Dietary Assessment Primer

Summary Tables: Recommendations on Potential Approaches to Dietary Assessment for Different Research Objectives Requiring Group-level Estimates

These summary tables provide an overview of potential approaches to dietary assessment and indicate which approaches we recommend for different research objectives, given the current evidence. The research objectives considered are limited to those that require group-level estimates of dietary intake, not those that require estimates of intake for a given individual (Learn More about Deriving Group-level Estimates from Individual-level Intakes). Further, our focus is mainly on objectives related to estimating [glossary term:] usual dietary intakes, though we do also consider estimating acute intake when relevant.

The top row of the table indicates a continuum across the approaches in terms of the type and amount of information provided, analytic flexibility, investigator and respondent burden, and cost. However, the recommendations themselves are driven by the quality and appropriateness of the approach for the specific research objective. In other words, the recommended approaches may reflect ideal circumstances in which practical considerations, such as cost, respondent burden and administration time, do not impose constraints.

Details of Recommendations and Further Considerations, which accompanies these tables, provides more details on our recommendations and discusses further considerations about data capture and data analysis issues involved in studies that address the major types of research objectives.

For more information on the types of instruments featured in the tables and discussed in Details of Recommendations and Further Considerations, see the Instrument Profiles. For an at-a-glance comparison of the major features of these dietary assessment instruments, see the Compare Dietary Assessment Instruments table.

For more information about determining the [glossary term:] accuracy of self-report instruments and issues in [glossary term:] measurement error - two critical issues in conducting dietary assessment research - see Key Concepts.

Recommendations on potential approaches to dietary assessment for different research objectives requiring group-level estimates a,b,c

Approach
More:
  • Information provided
  • Flexibility for analyses
  • Investigator and respondent burden cost
arrow indicating less towards the right of the table Less:
  • Information provided
  • Flexibility for analyses
  • Investigator and respondent burden cost

Return to top

Describing dietary intake
x(24HR) + FFQ
Multiple administration of 24HR plus FFQ on whole sample
x(24HR)
Multiple administration of 24HR on whole sample
24HR + x(24HR)
Single administration of 24HR on whole sample plus repeat(s) on subsample
FFQ/SCR + x(24HR)
FFQ/SCR on whole sample plus multiple administrations of 24HR on subsample
24HR
Single administration of 24HR on whole sample
FFQ
FFQ on whole sample
SCR
Screener on whole sample
1. Mean (µ) usual intake
More than necessaryd More than necessaryd More than necessaryd Acceptable, if calibrated to 24HRe Recommended Possiblei Possiblei
2. Difference between two groups in mean usual intake A - µB)
More than necessaryd More than necessaryd More than necessaryd Acceptable, if calibrated to 24HRe Recommended Possiblei Possiblei
3. Mean (µ), proportion of individuals (p) above/below some threshold, and percentiles (qp) of usual intake (usual intake distribution)
More than necessaryd Recommended Acceptable, if subsample large enough Acceptable, if calibrated to 24HRe Possibleh Not recommendedf Not recommendedg
4. Mean (µ), proportion of individuals (p) above/below some threshold, and percentiles (qp) of intake on a given day (acute intake distribution)
More than necessaryd More than necessaryd More than necessaryd Not recommended Recommended Not recommended Not recommended

Return to top

Examining association between diet as an independent variable and a dependent variable (e.g., health status)
x(24HR) + FFQ
Multiple administration of 24HR plus FFQ on whole sample
x(24HR)
Multiple administration of 24HR on whole sample
24HR + x(24HR)
Single administration of 24HR on whole sample plus repeat(s) on subsample
FFQ/SCR + x(24HR)
FFQ/SCR on whole sample plus multiple administrations of 24HR on subsample
24HR
Single administration of 24HR on whole sample
FFQ
FFQ on whole sample
SCR
Screener on whole sample
5. Regression coefficients (β) for prospective or cross-sectional studies
Ideal Recommended Recommended Acceptable, if calibrated to 24HR, using regression calibratione Not recommended j Possiblei Possiblei
6. Regression coefficients (β) for retrospective studies
Not possible Not possible Not possible Not possible Not possible Possiblef Possibleg

