Comprehensive FFQs have been developed to assess the total diets of children and adolescents in the United States , Norway , the United Kingdom  and Italy  with a limited number of validation studies undertaken. These studies demonstrated that for these populations, dietary intakes could be measured reasonably well using an FFQ. Differences in study methods, populations and between-person variation make it difficult to compare validity across studies internationally. However, it is worth noting that the results from the present study are similar to those of previous FFQ validation analyses for children and adolescents [25, 47, 49–53].
The validity of the ACAES FFQ was assessed by comparing the nutrient estimates from the average of the FRs with ACAES FFQ2 as well as comparing the average of the FRs with the average of ACAES FFQ1 and ACAES FFQ2. Although ACAES FFQ2 represents the conceptually appropriate time sequence, the process of keeping FRs might alter awareness of food intake and artificially improve accuracy in completing it . However, comparing ACAES FFQ1 with the FRs, would tend to underestimate validity because ACAES FFQ1 asked about diet prior to the study period. Therefore, the use of the average of ACAES FFQ1 and ACAES FFQ2 provides a combination of minimal and maximal estimates and is likely to be a more accurate indication of true validity .
Food frequency questionnaires tend to estimate higher nutrient intakes when compared to 24-hour recalls and FRs . This was shown in the present study, with the ACAES FFQ providing higher estimates of intakes for all nutrients, except polyunsaturated fat, carbohydrate, thiamin, niacin and vitamin C, when compared to assisted FRs.
The correlation results from this study are comparable to those for the YAQ . This was anticipated given that ACAES was modified from the YAQ and both studies had a similar design. When the correlation results for relative validity of the ACAES FFQ are compared to the YAQ , the correlation for carbohydrate is equivalent (r = 0.47), while the correlations for fat, saturated fat, folate, iron and riboflavin (Vitamin B2) are marginally lower. The correlations for beta-carotene, magnesium and zinc were higher for the ACAES FFQ compared to the YAQ. Correlations for all other nutrients were either lower (energy, protein, monounsaturated fat, fibre, thiamin, vitamin C, vitamin A, calcium), not significant (polyunsaturated fat, niacin and retinol) or not common to both studies (sugars, niacin equivalent).
The mean correlation coefficients for the common nutrients between both studies (without vitamin supplementation) were calculated, excluding those nutrients in our study with correlation coefficients that were not significant (>0.05) (polyunsaturated fat, niacin and retinol). Rockett observed a mean correlation of 0.42 between the 2nd YAQ and the mean of three 24-hour recalls , compared to 0.39 for this study, indicating similar correlations between the two methods in both studies.
When compared to the correlations for other comparative validity studies, the ACAES FFQ showed similar results. The ACAES FFQ had higher correlations than the Block Kids Questionnaire (BKQ) for fibre and calcium, but not for energy . Our study had almost identical correlations for calcium as two FFQs designed specifically to assess calcium intake in children and adolescents in the US and Italy [50, 48, 51]. As found in the present study, others have also shown poor agreement for vitamin A , polyunsaturated fat [52, 55] and protein [47, 53]. This is probably due to the large day-to-day variation in the intakes of these nutrients by children and adolescents, particularly girls. To reflect the usual intakes of vitamin A, polyunsaturated fat and protein, a greater number of recording days is necessary, with as many as 46 days recording required to estimate the usual vitamin A intake for females aged 5-17 years .
In order to detect associations between diet and disease, it is suggested that correlations need to be at least 0.3 or 0.4 . The present study had significant correlations greater than 0.3 for the deattenuated data of all nutrients except protein and vitamin A.
The reproducibility results of the YAQ  were similar to the present study. The mean correlation for the YAQ was 0.41 for the log-transformed, energy adjusted data for the 7 nutrients presented. For the 21 nutrients examined in the ACAES FFQ reproducibility analysis, the mean correlations were 0.46 for raw data, 0.44 for transformed, 0.32 for transformed, energy-adjusted data. In the present study, the correlations for many nutrients dropped following transformation and again following energy adjustment. The square-root transformation was applied to all nutrients due to its appropriateness for the majority of nutrients for the ACAES FFQ1 and FFQ2 data. However, it was not the most statistically appropriate transformation for some nutrients in ACAES FFQ2 and is likely to be the reason the correlation coefficients for these nutrients dropped following transformation. The reduction in the reproducibility correlations following energy adjustment has occurred in other studies of reproducibility [57–60] and is likely to be due to systematic errors of over- and under-estimation between ACAES FFQ1 and FFQ2. Although we have no way to assess this directly, adjustment for energy increases correlation coefficients when the variability in nutrient intake is related to energy intake, but results in lower correlations when the nutrient variability depends on systematic errors of overestimation and underestimation . The correlation coefficient for reproducibility for total fat intake in our study was higher for crude data (0.49) when compared to the reproducibility correlation (0.41) for an FFQ for children aged 2-5 years developed specifically to assess fat intake  demonstrating good validity for total fat intake.
