United States Off ice of May 1984
Environmental Protection Radiation Programs EPA 520/1-84-015
Agency Washington, D.C. 20460
Radiation
vxEPA An Estimation of the
Daily Food Intake Based on
Data from the 1977-1978
USDA Nationwid :
Consumption Survey
-------
AN ESTIMATION OF TEE DAILY FOOD INTAKE BASED ON DATA
FROM THE 1977-1978 USDA NATIONWIDE FOOD CONSUMPTION SURVEY
Christopher B. Nelson
You-yen Yang
May 1984
Office of Radiation Programs
U.S. Environmental Protection Agency
Washington, D.C. 20460
-------
FOREWORD
The Office of Radiation Programs carries out a national program to
evaluate the exposure of man to ionizing and nonionizing radiation and
to develop guidance, standards, and criteria for the protection of
public health and the quality of the environment.
This report describes a more refined methodology than heretofore
used for estimating the daily food intake as food utilization factors
in the assessment of environmental radionuclide intake by individuals
through food consumption.
Readers are encouraged to bring to our attention any questions
encountered in this report. Any comments and suggestions are also
welcome. These comments should be directed to Christopher B. Nelson,
Bioeffects Analysis Branch, Analysis and Support Division, Office of
Radiation Programs (ANR-461C), U.S. Environmental Protection Agency,
Washington, D.C. 20460.
Len L. Sjoblom, UJirector
Office of Radiation Programs
-------
ACKNOWLEDGEMENT
Numerous people helped in the preparation of this report,
but particular acknowledgement is due to Drs. Eleanor M. Pao and
Kathryn H. Fleming of USDA for their assistance and support relative
to the USDA study and to Mary Anne Culliton of EPA for her editorial
contributions.
-------
Contents
Foreword iii
Summary ......«•••• 1
I. Introduction 2
II. Survey Design 2
III. Food Classifications 3
IV. Data Analyses 5
V. Analytical Results 5
VI. Conclusions 10
References 15
APPENDICES
Appendix A—The USDA 1977-78 Nationwide Food Consumption Survey
Appendix B—Statistical Analysis
-------
An Estimation of the Daily Food Intake Based on Data
from the 1977-1978 USDA Nationwide Food Consumption Survey
Christopher B. Nelson
You-yen Yang
U.S. Environmental Protection Agency
Office of Radiation Programs
Summary
A Nationwide Food Consumption Survey was conducted by the
U.S. Department of Agriculture in 1977-78 to investigate the food
intake of various selected segments of the U.S. population and to
identify changes in U.S. food intake patterns. EPA has used data from
this survey to determine the average daily total intake of different
food items. This daily average can be used to assess radionuclide
intake by persons in various geographic categories within the United
States.
Through analytical processes, we were able to estimate the daily
intake of each food subclass, not only for the U.S. population as a
whole, but also for subpopulations classified according to their
geographical (census) characteristics. We also computed the standard
error of each estimate for statistical inference. These estimates are
weighted mean values of the daily intake and are adjusted for the age
and sex distribution of the 1969-71 U.S. stationary population, region,
division, primary sampling unit (PSU), and season.
All food items in the survey were grouped into 8 main classes
which were further divided into 22 subclasses. We analyzed the data
for these classes and subclasses to determine which factors in the
statistical model (Appendix B) had significant effects on food intake.
Region was the most important factor. It significantly affected
intake of practically all food subclasses. Liquid food, as opposed to
solid food, accounted for most of the differences among regions.
Urbanization (and the interaction of urbanization and geographic
division) played a trivial role in food intake. While season had
practically no significant effects on solid food intake, it affected
beverage, tap water, and soup intake significantly.
Different studies on caloric intake for the U.S. population were
compared and shown to be consistent with each other. The estimated
mean caloric intake in this study is 1863 kcal, which is about 84% of
the Recommended Dietary Allowances (RDA). This does not necessarily
indicate either low bias in either the USDA survey or the EPA analysis.
-------
I. Introduction
To assess the radionuclide intake by humans through food intake,
we need estimates of the daily food intake of different food items.
Historically, a series of ad hoc values for average and maximum food
intakes has been used. These values are generally the result of a
subjective interpretation of data reported in nutrition or marketing
studies; Rupp's reports (Rupp, 1980A-B) are good examples. None of
these reports, however, include significance tests for reported class
effects or confidence intervals for estimated parameters.
EPA's study has used the data from the U.S. Department of
Agriculture (USDA) 1977-78 Nationwide Food Consumption Survey (NFCS)
(see Appendix A for more information about the Survey) to estimate the
daily intake of different food classes which will be used for the
assessment of environmental radionuclide intake through food consump-
tion. We have also performed statistical analyses and computed the
standard error of each estimate for statistical inference.
The USDA survey objectives included investigating the dietary
intake of various segments of the U.S. population, determining the
adequacy of diets as compared to the Recommended Dietary Allowances,
and determining any changes in food intake patterns. Because EPA s
objective was different, we had difficulties with the survey design,
food classifications, and in applying the data to our statistical
analysis. To overcome these problems, we reorganized the original data
files so that we could classify food items according to the
characteristics of radionuclide transport and thus be able to analyze
the data for our purposes.
In this report, we use the term "daily intake" to mean "average
daily total intake." We will present statistical findings, interpret
their significance, and give the estimated mean daily intake of food
classes per caput. This is the first of a series of reports concerning
statistical analyses of the NFCS data. We discuss some geographical
effects on food intake in this report.
