Principles Underlying Recommendations

Self-report dietary assessment instruments can be categorized into two broad groups — [glossary term:] short-term instruments and [glossary term:] long-term instruments. Most studies using these instruments involve [glossary term:] usual dietary intake (i.e., long-run daily average) as the key variable of interest.

Short-term instruments include 24-hour dietary recalls (24HRs) and food records, which typically capture intake for a single day (referred to as acute intake) or a limited number of days. Short-term instruments can be used to estimate usual intake, so long as there are a sufficient number of administrations of the instrument and the administrations are appropriately spread across the time period of interest. Long-term instruments, such as food frequency questionnaires (FFQs) and screeners, aim to directly capture usual intake. They are not intended to be used to examine intake on a given day.

The primary focus of this Primer is on 24HRs as a short-term instrument and FFQs as a long-term instrument because most research conducted to date on [glossary term:] measurement error in dietary intake data has focused on these instruments.

The following principles are the foundation for the recommendations provided in this Primer.

All dietary studies should aim to collect data with as little bias as possible

All data collected using self-report dietary assessment instruments contain measurement error, which includes [glossary term:] random error and [glossary term:] systematic error. The most troubling of these is systematic error, also known as [glossary term:] bias, because of its potential for generating erroneous conclusions. Both random error and bias can be addressed with statistical methods. However, whereas random error can be addressed with repeat administrations, bias requires an additional, unbiased measure of dietary intake. This is a problem given that few unbiased measures exist. For this reason, it is preferable to use an instrument that collects dietary data with the least bias possible.

24-hour dietary recalls capture intake with less bias than do food frequency questionnaires

Because intake estimates from 24HRs appear to be less biased than are estimates from FFQs (see Key Concepts about Measurement Error), we recommend the use of 24HRs for many applications (including those related to both usual intake and acute intake). Estimates from 24HRs are also likely to be less biased than estimates from screeners, which share many characteristics with FFQs. From the perspective of estimating usual intake, data collected using 24HRs do contain [glossary term:] within-person random error (mostly driven by day-to-day variation), but this can be accounted for by using repeat administrations and appropriate statistical methods. Importantly, in the absence of a gold standard for [glossary term:] true intake for most dietary components, approaches developed for the use and analysis of 24HRs are based on the statistical assumption that the data are unbiased for true intake on the recalled day. [glossary term:] Recovery biomarkers, which provide an unbiased estimate of true intake, are known for a few dietary components and can be used to adjust for measurement error, including both random error and bias (Learn More about Biomarkers).

Food frequency questionnaires can be useful when efforts are made to reduce bias

Despite their bias, FFQs may be useful in certain situations, such as exploring relationships between diet and health status, especially when bias is adjusted for with the help of more accurate measures, including recovery biomarkers and less-biased dietary assessment instruments, such as 24HRs. Energy adjustment also can be useful in reducing bias in frequency data (Learn More about Energy Adjustment).

Combining different types of instruments is a valuable approach in some circumstances

Given that all self-report instruments have limitations, in some situations it may be advantageous to administer a combination of instruments rather than only one. The available evidence suggests that [glossary term:] combining instruments is a particularly useful approach for studies aimed at elucidating the [glossary term:] association of dietary [glossary term:] exposures with another variable (for example, a health outcome) (Learn More about Combining Instruments). Combining data from self-report instruments with recovery or [glossary term:] concentration biomarker data also is a possibility for mitigating the effects of measurement error on observed relationships between diet and another variable. Additional studies will help to inform the utility of combining different types of instruments for other research objectives.

Further research is needed to guide recommendations for the use of food records

Although food records share a number of characteristics with 24HRs, they have some important differences. For example, the real-time recording inherent in completing a food record can cause participants to change their dietary intakes in response to the act of recording (Learn More about Reactivity). On the other hand, food records are less dependent than 24HRs on a participant’s memory, as intakes may be recorded at the time they occur.

Most studies examining measurement error in dietary intake data conducted to date have focused on 24HRs rather than food records. As a result, further research is needed before specific guidance can be provided on the use of food records for different research objectives. Nonetheless, it is likely that many considerations applicable to 24HRs, such as the need for repeats on non-consecutive days for the estimation of [glossary term:] usual dietary intake distributions, also apply to food records (Learn More about Usual Dietary Intakes).