Examining the Association between an Independent Variable & Diet as a Dependent Variable

NOTE: Links for “Row” in the text below will take you to the relevant portion of the Summary Table.

Data Capture Considerations

Researchers are sometimes interested in examining the [glossary term:] association between one or more [glossary term:] independent variables (e.g., socioeconomic status) and diet as a dependent variable. Statistically, the association is examined through a regression model of the dietary intake on the [glossary term:] independent variables of interest. One can do this through [glossary term:] prospective, [glossary term:] cross-sectional, or [glossary term:] retrospective studies.

  • Regardless of the study design, one needs to consider the potential for [glossary term:] differential response bias (see Key Concepts about Measurement Error). These considerations also arise when Evaluating the Effect of an Intervention on Diet. Differential response bias is problematic because it can lead to spurious differences between the groups and/or reduced [glossary term:] statistical power to detect associations. This is an issue because this type of analysis involves making comparisons among groups that may differ in characteristics, such as body weight, that are known to be associated with dietary reporting error. Because of this, objective sources of data, such as [glossary term:] biomarkers, observation, or sales or purchasing records, should be considered in place of self-report data. Further research is needed to guide approaches in this area.
  • For prospective or cross-sectional studies, we recommend that you use 24HRs (see the 24-hour Dietary Recall Profile). A single administration may be acceptable, but collecting and averaging repeat administrations will help improve your statistical power to detect effects (Row 7). If the dietary intake of interest has a [glossary term:] recovery biomarker, and its use is possible, collecting recovery biomarker data from a subsample can aid in adjusting for [glossary term:] measurement error (Learn More about Biomarkers).
  • No matter what type of study design you use, collect 24HRs evenly over the period of time in which you are interested, whether that be an entire week or a full year.
  • If it is not possible to collect 24HRs from all study participants, use an FFQ as your main instrument and collect 24HR data (and/or recovery biomarker data, if feasible) from a subsample (see the Food Frequency Questionnaire Profile). This may allow you to use statistical techniques to adjust for [glossary term:] bias. However, as noted above, unless the [glossary term:] measurement error in the 24HR and FFQ data are independent of all the [glossary term:] covariates in the model, differential response bias is likely to be an issue. In such cases, extreme care must be taken in interpreting the results. If an FFQ must be used without additional information from a less-biased instrument, estimates of the association will be prone to the bias that results from the biasedness of the FFQ, over and above any bias resulting from differential response bias.
  • A screener also may be acceptable if you are interested only in one or a few specific dietary components that are concentrated in relatively few food sources and if you can adjust estimates for bias using [glossary term:] scoring algorithms developed using less-biased intake data, such as 24HRs (see the Screeners Profile and Learn More about Scoring Algorithms for Screeners).
  • Carefully select your FFQ or screener to ensure that it is tailored to the population and captures the majority of food and beverage sources within the food supply for the dietary component of interest.
  • For retrospective studies, an instrument querying about past diet is the only choice possible (Row 8). Possibilities include an FFQ or screener. In this case, energy adjustment can be used only with an FFQ to aid in adjusting for measurement error.

Data Analysis Considerations

  • Differential response bias in self-reported dietary intake can be an issue in analyses aimed at examining the effects of one or more variables on diet if the variables of interest are associated with reporting error in diet. Examples include body mass index and education. It is a good idea to consider whether any [glossary term:] independent variables known to be associated with dietary misreporting are included in your analysis. If such independent variables are being considered, then rather than making comparisons between different categories of these variables, they should be used to adjust the comparisons for other variables, either by stratification or by inclusion as a confounding variable in the model.
  • In studies in which you have multiple 24HRs, averaging the 24HRs will increase [glossary term:] power to detect effects.
  • If you did not account for [glossary term:] nuisance effects, such as day of week or season, in your study design, control for them in your analysis (Learn More about Day-of-Week Effect and Learn More about Season Effect). Even if you did try to spread the recalls evenly, that may not have worked out perfectly and controlling for them is advisable.
  • Less-biased instruments, such as 24HRs or [glossary term:] recovery biomarkers, may be used in conjunction with FFQs to adjust for bias. However, how well this works in practice is not known and more research is needed.
  • Energy adjustment of estimates obtained from an FFQ also may reduce bias, but energy adjustment is not possible with screeners because they do not capture total intake.

References and Resources

The following references and resources provide additional information on the topics discussed in this section.


National Cancer Institute. Dietary Screener Questionnaire in the NHANES 2009-2010.

National Cancer Institute. Short Dietary Assessment Instruments.

National Cancer Institute. Usual Dietary Intakes.

National Health and Nutrition Examination Survey (NHANES). NHANES Dietary Web Tutorial.