INGESTIVE BEHAVIOR AND OBESITY

Eating Behavior: Lessons From the Real World

of Humans
John M. de Castro, PhD
From the Department of Psychology, Georgia State University, Atlanta, Georgia, USA
Food intake by normal humans has been investigated both in the laboratory and under free-living
conditions in the natural environment. For measurement of real-world intake, the diet-diary technique is
imperfect and tends to underestimate actual intakes but it appears to be sensitive, can detect subtle
inuences on eating behavior, and produces reliable and valid measures. Research studies in the real world
show the multivariate richness of the natural environment, which allows investigation of the complexities
of intake regulation, and even causation can be investigated. Real-world research can overcome some of
the weaknesses of laboratory studies, where constraints on eating are often removed or missing,
facilitatory inuences on eating are often controlled or eliminated, the importance of variables can be
overestimated, and important inuences can be missed because of the short durations of the studies.
Real-world studies have shown a wide array of physiologic, psychological, and social variables that can
have potent and immediate effects on intake. Compensatory mechanisms, including some that operate
with a 2- to 3-d delay, adjust for prior excesses. Heredity affects all aspect of food-intake regulation, from
the determination of body size to the subtleties of the individual preferences and social proclivities and
the extent to which environmental factors affect the individual. Hence, real-world research teaches
valuable lessons, and much more is needed to complement laboratory studies. Nutrition 2000;16:
800813. Elsevier Science Inc. 2000
Key words: intake regulation, patterns, diet diaries, food intake
INTRODUCTION
Even though eating behavior has been intensively studied over the
past century, an understanding of how it is regulated has been
elusive. The research has not yielded practical results, and as a
result science has failed to produce effective therapeutic interven-
tions for over- and under-weight. In fact, the problem of over-
weight and obesity, rather than being resolved, has increased
dramatically.
1
It is the thesis of this presentation that part of the
problem resides with the focus on laboratory research. Some of the
strategies employed in laboratory research produce results that are
valid for the laboratory but miss essential variables that differ
between the laboratory and the real world. This results in an
overestimation of the importance of some variables, the underes-
timation of the importance of others, and missing altogether other
salient inuences.
The differences between laboratory and naturalistic research
are primarily due to differences in the degree of control over
sources of variance. The high degree of control that occurs in the
laboratory is both its virtue and its curse. The control allows the
researcher to isolate independent and dependent variables such that
causal connections can be readily discerned. However, the re-
searcher has to know ahead of time which variables are important
to investigate and which need to be eliminated or controlled. In a
situation where the important contributors to the behavior may be
unknown, the researcher may well eliminate or control very salient
variables and thereby miss some important contributors to the
behavior. Also, the fact of control can create an articial situation
that humans can respond to in ways that do not reect their typical
behavior. In addition, because laboratory studies eliminate most
sources of variance other than the independent variable, the pro-
portion of the variance accounted for by this variable can appear
articially large.
In the real world most behaviors are affected by multiple causal
factors. This multivariate environment allows for compensations,
modulation, and interactions between variables that are difcult
to detect in the laboratory. A variable that may signicantly alter
behavior in the laboratory may have little or no effect in the real
world due to a compensatory reaction or a modulating inuence of
another variable. These compensatory or modulating variables
may well be eliminated or controlled in the laboratory and thus the
importance of the independent variable may be grossly overesti-
mated. For example, a modulating variable that is present in the
real world is the cost of food. This can markedly reduce the
inuence of otherwise salient variables such as palatability or even
hunger. Also, as will be reviewed later, intake may be affected by
a short-term variable on the day that it is studied in the laboratory.
The participants intake may be markedly increased or decreased
by a potent environmental variable and no immediate compensa-
tion is apparent. However, a compensatory reaction occurs 2 to 3 d
later.
2
This is missed in the laboratory because 2- to 3-d periods are
normally not investigated.
Of course, the lack of control in natural-environment studies
presents a separate set of challenges. The lack of control makes the
identication of causal factors difcult and the large numbers of
variables that are present and simultaneously inuencing the be-
havior makes it difcult to discern the inuence of subtle factors.
However, the development of modern multivariate techniques has
markedly relieved this problem. In fact, although variables are not
physically controlled, their inuence can be mathematically con-
trolled and their inuence extracted from the variance in the
Supported in part by grant R01-DK39881 from the National Institute of
Diabetes and Digestive and Kidney Diseases and by a grant from the
Georgia State University Research Program Enhancement Fund.
Correspondence to: John M. de Castro, PhD, Department of Psychology,
Georgia State University, University Plaza, Atlanta, GA 30303-3083,
USA. E-mail: jdecastr@[Link]
Date accepted: May 24, 2000.
Nutrition 16:800813, 2000 0899-9007/00/$20.00
Elsevier Science Inc., 2000. Printed in the United States. All rights reserved. PII S0899-9007(00)00414-7
dependent variable before the effect of an independent variable is
analyzed. Linear-structural modeling techniques allow for assess-
ment of subtle interactive inuence and consideration of theoret-
ical structures of causation.
One of the most difcult challenges for real-world research is
that in the natural environment it is difcult to devise ways to
validly and reliably measure the variables of interest. It is this
problem that dissuades most researchers from venturing into the
real world. It is a thesis of this presentation that the measurement
difculties are overestimated and it is, in fact, not difcult to
reliably and validly measure behaviors of interest. The methods are
not perfect, but their imperfections can be understood and factored
into the reasoning process, such that meaningful conclusions can
be reached from the data. Hence, I rst present and discuss the
methodology employed to investigate eating behavior in the real
world along with its strengths and its shortcomings.
MEASURING FOOD INTAKE IN THE REAL WORLD
The Diet-Diary Method
There are many methods used to estimate the nutrient intakes of
humans (for review, see Thompson and Byers
3
). However, these
methods are designed to measure overall usual-intake levels and
do little or nothing to measure actual eating behavior. This limits
the ability to identify the factors that affect intake. I and my
colleagues have been developing and using the 7-d diet-diary
method
4,5
to measure the detailed eating behavior by humans in
their natural environment and have included measures of contex-
tual and psychological variables. The participants are asked to
record for 7 consecutive d, in a pocket-sized diary, their eating
behavior, feelings, and the environmental context. The 7-d period
has been found to be an acceptable duration for maintaining the
subjects motivation and tolerance of diary recording and still
produce regular, stable, and interpretable data. This procedure has
good reliability, with reasonable agreement between records re-
peated after as much as 2 y.
610
Even though the diet-diary method is a self-report, it has been
found to have a high degree of validity when tested empirically.
Gersovitz et al.
11
had elderly subjects keep a 7-d diet diary.
Unknown to the participants, the experimenters surreptitiously
measured the actual amount of food consumed at lunch. They
found good agreement between the diary records and the actual
amounts eaten. Also, Krantzler et al.
12
using a similar technique
with college students found 87% agreement between 7-d records
and surreptitious measures of intake. The 13% disagreement usu-
ally involved non-reporting of small items, especially condiments.
Hence, there is reasonable evidence that the diet-diary reports of
intake are imperfect but reasonably reliable and valid.
To help ensure that reliable and valid records will be kept, we
offer a contingent reinforcer; a detailed nutritional-composition
analysis of the reported diets. The subjects usually express great
interest in this report and have participated willingly with this
analysis as the only incentive for participation.
13,14
The partici-
pants are informed that record keeping often produces a change in
nutrient intake and they should eat normally because their dietary
nutrient composition report will be of use to them only if it
accurately reects their normal intake. As another motivator and
validity check, we contact two individuals who ate with the subject
during the recording period and ask them to verify the subjects
diary entries. The subjects are aware that this check will be
performed. After more than 1000 verications, we have found that
45% of the meals are fully validated without any prompting, 50%
are veried with prompting, and 5% cannot be recalled.
Recently, we have added a new wrinkle.
5
Subjects are given
cameras and are asked to take a picture of the foods that they are
eating before ingesting them. The diaries are then coded as usual,
without the benet of the pictures, and then independently using
the photographs to aid in the coding. The inclusion of the pictures
increased the estimates of the amounts of the macronutrients
ingested in the meals by 6% to 9%. Hence, this procedure would
appear to help overcome a potential problem with underreporting.
In addition, it is possible that the fact of taking the pictures
increases the accuracy of the written records because the subjects
know that the pictures will be seen with their entries.
Underreporting and Reactivity
Even though the diet-diary procedure appears to be reliable and
valid, there is considerable evidence that reported intakes are about
20% below what should be ingested by weight-stable subjects of
comparable activity levels. Diet-diary records produce estimates of
energy intakes that are 18%
15
to 23%
16
below the levels that have
been empirically determined to be required to maintain body
weight. The doubly labeled water technique provides an accurate
measure of energy expenditure. In some studies, good correspon-
dence has been found between these estimates and those produced
by diary self-report methods.
