Estimating energy intake of urban women in Colombia: Comparison of diet records and recalls

Estimating Energy Intake of Urban Women in Colombia:Comparison of Diet Records and Recalls

DARNA L. DUFOUR,1,5* LISA K. STATEN,2 CAROL I. WASLIEN,3JULIO C. REINA,4 AND G.B. SPURR5,6,7

1Department of Anthropology, University of Colorado,Boulder, Colorado 80309–02332Department of Family and Community Medicine, University of Arizona,Tucson, Arizona 857123University of Hawaii School of Public Health, Honolulu, Hawaii 968224Department of Pediatrics, Universidad del Valle, Cali, Colombia5Department of Physiological Sciences, Universidad del Valle,Cali, Colombia6Department of Physiology, Medical College of Wisconsin,Milwaukee, Wisconsin 532267Research Service, Zablocki Veterans’ Administration Center,Milwaukee, Wisconsin 53226

KEY WORDS nutrition; methods; 24 h recalls; estimated records;food groups; developing country

ABSTRACT As part of a larger study of energy-nutrition, we comparedthe performance of 24 h diet recalls with estimated diet records kept bytrained observers. The subjects were economically disadvantaged women (n 585) in the city of Cali, Colombia. A 24 h recall and an estimated diet recordwere collected for each woman at 0 and approximately 3 and 6 months.Energy intake obtained from the estimated dietary records was validatedagainst energy expenditure and used as the reference method. Energy andmacronutrient intake were calculated from published food composition tablesand proximate analyses of common foods. The number of food items consumedper woman per day, total and in each of 16 food groups, was tabulated. Energyand macronutrient intakes were 11–13% lower in the 24 h recalls. Thediscrepancy energy could be largely accounted for by the lower number of fooditems in the recalls. The number of food items in eight of 16 food groups wassignificantly lower in the recalls compared to the records. Underreporting onthe recalls was a general tendency in these subjects and not clearly related toaverage energy intake. We conclude that 24 h diet recalls underestimateenergy and nutrient intake in this population and are not suitable for studiesof human energetics.Am J PhysAnthropol 108:53–63, 1999. r 1999 Wiley-Liss, Inc.

Estimating the diet intake of a free-livingpopulation is important in addressing manyanthropological questions of nutritional ad-aptation, energetics, and epidemiology (Huss-Ashmore, 1996), but it is a difficult task. Thereare a number of methods available, each ofwhich has advantages and disadvantages(Bingham and Nelson, 1991). The 24 h recallmethod, in which the subject is asked to recallall food eaten in the past 24 h, has probablybeen the most widely used in field situations. It

is a method that does not depend on subjectliteracy and has a number of advantages interms of ease of administration and low cost. Amajor disadvantage is its dependence on therecall ability of subjects.

Grant sponsor: NIH; Grant number: 5-R22-DK39734.*Correspondence to: Darna L. Dufour, Anthropology Depart-

ment, University of Colorado, Boulder, CO 80309–0233.E-mail: [email protected]

Received 26 February 1998; accepted 4 October 1998.

AMERICAN JOURNAL OF PHYSICAL ANTHROPOLOGY 108:53–63 (1999)

r 1999 WILEY-LISS, INC.

A number of studies have shown that 24 hrecall data tends to underestimate nutrientintake obtained by weighed records, whichare generally considered the gold standard(Bingham, 1987). Other studies, however,have shown that 24 h recalls can yieldsimilar and even higher nutrient intakes(Blake and Durnin, 1963; Greger and Et-nyer, 1978; Ohlson et al., 1950). In trying tounderstand these differences, most research-ers have focused on problems associatedwith the estimation of food portion size.However, the first thing a subject has to doin completing a 24 h recall is remember thata particular food item was consumed. Con-ceptualizing the size of the food item andeffectively communicating that informationto the interviewer are the second and thirdsteps, respectively. Although errors are pos-sible in each or all three of these steps,surprisingly few studies (Acheson et al.,1980; Linusson et al., 1974; Thomson 1958)have considered the accuracy of the firststep, the recall of food items, upon whicheverything else depends.