II. Survey Design
In the Nationwide Food Consumption Survey (see Appendix A), the
sampling unit was a household. For each household there were two parts
to the survey: the household part investigated the quantities and
sources of foodstuffs used in the preparation of meals served in the
household, while the individual part surveyed the intakes of individual
household members at each eating occasion over a three-day period. The
data for the individual part of the survey were analyzed for this
report. Since members of the same household would usually be eating
substantially from the same food supply and would often have similar
eating patterns, we could not expect data for members of the same
household to be independent. Furthermore, the households in each
segment of at least 100 households within a primary sampling unit (PSU)
-------
were allocated to four different quarterly samples—no samplings were
performed for each season, and the households selected for the study
were not statistically independent. For these reasons, we did not use
the mean squares of "individual" or "household" as an estimate of
variance of sampling error for hypothesis testing in the analysis of
variance. We used, instead, the mean square of the PSU, or precisely,
PSU nested in region, division, and urbanization, as an estimate of the
variance of sampling error for the analysis. Thus, the analytical
results we obtained are somewhat conservative.
Because of our different objectives, we encountered the following
problems:
1) The protocol (USDA, 1981) for the spring quarter was
different from that for the remaining quarters. In the
spring quarter, every member in a sampled household was
interviewed, but in the remaining quarters, all persons
younger than 19 and one-half of all persons 19 and older
were interviewed, i.e., a part of the household was
sub sampled. If there was only one person in the
household, he or she was interviewed with certainty.
2) The protocol reads: "In order to provide seasonal
information, the sample design called for identification
of four independent interpenetrating samples to be
implemented in consecutive quarters of data collection, "
i.e., households in each area segment were allocated to
four quarterly samples. While this sampling procedure
would be more economical than a random sample, it is very
doubtful that these four quarterly samples would indeed be
independent as stated in the protocol, because the samples
were drawn systematically from the same area segment.
3) Because part of the data was collected through
recollection, the data could be low-biased.
III. Food Classifications
Because concentrations of radionuclides in foods can vary widely,
we classified foods by categories for which we can measure or calculate
concentrations of radioactivity and which we know comprise significant
dietary intake. We divided foods into eight classes; in turn, each
class was divided into subclasses as shown in Table 1 below.
For example, we classified produce into three subclasses: leafy
vegetables, produce exposed to the atmosphere, and protected produce.
Leafy vegetables, such as lettuce, have a broad flat leaf surface for
direct interception of atmospherically deposited material. The edible
portion of the plants primarily consists of leaves and stems. Exposed
produce, tomatoes for example, also intercepts atmospherically
deposited material, but surface areas (for exposure) are typically
-------
smaller than those of leafy vegetables. Edible parts are usually
reproductive organs which accumulate additional radionuclides through
soil uptake to a smaller extent than vegetative parts. Protected
produce, such as potatoes, are not directly, exposed to atmospherically
deposited material. Edible portions are principally reproductive or
storage organs which either grow underground or are protected by pods,
shells or nonedible peels. The accumulation of radionuclides takes
place through uptake of radionuclides from soil or transfer from
nonedible portions.
Similar considerations were extended to other classifications of
foods so that each classification is pertinent to specific radionuclide
pathways.
Every food item in the NFCS was assigned a 7-digit code (USDA,
1979A). The first three digits represent food groups; the last four
are associated with ingredients added to, or the preparation of, the
food items. The following illustrates this coding system for the first
two digits.
First digit = major food groups.
1 = milk and milk products,
2 = meat, poultry, fish, and mixtures,
3 = eggs, mixtures, substitutes,
etc.
First and second digits = major food subgroups.
11 = milk and milk drinks,
12 = cream and cream substitutes,
13 = milk desserts,
14 = cheeses,
21 = beef,
22 = pork,
23 = lamb, veal, game, other carcass meat,
24 = poultry;
etc.
We classified all 3,735 food items in the survey into our eight
major classes; each class again was divided into two to four
subclasses. There were many occasions where the appropriate
classification was not clear. To classify them, we made a judgment
based on the objectives of the study and the characteristics of the
food items under consideration. Reconstituted milk, for example, could
be classified as other dairy products or as water-based drinks. We
considered the item from a radiological assessment perspective: what
were the ingredients of the food items and what was the most important
ingredient; had it been processed or stored for some time before it was
served? After making those considerations, we classified reconstituted
milk into the other dairy products subclass.
Stew, for another example, consists of meat, vegetables, starch,
water, and unspecified items. We considered the meat to be the most
-------
significant ingredient and identified the meat in the stew by using the
USDA coding system. We separated bread and filling for sandwiches,
including hamburgers, hot dogs, submarines, and tortillas. All 3,735
food items in the survey data file were classified as shown in Table 1.
Because there are so many varieties of vegetables and fruits, the
following indicate important produce belonging to each subclass.
Leafy vegetables: cabbage, cauliflower, broccoli,
celery, lettuce, spinach.
Exposed produce: apples, pears, berries, cucumber,
squash, grapes, peaches, apricots,
plums, prunes, string beans, pea
pods, tomatoes.
Protected produce: carrots, beets, turnips, parsnips,
citrus fruits, sweet corn, legumes
(peas, beans, etc.), melons,
onions, potatoes.
Other produce: Unspecified vegetables or fruits
and mixtures of vegetables or
fruits.
IV- Data Analyses
USDA recorded the information on food intake for every eating
occasion during the three-day survey period. Since the objective of
the EPA study was to estimate the daily intake of food classes and
subclasses for each sampled individual, we added the quantities of a
food subclass or class consumed on each occasion throughout the day,
then took an average over the number of days the individual had
participated in the survey.
If an individual had participated two days, for example, the
average daily total would be obtained by dividing the total quantity by
two for that food subclass or class. Most of the survey data had a
complete three-day set of observations (93%), very few had a one-day
(5%) or a two-day (2%) set. A total of 30,770 persons participated in
the survey. Given the high proportion of individuals with three-day
sets of observations, we chose not to weight the observations on the
basis of the numbers of days of data (see Appendix B for more
information).
