Measuring Food Intake 271with other sociodemographic factors of the type nor-
mally included in nutrition studies, such as gender,
age, education, socioeconomic status, and smoking,
were inconsistent.
10.7 Evaluation of food intake data
Recognizing the impact of underreporting
As indicated in Section 10.5, dietary studies are often
conducted in order to compare food and nutrient
intake between different groups in the population, to
determine the proportion of individuals at risk of
dietary inadequacy or excess, or to determine the
habitual intake of individuals.
In each case it is important fi rst to assess the valid-
ity of the data. For most investigators the use of the
Goldberg cut-offs is currently the most practical
option to indicate whether, and to what extent, the
results are likely to be biased. However, to use the
Goldberg cutoffs effectively dietary studies need to
include:
● measurements of weight and height, to be able to
estimate BMR from equations
● questions on activity level to provide guidance on
suitable PALs for evaluation of both mean and indi-
vidual data.
While the characteristics of “true” underreporters
(as opposed to LERs identifi ed by a single EI:BMR
cut-off ) remain to be confi rmed, the associations
consistently observed between high BMI, weight con-
sciousness, and low energy reporting suggest that, in
addition, questions on self-perception of body shape,
dieting, and dietary restraint may also help in identi-
fying true underreporters.
It cannot be overemphasized that it is always
important to examine all dietary intake data critically
because false conclusions generate false hypotheses
that may take years to be disproved.
A classic case was the luxus konsumption hypoth-
esis, namely that lean individuals are energy prodigal
and obese individuals energy effi cient. This hypothe-
sis was generated by studies apparently showing that
obese persons did not consume more energy than
their lean controls. Subsequently, DLW studies dem-
onstrated beyond doubt that obese persons recruited
for studies of obesity grossly underreported their
food intake.
Allowing for the effects of underreporting
Although techniques for handling biased dietary data
have been developed, most are complex However,
the following suggestions serve to promote critical
examination of data and wariness in drawing
conclusions.
If the proportion of individuals who report implau-
sibly low intakes of energy differs between population
subgroups of interest, then any comparisons between
them that do not take this into account will be biased.
One way to draw attention to the possibility of bias
between groups is to report not only the mean or
median energy intake of the groups being compared
but also the EI:BMR ratio. If differences are evident
then the groups should be compared both with and
without LERs included. One problem that arises is
that by subdividing the groups the sample size is
reduced and imprecision increased, so that a differ-
ence of biological signifi cance may be missed, not
because it does not exist but because the sample size
is too small to detect it statistically.
When dietary inadequacy or excess is the question
of interest, it is again important to consider
LERs separately. Energy intake is highly correlated
with the intake of many nutrients and, consequently,
intake of nutrients is also likely to be underestimated
in underreporters and more likely to indicate
inadequacy relative to recommendations for nutrient
intake. An alternative approach is to compare
nutrient intake per unit energy for both groups. If
this differs between LERs and the rest of the popula-
tion it provides evidence that the reporting of food
intake is also likely to be selective. The nutrients for
which signifi cant differences are observed can also
provide clues as to the types of food likely to be
involved.
Studies that have examined macronutrient intake
between respondents above and below a given value
of EI:BMR have generally found that the percentage
of energy derived from protein was higher and that
from fat lower in LERs than in non-LERs. Results for
carbohydrate have been more variable, but, when
separated into starch and sugars, energy from starch
tended to be higher and energy from sugars lower in
LERs. Nutrient density also tends to be higher for
most nutrients in LERs than in non-LERs, providing
further indication of differences in food patterns
between the two groups.