In addition to problems of memory, re-ports of food intake obtained using the 24 hrecall protocol may be biased in severalother ways. For example, subjects may notaccurately report intake for reasons relatedto the social desirability of consuming spe-cific foods (Thompson and Byers, 1994) orpopulation specific conceptions of how muchone should eat. With regard to the latter, twostudies in the USA have reported a tendencyfor individuals with relatively high intakeson diet records to underreport their intakeon 24 h recalls and visa versa (Linusson etal., 1974; Maddan et al., 1976; Gersovitz etal., 1978 ). This phenomenon is known as theflattened slope syndrome since regressionsof intake obtained by recall on intake ob-tained by records typically exhibit slopes ofless than one. Whether this phenomenonoccurs in other populations is not known.

The objective of this paper is to evaluatethe information on diet intake obtained us-ing 24 h recalls in a population of low-income urban women living under condi-tions of economic disadvantage in Cali,Colombia. In evaluating the recall data, weused the estimated diet records as a refer-ence since diet records are assumed to in-

volve fewer sources of error than recalls(Bingham and Nelson, 1991), and energyintake obtained from estimated records wasvalidated against energy expenditure (Spurret al., 1996). In this paper we test threehypotheses: 1) that energy and macronutri-ent intake obtained by recall is lower thanthat obtained from diet records, as is thenumber of food items; 2) that all foods arerecalled with the same degree of accuracy;and 3) that there is no difference in degree ofunderreporting on recalls in women withlow-average vs. high-average intakes.

The research reported here was done aspart of a larger study of energy nutrition ofurban women living in poor neighborhoodsof Cali, Colombia. One objective of the largerstudy was to use energy intake as an indica-tor of undernutrition. Because of the lowlevel of literacy of some of the subjects, wedecided to rely on food records estimated bytrained observers. A 24 h recall protocol wasincluded because we were unable to obtaindiet records covering weekend days andwere interested in knowing if there wereweekday-weekend differences in energy andmacronutrient intake.

SUBJECTS AND METHODSSubjects and setting

The subjects were 85 nonpregnant, nonlac-tating women (aged 19–43 years), all of thewomen were volunteers and had been re-cruited through word of mouth. All subjectsreceived a physical exam by a physicianbefore being admitted into the study. In-formed consent was secured from all sub-jects. The protocols were approved by theHuman Research Review Committee of theMedical College of Wisconsin and the Re-search Committee of the Universidad delValle.

The women lived in poor neighborhoodson the periphery of Cali, a city of ,2.2million people. Living conditions are de-scribed in Dufour et al. (1997). The womenpurchased most food at the small stores intheir neighborhood. Some women purchaseddry goods (grains, legumes, sugar, coffee,etc.) weekly and fresh products (bread, meat,fruits, vegetables) daily, and others pur-chased all food daily or by the meal. Thesubjects themselves did most of the food

54 D.L. DUFOUR ET AL.

preparation in their own households and atemost meals at home.

Study design

Diet intake was assessed in three measure-ment rounds at 0 and approximately 3 and 6months. Each measurement round lasted 1week, during which diet intake was as-sessed for 4 different days. Estimated dietrecords were used on 2 consecutive week-days, and 24 h recalls were used on 2 otherdays, the second of which was always aweekday. Other than these constraints, thespecific days of dietary assessment wereassigned randomly to individual subjects. Inthis paper, we compare mean values ob-tained from 1 day of record (day 2) and 1 dayof recall (day 2) from each measure round.The second day of records and recalls in eachround was used to eliminate potential week-end-weekday differences and to minimizeany observer effect. Data were collected be-tween August 1990 and May 1995. Days offood intake were representative of normalintake as far as we were able to determine.

Anthropometry and sociodemographiccharacteristics

Measurements of body weight and heightwere taken by a trained technician followingstandard techniques described by Lohmanet al. (1988). Body weight was measured inkilograms (Homs Beam Balance, DouglasHoms Corp., Belmont, CA) (625 g) with thesubjects lightly clothed and without shoes.Height was measured in meters using a wallstadiometer. Body mass index (BMI) wascalculated as weight/height2. Sociodemo-graphic information was obtained by struc-tured interview and observation.

Observer training and accuracyin estimating portion size

All dietary data were collected by trainedobservers. The observers were young women(aged 18–25 years) who lived in the sameneighborhoods as the subjects and weretrained to unobtrusively record both physi-cal activity and diet intake. They worked inpairs and alternated 4 h shifts scheduled tocover the awake portion of subjects’ days(about 16 h).