V. Analytical Results
The general linear models procedure, GLM, in the Statistical
Analysis System (1979), commonly known as SAS, was used to analyze the
data. The results of analysis of variance are given in Table 2.
-------
Table 1. Classification of the Food Items in the NFCS
into Major Classes and Subclasses
Major Class
Subclass
Dairy products
Eggs
Meats
Fish
Produce
Grains
Beverages
Miscellaneous
Fresh cow's milk,
Other dairy products (dry milk,
butter, cheese, etc.).
Fresh eggs,
Other egg products (powdered
eggs and other prepared
egg products).
Beef, veal,
Pork,
Poultry (chicken, duck,
other birds),
Other meats (game, organ
meats, meat mixtures,
etc.).
Fin fish,
Shellfish,
Other seafood (mixtures and
unspecified fish products).
Leafy vegetables,
Exposed produce,
Protected produce,
Other produce (unspecified
vegetables or fruits; mix-
tures of vegetables or fruits).
Breads, pasta, cakes, etc.,
Cereals,
Other grains (wheat, rice,
raw millet, raw rye, etc.).
Tap water,
Water-based drinks
(coffee, tea, etc.),
Soups,
Other drinks (soft drinks,
fruitades, alcoholic drinks).
No subclasses (chocolate,
sugar, salad dressings, etc.).
-------
The F statistics given in Table 2 can be expressed by:
F = MS(A)/MS(PSU),
where MS(A) and MS(PSU) are the mean squares of A and PSU, respectively.
A" represents any factor under hypothesis testing, such as region,
division nested in region, or urbanization nested in region. The test
statistic, under the null hypothesis, has an F-distribution with
degrees of freedom f^ and fpsu> i«e.« F(f^, fpsiP' ^or a §iven
level of significance, say a, when
MS (A)/MS ( PSU) ^ Fln]i(fA, fpSu>>
we reject the null hypothesis and declare that the test result is
(statistically) significant at the a level.
For simplicity, when we mention division, urbanization, or season,
we will mean division nested in region, urbanization nested in region,
and so on. Note that since the quantities of other egg products and
other fish products consumed were negligibly small, less than one
gram/day, we have excluded them from the statistical analysis.
Census region and geographical division (see Appendix A)
significantly affected food intake. Urbanization, interaction of
division and urbanization are relatively unimportant on food intake.
Season had practically no effect on solid food intake, but it
significantly affected beverages, tap water, and soup intake.
The intake of two major food classes, dairy products and
beverages, varied significantly by both region and geographical
division (Table 2); for other food subclasses, it varied significantly
only by region or division. Table 2 also shows that food intake
patterns of food subclasses were overwhelmingly regional. In all but
three subclasses, leafy vegetables, other grain products and other
beverages, significant regional effects were detected. When hypothesis
tests were performed for the major classes, significant effects could
appear or disappear compared with the results obtained for subclasses.
For example, all four subclasses of meats showed significant effects
for region yet the class meats did not show a significant effect for
region. Conversely, only one of the four subclasses of beverages
showed a significant effect for geographical division yet beverages as
a whole showed a significant effect for geographical division. This is
because the aggregation can cause effects to be magnified or
diminished. A few significant results were detected for season,
urbanization, and the interaction of urbanization and geographical
division.
The estimated mean daily intake, weighted by age, sex, region,
division, PSU, and season, are given in Table 3. The values given in
-------
Table 2. Analysis of Variance by Food Class and Subclasses
(F values for sources of variation)
Source of variation
Food class or
Subclass
Degrees of Freedom
DAIRY PRODUCTS
Fresh cow1 s milk
Other
EGGS
MEATS
Beef & veal
Pork
Poultry
Other
FISH
Fin fish
Shellfish
PRODUCE
Leafy
Exposed
Protected
Other
GRAIN
Breads
Cereals
Other
BEVERAGES
Tap water
Water based
Soups
Other
MISCELLANEOUS
Region
(3)
1 7 . 40**
11.54**
22.68**
17.97**
1.59
7.12**
8.75**
8.57**
14.63**
6.23**
4.12**
3.65*
1.19
1.03
7.32**
5.10**
8.63**
0.43
7.90**
2.86*
2.57
17.21**
30.91**
3.32*
9.80**
1.26
9.14**
Division
(5)
3.91**
4.38**
1.00
1.79
2.42*
5.17**
1.61
.97
1.42
2.14
2.03
1.32
3.99**
3.07*
4.22**
3.28**
1.69
2.47*
2.27
1.14
3.51**
3.85**
7.93**
.61
.94
1.89
2.31
Urbani-
zation
(8)
1.94
1.76
1.69
2.07*
1.15
1.18
.71
1.87
2.86**
.78
.65
.87
.47
2.64*
1.00
.96
1.12
.67
1.17
.38
1.67
.30
1.57
2.27*
1.07
3.34**
3.89**
Division
Urbani-
zation
(10)
.82
.83
.90
1.34
1.07
.76
1.21
1.40
1.86
1.06
1.00
1.08
.25
.86
.87
.42
2.08*
.67
1.86
.57
1.44
.81
2.55**
1.66
.98
.51
1.51
Season
(12)
1.29
1.67
.42
.45
.42
.66
.56
.53
.71
.57
.61
.93
1.35
1.08
7 . 09**
.42
1.33
1.20
1.84
.81
.12
2.33*
2.04*
.63
4.51**
1.18
.54
Sampli-
ing
Error
(gram)
(84)
807.3
753.7
179.4
75.7
252.3
176.0
85.7
115.9
66.1
81.5
67.5
35.3
538.0
116.8
232.4
348.6
43.2
451.5
218.9
194.3
259.0
2098.5
1513.0
1034.0
186.0
724.6
107.7
Notes: *: .01 < P < .05
P < .01.
