterns, either a hypothesis-driven a priori approach, or an empirical a posteriori approach(Reference Michels and Schulze1, Reference Newby and Tucker29, Reference Schulze and Hoffmann32). In the hypothesis-driven a priori approach, the diet is given a score based on the degree to which it conforms to dietary recommendations such as the Healthy Eating Index(Reference Kennedy, Ohls and Cralson33), the Diet Quality Index(Reference Patterson, Haines and Popkin34) or the Mediterranean Diet Score(Reference Trichopoulou, Kouris-Blazos and Wahlqvist35) and this score is used as an exposure variable. The score does not reflect overall diet, but rather specific aspects of it that are thought to represent a healthy diet. This method, however, is limited by current knowledge of diet–disease relationships and the choice of which foods and/or nutrients to include in the score and what cut-off points to apply. The empirical approach relies entirely on data, and patterns are derived using statistical techniques. PCA is the most popular of these approaches but cluster analysis is also commonly used. Cluster analysis differs from PCA in that it aggregates individuals into mutually exclusive subgroups or clusters with similar diets rather than grouping foods that are consumed together. Individuals can only belong to one cluster; therefore, there can be a risk of misclassification which does not exist for PCA. Further analysis such as comparing profiles across clusters is required to interpret these patterns, whereas in PCA the patterns are directly interpretable. A more recent addition to dietary pattern methodology is reduced rank regression(Reference Hoffmann, Schulze and Schienkiewitz36). This is of most use when studying diet–disease relationships, as it takes account of biological pathways by identifying patterns associated with biomarkers of disease. It is therefore considered a combination of a priori and a posteriori methods. However, it requires information on disease biomarkers.
Using PCA to generate dietary patterns shows which foods tend to be eaten together, and relating these patterns to other factors such as demographics, lifestyle and health helps to tailor and set priorities for health promotion and also to better understand the role of diet in relation to disease risk.
In conclusion, distinct dietary patterns could be identified in school-age children, which differed slightly between age and sex groups. The predominant dietary pattern seemed to be a ‘healthy’ one containing vegetables, in combination with fruit for the younger children. This pattern was consistently associated with higher socio-economic status (less deprivation, higher income and higher education level of the main food provider) and less time spent in inactivity. Although clear associations were found between dietary patterns and socio-economic factors, it should be noted that this does not necessarily translate into large differences in nutrient intakes between socio-economic groups; therefore analyses of dietary patterns should be combined with analyses of nutrients, as studying either one alone could be misleading. This may have important implications for targeting health promotion messages at subgroups in terms of changes required to current food intakes. Clear associations between ‘unhealthy’ dietary patterns and inactivity were found but there were no consistent associations between dietary patterns and physical activity except for the suggestion in younger children that more unhealthy ‘snacks’ patterns were positively associated with physical activity. This suggests that ‘healthy’ lifestyle factors do not necessarily cluster together in the same children but ‘unhealthy’ ones do tend to cluster together. This also has implications for health promotion strategies in children which should focus on ‘healthy’ lifestyles including both diet and physical activity messages.
Acknowledgements
The original study and secondary analysis were funded by the Food Standards Agency Scotland (contracts S14029 and S14039).
The authors would like to acknowledge the statistical advice from Dr Andrea Sherriff, University of Glasgow and Dr Lorna Aucott, University of Aberdeen.
G. McN. and L. F. M. conceived of the original study, with L. C. A. C., J. I. M. and B. A. H. providing input. L. C. A. C. conducted the statistical analysis and drafted the manuscript, which was reviewed by all authors.
The authors state that there are no conflicts of interest.
Appendix 1. Factor loadings for principal components analysis
Components for 5- to 11-year-olds
* Factor loadings < − 0·3 and >0·3.
Components for 12- to 17-year-olds
* Factor loadings < − 0·3 and >0·3.
† Consumed by < 5 % of girls.
‡ Consumed by < 5 % of boys.