Return to top

Examining association between an independent variable (e.g., self efficacy or socioeconomic status) and diet as a dependent variable
x(24HR) + FFQ
Multiple administration of 24HR plus FFQ on whole sample
x(24HR)
Multiple administration of 24HR on whole sample
24HR + x(24HR)
Single administration of 24HR on whole sample plus repeat(s) on subsample
FFQ/SCR + x(24HR)
FFQ/SCR on whole sample plus multiple administrations of 24HR on subsample
24HR
Single administration of 24HR on whole sample
FFQ
FFQ on whole sample
SCR
Screener on whole sample
7. Regression coefficients (β) for prospective or cross-sectional studies
More than necessary Recommended Acceptable Acceptable, if calibrated to 24HR, using regression calibration Acceptable Possiblei Possiblei
8. Regression coefficients (β) for retrospective studies
Not possible Not possible Not possible Not possible Not possible Possiblef Possibleg

Return to top

Examining the effect of an intervention k
x(24HR) + FFQ
Multiple administration of 24HR plus FFQ on whole sample at each point in time
x(24HR)
Multiple administration of 24HR on whole sample at each point in time
24HR + x(24HR)
Single administration of 24HR on whole sample plus repeat(s) on subsample at each point in time
FFQ/SCR + x(24HR)
FFQ/SCR on whole sample plus multiple administrations of 24HR on subsample at each point in time
24HR
Single administration of 24HR on whole sample at each point in time
FFQ
FFQ on whole sample at each point in time
SCR
Screener on whole sample at each point in time
9. Determining change, between two points in time, in mean usual intake of group post - µpre)
More than necessaryd More than necessaryd More than necessaryd Acceptable, if calibrated to 24HRe Recommended Possiblei Possiblei
10. Determining difference between two groups in mean usual intake change (µ(post - pre)i)A - (µ(post - pre)i)B
More than necessaryd More than necessaryd More than necessaryd Acceptable, if calibrated to 24HRe Recommended Possiblei Possiblei
11. Determining difference between two groups in post-intervention mean usual intake A, post - µB, post)
More than necessaryd More than necessaryd More than necessaryd Acceptable, if calibrated to 24HRe Recommended Possiblei Possiblei
12. Determining change, between two points in time, in proportion of individuals above/below some threshold (p post - ppre)
More than necessaryd Recommended Acceptable, if subsample large enough Acceptable, if calibrated to 24HRe Possibleh Possiblei Possiblei
13. Determining difference between two groups in change in proportion of individuals above/below some threshold
(pA, post - pA, pre) - (pB, post - pB, pre)
More than necessaryd Recommended Acceptable, if subsample large enough Acceptable, if calibrated to 24HRe Possibleh Possiblei Possiblei
14. Determining difference between two groups in post-intervention in proportion of individuals above/below some threshold (pA, post - pB, post)
More than necessaryd Recommended Acceptable, if subsample large enough Acceptable, if calibrated to 24HRe Possibleh Possiblei Possiblei

Return to top

24HR: 24-hour recall
FFQ: food frequency questionnaire
SCR: screener

a This table refers to the assessment of any food or nutrient consumed as part of diet (not including supplements).

b For approaches involving more than one instrument, the first instrument listed is considered the main instrument (Learn More about Combining Instruments).

c Throughout the table, whenever 24-hour recalls are mentioned, food records could be employed in the same fashion because they also capture short-term intake with more detail than a frequency instrument. However, there is much less evidence available on the measurement error properties of food records, and the act of recording is known to cause reactivity.

d Approach will work, but provides more data than necessary to provide estimate for this objective.

e Energy-adjustment is recommended but is not possible with screeners.

f Approach is possible, but not recommended because absolute values are likely to be [glossary term:] biased. Energy-adjusted values may be less biased.

g Approach is possible, but not recommended because absolute values are likely to be biased, and energy adjustment is not possible with screeners.

h Estimated distributions will be too wide unless an estimate of the ratio of within-person to the sum of the within- plus between-person variation from an external study with a comparable sample is available.

i Estimates are likely to be biased but this bias can be corrected to some degree if frequency instrument is calibrated to less-biased instrument through an external calibration study.

j Approach is possible, but measurement error correction is not possible.

k Intervention associated recall bias, if present, contradicts the assumption that there is no [glossary term:] differential bias between control and intervention groups. In addition, [glossary term:] reactivity bias in food records may be particularly problematic for evaluating the effect of interventions.