-------
Table 3. Mean and Standard Error for the Daily Intake of
Food Class and Subclass Region (in grams)
Food class/
Subclass
DAIRY PRODUCTS
Fresh cow's milk
Other
EGGS
MEATS
Beef & veal
Pork
Poultry
Other
FISH
Fin fish
Shellfish
PRODUCE
Leafy
Exposed
Protected
Other
GRAIN
Breads
Cereals
Other
BEVERAGES
Tap water
Water based
Soups
Other
MISCELLANEOUS
U.S.
Population
308.6+5.3
253.5+4.9
55.1+1.2
26.9+ .5
172.2+1.6
87.6+1.1
28.2+ .6
31.3+ .8
25.1+ .4
17.5+ .5
14.7+ .4
2.6_+ .2
282.6+3.5
39.2+ .8
86.0+1.5
150.4+2.3
7.0+ .3
200.0+3.0
147.3+1.4
29.9+1.3
22.9+1.7
1434. +13. 7
662.5+9.9
457.1+6.7
45.9+1.2
269.9+4.7
34.6+ .7
Region
Northeast
318.6+10.4
256.1+9.7
62.5+2.3
23.8+1.0
169.9+3.3
82.3+2.3
28.8+1.1
31.7+1.5
27.1+_ .9
20.5+ 1.1
16.7+ .9
3.6+ .5
270.6+6.9
38.1+1.5
88.5+3.0
137.2+4.5
6.9+ .6
203.5+5.8
153.1+2.8
24.6+2.5
25.9+_3.3
1288. +27.0
520.9+19.5
418.5+13.3
52.2+ 2.4
297.2+ 9.4
31.4+ 1.4
North
Central
336.1_+10.0
279.7+9.4
56.5+2.2
23.5+ .9
176.9+3.1
92.9+2.2
29.6+1.1
26.6+1.4
27.8+, .8
14.7+1.0
13.1+ .8
1.5+ .4
282.4+6.7
37.1+1.5
87.8+2.9
150.1+4.3
7.3+ .5
192.8+5.6
150.9+2.7
28.7+2.4
13.3+3.2
1448. +26.0
659.8+18.8
476.0+12.8
48.9+ 2.3
264.5+ 9.0
37.0+ 1.3
South
253.6+8.4
211.0+7.8
42.6+1.9
31.0+ .8
171.9+2.6
84.0+1.8
30.1+ .9
36.5+1.2
21.3+ .7
17.0+ .8
14.2+ .7
2.6+ .4
280.7+5.6
38.4+1.2
76.9+2.4
160.1+3.6
5.4+ .4
202.2+4.7
143.9+2.3
34.6+2.0
23.7+2.7
1513. +21. 8
748.5+15.7
470.5+10.7
36.8+ 1.9
258. 1+ 7.5
32.3+ 1.1
West
348.1+12.3
283.5+11.5
64.6+2.7
29.1+1.2
168.6+3.9
92.9+2.7
22.1+1.3
28.9+1.8
24.7+1.0
18.5+1.2
15.2+1.0
2.8+ .5
303.1+8.2
45.3+1.8
95.5+3.6
152.5+5.3
9.8+ .7
202.6+6.9
139.5+3.3
30.9+3.0
32.1+4.0
1480. +32.1
714.4+23.1
458.2+15.8
48.1+ 2.9
260.6+11.1
39.2+ 1.6
-------
the second column, U.S. Population, of Table 3, are the expectation of
the 1969-71 U.S. stationary population. It is interesting to note that
the estimated standard error is small compared to the correspond-
ing mean, although we have used a conservative sampling error in the
statistical analyses. The age and sex distribution assumed is that of
the 1969-71 U.S. stationary population (see Appendix B).
VI. Conclusions
Of the effects tested by analysis of variance, the regional factor
is by far the most important. Except for leafy vegetables, other grain
products, and other beverages, it significantly affects the intake of
all food items under consideration. Food intake patterns are diverse
as one can see in Table 3. The division factor is the second most
important. It directly affects the intake of fresh cow's milk, beef
and veal, leafy vegetables, exposed produce, protected produce, other
grain products, and tap water. Urbanization and the interaction
between division and urbanization do not significantly affect food
intake. Almost no significant seasonal changes in solid food intake
were detected.
All food subclasses can be classified as liquid or solid foods
(Table 4). Of the total daily intake per caput (2478 grams), 68% (1689
grams) is liquid and 32% (789 grams) is solid, or the ratio of liquid
foods consumed to solid foods is approximately 2 to 1. The total solid
food intake is fairly constant across the regions whereas the liquid
food intake has a wider range of variation. Overall intake is greatest
in the West and least in the Northeast region.
When the regional intake is normalized by the national population
intake for each major class and subclass, most regions are within ± 10%
of the national levels. (Table 5 indicates the exceptions.)
The intake of shellfish in the Northeast region is 38% above the
national average, while in North Central it is 42% below. The intake
of other grains in the West is 40% above the national average; in North
Central it was 42% below the national average. One may notice that the
quantities of these two subclasses are small (see Table 3). It is
interesting to note that only one subclass (pork) is more than 10%
below the national average in the West, while there were five such
subclasses in the South.
We have attempted to compare our findings to those of USDA (Pao,
et al., 1982) and those of Rupp (Rupp, 1980). It is very difficult to
make direct comparisons among these three sets of results. The USDA
report presents average intakes per day and per eating occasion by sex
and age. These averages are for individuals actually consuming the
given food items; therefore, the values in the report are often quite
different from the customary averages for the entire U.S. population.
10
-------
Table 4. Average Total Daily Food Intake
by Region (in grams)
Region
Food type
Solid food
Liquid food*
TOTAL
Northeast
782
1545
2327
North
Central
784
1729
2513
South
111
1725
2502
West
825
1765
2590
U.S.