1720
However, a number of studies
have suggested a serious, approximately 20%, difference between
measured expenditures and the intake estimates produced by
self-reports.
9,17,18,21,22
To assess the degree to which the intakes reported in our diet
diaries were representative of the participants typical daily in-
takes, reported intakes were compared with the estimated basal
metabolic rate (BMR
est
) for each participant. Basal metabolic rate
was estimated from the participants weight by considering age
and sex according to the procedure outlined by Schoeld et al.
23
The ratio of the reported daily food-energy intake (EI) to the
BMR
est
was calculated for each participant (EI:BMR
est
)
.
A rea-
sonable cutoff for identifying unrepresentative intake is EI:BM-
R
est
1.1.
24,25
This includes participants whose intake is at least
10% above their BMR
est
. For the entire sample, the mean EI:
BMR
est
was 1.28 (0.013). This mean was about 18% below the
expected level of 1.55 for typical intake
24
and was commensurate
with the estimates of diet-diary reported-intake differences from
expenditure estimated with the doubly labeled water tech-
nique.
9,17,18,21,22
Almost exactly two-thirds of the participants were
above the cutoff (67.9%). These results suggest that the intake
estimates that were obtained with the diet-diary technique are on
the order of 18% below the level that would be expected for the
participants to be maintaining their body weights.
These low levels may or may not present an interpretive prob-
lem, depending on the nature of the questions being addressed in
the research. When the absolute amounts of nutrients ingested are
required, these levels can be a problem. However, in most research
applications, the quantities ingested by one group or condition are
compared with those ingested by another group or condition.
Although subjects may differ in the degree to which they tend to
report low intake, these individual differences would be expected
to be distributed randomly among groups. Hence, the error created
by low estimation would be expected to affect different groups
approximately to the same extent and thus would not systemati-
cally change relative values. Hence, the errors should not produce
confounding probabilites but rather increase type I error probabil-
ities.
26
Indeed, the relative differences between males and females
remain the same even with the low estimation of absolute in-
takes,
15
as would also appear to be true for the differences between
obese and lean subjects.
16
The literature does suggest that caution
must be exercised when comparing adolescents or obese subjects
with other groups because the estimation may be lower in these
groups.
17,18,21
Hence, the technique is not recommended for mea-
surement of absolute intakes among groups but appears to produce
reliable, valid, and reasonably accurate intake records when ap-
plied to nonobese adult subjects.
These studies suggest that, regardless of the veracity of the
reporting in the diaries, diary recording can produce reports of
Nutrition Volume 16, Number 10, 2000 801 Real-World Eating Behavior
intake levels that are not representative of the individuals typical
daily intake. This may be due to underreporting of ingested nutri-
ents, reactivity to the measurement procedure, or a combination of
both. Underreporting involves an inaccurate report of actual in-
take. Conversely, reactivity involves an unintentional or inten-
tional decrease in intake during the recording period. Goris and
Westerterp
27
presented evidence that reactivity-producing under-
eating is responsible for low-intake estimates in lean women but
that a mix of underreporting of fat intake and undereating is
responsible in obese men.
28
It is not possible within the current
data set to ascertain whether the low-intake reports were due to
underreporting or to reactivity-induced undereating.
To investigate the extent to which the diet-diary meal intake
results might have been biased by the inclusion of participants who
reported unrealistically low intakes, an analysis was performed
comparing the intakes from participants who were above and
below the EI:BMR
est
1.1 cutoff.
24,25
The participants who
reported intakes that were above the cutoff had average meal sizes
that were 6.6% larger than those who reported intake below the
cutoff value. However, the relation between meal size and other
signicant variables such as palatability and social facilitation did
not differ between groups.
The correlations between six different variables that are
known to affect meal size and the size of the meals ingested
were calculated for both the low- and acceptable-reporting
groups (Fig. 1). These correlations were calculated separately
for each subject and then averaged across groups. The correla-
tion between the number of people present and the meal size,
reecting social facilitation,
2931
was positive and signicant
and did not differ between groups. The correlation between the
time of day that the meal occurred and the meal size, reecting
diurnal rhythms,
32
was positive and signicant and did not
differ between groups. The correlation between self-reported
hunger and the meal size, reecting motivational state,
33,34
was
positive and signicant. The low reporters correlations were
slightly but signicantly smaller than those of the acceptable
reporters. The correlation between self-rated palatability and
meal size, reecting hedonic responsiveness,
35,36
was positive
and signicant and did not differ between groups. The correla-
tion between the duration of the interval before the meal and the
meal size, reecting responsiveness to deprivation,
14
was pos-
itive and signicant and did not differ between groups. The
correlation between the estimated content of the stomach at the
beginning of the meal and the meal size, reecting responsive-
ness to prior energy intake,
14,37
was negative and signicant and
also did not differ between groups.
The evidence presented and reviewed indicates that the diet-
diary method produces reliable and valid results and that the errors
that tend to occur are not a problem in most research applications.
It is clear from the current analysis that the relations between
signicant variables that affect eating do not appear to be changed
by low reporting. The correlation data demonstrate that the rela-
tions between signicant variables that affect eating are not
changed by low reporting, suggesting that this measurement prob-
lem does not affect the nature of the conclusions that can be
reached about the factors that affect eating.
Nevertheless, caution must be exercised with regard to inter-
preting the absolute levels of nutrients reported because these tend
to underestimate customary intakes due to undereating and/or
underreporting. In addition, there are potential problems when
using the technique to compare certain groups, in particular obese
and adolescent subjects, because these groups tend to produce
estimates that underestimate customary intake to a greater extent
than typical comparison groups.
17,18,21
The technique is only use-
ful if the subjects are highly motivated and competent. This re-
quirement restricts the usefulness of the technique for investigating
the intake of uncooperative groups. We were able to apply the
technique to investigating the meal patterns of head-injured sub-
jects by having the caregivers maintain the records, but this was
laborious and it was difcult to motivate the caregivers to accu-
rately maintain the records.
38
INTAKE IN THE REAL WORLD VERSUS THE
LABORATORY
With the diet-diary technique, we have been able to assess intake
regulation in free-living normal humans in their natural environ-
ments. By comparing what has been learned with this technique
with the results of laboratory studies, it has become apparent that,
in comparison with real-world research, there are a number of
signicant shortcomings with laboratory research. In particular,
real-world constraints on eating are often removed or missing in
the laboratory, real-world facilitatory inuences on eating are often
controlled or eliminated in the laboratory, the importance of vari-
ables can be overestimated in the laboratory, and important inu-
ences can be missed because of the short durations of laboratory
studies. In addition, real-world studies can be used to ascertain
causation, and the multivariate richness of the real world is an asset
in the study of intake regulation. Each of these issues is discussed
in the following sections.
Real-World Constraints on Eating Are Often Removed or
Missing in the Laboratory
There are circumstances in the real world that constrain the amount
of intake at a meal and the timing of meals that are often absent in
the laboratory setting. A major factor that profoundly inuences
real-world intake is the cost of food. In animals it has been
demonstrated that the cost of intake can have a profound effect
on the intake pattern.
39
This constraint is most often missing in the
laboratory environment. This free feeding can lead to marked
overingestion or gorging in laboratory research. For example, with
the diet-diary technique we nd that a typical lunch meal averages
2.4 MJ.
40
Adjusting for an underestimation of 20% yields 2.9 MJ.
Yet in laboratory studies where subjects are allowed free access to
foods, intakes are sometimes seen at over double the levels that we
observe, from 6.1 to 6.9 MJ,
41
4.6 to 6.4 MJ,
42
6.7 to 7.3 MJ,
43
or
4.3 to 5.3 MJ.
44
For snacks, we typically nd, using the diet-diary
technique, that an average snack is about 1.7 MJ.
40
In contrast, in
FIG. 1. The mean correlations between the amounts ingested in the meals
and the number of other people present, the minute of the day of meal
initiation, the self-ratings of hunger, the self-ratings of palatability, the
amount of time since the last meal, and the estimated stomach content of
food energy for subjects whose intake was less (black bar) and greater (gray
bar) than 10% above their estimated basal metabolic rate (BMR).
802 de Castro Nutrition Volume 16, Number 10, 2000
laboratory settings, where snack foods are free, levels of 2.8 to
4.4 MJ are seen.
45
Hence, the laboratory environment, by remov-
ing the constraint of cost on intake, can result in abnormally high
intakes. In fact, the cost of food, or the lack thereof, may be the
most potent variable present in these studies, but it is rarely
recognized as such.