Training occurred over a 5 week periodand involved 1) the development of consen-

sus on the names of foods and food prepara-tions, names of common serving utensils,sizes of food items (small, medium, large,etc.), and descriptions of the fullness ofserving utensils (half-full, full, heaping, etc.);2) numerous practice sessions in sizing fooditems and estimating fullness of servingutensils; and 3) practice in the collection ofweighed recipes from local women. As ob-server competence advanced, training incor-porated the development of a data tablecontaining average weights and edible por-tions of common food items. The table wasdeveloped by purchasing representativesamples of foods in local shops and weighingthem in the laboratory, and an averageweight was established for each type of foodand size category. Breads were sized bymonetary units (i.e., a 50 peso bread), asthat is the way they are named. Since breadsize tended to change over time, a represen-tative sample of breads was obtained fromvendors every 6–12 months and weighed inthe laboratory. For foods served using spoonsor cups, the conversion of serving sizes togram weights was based on an extensiveseries of repeat measurements of commonfoods done in the laboratory using a varietyof common household utensils. All foodswere weighed on a Sunbeam electronic bal-ance (Oster, Mcminnvile, TN) (63 g) cali-brated periodically with standard weights.

The accuracy of the observers in estimat-ing serving sizes was assessed periodicallyby the following procedure. Local womenwho had participated in the project werevisited at meal time and asked to serveplates of food as they normally did. One of us(D.L.D.) weighed the plates on a dietarybalance (Sunbeam electronic, 63 g) as eachfood was added. At the same time, theobservers recorded the serving size of eachfood item in household measures. The scaleweights were not visible to the observers atany time. The estimated weights were laterconverted to metric weights as describedbelow.

Estimated records

In recording food intake, observers esti-mated the serving size of food items as theywere served to or selected by the subject. Forfoods served by the piece, such as pieces of

55ENERGY INTAKE IN URBAN WOMEN

meat or fruit, the observers recorded a de-scription of the food and its relative size(small, medium, large). For foods servedusing spoons or cups, observers recorded thetype of spoon or cup used and the degree offullness. Descriptions of food item sizes werelater converted to gram weights using thefood portion size data table described above.In addition, volumetric measurements ofserving utensils and drinking vessels werecompleted in each subject’s home beforeobservations of the food intake for thatsubject began, and these measurements wereused as appropriate. Observers were alsotrained to obtain serving size weights of allnew and unusual foods and common foodsserved with unusual utensils and to add thisdata to the portion size table as the studyprogressed.

Twenty-four hour recalls

The same trained observers responsiblefor the diet records also administered the 24h recalls. The diet records and 24 h recallswere done during the same week in eachmeasurement round but covered differentdays. Estimated food quantities were con-verted to gram weights as described above.

Calculation of nutrient intake

Energy and nutrient composition of foodswere obtained from published food composi-tion tables (Instituto Nacional de Nutricion,1988; Pennington and Church, 1985; USDA,1994) and proximate analyses of samples ofthe most commonly consumed foods. Thelatter were completed by the Instituto Nacio-nal de Nutricion in Bogota. Recipes (in house-hold units) were collected for compositedishes and beverages. For the more commondishes and beverages, weighed recipes werecollected, the food was prepared in the labo-ratory to determine the final cooked weightof the recipe, and nutrient composition wascalculated from published tables or proxi-mate analyses. For packaged snack foods,nutrient composition was obtained from in-formation on the package. Energy and nutri-ent intakes were calculated for all intakedata using custom software programs writ-ten by G.B.S.

Types of foods consumed

The food items listed in both the recordsand recalls were aggregated into 16 foodgroups on the basis of shared characteris-tics. The food groups are as follows:

1. Alcohol2. Bread, pastry, cookies3. Candy and other sweets4. Coffee, hot chocolate, aguapanela (a

sugar-water drink)5. Fruit juices (made with water or milk,

and added sugar)6. Fruit7. Legumes8. Meat, fish, offal, canned meats and fish,

mixed dishes with meat or fish9. Milk and milk products

10. Mixed vegetable dishes11. Other (condiments, added fats, uncom-

mon foods)12. Rice, rice-based dishes, pasta, pasta-

based dishes13. Roots, tubers, plantains14. Salads, cooked vegetables15. Soft drinks, other drinks (powdered

drink mixes, starch thickened drinks,corn-based drinks)

16. Soup, stew

In the following text, these groups are re-ferred to by the first item, which was themost frequently consumed item in the group.