Population
789
1689
2478
*Including fresh cow's milk.
Table 5. Exceptions to National Average Levels of
Food Intake by Region
Northeast
Other dairy
products
Fish*
Fin fish
North Central
More Than 110% of
Fresh cow's milk
Other meats
South
National Average
Eggs*
Poultry
Cereals
Tap Water
West
Dairy
products*
Fresh cow's
milk
Other dairy
products
Shellfish
Other grains
Soups
Other beverages
Leafy
vegetables
Exposed
produce
Other produce
Other grains
Miscellaneous
Less Than 90% of National Average
Eggs*
Cereals
Beverages*
Tap water
Eggs*
Poultry
Fish*
Fin fish
Shellfish
Other grains
Dairy products*
Fresh cow's
milk
Other dairy
products
Exposed produce
Other produce
Soups
Pork
*Major class.
11
-------
Table 6. A Comparison of Studies by Rupp and EPA
(Mean daily intake per caput, in grams)
Rupp
Food class (Over 18 yrs) (All ages)
Dairy products 306 309
Meats 258 172
Produce 438 283
Grains 97 200
Fluid 1351* 1388**
*0ther than soups and fresh cow's milk.
**0ther than fresh cow's milk, but including soups
Rupp's results, which are averages over a population, were
obtained through a review of the literature and personal
communications. Because Rupp's results are similar to ours, we
summarize them in Table 6 for important major classes. One may note
that the averages given by Rupp are for those individuals over 18.
These values would normally be expected to be greater than those for
all ages. There are considerable differences in intakes of meats,
produce, and grains.
A difficult question to resolve in studies such as this is the
possibility of bias. We recognize that even though study participants
may have conscientiously recorded their best estimates of their
intakes, the potential for recall bias clearly exists. Individuals may
forget items they have eaten or systematically underestimate intake
quantities. To our knowledge, no study has been performed to assess
this effect.
One way to evaluate dietary intakes is to compare the resultant
caloric intake with the food energy needs of the corresponding class of
individuals. Zach, et al., have taken this approach and have concluded
that "... the USNRC77 diets for maximum and average adults and teens
fall well short of the expected caloric intake rates" (Zach, 1983).
Their standard for comparison is primarily the dietary standard for
Canada. We have estimated the daily caloric intake for an average U.S.
individual using the caloric factors in Table 7.
These factors (col. 2) are based on caloric data reported in
(USDA, 1979B). The resultant daily food intake for an average
individual is 1853 kcal. For comparison, we have calculated the
average food energy intake for the Health and Nutrition Examination
Survey. 1971-74 (HANES)(USDHEW, 1979), the Nationwide Food Intake
Survey-1977 (NFCS) (USDA, 1984), and the 1980 National Academy of
Sciences Recommended Daily Allowances (USDA, 1980).
12
-------
Table 7. Caloric Factors and Food Energy
Food Class
Milk
Other dairy products
Eggs
Beef
Pork
Poultry
Other meat products
Fin fish
Shellfish
Produce
Grains
Water based beverages
Soups
Other beverages
Caloric Factor
(kcal/gram)
0.61
1.76
1.60
2.85
2.87
1.85
2.43
1.90
0.95
0.81
3.42
0.0074
0.42
0.40
Daily Intake
(gram)
253.5
55.1
26.9
87.6
28.2
31.3
25.1
14.7
2.6
282.6
200.0
457.1
45.9
269.9
Food Energy
(kcal)
155
97
43
250
81
58
61
28
2
229
684
3
19
108
Miscellaneous
Total
1.00
34.6
35
1853
Since our estimate is for an average individual in the 1970 stationary
population, we have weighted the values according to age and sex in the same
way for each set of data. HANES considers only ages 1 through 74. We have
adjusted the value to all ages by presuming that the ratio of these two values
would be the same for HANES as for NFCS (1897:1849). The results of this
comparison are summarized in Table 8.
Caloric intakes in 1970s declined somewhat compared with those in the
1960s (USDA, 1980). The HANES result is consistent with this observation. In
documenting HANES, USDHEW notes: "The recommended dietary allowances are
designed to guide dietitians in formulating diets for the maintenance of good
nutrition in healthy persons. They allow for some margin above what is really
needed by most individuals, with the objective of maintaining good health"
(USDHEW, 1979). It is interesting to note that the dietary standard for
Canada (Zach, 1983), 2150-5350 kcal for adult males and 1750-4400 kcal for
adult females, appears to be greater than for the United States, perhaps
13
-------
Table 8. Comparisons of Caloric Intakes of the
U.S. Population
RDA*
HANES
NFCS
EPA
Year
kcal/day
Relative to RDA
1980
2194
"
1971-74
1925
0.88
1977-78
1849
0.84
1977-78
1853
0.84
*Recommended Dietary Allowances (see text).
refecting a somewhat colder climate than the U.S. average. We conclude
that our result, 84% of the RDA, does not necessarily indicate either
low bias or inadequate nutrient intakes-
Based upon the comparisons of different dietary parameters with
the studies discussed above, we conclude that it is appropriate to use
these estimated means of daily food intake as food utilization factors
in the assessment of environmental radionuclide intake by individuals.
14
-------
REFERENCES
USDA, (1979A), U.S. Department of Agriculture. Coding Manual to Handle
Data from Nationwide Survey of Individuals. Spring 1977-78.
Vols. I & II. National Technical Information Service, Springfield,
Va.
USDA, (1979B), U.S. Department of Agriculture. Food Composition Data
Values. CFE(Adm.)-347. Science and Education Administration,
Consumer and Food Economics Institute.
USDA, (1980), U.S. Department of Agriculture. Food and Nutrient Intakes
of Individuals in 1 Day in the United States, Spring 1977. Nation-
wide Food Consumption Survey 1977-1978, Preliminary Report NO. 2.