Bland Altman analyses have not often been reported in studies comparing FFQs with diet records in children and adolescents . In the only other study reporting Bland Altman results for children (11-13 years) , the findings were similar. The mean differences were positive and of very similar magnitude to the current study, with the limits of agreement wide for all nutrients presented (energy, fat, sugar, calcium and protein). Similarly, the Bland Altman plots from the raw data showed strong trends of increasing difference with increasing intakes. The Bland Altman results for the present study show that the ACAES FFQ is not suitable for estimating absolute intakes for children and adolescents, but is appropriate for ranking intakes. The similarity of results between the present study and that of Lietz et al , suggests that large variation in the agreement between methods may be characteristic of child/adolescent populations and that these results may be due to the usual variability of dietary intake of children and adolescents. Agreement between FFQs and 24-hour recalls or food records may be lower in children and adolescents than in adults due to the greater day to day variation in their dietary intake . Adolescents have highly variable food patterns, with possibly half of the foods they eat varying greatly .
The reference method of choice for FFQ validation studies is weighed food records or diet records . Although 24-hour recalls have less respondent burden, their sources of error tend to be more correlated with the error in an FFQ due to reliance upon memory, conceptualisation of portion sizes and distortion of reported diet . There was a high proportion of children aged 9-12 years (71.7%) in the sample and it was expected that some of these participants would be less likely to complete a food record. The 24-hour recall method was undertaken with participants that had not completed their food records due to it's suitability to participants with limited cooperation or literacy . Interestingly, approximately the same proportion of younger children (9-12 years) and older children (13-16 years) completed 24-hour recalls (36.8% and 38.1% respectively) because they had not completed a food record.
Standard portion sizes were applied to the ACAES FFQ because frequency has been found to be more discriminatory than portion size [63, 64]. Adding open-ended questions regarding portion size can actually reduce validity of an FFQ due to sources of error in conceptualising serve sizes and large, within-person variations in serving sizes when the same food is consumed on different occasions .
A general limitation of validation studies is that the results are not necessarily transferable to another population . A sample size of at least 50 is desirable for each demographic group , and ideally between 100 and 200 participants . The sample size in the present study was inadequate to compare the validity and reproducibility of subsets for age, gender or BMI category. Performance of the ACAES FFQ would need to be tested in populations of different SES and ethnicity.
A total of 43 participants were excluded due to implausible energy intakes reported on the ACAES FFQ, representing 19% of the number recruited. A total of 36 of these were for high intakes, while seven were for implausibly low intakes. This is a high proportion and we feel that the main reason was the completion of the ACAES FFQ in a class environment. It seemed that by completing the FFQ as a group, there may have been more distractions compared to completing it on their own or with a parent present. Some of the excluded students marked the highest intake category for almost all food items. In a subsequent study by the same research team using the ACAES FFQ with primary school children, only 2 out of 60 participants (3.3% of the sample) were excluded due to implausible energy intakes (data not shown). These children completed the FFQ independently at the school, rather than in a class environment. In future studies, it is recommended that the ACAES FFQ is completed by the child independently, rather than in a class environment.
Despite these limitations, there are many aspects of the study design and analysis that are likely to contribute to an underestimation of the true validity and reproducibility of the ACAES FFQ. Food records were used predominantly (63% of records) and combined with 24-hour recalls (37% of records). Due to differences in the biases of food records and FFQs, the predominant use of food records is likely to underestimate validation coefficients when compared to 24-hour recalls as the reference method . The day of the week is also likely to influence the results of validation studies. In the present study, the proportion of records collected on weekdays (75%) and weekends (25%) is close to the actual proportion of weekdays and weekend days (71% and 29%, respectively). When comparing 14 days versus 2 days of food records in Australian children, Jenner et al found that correlation coefficients were slightly higher when both days of the two day records were taken on weekdays . It is possible that this is due to a perceived 'usual' intake during the week and greater variation on weekends. Participants reported a 'usual' intake for 64% of the records collected. Of the remaining 36% of records, half were reported as 'less than usual' with the other half reported as 'more than usual'. This large proportion of 'more' or 'less' than usual intakes may be typical of the intakes of this age group, but is likely to contribute to reduced agreement with the FFQ, where they record their perceived 'usual' intake. The conservative approach to excluding records from the data analysis also contributes to underestimating the validity of the FFQ. For example, respondents with food records that were inconsistent with their FFQs were not excluded if their responses were plausible. The most extreme examples of these were intakes of liver in the FFQ, but none reported on the FR days (resulting in particularly poor agreement for vitamin A and retinol) and reporting items in their food records that were not on the FFQ (eg. oysters, popcorn, slurpee, added sugar, chocolate topping, sherbet).
Although 24-hour recalls capture rich information on food consumption, they measure episodically consumed foods poorly. Recent statistical modelling has suggested that combining FFQ data with a limited number of 24-hour recalls may provide the most accurate method of estimating usual dietary intake at the individual level, supporting the use of both methods in national surveillance .