In the real world there are constraints not only on the amount
of food ingested in a meal but also on the timing of meals. This
produces subtly different patterns of intake in the laboratory than
seen in the real world. The so-called postprandial relationship is a
case in point. Many years ago it was discovered that rats eating ad
libitum in the laboratory delay their next meal by an amount that
is proportional to the amount of food eaten in the preceding meal
(Fig. 2). This was shown by a signicant correlation between the
size of the meal and the following interval until the next meal.
4648
In contrast, the duration of the interval before the meal is not
related to the size of the meal ingested. This relation between meal
size and after-meal interval has also been observed in humans
eating under ad libitum conditions in the laboratory.
49
This led to
the conclusion that regulation occurs by adjusting the time be-
tween meals, meal frequency, and not meal size.
However, it is clear that in the real world humans adjust intake
primarily by adjusting the amount eaten in the meals and not the
timing or frequency of eating.
50
In addition, in the natural, normal,
human environment, the meal size is found to be predicted by the
prior interval and the relation between meal size and after-meal
interval is non-signicant.
13,14
In other words, for humans in the
real world, meal size is adjusted based on the period of time since
the previous meal, but how long they will wait till the next meal is
not affected by how much they have just eaten. This is the
complete opposite pattern to that observed in the laboratory
(Fig. 2).
The key difference between the laboratory and the real world
that is probably responsible for the observed difference in the
patterns of intake is that in the natural, normal, human environ-
ment, the timing of meals is constrained by external schedules
dictated by work and social commitments. The human is fre-
quently not free to adjust eating time. Thus, regulation occurs by
adjusting how much is eaten, which suggests that the laboratory
ndings of a relation between meal size and after-meal interval
should be observed only under completely ad libitum conditions of
testing and, conversely, should not be observed when the timing of
eating is constrained. We tested this notion in the laboratory by
constraining rats to eat only at particular times of the day. When
these constraints were imposed, the real-world humanlike pattern
was evident, such that the interval before the meal signicantly
predicted the meal size, whereas the relation between meal size
and after-meal interval was not signicant.
5152
It is interesting that this relation between prior interval and
meal size for humans was observed for North Americans. In
French culture, dining is a much more important event, and work
and socializing are often scheduled around eating rather than the
reverse, which is the predominant case in North American culture.
In a sense, eating is less constrained by outside schedules in France
than in North America. This cultural difference is reected in a
difference in the meal pattern. The meal intake of free-living
French subjects under ad libitum conditions shows a signicant
relationship between the meal size and the following interval.
35,53
In total, these results suggest that regulation is exible and can
occur by adjustments to either meal size or frequency, depending
on the environmental and social contexts. Because the environ-
mental and social contexts are frequently eliminated or held con-
stant in the laboratory, the inuence of these very important
variables on intake are rarely observed or recognized.
39
Real-World Facilitatory Inuences on Eating Are Often
Controlled or Eliminated in the Laboratory
Studies of intake by humans in the laboratory are usually sched-
uled around the convenience of the experimenter and the partici-
pants. As such, intake is usually studied during the daytime and
much research occurs around lunch time. Unfortunately, this strat-
egy may miss important diurnal or circadian inuences on intake.
Studies using the diet-diary technique
54
have demonstrated that
there are substantial and important changes in ingestive behavior
that occur over the course of the day. In the real world, as the day
progresses, meal sizes increase. At the same time, the following
interval until the next meal gets shorter and shorter (Fig. 3).
The satiety ratio is dened as the duration of the following
interval divided by the meal size (min/MJ) and gauges the period
of satiety produced per unit of food energy ingested. This satiety
ratio shows a marked decline over the course of the day and
becomes quite low during the late evening. Such a pattern can also
be discerned in rats in the laboratory.
55,56
These results suggest
that the satiating effects of food changes with the time of day. If
one is worried about the overingestion of food, then noontime is
not a good choice as the time for testing because at noon the
satiating effects of food remain rather high. It is only in the late
evening that the participant becomes much less satised by intake
and may be more vulnerable to overeating. This important inu-
ence can be entirely missed in laboratory research, where time of
day is held constant. It should be noted that these effects are
culturally dependent. The pattern of results described appears to be
true for North Americans and the Dutch. However, the French
FIG. 2. Schematic diagram of the intermeal intervals and the meal sizes
that were used to calculate the pre- and postprandial correlations.
FIG. 3. The mean meal sizes (left), interval until the following meal
(middle), and the satiety ratio (right) observed during the morning, after-
noon, and evening periods.
Nutrition Volume 16, Number 10, 2000 803 Real-World Eating Behavior
show a slightly different pattern in that noontime meals tend to be
the same size as evening meals.
53
In the laboratory, humans are normally studied on weekdays.
However, using the diet-diary technique to study real-world intake
has shown that humans eat differently on the weekends than during
the week (Fig. 4). In general, 8% more is ingested on weekends
than on weekdays.
57
It is doubtful that this represents anything
other than environmental and cultural effects,
58
and it is doubtful
that different effects would be observed in the laboratory if par-
ticipants were studied on the weekends. However, an 8% change in
intake is rather substantial and not observed in laboratory studies.
If the goal is to understand overeating, then studying the factors
that alter intake on the weekend may be a fruitful approach to
identifying factors promoting the overingestion of nutrients.
Another major inuence on intake that is controlled or removed
in laboratory research is the participants expectancies. The indi-
viduals beliefs about the foods presented can have a marked
inuence on intake. Shide and Rolls
59
labeled yogurt preloads that
did not differ in fat content as either low fat or high fat. The
participants who believed that the yogurt was low fat ate more at
a subsequent meal than those who believed it was high fat. In the
real world, such expectancies are manipulated by package labeling
and advertising and may be a major inuence on eating. Caputo
and Mattes
60
investigated this manipulation by supplying free-
living participants with a daily lunch for 6-d periods and instructed
them that it was normal, low fat, or high fat. In fact, the lunches did
not differ in fat content. However, the participants ate signicantly
more during the rest of the day when they thought it was low fat
than when they were led to believe that it was high in fat. In most
laboratory research projects, expectancies are controlled; in the
real world, they are not. These expectancies about the foods that
people ingest may have a more potent effect on intake than the
variable studied and manipulated in the laboratory and may over-
ride the effects of these variables in the real world.
A very signicant facilitatory inuence on intake that is usually
controlled or eliminated in the laboratory is the effect of the
presence of other people. Normally, in laboratory research with
both humans and non-humans species, the individual is isolated
and behavior is studied. In the real world, using the diet-diary
method, it has been demonstrated that this procedure eliminates a
very potent inuence on intake, i.e., social facilitation. To inves-
tigate social effects on intake, we simply looked at the amounts
ingested in meals eaten alone versus those eaten with other people
present. It was found that the meals eaten with other people present
were on the average 44% larger than meals eaten alone,
31
includ-
ing larger amounts of carbohydrate, fat, protein, and alcohol.
This social effect appears to occur as a result of an increase in
the duration of meals rather than an increase in the rate of intake
at the meal.
6162
Also, who the individual eats with inuences the
amount ingested in the meal.
63
Females eat signicantly more
(13%) when eating with a male than when eating with another
female, whereas males eat the same amount regardless of the sex
of their companion. Meals ingested with spouse, family, or friend
are signicantly larger (22%, 23%, and 14%, respectively) than
meals ingested with others but without family or friends present,
whereas meals ingested with coworkers are signicantly smaller
(16%).
In addition, it was shown that social facilitation is an orderly
phenomenon that can be measured with the correlation between
the number of people present and the amount eaten in the meal.
This social correlation has been found to be signicant and posi-
tive under many conditions and with many different groups. Social
facilitation can also be described as a functional relation between
meal size and the number of people present (Fig. 4). Meals eaten
with one other person present were 33% larger than meals eaten
alone, whereas 47%, 58%, 69%, 70%, 72%, and 96% increases
were associated with two, three, four, ve, six, and seven or more
people present, respectively. These data can be t nicely by a
power function,
30
which is characteristic of the social-facilitation
phenomena in general.
64
Much has been learned about this phenomenon, but it has been
learned primarily with observational studies. Subsequent to these
real-world observations, social facilitation was veried in the
laboratory.
65
However, controlling and eliminating social effects
in the laboratory by isolating the participant for study disallows
this very salient inuence on intake from being observed or doc-
umented. Only when the constraints on variables, such as the
social context, time of day, and subject expectancies are removed,
such as occurs in the real world, can the true characteristics of
these salient inuences on intake be revealed.
Causation Can Be Established in Real-World Studies
A purported shortcoming of real-world research is that causation
cannot be established. It is true that it is more difcult to establish
causation in real-world studies, but it can be done. To demonstrate
causation, three prerequisites must be met: correlation, time pre-
cedence, and nonspuriousness. Obviously, in the case of the num-
ber of people present, the rst prerequisite, correlation, is satised.