Science and Education Administration, Consumer and Food Economics
Institute.
USDA, (1981), U.S. Department of Agriculture. Basic Sample Design.
Personal communication from Brucy Gray.
USDA, (1984), U.S. Department of Agriculture. Food Intakes:
Individuals in 48 States, Year 1977-78. Nationwide Food
Consumption Survey 1977-78, Report 1-1. Human Nutrition
Information Service.
USDHEW, (1979). Caloric and Selected Nutrient Values for Persons 1-74
Years of Age: First Health and Nutrition Examination Survey,
United States, 1971-1974. Office of Health Research, Statistics,
and Technology; National Center for Health Statistics, U.S. Depart-
ment of Health, Education, and Welfare.
Pao, Eleanor M., Kathryn H. Fleming, Patricia M. Guenther, and Sharon J.
Mickle, (1982). "Foods Commonly Eaten by Individuals: Amount Per
Day and Per Eating Occasion." U.S. Department f Agriculture, Home
Economics Research Report, No. 44.
Rupp, E. M., (1980A). "Age Dependent Values of Dieta Intake for
Assessing Human Exposure to Environmental Pollut .its," Health
Physics 9, 151-63.
Rupp, E. M., Forest L. Miller, C. F. Baes, III, (1980B). "Some Results
of Recent Survey of Fish and Shellfish Consumption by Age and
Region of U. S. Residents," Ibid, 165-75.
SAS, (1979). SAS User's Guide, 1979 Edition, pp 244-263. SAS Institute
Inc., Raleigh, N. C.
Zach R. and Mayoh, K.R., (1983). "Caloric Contents of Diets for
Environmental Assessment Models: A Necessary Condition." Health
Physics Vol. 45, No. 3, 786-789.
15
-------
APPENDIX A
THE USDA 1977-78 NATIONWIDE FOOD CONSUMPTION SURVEY
-------
APPENDIX A
THE USDA 1977-78 NATIONWIDE FOOD CONSUMPTION SURVEY
A.1 Design of the USDA Natiowide Food Consumption Survey
The basic sample design of the USDA Nationwide Food Consumption
Survey (NFCS) is a multistage, stratified design covering a universe of
all private households in the conterminous United States. This basic
sampling design consists of four stages (USDA, 1981): stratification,
formation and selection of primary sampling units (PSU), selection of
area segments within PSUs, and sampling of housing units.
A.1.1. Stratification
The basic design consists of three levels of stratification:
First Level: Census Region
The conterminous United States, including the District of Columbia,
was divided into four census regions: Northeast, North Central, South,
and West. The following is the geographic grouping of states used by the
Bureau of Census (1970):
Census Region Geographic Division
Northeast New England
Middle Atlantic
North Central East North Central
West North Central
State
Maine, New Hampshire, Vermont,
Connecticut, Rhode Island.
New York, New Jersey,
Pennsylvania.
Ohio, Illinois, Indiana,
Wisconsin, Michigan.
Minnesota, Iowa, Missouri,
North Dakota, South Dakota,
Nebraska, Kansas.
A-l
-------
South South Atlantic Maryland, Delaware, District of
Columbia, Virginia, West Virginia,
North Carolina, South Carolina,
Georgia, Florida.
East South Central Kentucky, Tennessee, Alabama,
Mississippi.
West South Central Arkansas, Louisiana, Texas,
Oklahoma.
West Mountain Montana, Idaho, Wyoming, Utah,
Colorado, New Mexico, Arizona,
Nevada.
Pacific Washington, Oregon, California.
The states within each of these divisions are, for the most part,
fairly homogeneous in physical characteristics as well as in the
characteristics of their population, economic, and social conditions; on
the other hand, each division differs more or less sharply from most
others in these respects.
Second Level: Urbanization
Each housing unit was classified by urbanization as: central
city—in the central city of a standard metropolitan statistical area
(SMSA); suburban—within an SMSA, but not in the central city; or
nonmetropolitan—not within an SMSA.
A standard metropolitan statistical area consists of one or more
entire counties, economically and socially integrated, that have a large
population nucleus. The definition of an SMSA involves two considera-
tions; first, a city or cities of specified population to constitute the
central city and to identify the county in which it is located as the
central county; second, economic and social relationships with contiguous
counties which are metropolitan in character, so that the periphery of
the specific metropolitan area may be determined. Standard metropolitan
statistical areas may cross state lines to include qualified contiguous
counties (Shryock, et al., 1976).
Third Level: Sampling Stratum
The third level of stratification groups census units with like
geographic divisions and urbanization to produce sampling strata. One
hundred fourteen (114) (Table A-l) strata were formed incorporating all
housing units in the entire conterminous United States. The average
stratum size is approximately 600,000 housing units.
A-2
-------
Table A-l. Number of Strata in the Nationwide Food Consumption Survey
by Region, Division, and Urbanization
Region
Northeast
North
Central
South
West
Division Central
City
New England
Middle Atlantic
East North Central
West North Central
South Atlantic
East South Central
West South Central
Mountain
Pacific
2
8
8
2
4
2
4
2
6
Sub-
urban
3
9
8
2
6
1
2
1
7
Nonraetro-
politan
2
4
6
5
7
4
5
2
2
Total
7
21
22
9
17
7
11
5
15
TOTAL
38
39
37
114
A.1.2. Formation and Selection of PSUs
Every stratum was divided into one or more primary sampling units,
or PSUs, formed from cities, parts of cities, or counties, containing at
least 10,000 housing units each. One PSU per stratum was randomly
selected according to the following probablistic scheme:
Pii =
i = 1,2,..., 114;
i = l? \t • •
J 1 » ^ > • • • j K-l >
(A.I)
where
p^^ = the probability of the jth PSU in the ith stratum being
chosen,
nj; = number of housing units in the jth PSU in the ith stratum,
k£ = number of PSU in the ith stratum,
N.: = total number of housing units in the ith stratum.
