Background: Dietary surveys often rely upon subjects' ability to estimate their portion sizes instead of weighing them, but information on the validity of such estimates is lacking. This study investigated the validity of estimated portion sizes from a self-administered food frequency questionnaire including photos.
Methods: Validity was evaluated by comparison with portion sizes from 14-day weighed food records. Eight sets of photos were included in the study (three meat dishes, one fish dish, three vegetables, potatoes), each set with four options. Subjects (59 men, 85 women) were 40-64 years.
Results: The majority of subjects (85-95%) selected the most correct photo or a neighbouring photo. However, regression analyses showed that the relationship between the estimated and the measured portion sizes was relatively weak for most of the foods tested, and correlations were only significant for three of the eight foods. This apparent contradiction was partly due to the pattern of the error: subjects selecting small portions had underestimated their actual portions while subjects selecting large portions had overestimated them. This error pattern could not be explained by the theoretical error resulting from the use of four discrete portion sizes instead of continuous weights, neither was it a simple consequence of the unidirectional errors at the extreme ends. Actual portion sizes varied from day to day, with intra-individual coefficients of variation of 34-40%.
Conclusions: The photos used in the present study were of limited value for ranking individuals correctly according to their actual portion sizes. Whether this is a common feature of photos used in food frequency questionnaires remains to be answered.