Because the people eating with the subject are there from the
beginning of the meal, the time precedence prerequisite is also
satised. The nal criterion, nonspuriousness, is more difcult to
deal with in the real world. With the lack of control over variables,
it is difcult to demonstrate that a third factor may not be respon-
sible for the observed covariation.
Spurious relationships can be investigated in real-world con-
texts. For example, the relation between the presence of other
people and meal size may result from a covariation produced by a
third factor. In particular, the time of day, alcohol intake, snacks,
locations such as restaurants, or weekends are possible covariates.
People could simply be eating more on occasions and in circum-
stances where more people tend to be present. To investigate these
potential artifactual explanations, meals were identied that oc-
curred under specic conditions. It was demonstrated that, al-
though the covariances existed, they did not account for the social
correlation. Strong, positive, and signicant correlations between
meal size and the number of other people present, social correla-
tions, were found separately for meals eaten during the breakfast
period, the lunch period, or the dinner period, eaten in restaurants,
at home, or elsewhere, eaten accompanied by alcohol intake or
without alcohol, for only snacks or only meals,
40
or for meals eaten
during weekdays or during weekends.
58
FIG. 4. The mean meal sizes observed over the 7 d of the week (left) and
in relation to the number of other people present (right).
804 de Castro Nutrition Volume 16, Number 10, 2000
Even with this evidence, because of the observational nature of
the research, it is not acceptable to conclude that the presence of
other people is the cause of the increased intake. To establish
causation, the number of other people present was actively manip-
ulated by instructing subjects to eat only by themselves for a 5-d
period, eat normally for another 5-d period, and eat only with other
people for a third 5-d period. The order of these periods was
randomized. In comparison with the normal instruction period, the
subjects ingested on average 212 kcal, or 11%, less per day when
instructed to eat alone.
66
This result suggests that the presence of
other people is indeed the cause of the increase in intake at meals.
Subsequent laboratory studies have supported the idea of a causal
connection between the presence of other people and increased
meal size. Clendennen et al.
65
demonstrated that, when subjects
were required to eat a test meal with one or three other subjects,
they ate signicantly more than when alone.
It should be noted that the magnitude of the effect of eating
alone, 11%, is considerably smaller than the magnitude of the
social-facilitation effects observed in unmanipulated contexts, as
reported above. In fact, the meal sizes reported during the
manipulated-alone condition were 20% larger than the alone meals
during the normal condition,
66
which may indicate that separating
naturally occurring meals that just happen to be eaten alone from
those that happen to be eaten with others may overestimate the
impact of social facilitation of eating. Alternatively, it may suggest
that the subjects compensate by increasing meal size in the alone
condition to bring overall intake to more nearly normal levels.
Another example of the investigation of causal connections in
real-world research involves the effect of alcohol ingestion on
overall food-energy intake. Alcohol is an important contributor to
human energy intake that is frequently ignored in laboratory re-
search. Alcohol intake in the real world appears to supplement
rather than displace food energy in the diet.
67
Using the diet-diary
technique to look at normal humans who did not ingest alcohol
versus moderate alcohol drinkers showed that the total food-
energy intakes of drinkers and nondrinkers are not substantially
different for carbohydrate, fat, or protein. However, drinkers in-
gested more total food energy because of the alcohol intake. In
addition, drinkers ingest comparable amounts of carbohydrate, fat,
or protein on days that they drink and on days that they do not
drink alcohol. However, on drinking days, more total food energy
is ingested because of the additional energy from alcohol.
These observations about alcohol intake could be due to a
causal connection between alcohol intake and total energy inges-
tion. However, this cannot be determined by the strictly observa-
tional studies. To investigate causation, Orozco and de Castro
68
simply required moderately alcohol-consuming participants to re-
frain from alcohol intake for a 5-d period. During the 5-d period of
normal alcohol intake, the participants ingested comparable
amounts of carbohydrate, fat, or protein as they did during the 5-d
abstinence period. However, due to the added alcohol food energy,
the participants ingested 9.2 MJ versus 7.6 MJ during the absti-
nence period.
These studies demonstrate that causation can be investigated in
real-world contexts. They demonstrate that the presence of other
people is a causal factor that increases the amount ingested at
meals. They also show that alcohol intake is a causal factor that
increases overall food-energy intake over the course of the day.
Hence, not only can observations be made in the real world to
uncover important variables in the control of intake, but causation
can be established by manipulating these variables.
The Importance of Variables Can Be Overestimated in the
Laboratory
One of the most frequent criticisms of the results of real-world
research projects is that the variables studied only account for a
very small proportion of the variance in intake. In general, it is true
that only small proportions of the variance can be accounted for by
the independent variable. However, this is actually a strength of
the method because the importance of the variable to actual intake
is demonstrated, whereas in the laboratory, large proportions of the
variance can be accounted for, but in such an articial manner that
the real importance of the variable can be misrepresented. In
laboratory studies, most if not all of the sources of variation in the
dependent variable are controlled or eliminated. As a result, be-
cause there are so few other inuences producing variance, the
independent variables in laboratory studies can appear to account
for large proportions of the variance (Fig. 5). This can result in an
overestimation of the importance of the variable. Conversely, in
the natural environment, with many other salient variables opera-
tive, the independent variable is only one of many inuences. It
may then only account for a small part of the total variation. This
difference can lead to an overestimation of the importance of
variables investigated in the laboratory and an underappreciation
of the salience of variables observed in real-world contexts.
The regulation of water balance is an essential physiologic
requirement. It occurs via the balance of intake with utilization and
excretion. Laboratory research has established that excretion is
curtailed and intake facilitated when uid homeostasis is chal-
lenged by either an increase in plasma osmolarity,
69,70
or a de-
crease in blood volume.
71,72
As a result, these stimuli have been
postulated as the controlling inuences on thirst. However, uid
intake usually occurs without any clearly dened decit in either
of these factors.
73
We have produced evidence with the dietary-diary technique
that uid intake occurs in humans primarily in association with
eating and that the amount and timing of uid ingestion are
determined primarily by eating.
74
In addition, it was found that
uid intake and thirst in real-world contexts were not related to the
depleting characteristics of ingested nutrients.
75
Sodium intake,
which would be expected to raise plasma osmolarity and should
increase thirst and uid intake, was found to be unrelated to intake.
In contrast, carbohydrate intake, which has only a very small
osmotic impact and its metabolism results in the production of
water, would be expected to be related negatively to uid intake.
In fact, the opposite was true, with carbohydrate intake being one
of the strongest and clearest positive predictors of uid ingestion.
Hence, the observations of people ingesting uids in their natural
environment suggests that the two stimuli that were identied as
the most important inuences in laboratory research were in fact of
little or no consequence in the control of everyday intake. In fact,
FIG. 5. Venn diagram illustrates the proportions of the variance accounted
for in laboratory and real-world research.
Nutrition Volume 16, Number 10, 2000 805 Real-World Eating Behavior
uid intake seems to occur primarily in conjunction with eating, in
excess of what is needed, and the remainder is eliminated in the
urine. The fact that humans mostly produce copious quantities of
dilute urine would support this contention.
76
Phillips et al.
77
ob-
served water intake and the contents of the blood and the urine in
healthy men during their working hours and could nd no changes
in body-uid variables associated with the spontaneous ingestion
of water. Hence, it would appear that, in the real world, unlike the
laboratory, osmotic and volumetric stimuli are rarely inuential in
the control of uid intake and thirst. Laboratory research resulted
in a remarkable overestimation of the importance of these
variables.
Another instance of a variable whose importance was overes-
timated in the laboratory is the hedonic pleasure derived from
food, i.e., palatability. The hedonic properties of food have been
shown in the laboratory to have a major impact on the amounts and
types of foods ingested,
7882
with the amounts ingested markedly
altered by manipulation of the avor, texture, or appearance of
food. However, in the real world, palatability has much less of an
impact.
We investigated palatability inuences on the ad libitum eating
behavior by analyzing the 7-d diet diaries of more than 500
free-living humans.
36
The participants recorded their intake and a
global rating of the palatability of the entire meal on a 7-point
scale. The results are summarized in Figure 6. Meals that were
rated the highest in palatability were 44% larger than meals that
were rated low in palatability (Fig. 6, right). This large impact is
commensurate to that observed in the laboratory. However, it was
found that most meals that are self-selected are palatable and that
only 9.3% are rated as unpalatable, with ratings below neutral
(rating of 4 in Fig. 6, left). As a result, the relation between
palatability and intake only accounted for approximately 4% of the
variance in meal sizes. These results were very similar to those
observed for the French and suggest that palatability operates
similarly on intake regardless of culture.