A. 1.3. Selection of Area Segments within PSUs
Each selected PSU was a. d into small clusters of housing units
called area segments. The i ts, designed to contain 100 or more
housing units, were based o 1970 Census and usually consisted of one
or more city blocks in urba jas and a part of a Census Enumeration
District elsewhere. From all PSUs, 2550 segments were randomly selected
A-3
-------
with the number of segments in a PSU being proportional to the size of
the stratum in which the PSU was located. The probability that
individual segments were drawn in a PSU was proportional to the ratio of
the number of units in the segments to the total number of units in the
PSU. There was one exception. Three strata, New York City, Chicago, and
Los Angeles County, each with size well in excess of 600,000, were
selected with certainty. Each of these was defined as two PSUs.
Sampling from these three strata was done directly without further
stratification by area segment.
A.1.4. Sampling of Housing Units
All 2550 area segments were prelisted to determine the number of
occupied housing units. Then, the national increase or decrease in the
number of housing units from 1970 to 1977 was estimated. This
information, along with estimates of occupancy and completion rates,
permitted sampling rpf:i>t, co be calculated for the area segments that
would yield a total ot 3750 households per quarter.
For the first two quarters, an average sampling ratio of 2.3
households per segment was used. The housing units were ordered within
their respective segments. For each quarter, a sample was systematically
selected from each segment without replacement after a random start. By
the end of the second quarter, the estimated completion rate had been
adjusted, and an average sampling ratio of 2.86 households per segment
was used for the last two quarters. Note that once the segments were
chosen, different households would be sampled from the same segments for
each season.
A.2. Household and Individual Survey
The Nationwide Food Consumption Survey consisted of two parts:
household and individual.
Household Survey
For every sampled household, the eligible respondent was the person
who usually planned and prepared the meals in the seven days before the
interview. He or she was asked to keep dietary records, such as grocery
shopping lists and receipts, menus, and labels from canned and packaged
foods used for the meals. The information, in part, included:
-number of meals eaten during the past 7 days by each
household member—from home supplies, not from home
supplies,
-number of meals and refreshments served at home to
nonhousehold members during the past 7 days—number,
type, sex, age of consumers,
A-4
-------
-household food consumption during the past 7
days—description of food, form, quantity, and money
values, etc.
Individual Survey
Individuals from the sampled household were asked, as in the 7-day
household survey, to keep records of their food intake. However, this
survey was different from the household part in many respects:
-this was a 3-day instead of a 1-day survey. It included a
one-day recall (yesterday), and a 2-day record (today and
tomorrow) of food intake.
-in the spring quarter, every member of a sampled household was
interviewed, but in the remaining quarters, all persons younger
than 19, and one-half of all persons 19 and older were inter-
viewed. If there was only one person in the household, he or she
was included with certainty.
-food intake records were kept for all foods and beverages
consumed.
The food intake information included:
-for each eating/drinking occasion, whether food was consumed at
or away from home, information was collected for each item as to
when, with whom, in what quantity and from what source. For
away-from-home food, data on type of service and cost were also
collected.
-intake of water.
A-5
-------
REFERENCES
USDA (1981). U.S. Department of Agriculture. Basic Sample Design.
Personal communication with Brucy Gray.
Department of Commerce, (1970). "Pocket data book, U.S.A. 1970."
Shryock, H. S. and J. S. Siegel and Associates (1976). "The Methods
and Materials of Demography," Academic Press. New York.
A-6
-------
APPENDIX B
STATISTICAL ANALYSIS
-------
APPENDIX B
STATISTICAL ANALYSIS
B.I. Statistical Model
The data collected by USDA in this survey were very extensive.
Many individuals were interviewed regarding their food intake; each was
asked many questions. The data consisted of numerous variables and
factors for a large number of people. For example, in addition to
questions related to food consumption, there were 21 variables. Of
these, seven, for the purpose of the EPA1 s study, might be contributing
factors of food consumption. We attempted to discern whether there
were any relationships between these factors and food intake by fitting
a linear model to food intake incorporating these factors.
Data of this type must be subjected to careful preliminary
examination before trying to fit a model involving many factors with
several levels. A model that would normally have main effects could
also include many high order interactions. As a result, the model
could be very lengthy and complex. Before including a possible
interaction we considered its practical significance and also the
practicability of incorporating large numbers of (main) factors and
interactions into the analysis.
After considering the design of the survey and the objective of
our study, we adopted the following statistical model:
y. . , = M +R. + D., . , + U, , . v +DU,., + S ,. ,
•'ijklmpqr i j(i) kU) jkd) Id)
+ P (. .. •. + Sex + A + (Sex * A) + e . ., . (B.l)
rndjk; p q pq ijklmpqr
where :
= t'ie f°o<* intake of the rth individual in the qth age group
of pth sex (male: p = 1; female: p = 2) in the m(ijk)th PSU and in
the 1th season;
R^ = the ith census region, i = 1,2,3,4;
D-(J) = the jth geographic division in the ith census region,
j = l,2,...,n£ (since there are a total of 9 divisions,
Z i»i = 9);
B-l
-------
Uk(i) = tne ktn urbanization in the ith region, k - 1,2,3;
DUjk(i) = the interaction effect of the jth division with the
kth urbanization in the ith region;
£>l(i) = the 1th season in the region i, 1 = 1,2,3,4;
Pm(ijk) = the mth PSU in the ^-th regi°n> tne Jth division and in
the kth urbanization,
m
= l,2,...,Tiik (since there are 111 PSUs , Z T^ = 111);
Sex = male (p = 1) or female (p = 2)
A = the qth age group, q = 1,2,..., 10.
eijklmpqr = the random error associated with yijkimnpqr.
