Insights into dietary recall from a longitudinal study: accuracy over four decades.

We studied the validity and consistency of memory of foods consumed decades earlier in a longitudinal population of 91 persons. The memory of middle-aged persons for food intake years earlier did not decline invariably over time, although median correlations between actual and recalled consumption decreased. Time-related memory loss varied greatly from food to food. Neither analyses of group mean differences and SDs over all foods nor those for individual foods provided clearcut evidence of time-related memory loss. Systematic biases in memories of dietary intakes in the distant past were evident, depending on current diet and varying according to individual foods and food groups. These discrepancies may represent either misremembering or the effects of inferential processes. Examination of the consistency (reliability) and validity of reports of food intake indicated that validity estimates provided by mean differences between recalled and actual intakes were more revealing than correlation coefficient. Consistent reports were not necessarily valid, as indicated by memories of food intake. Correct inferences may have been made even when the actual memory was lost.