35
Palatability appears, from both the laboratory and the real-
world analysis, to be a potent inuence on intake. However, this
analysis markedly overestimates its real impact on everyday
ingestive behavior. It only accounts for a small fraction in the
variance in the amount eaten. One reason for the seeming
contradiction may have to do with a range restriction on the
ratings produced by the self-selection of foods for ingestion in
the natural environment. In the laboratory, the subject has much
less control over the palatability of the food and unappealing
food can be presented for ingestion. This can maximize the
effect of palatability on intake.
78,81,82
In the natural environ-
ment, this would rarely happen. People simply do not select to
eat bad food. In fact, it might be argued that they strive to
maximize the palatability of their meals. Hence, in the natural
environment, there is less opportunity for the inuence of
palatability to be expressed. As a result, palatability appears to
be only one, rather small, inuence in complex array of many
inuential factors. Thus, palatability accounts for only a small
proportion of the variance in intake.
FIG. 6. The number of meals per day and the proportion of subjects who used each of the seven levels of palatability ratings (left) and the relationship
between the palatability ratings and the mean meal intakes of the macronutrients (right).
806 de Castro Nutrition Volume 16, Number 10, 2000
Important Inuences Can Be Missed Because of the Short
Durations of Laboratory Studies
One of the drawbacks of laboratory research is the great difculty
in studying the behavior of the participant for any extended period.
As a result, the vast majority of laboratory studies are very short
term and involve only a few hours, at most, of study. This does not
allow the researcher to view responses and adjustments that may
entirely compensate for the effect of the manipulated variable.
Real-world studies afford the opportunity to view behavior over a
much longer time frame and thus can capture how the system
reacts to and adjusts for short-term uctuations in foods or the
environment.
As an example, Raben et al.
83
compared the effect of pregela-
tinized starch versus that of raw-potato starch in a test meal on
plasma glucose, lipids, and hormones and subjective satiety. They
found substantially lower hormone, glucose, and lipid levels after
ingestion of the resistant starch. In addition, the resistant starch
resulted in signicantly lower sensations of satiety. However,
when this manipulation was performed over a 4-wk period, total
food energy and macronutrient intake were not affected.
84
Hence,
the manipulation that was effective in suppressing satiety on the
short term had no impact on longer-term intake.
It is unusual for a laboratory research project to investigate
intake for longer than a single day. As a result, compensation for
intake of the previous day is not viewed or investigated, but this is
important for the understanding of intake regulation. If the short-
term mechanisms that are studied in the laboratory were the only
mechanisms responsible for regulation, then the total amount of
food energy ingested in a day would be relatively constant, pro-
vided that activity levels also were relatively constant. However,
this is not the case because food-energy intake changes consider-
ably from day to day.
8587
Hence, for regulation to occur, there
must be a compensatory mechanism available to increase intake in
response to a decit from the previous day and/or decrease intake
in response to a surfeit from the previous day. Negative feedback
from intake of 1 d must, in some way, occur to affect intake on
subsequent days. This should produce a negative correlation be-
tween food-energy intake on 1 d and that ingested on the next day.
However, the predicted signicant negative autocorrelations have
not been observed. They have been found to be small and pre-
dominantly positive.
86,87
We investigated this issue by using the daily intakes reported in
the 7-d diet diary that we had obtained in previous studies.
2
Autocorrelations were calculated between the amounts ingested in
1 d and during each of the 4 subsequent d. As in previous studies,
the correlations between food-energy intake on 1 d and that oc-
curring on the next day were not signicant. However, the auto-
correlations with the 2- and 3-d delays were signicant (Fig. 7).
The 2-d lag autocorrelations were signicantly stronger than with
other delays. Interestingly, there appeared to be macronutrient-
specic effects. Carbohydrate intake had a signicantly larger
negative correlation than did either fat or protein, with the amount
of carbohydrate ingested either 1 or 2 d later. Fat intake had a
signicantly larger negative correlation than either carbohydrate or
protein. with the amount of fat ingested either 1 or 2 d later.
Similarly, protein intake had a signicantly larger negative corre-
lation than either carbohydrate or fat, with the amount of protein
ingested either 1 or 2 d later. Hence, carbohydrate appeared to
maximally affect carbohydrate, fat, and protein 2 d after ingestion.
In addition, the individual macronutrients appeared to have
macronutrient-specic negative feedback, with a 2-d delay.
2,88
Autocorrelation assumes that the error terms from each of the
variables in the correlation are themselves uncorrelated. Violations
of this assumption can result in an overestimation of the proportion
of the variance that is accounted for by the regression. This
assumption is nearly always violated with autocorrelations. How-
ever, using a simplex linear-structural modeling analysis that does
not make this assumption supports the conclusions from the auto-
correlational analysis that the negative effect of ingestion of a
macronutrients is greatest on the subsequent intake of that partic-
ular macronutrients 2 d later. Hence, although there is a slight
negative feedback that tends to restrain intake 1 d later, it is not
until 2 d later that the effect is maximal. It continues on the third
day but disappears by the fourth day. This interesting and impor-
tant feedback mechanism was apparent only in the real-world data
because of the ease with which extended time periods can be
investigated.
In another example of the utility of the time periods available
in real-world data, we were able to investigate seasonal effects on
eating behavior. We simply looked at the daily intakes of partic-
ipants who happened to be recording in the diaries during the four
seasons. For most of the year, intake was not signicantly different.
However, in the autumn, intake increased by 11% to 14% com-
pared with the other seasons (Fig. 8). This increase in overall
intake occurred as a result of larger meal sizes, especially of
carbohydrate. Once again, the utility of real-world data is evident
for the exploration of events that are difcult to view with labo-
ratory studies.
The Multivariate Richness of the Real World Is an Asset
The real-world environment contains a vast array of variables that
inuence intake. This can be viewed as a noisy environment that
dees systematic analysis or as a rich mosaic lled with opportu-
nities to discover surprising and unanticipated relations. The real
world can reveal what is important. All the researcher has to do is
observe and measure. There is no need to anticipate in advance
what may or may not be important and then arrange the environ-
ment to isolate that factor. Rather, nature can be allowed to tell the
researcher what is important.
The difculty arises, however, with data from complex real-
world environments in separating the inuence of multiple inter-
FIG. 7. Autocorrelations calculated between the amounts ingested on 1 d
and the amounts ingested 1, 2, 3, and 4 d later.
Nutrition Volume 16, Number 10, 2000 807 Real-World Eating Behavior
correlated inuences. Fortunately, the development of multivariate
statistical techniques can allow for the separation of these inu-
ences. The analysis of the effects of prior macronutrient intake on
subsequent intake demonstrates how multiple linear regression can
be used to separate effect. The before-meal contents of the stomach
have a restraining effect on the amount eaten during the meal. This
is indexed by a signicant negative correlation between the food
energy estimated to be present in the stomach at the onset of a meal
and the meal size.
13,14
In other words, the more there is in the
stomach, the less food that will be eaten.
To examine whether the macronutrients have equivalent effects
in restraining intake, we correlated the amounts of protein, carbo-
hydrate, and fat that were estimated to be present in the stomach at
the beginning of the meal with the meal size. These correlations
are presented on the left panel of Figure 9. Clearly, all three
macronutrients were negatively correlated with subsequent intake.
Unfortunately, they were highly intercorrelated. When a large
amount of food is present in the stomach, it is likely that there will
be large amounts of each of the macronutrients and vice-versa with
a small amount in the stomach. To look at the individual contri-
butions of each macronutrient, we used multiple linear regression,
where it is possible to view the effects of each while taking into
consideration, and mathematically holding constant, the effects of
the other two. The coefcients for the three macronutrients from
the analysis are presented on the right side of Figure 9. Clearly,
protein in the stomach has a large negative effect on intake,
whereas carbohydrate and fat have little or no effect. This suggests
that the effectiveness of the before-meal stomach contents on
suppressing subsequent intake is due to the effect of protein and
not of fat or carbohydrate. From the perspective of the current
discussion, it is clear that multivariate statistics can be used to
isolate and study individual variables within the complex intercor-
related matrix of inuences present in real-world research.
Multiple regression can be used to investigate the independence
of multiple variables inuence on intake. For example, ve vari-
ables that have been shown to inuence intakethe time of the
meal, the number of people present, self-rated hunger, the duration
of the prior interval, and the stomach contentwere used as
predictors of the meal size. The univariate correlations are pre-
sented in the left panel of Figure 10. A multiple regression on these
same variables as predictors of meal size produced the coefcients
presented in the right panel of Figure 10.