The geographic division in the nested classification was obvious;
however, the urbanization and season in the nested classification were
not that clear. Since each census region covers vast areas, winter,
for example, in one region may not be same as in the others. The same
is true for degree of urbanization. P (PSU) was treated as a random
factor. Respondents were divided into 10 age groups by year: under 1,
1 to 4, 5 to 9, 10 to 14, 15 to 19, 20 to 24, 25 to 29, 30 to 39, 40 to
59, 60 and over. The random error, e, is assumed to be normally
distributed with mean 0 and variance, a .
One of the assumptions for the analysis of variance is that y and
e in the model are independent and normally distributed. With this and
other conditions, tests of hypothesis are possible, and confidence in-
tervals can be derived. In an analysis of variance we chose the mean
square of "PSU," instead of "segment," "household," or "individual," as
an estimate of the variance of sampling error. The reasons were as
follows :
1) Individuals were subsampling units in the summer, autumn, and
winter quarters, but not in the spring quarter.
2) The intake data of individuals from the same household were
correlated.
3) For each quarter, the sample of housing units was system-
atically selected, with a random start, from each area
segment without replacement, i.e., the housing units were
sampled repeatedly from the same segment for the different
quarters. The housing units from the same segment in the
different season were not totally independent.
B-2
-------
4) The segments from which the housing units were selected were
highly clustered. The segments selected were not randomly
scattered in the PSU ; therefore, neighboring segments were
correlated .
Since, as described in A. 1.2, PSUs were randomly selected we have
used the mean square of PSU as an estimate of the sampling error; the
analytical results we obtained would be somewhat conservative.
Because the original design was unbalanced, traditional analysis
of variance methods were not applicable. Instead, we have used the
general linear model procedures (Searle, 1971). The model in equation
(B.I) can be rewritten in matrix form:
I = Xb. + e (B.2)
where :
Y = an N x 1 vector of observations, y, N is the total number of
observations in the survey. N = 30770 in the present case. If is
normally distributed with mean Xb_, and variance, a _I, _I is an
N x N identity matrix.
b_ = a p x 1 vector of parameters; it is the vector of all param-
eters of the model.
X = an N x p design matrix, Os and Is throughout the matrix rep-
resent the incidence of terms of the model among the observations,
i.e., the incidence of the classifications in which the observa-
tions lie.
e_ = an N x 1 vector of random error terms; e_ is normally distrib-
uted with mean J) and variance a _I.
Note that X does not have full column rank, neither does X'X. The
normal equations corresponding to the model can be derived by least
squares, we obtain
X'jSb0 = X'Y. (B.3)
Since X'X is not of full rank, X'X has no unique inverse, and the
normal equations (B.3) have no unique solution. There are many possi-
ble solutions. To get any of them, we find a generalized inverse C- of
t'X, and write
_ (B.4)
Define
H = GX'X. (B.5)
B-3
-------
The b_° in equation (B.4) for a solution to normal equation (B.3)
emphasizes that what is derived by solving equation (B.3) is only a solu-
tion to the equations and not an estimator of b_. This point cannot be
overemphasized. In a general discussion of linear models that are not of
full rank, it is essential to realize that what is obtained as a solution
of the normal equations is just a solution and nothing more. It is mis-
leading and in most cases quite wrong for b_° to be termed an estimator of
b_. The particular solution obtained for b_° depends entirely on which JS
is used.
Although b_° is not an estimator of b_, it is, in a practical sense,
closely related to b_. In order to establish this relationship, the con-
cept of estimable functions is essential. A linear function of the para-
meters, q_'b_, is defined as estimable if it is identically equal to some
linear function of the expected value of the vector of observations
Y_, i.e., for any g_, if there exists a vector t_ such that t/E(Y) = q_'b_,
then
-------
2) Seasons
Each season had the same length of duration, i.e., three months.
Their distribution is, therefore, uniform.
3) Age and sex
We generated distributions of age and sex based on the 1969-71
U.S. stationary population (National Center for Health Statistics,
1975,) with age- and sex-specific mortality rates and a constant male-
to-female birth ratio of 1.051 (Shryock, et al., 1976). The distribu-
tions of the resulting stationary population by age and sex are given
in Table B-l.
The stationary population was based on the life table technique.
A stationary population is a population whose total number and distri-
bution by age do not change with time. The birth rate remains constant
for a long period of time, and each cohort of births experiences the
current observed mortality rate throughout life. This hypothetical
population provides age weighted values which are substantially cor-
rected for those effects such as fluctuating birth or migration rates
which affect the age distribution of a population but not its age-
specific mortality.
The weighting factors described above were applied to Tables A-l
and B-l to estimate the individual means shown in Table 3.
Table B-l. Percent Distribution of the Stationary Population
by Age and Sex
Age
(years)
Total
Male
Female
TOTAL
Under 1
1 to 4
5 to 9
10 to 14
15 to 19
20 to 24
25 to 29
30 to 39
40 to 59
60 and over
100.00
1.40
5.53
6.89
6.88
6.86
6.81
6.75
13.34
24.89
20.65
48.56
.72
2.83
3.52
3.51
50
46
3.42
6.73
12.31
8.56
51.44
.68
2.70
3.37
37
36
35
3.33
6.61
12.58
12.09
B-5
-------
REFERENCES
National Center for Health Statistics, (1975). "United States Life
Tables, 1969-71", Vol. 1, No. 1. U.S. Department of Health, Educa-
tion and Welfare.
Searle, S. R., (1971). "Linear Models," John Wiley & Sons, Inc.
New York.
Shryock, H. S. and J. S. Siegel and Associates, (1976). "The
Methods and Materials of Demography," Academic Press. New York.
5-6
------- |