Direct observation of dietary intake is often undertaken as a means of validating a dietary assessment method by providing an objective measure of dietary intake. Eating behaviours including items and amounts of foods consumed, received or given away, or spilled can all be noted . However it is not a feasible method for obtaining habitual diet data at either a group or an individual level. Dietary dimensions assessed by direct observation are summarised in table D.3.1.
Table D.3.1 Dietary dimensions assessed by direct observation.
|Dietary dimension||Possible to assess?|
|Energy and nutrient intake of a total diet||Yes|
|Intake of specific nutrients or food||Yes|
|Infrequently consumed foods||Yes|
|Frequency of eating/meal occasions||Yes|
|Adult report of diet at younger age||Yes|
Observation of diet should be carried out in a systematic way using methodology that is standardised across observers and observation periods. Researchers observe participants directly and note down their eating behaviour during a defined time period (e.g. during school lunch) .
The observation period
Observation typically takes place over a relatively short period, for example a one hour lunch break. However, some studies have reported much longer observation periods, such as 12 hours of continuous monitoring with participants being followed throughout this time .
The recording form
In structured settings such as school canteens and hospitals it may be possible to know in advance the foods to be served and the standard portion sizes. In these instances the recording form must list the foods served with space for the observer to note the serving size provided and the percentage of that serving which is actually consumed . It is often necessary to account for the amount of food dropped or shared, especially in children. In more free-living scenarios the recording form must be more flexible in order to capture the wide variety of possible intakes, including spaces for :
Portion sizes can be obtained by weighing plate waste, visual estimation or use of likert-scale ratings. For example, Baglio et al.  used the following ratings:
Minimising the impact of observation on normal behaviour
It is important that the observation takes place with minimal intrusion on the participant(s) so as to reduce the effect of reactivity bias. Strategies include:
It is crucial that observers undergo extensive training and that quality control procedures are in place to ensure accurate and consistent recording and data processing, especially if portion sizes are to be recorded. It is important that observers record intake in the same way, as weights/volumes can be difficult to reconcile. Such training can include :
It is essential that inter-observer reliability is assessed and clearly reported in research outputs. If it is not, it remains unclear whether differences between individuals represent true variation or depend upon who was conducting the observation . Figure D.3.1 presents an example of calculation of inter-rater reliability.
Figure D.3.1 Example of calculation of inter-rater reliability for a single participant.
Inter-rater reliability calculation:
Number of items = 11
Number of items with agreement within ¼ serving = 7
Percent agreement = (7 ÷ 11) * 100 = 63.6%
Direct observation can be used in the following situations:
Note inference of dietary intake is not always needed. For example, a research question could be: “How many children in a school select fried products and sugar-sweetened products in a buffet-style school lunch?” Then, direct observations in a school dinner hall could provide the answer to this question without any estimation or additional inferential steps.
Key characteristics of direct observation are summarised in Table D.3.2.
Table D.3.2 Characteristics of direct observation.
|Number of participants||Small|
|Cost of development||Low|
|Cost of use||High|
|Researcher burden of data collection||High|
|Researcher burden of coding and data analysis||High|
|Risk of reactivity bias||Yes|
|Risk of recall bias||No|
|Risk of social desirability bias||Minimised|
|Risk of observer bias||Yes|
|Participant literacy required||No|
|Suitable for use in free living||No|
|Requires individual portion size estimation||Yes|
Direct observation can be implemented for anyone as long there are no ethical restrictions. Considerations relating to the use of direct observation for assessing diet are summarised in Table D.3.3.
The target can be a population as a whole, rather than each individual in that population. For example, as above mentioned, if a research question is about a proportion of children who select fried or sugar-sweetened products in a school lunch, the output can be produced without any estimation of individuals’ diet.
Table D.3.3 Diet assessment by direct observation in different populations.
|Infancy and lactation||Suitable.|
|Toddlers and young children||Records of foods shared/dropped more likely to be required.|
|Adolescents||Records of foods shared/dropped more likely to be required.|
Informed consent needs to be obtained as usual by providing detailed, but succinct, information on the direct observation method. The burden could be greater than an individual anticipated at the time of providing consent. Thus, individuals should be informed with emphasis about the freedom of requesting termination of direct observations.
A method specific instrument library is being developed for this section. In the meantime, please refer to the overall instrument library page by clicking here to open in a new page.