Comparing the univariate with the multivariate outcomes, the
number of people and hunger have essentially the same effects,
which suggests that these variables affect meal intake indepen-
dently of the other factors. However, the prior interval and the
stomach content, which have highly signicant univariate correla-
tions, become nonsignicant factors in the multiple regression,
suggesting that they are only seemingly effective variables because
of their covariation with another factor. That factor would appear
to be hunger,
34
and this suggests that the duration of the prior
interval is associated with meal size because the individual feels
hungrier as the interval increases. Similarly, the stomach content
appears to be associated with meal size because the individual
feels hungrier as the stomach becomes emptier. Furthermore, the
time of the meal, which was a nonsignicant univariate factor,
becomes highly signicant in the multivariate analysis. This sug-
gests that the inuence of time of day is suppressed by a negative
covariation with another factor, possibly the before-meal stomach
contents, and only after its effects are mathematically extracted
does the true impact of time of day become apparent. This analysis
demonstrates the value of using multivariate techniques on real-
world data. The interdependencies and interactions between mul-
tiple factors can be sorted out, primary effect can be distinguished
from secondary effect, and otherwise suppressed factors can be
unmasked.
A nal example of how the richness of the real-world environ-
ment can be used to detect subtle and interesting inuences on
intake regulation is the research on the inheritance of food-intake
regulation. Because it has been well established that inheritance is
a signicant factor in the determination of body size, which is
inuenced by food intake, it stood to reason that food intake must
in some way be affected by the genes. To investigate this idea, we
studied the real-world intake behavior of free-living twins. Diet-
diary nutrient intake data were collected for a 7-d period from 110
identical and 102 fraternal same-sex adult twins and 53 pairs of
other-sex fraternal twins who were living independently.
8992
The
diary entries were analyzed with linear-structural modeling tech-
niques of heritability analysis. These techniques are sensitive,
allow the assessment of genetic and early familial environmental
effects, allow for the assessment of true sex differences in genetic
and environmental effects, and allow the inclusion of other vari-
ables in the models to assess where in the complex interactive
chains heredity affects the behaviors.
93,94
It was found that genetic inuences permeate all aspects of
ingestive behavior. Not only body size was affected by inheritance,
but also the amount eaten and the macronutrients and alcohol
composition were affected.
8991
In fact, 65% of the variance in
daily energy intake, 44% of the variance in meal frequency, 65%
of the variance in average meal size could be attributed to heredity.
In contrast, the analysis indicated that the shared familial environ-
ments in which the twins were raised had no signicant impact on
the levels or pattern of intake in adulthood. A possible alternative
explanation for these apparent genetic effects is that what is
inherited is body size and that body size inuences the amount of
intake. However, linear-structural modeling showed (Fig. 11) that
the heritability of daily food energy and macronutrient intake were
independent of body size.
91
Heredity accounted for 42% of the
variance in daily intake even when the inuence of body size was
extracted in the model.
That there are genetic inuences on the amounts ingested
implies that in some way the levels of intake must be inuenced or
FIG. 8. Seasonal rhythm of the mean total daily intake of food energy. The
average intakes for the four seasons are double plotted to emphasize the
rhythmicity.
808 de Castro Nutrition Volume 16, Number 10, 2000
regulated. Ultimately, the genes affect molecular and physical
structure. However, we have obtained evidence that the genes may
be also affecting the individuals psychological state, the individ-
uals responsiveness to this state, and even the environment in
which the individual is immersed and the individuals reactions to
the environments.
29,33,37
The psychological appreciation of hunger is a case in point.
How hungry an individual is at the beginning of a meal has a
signicant positive relation to how much will be eaten during the
meal.
34
It has been shown that the level of hunger, its correlation
with meal size, and the slope of the relation are signicantly
heritable, with the genes accounting for 31%, 25%, and 27% of the
mean, correlation, and slope, respectively.
33
These results indicate
that the genes have encompassing effects on this relation in af-
fecting how hungry an individual reports to be, how well that
hunger predicts meal size, and the degree of responsiveness of the
individual to that hunger.
The estimated amount remaining in the stomach at the begin-
ning of the meal has a signicant negative relation with meal
size.
13,14
We demonstrated that the stomach content, its correlation
with meal size, and the slope of the relation between stomach
content and meal size are signicantly heritable, with the genes
accounting for 40%, 29%, and 45% of the mean, correlation, and
slope, respectively.
37
These results indicate that the genes also
have encompassing effects on this relation in affecting the level of
stomach fullness or emptiness that is associated with meal initia-
tion, how well that stomach content predicts meal size, and the
degree of responsiveness of the individual to the level of stomach
fullness or emptiness.
The number of other people present at the meal has a signicant
positive relation with meal size.
30,31,40,62,66
It was shown that the
number of people eating with the individual, its correlation with
meal size, and the slope of the relation between the number of
people present and meal size are signicantly heritable, with the
genes accounting for 20%, 27%, and 30% of the mean, correlation,
and slope, respectively.
29
This suggests that the genes affect all
aspects of this relation in affecting how many people the individual
eats with, how well that number of people predicts meal size, and
the degree of responsiveness of the individual to that social
facilitation.
The time of day at which a meal is eaten has a signicant
positive relation with meal size.
32
It has been shown that the time
of day at the beginning of the meal, its correlation with meal size,
and the slope of the relation between meal time and meal size are
signicantly heritable, with the genes accounting for 37%, 21%,
and 18% of the mean, correlation, and slope, respectively. This
suggests that the genes inuence all facets of this relation in
affecting the time of day that meals occur, how well that time of
day predicts meal size, and the degree of responsiveness of the
individual to that diurnal rhythm factor.
The results of these behavioralgenetic studies indicate that
signicant genetic inuences on food intake can be detected with
real-world studies of normal free-living humans. The detected
heritabilities involve a variety of aspects of intake regulation and
suggest that there are subtle inuences of inheritance that extend
beyond the simple encoding of gross anatomy. The data imply that
there are heritable factors that not only inuence behavioral ten-
dencies and preferences but also the individuals responsiveness to
environmental factors. This likely involves the inheritance of sub-
tle nuances in the structures of the nervous system that underlie
personal preferences, habits, dispositions, and sensitivities. These
in turn affect the amounts of foods and uids ingested.
DISCUSSION
It should be evident from this presentation that real-world behavior
can be well measured and analyzed with self-reports. These tech-
FIG. 9. The relationship between the estimated before-meal stomach contents of the macronutrients and the amount of total food energy ingested in the meal
expressed as univariate correlations (left) or as mean coefcients from the multiple linear regressions (right).
Nutrition Volume 16, Number 10, 2000 809 Real-World Eating Behavior
niques are not without problems and errors. The diet-diary tech-
nique produces underreporting and reactivity that articially re-
duce the estimates of the absolute values of intake. However, that
such phenomena as the inheritance of food intake, seasonal
rhythms, and social facilitation can be discerned and analyzed with
diet-diary data suggest that it is sensitive enough, even with the
errors, to detect and document the inuences of important and
subtle variables on intake. The errors produced by the technique,
when properly considered in the logical process of interpreting the
data, are manageable and of less importance than previously
thought. Hence, there is a simple, reliable, and valid methodology
for studying real-world behavior.
It should be clear that there is a large number of inuential
factors that affect food intake in humans, including environmental,
psychological, and social variables. It should also be apparent that
these inuences can be studied in normal humans who are living
freely in their natural environments. The array of operative factors
in the real world is not necessarily noise that contaminates the
research; rather, it is a rich source of information on the nature and
importance of real inuences on eating. It is even possible to
investigate causal linkages in real-world studies. By employing
self-observational techniques and multivariate statistics, the com-
plexity of the real world can be harnessed to produce a clearer
understanding of the nature of intake regulation as it actually
unfolds.
It should also be apparent that laboratory research does not
necessarily lead to a clear understanding of the controls of intake.
In the laboratory, many factors that tend to restrain eating are often
removed or missing, which can lead to overingestion and an
altered pattern of intake. Many real-world facilitatory inuences
on eating are controlled or removed in laboratory research, which
can result in missing important variables in the control of intake
such as daily, weekly, or even seasonal rhythms, expectancies, and
social facilitation. As a result of removing, in the laboratory, many
of the real-world sources of variance in intake, the importance of
the studied variables such as palatability can be overestimated.
This can also result in a misunderstanding of how intake is nor-
mally regulated as in the case of osmotic and volumetric thirst. In
addition, the short durations of laboratory studies make difcult
the detection of many important inuences on intake that operate
over longer time frames, such as delayed compensations or sea-
sonal effects.
