Measure selection is critically important but is only one part of the process in terms of research related to dietary behavior. Considerations regarding how data will be analyzed should come into play early in the study design process. This is fundamental due to interconnections between the most appropriate analyses to arrive at the desired estimates, the measures used, and parameters for their administration (e.g., number of repeat measures of short-term instruments, timing of repeat measures within an intervention study, incorporation of biomarkers). Due to this link between data collection and analytic techniques, collaborating with statisticians early in the process is encouraged.

Detailed recommendations regarding the collection of dietary data for different types of studies are outlined in the National Cancer Institute’s (NCI’s) Dietary Assessment Primer.^1,56 For example, the Primer notes considerations relating to whether or not addressing the research question of interest requires estimation of usual intake distributions. If so, it is necessary to collect repeat recalls or records on at least a subsample of the target population and to conduct statistical modeling to account for day-to-day variation. In cohort studies aimed at enabling analyses between dietary exposures and outcomes, it is advisable to collect data using a recovery biomarker if possible, or a less-biased self-report measure compared to the main dietary instrument, among a subsample. Data from this calibration sub-study can then be used to conduct regression calibration to reduce error in data from the main dietary instrument. For epidemiologic studies, collecting concentration biomarker data also may be helpful in mitigating error. In intervention studies, it is important to consider the potential for differential biases in that those exposed to some intervention designed to alter eating patterns may misreport diet differently than those who were not exposed. Means of dealing with differential bias are not well developed. Thus, it is advisable to collect objective measures to corroborate dietary intake data whenever possible within intervention studies. This differential error can also come into play in observational studies in which groups compared differ with respect to factors that might affect reporting error. Consulting the Dietary Assessment Primer in combination with the Measures Registry can help researchers identify the broad range of considerations that should be taken into account when planning a study, choosing measures, and analyzing and interpreting data.

Once a measure has been selected, it is critical to consider how to optimize the data captured to the extent possible, again keeping in mind the particular challenges faced in characterizing dietary intake among children (see Section 4). Also of import are databases linked to measures of dietary intake to arrive at estimates of nutrient and food consumption. These databases should be current and comprehensive, to the extent possible given that this is usually out of the control of researchers. Key limitations for the estimation of certain dietary components should be outlined when reporting research results. Other issues related to processing of data, such as dealing with outliers, are highlighted in NCI’s Dietary Assessment Primer.

Increasingly, when investigators measure dietary intake, they are attempting to assess overall dietary or eating patterns rather than quantifying consumption of particular aspects of diet, such as fat intake. As defined by the 2015–2020 Dietary Guidelines for Americans, eating patterns “represent the totality of what individuals habitually eat and drink.”²⁴ There is growing recognition that dietary components act synergistically and that eating patterns may be more strongly related to health than individual foods or nutrients. Measuring eating patterns is complex because they are characterized by multidimensionality and dynamism.¹¹⁸ In other words, individuals eat and drink many different foods and beverages (i.e., multidimensionality), all of which have their own profiles in terms of nutrients and other dietary components such as phytochemicals. For some individuals, this complexity is compounded by the contributions of vitamin and mineral supplements to total intake. Further, eating patterns vary temporally (i.e., dynamism)—within a day, across days, across seasons, and across the lifecycle—possibly in relation to critical points, such as infancy or the transition to adolescence. Various methods for capturing patterns of dietary intake have been developed; these include the use of investigator-defined indices identified a priori to assess the quality of diet relative to some pre-determined criteria.^119,120 For example, the Healthy Eating Index-2010 assesses the alignment between dietary intakes and the Dietary Guidelines for Americans.^121–123 Data driven, or a posteriori approaches, include the use of statistical techniques, such as cluster and factor analysis, to look for patterns in data and relate these to health or disease outcomes.^119,120 Methods for capturing patterns is an area of ongoing inquiry, and approaches that embrace the true extent of multidimensionality and temporality¹²⁴ inherent in dietary patterns require further development.

Considerations regarding data interpretation and reporting are also important to contribute to a robust body of evidence with which to inform interventions for childhood obesity prevention. In interpreting the results of studies making use of measures of dietary behavior, it is key that measurement error, which is unavoidable in self-report measures, and its implications for study results are considered and discussed. Despite the fact that error implicit in the measurement of dietary intake among children has been long recognized, this is often not indicated when data are reported and inferences based upon them made.³¹ This can lead to a confusing and contradictory body of literature. In the context of obesity, it is critical to consider potential interactions between body weight, or factors linked with body weight, and self-reporting of dietary behavior and the potential implications for the results (as well as whether a particular analysis is advisable given the likelihood of differential bias). In developing publications to share research findings, attention to the Strengthening the Reporting of Observational Studies in Epidemiology–Nutritional Epidemiology (STROBE-nut) guidelines¹²⁵ may assist in achieving improved transparency in terms of measures used, how they were administered, their psychometric properties in relation to the target population, and other salient issues necessary for the critical appraisal of any study making use of dietary behavior measures.