To some extent, this discussion has been an overstatement of
the case for real-world studies. In fact, both laboratory and real-
world studies are important in the unraveling of the complexities
of food-intake regulation. Each has its strengths and its weak-
nesses; fortunately, the weaknesses of one type of research are
often the strengths of the other. However, in the past, scientic
analysis of food intake has overemphasized laboratory research
and downplayed the importance or meaningfulness of real-world
research. In fact, basic scientists often do not adequately value the
results of real-world research because of the lack of the tight
controls that scientists have been taught to treasure. What is
needed is a balance, but, because the predominance of studies are
controlled laboratory investigations, to achieve that balance, far
more real-world research is needed. The primary underlying mes-
FIG. 10. The relationship of the meal size with the minute of the day of meal initiation, the number of other people present at the meal, the hunger self-ratings,
the amount of time since the previous meal, and the estimated before-meal stomach content of food energy expressed as univariate correlations (left) or as
mean coefcients from the multiple linear regressions (right).
810 de Castro Nutrition Volume 16, Number 10, 2000
sage of the present treatise is that real-world research can be
reliably and validly performed and that it can teach valuable
lessons that are difcult to obtain in the laboratory.
From the perspective of the regulation of intake, the real-world
data teach the lesson that intake is affected on the short term by a
wide array of physiologic variables such as stomach lling and
diurnal and seasonal rhythms, psychological variables such as
hunger palatability, restraint, and expectancies, and social vari-
ables such as social facilitation and cultural inuences. These
factors can have potent and immediate effects on intake on a
meal-to-meal basis. There are, however, long-term compensatory
mechanisms that operate to provide negative feedback and adjust
for prior excesses, but these compensatory mechanism take 2 to 3 d
to affect intake. The data also suggest that heredity affects all
aspects of food-intake regulation, from the determination of body
size to the subtleties of the individual preferences and social
proclivities and the extent to which environmental factors affect
the individual. Hence, the research on real-world eating behavior
by humans has shown that it is an intricate, multivariate process
whose understanding will require an appreciation for the totality of
the context in which it occurs.
ACKNOWLEDGMENTS
The author acknowledges the substantial contributions of Dixie K.
Elmore, Sara Orozco, Marie (Brewer) Redd, Elizabeth Shuler,
Heide Heck, Sandor Goldstein, Sharon Pearcey, Stephanie Plun-
kett, and Margaret Pedersen, without whose assistance this work
could not have been performed.
REFERENCES
1. Flegal KM. The obesity epidemic in children and adults: current evidence and
research issues. Med Sci Sports Exer 1999;31(11 suppl):S509
2. de Castro JM. Prior days intake has macronutrients specic delayed negative
feedback effects on the spontaneous food intake of free-living humans. J Nutr
1998;128:61
3. Thompson FE, Byers T. Dietary assessment resource manual. J Nutr 1994;124:
2245S
4. de Castro JM. Methodology, correlational analysis, and interpretation of diet
diary records of the food and uid intakes of free-living humans. Appetite
1994;23:179
5. de Castro JM. Measuring real world eating behavior. Prog Obes Res 1999;8:215
6. Adleson SF. Some problems in collecting dietary data from individuals. J Am
Diet Assoc 1960;36:453
7. Block G. A review of validations of dietary assessment methods. Am J Epidemiol
1982;115:492
8. Heady JA. Diets of bank clerks: development of a method of classifying the diets
of individuals for use in epidemiological studies. J R Stat Soc Ser A 1961;124:
336
9. Livingstone MB, Prentice AM, Strain JJ, et al. Accuracy of weighed dietary
records in studies of diet and health. Br Med J 1990;300:708
10. St. Jeor ST, Guthrie HA, Jones MB. Variability on nutrient intake in a 28 day
period. J Am Diet Assoc 1983;83:155
11. Gersovitz M, Madden JP, Smicikalas-Wright H. Validity of the 24-hour dietary
recall and seven-day record for group comparisons. J Am Diet Assoc 1978;73:48
12. Krantzler NJ, Mullen BJ, Schultz HG, et al. The validity of telephoned diet recalls
and records for assessment of individual food intake. Am J Clin Nutr 1982;36:1234
FIG. 11. Linear-structural model and the most parsimonious model tting the twin data for height, weight, and total daily intake are represented by the solid
path lines and path coefcients. Dotted lines represent non-signicant paths. For all remaining parameters, removing any one leads to a statistically
signicant reduction in the models account of the observations.
Nutrition Volume 16, Number 10, 2000 811 Real-World Eating Behavior
13. de Castro JM, Kreitzman SN. A microregulatory analysis of spontaneous human
feeding patterns. Physiol Behav 1985;35:329
14. de Castro JM, McCormick J, Pedersen M, Kreitzman SN. Spontaneous human
meal patterns are related to preprandial factors regardless of natural environmen-
tal constraints. Physiol Behav 1986;38:25
15. Mertz W, Tsui JC, Judd JT, et al. What are people really eating? The relation
between energy intake derived from estimated diet records and intake determined
to maintain body weight. Am J Clin Nutr 1991;54:291
16. Lissner L, Habicht J-P, Strupp BJ, et al. Body composition and energy intake: do
overweight women overeat and underreport? Am J Clin Nutr 1989;49:320
17. Livingstone MB, Prentice AM, Coward WA, et al. Validation of estimates of
energy intake by weighted-dietary record and diet history in children and ado-
lescents. Am J Clin Nutr 1992;56:29
18. Prentice AM, Black AE, Coward WA, et al. High levels of energy expenditure in
obese women. Br Med J 1986;292:983
19. Riumallo JA, Schoeller DA, Barrera G, Gattas V, Uauy R. Energy expenditure in
underweight free-living adults: impact of energy supplementation as determined
by doubly labeled water and indirect calorimetry. Am J Clin Nutr 1989;49:239
20. Schulz S, Westerterp KR, Bruck K. Comparison of energy expenditure by the
doubly labeled water technique with energy intake, heart rate, and activity
recording in man. Am J Clin Nutr 1989;49:1146
21. Bandini LG, Schoeller DA, Cyr HN, Dietz WH. Validity of reported energy
intake in obese and nonobese adolescents. Am J Clin Nutr 1990;52:421
22. Goran MI, Poehlman ET. Total energy expenditure and energy requirements in
healthy elderly persons. Metabolism 1992;41:744
23. Schoeld WN, Schoeld C, James WPT. Basal metabolic ratereview and
prediction, together with an annotated bibliography and source material. Hum
Nutr Clin Nutr 1985;39c(suppl 1):1
24. Black AE, Goldberg GR, Jebb SA, et al. Critical evaluation of energy intake data
using fundamental principles of energy physiology: 2. Evaluating the results of
published surveys. Br J Clin Nutr 1991;45:583
25. Goldberg GR, Black AE, Jebb SA, et al. Critical evaluation of energy intake data
using fundamental principles of energy physiology: 1. Derivation of cut-off limits
to identifying under-recording. Br J Clin Nutr 1991;45:569
26. Borrelli R. Collection of food intake data: a reappraisal of criteria for judging the
methods. Br J Nutr 1990;63:411
27. Goris AH, Westerterp KR. Underreporting of habitual food intake is explained by
undereating in highly motivated lean women. J Nutr 1999;129:878
28. Goris AH, Westerterp-Plantenga MS, Westerterp KR. Undereating and underre-
cording of habitual food intake in obese men: selective underreporting of fat
intake. Am J Clin Nutr 2000;71:130
29. de Castro JM. Inheritance of social inuences on eating and drinking in humans.
Nutr Res 1997;17:631
30. de Castro JM, Brewer EM. The amount eaten in meals by humans is a power
function of the number of people present. Physiol Behav 1992;51:121
31. de Castro JM, de Castro ES. Spontaneous meal patterns in humans: inuence of
the presence of other people. Am J Clin Nutr 1989;50:237
32. de Castro JM. Circadian rhythms of the spontaneous meal patterns, macronutrient
intake, and mood of humans. Physiol Behav 1987;40:437
33. de Castro JM. Inheritance of hunger relationships with food intake in free living
humans. Physiol Behav 1999;67:249
34. de Castro JM, Elmore DK. Subjective hunger relationships with meal patterns in
the spontaneous feeding behavior of humans: evidence for a causal connection.
Physiol Behav 1988;43:159
35. de Castro JM, Bellisle F, Dalix AM. Palatability and intake relationships in
free-living humans: measurement and characterization in the French. Physiol
Behav 2000;68: 271
36. de Castro JM, Bellisle F, Dalix AM, Pearcey S. Palatability and intake relation-
ships in free-living humans: characterization and independence of inuence in
North Americans. Physiol Behav 2000(in press)
37. de Castro JM. Inheritance of pre-meal stomach content inuences on eating and
drinking in free-living humans. Physiol Behav 1999;66:223
38. Henson MB, de Castro JM, Johnson CJ, Stringer A. Food intake by brain injured
humans who are in the chronic phase of recovery. Brain Inj 1993;7:169
39. Collier G. The dialog between the house economist and the resident physiologist.
Nutr Behav 1986;3:9
40. de Castro JM, Brewer EM, Elmore DK, Orozco S. Social facilitation of the
spontaneous meal size of humans is independent of time, place, alcohol, or
snacks. Appetite 1990;15:89
41. de Graaf C, Hulshof T. Effects of weight and energy content of preloads on
subsequent appetite and food intake. Appetite 1996;26:139
42. Rolls BJ, Kim S, McNelis AL, et al. Time course of effects of preloads high in
fat or carbohydrate on food intake and hunger ratings in humans. Am J Physiol
1991;260(4, Pt 2):R756
43. Shide DJ, Caballero B, Reidelberger R, Rolls BJ. Accurate energy compensation
for intragastric and oral mitrients in lean males. Am J Clin Nutr 1995;61:754
44. Rolls BJ, Castellanos VH, Halford JC, et al. Volume of food consumed affects
satiety in men. Am J Clin Nutr 1998;67:1170
45. Green SM, Blundell JE. Effect of fat- and sucrose-containing foods on the size of
eating episodes and energy intake in lean dietary restrained and unrestrained
females: potential for causing overconsumption. Eur J Clin Nutr 1996;50:625
46. de Castro JM. Meal pattern correlations: facts and artifacts. Physiol Behav
1975;15:13
47. LeMagnen J, Tallon S. La periodicite spontanee de la prise daliments ad libitum
du rat blanc. J Physiol (Paris) 1966;58:323
48. LeMagnen J, Tallon S. Leffect du jeune preable sur les caracteristiques tempo-
relles de la prise daliments chez le rat. J Physiol (Paris) 1968;60:143
49. Bernstein IL, Zimmerman JC, Czeisler AC, Weitzman ED. Meal patterning in
free-running humans. Physiol Behav 1981;27:621
50. de Castro JM. Social, cultural and genetic correlates of meal size and frequency
in normal and free-living humans with diabetes. Can J Diab Care 1999;23(suppl
2):61
51. de Castro JM. Regulatory alchemy: an attempt to make rat eating patterns
humanlike and human eating patterns ratlike. Appetite 1986;7:249
52. de Castro JM. The meal pattern of rats shifts from postprandial regulation to
preprandial regulation when only ve meals per day are scheduled. Physiol
Behav 1988;43:739
53. de Castro JM, Bellisle F, Feunekes GIJ, Dalix AM, De Graaf C. Culture, and
meal patterns. A comparison of the food intake of free-living American, Dutch,
and French students. Nutr Res 1997;17:807
54. de Castro JM. Circadian rhythms of the spontaneous meal patterns, macronutrient
intake, and mood of humans. Physiol Behav 1987;40:437
55. LeMagnen J, Devos, M. Meal to meal energy balance in rats. Physiol Behav
1984;32:39
56. Rosenwasser AM, Boulos Z, Terman M. Circadian organization of food intake
and meal patterns in the rat. Physiol Behav 1981;27:32
57. de Castro JM. Weekly rhythms of spontaneous nutrient intake and meal pattern
of humans. Physiol Behav 1991;50:729
58. de Castro JM. Social facilitation of the spontaneous meal size of humans occurs
on both weekdays and weekends. Physiol Behav 1991;49:1289
59. Shide DJ, Rolls BJ. Information about the fat content of preloads inuences
energy intake in healthy women. J Am Diet Assoc 1995;95:993
60. Caputo FA, Mattes RD. Human dietary responses to perceived manipulation of
fat content in a midday meal. Int J Obes Rel Metab Disord 1993;17:237
61. de Castro JM. Social facilitation of duration and size but not rate of the
spontaneous meal intake of humans. Physiol Behav 1990;47:1129
62. Feunekes GI, de Graaf C, van Staveren WA. Social facilitation of food intake is
mediated by meal duration. Physiol Behav 1995;58:551
63. de Castro JM. Family and friends produce greater social facilitation of food
intake than other companions. Physiol Behav 1994;56:445
64. Latane B. The psychology of social impact. Am Psychol 1981;36:343
65. Clendenen VI, Herman CP, Polivy J. Social facilitation of eating among friends
and strangers. Appetite 1994;23:1
66. de Castro JM, Orozco S. Moderate alcohol intake and the spontaneous eating
patterns of humans: evidence of unregulated supplementation. Am J Clin Nutr
1990;52:246
67. Redd EM, de Castro JM. Social facilitation of eating: effects of instructions to eat
alone or with others. Physiol Behav 1992;52:749
68. Orozco S, de Castro JM. Effect of abstinence on spontaneous feeding patterns in
moderate alcohol consuming individuals. Pharmacol Biochem Behav 1991;40:
867
69. Fitzsimons JT. Drinking of rats depleted of body uid without increase in osmotic
pressure. J Physiol (Lond) 1961;159:297
70. Gilman A. The relation between blood osmotic pressure, uid distribution and
voluntary water intake. Am J Physiol 1937;120:323
71. Fitzsimons JT. Drinking by nephrectomized rats injected with various substances.
J Physiol (Lond) 1961;155:563
72. Striker EM. Some physiological and motivational properties of the hypovolemic
stimulus for thirst. Physiol Behav 1968;3:379
73. Kraly FS. Physiology of drinking elicited by eating. Psychol Rev 1984;91:478
74. de Castro JM. A microregulatory analysis of spontaneous uid intake by humans:
evidence that the amount of liquid ingested and its timing is mainly governed by
feeding. Physiol Behav 1988;43:705
75. de Castro JM. The relationship of spontaneous macronutrient and sodium intake
with uid ingestion and thirst in humans. Physiol Behav 1991;49:513
76. Rolls BJ, Rolls ET. Thirst. Cambridge: Cambridge University Press, 1982
77. Phillips PA, Rolls BJ, Ledingham JG, Morton JJ. Body uid changes, thirst and
drinking in man during free access to water. Physiol Behav 1984;33:357
78. Bobroff EM, Kissileff HR. Effects of changes in palatability on food intake and
the cumulative food intake curve in man. Appetite 1986;7:85
79. Guy-Grand B, Lehnert V, Doassans M, Bellisle F. Type of meal affects palat-
ability and eating style in humans. Appetite 1994;22:125
812 de Castro Nutrition Volume 16, Number 10, 2000
80. Spiegel TA, Shrager EE, Stellar E. Responses of lean and obese subjects to
preloads, deprivation, and palatability. Appetite 1988;13:45
81. Yeomans NR. Palatability and the micro-structure of feeding in humans: the
appetizer effect. Appetite 1996;27:119
82. Yeomans NR, Gray RW, Mitchell CJ, True S. Independent effects of palatability
and within-meal pauses on intake and appetite ratings in human volunteers.
Appetite 1997;29:61
83. Raben A, Tagliabue A, Christensen NJ, et al. Resistant starch: the effect on
postprandial glycemia, hormonal response, and satiety. Am J Clin Nutr 1994;60:
544
84. de Roos N, Heijnen ML, de Graaf C, Woestenenk G, Hobbel E. Resistant starch
has little effect on appetite, food intake and insulin secretion of healthy young
men. Eur J Clin Nutr 1995;49:532
85. Balogh M, Kahn HA, Medalie JH. Random repeat 24-hour dietary recalls. Am J
Physiol 1971;24:304
86. Morgan KJ, Johnson SR, Goungetas G. Variability of food intakes: An analysis
of a 12-day data series using persistence measures. Am J Epidemiol 1987;126:
326
87. Tarasuk V, Beaton GH. The nature and individuality of within-subject variation
in energy intake. Am J Clin Nutr 1991;54:464
88. de Castro JM. Macronutrient selection in free feeding humans: evidence for long
term regulation. In: Berthoud H-R, Seeley RJ, eds. Neural control of macronu-
trient selection. New York: CRC Press, 2000:43
89. de Castro JM. Genetic inuences on daily intake and meal patterns of humans.
Physiol Behav 1993;53:777
90. de Castro JM. A twin study of genetic and environmental inuences on the intake
of uids and beverages. Physiol Behav 1993;54:677
91. de Castro JM. Independence of genetic inuences on body size, daily intake, and
meal patterns of humans. Physiol Behav 1993;54:633
92. de Castro JM. Genes and environment have gender independent inuences on the
eating and drinking of free-living humans. Physiol Behav 1998;63:385
93. Heath AC, Neale MC, Hewitt JK, Eaves LJ, Fulker DW. Testing structural
equation models for twin data using LISREL. Behav Genet 1989;19:9
94. Neale MC, Cardon LR. Methodology for genetic studies of twins and families.
Dordrecht: Kluwer Academic Publishers, 1992
Nutrition Volume 16, Number 10, 2000 813 Real-World Eating Behavior