Number of food items

The number of food items in the recordsand recalls, total and subtotal for each foodgroup, was tabulated and the average num-ber of items per woman per day calculated. Afood item was defined as a food recorded orrecalled as being consumed as a single unitor serving at a given time. For example, aserving of rice (made up of one or morespoonfuls) at a given meal would be recordedas a single food item, a glass of juice as asecond food item. Another serving of riceeaten at the same meal, or later the sameday, would be recorded as a third food item.

Data analysis

Random samples of data were checkedperiodically for coding errors. For all sub-jects with unusually high or low energy


intakes, the original data sheets werechecked for possible coding errors and hand-written comments on the subject that mightclarify unusual values.

To evaluate the accuracy of energy valuescalculated from observer estimated weights,we computed Pearson correlation coeffi-cients between each observer and the scaleweight and among the three observers. Tocontrol for Type I error across six correla-tions, we used the Bonferroni procedure,and a P value of ,0.008 (equal to 0.05divided by the number of correlations [n 56]) was considered significant. Agreementbetween energy values calculated from esti-mated and actual food weights was alsoevaluated graphically using the method ofBland and Altman (1986), in which thedifferences between the values obtained withdifferent methods are plotted against theirmean. A similar analysis was done for theactual and estimated food weights them-selves.

Two-way within-subject analysis of vari-ance (ANOVA) with repeat measures wasused to evaluate the effect of method andtime on macronutrient intake. The depen-dent variable was intake in kilojoules orgrams; the within subjects factors weremethods (record vs. recall) and time (threelevels [i.e., rounds 1, 2, 3]). Wilks’s lambda(L) was used as the criterion of significance.Paired t-tests were used to compare themean number of food items/woman/day inthe diet records and recall and the meannumber of food items in each of the 16 foodgroups in the records and recalls. The Bonfer-roni procedure was used to control for Type Ierror across multiple tests, and a P value of.0045 (.05/11) was considered significant.Linear regression analysis was used to evalu-ate the prediction of total energy intakefrom the number of food items/woman/day.Pearson correlation coefficients were calcu-lated to evaluate the linear relationshipbetween sociodemographic characteristics(age and education) and the magnitude ofthe discrepancy between the methods.

To test the hypothesis that subjects withlow-average energy intakes will tend to over-report on the recall and that subjects withhigh-average intakes will tend to underre-port intake on the recalls, we divided the

subjects into tertiles based on mean energyintake in the records. The difference be-tween energy intake in the records (mean of3 days) and recalls (mean of 3 days) wascalculated for each subject. Mean values for3 days were used to reduce some of theday-to-day variability in intake. The pooledmean energy intakes of the three tertileswere compared using one-way ANOVA.

All statistical analyses were completedusing SPSS 7.5 (SPSS Inc., Chicago IL), andsignificance was set at P 5 0.05, exceptwhere noted.

RESULTSObserver accuracy in estimating

food weights

The energy values calculated from actualweights of foods were compared with thosecalculated from estimated weights for 38food items served at midday meals in differ-ent households. The food items were rice(n 5 12), pasta (n 5 5), beans (n 5 3), soupand stew (n 5 7), meat (n 5 4), fried plantainslices (n 5 4), and salad (n 5 4). Thecorrelation between the energy values calcu-lated from scale weights and estimatedweights was high (r 5 0.82–0.88, P , 0.008),as were correlation coefficients between ob-servers (Table 1). The total energy value ofall 38 food items calculated from scaleweights was not significantly different fromthat calculated from the estimated weights(paired t-test, t 5 - 0.21, P 5 0.837). Coeffi-cients of variation (CV) were also calculatedfor the scale and estimated weights (mean ofthree observers) for each of the foods. Thesevalues ranged from 0.4–42%, with a mean of14 6 11.4%.

Results of the Bland-Altman analysis areshown as two scatterplots, one of the weightsof food items (Fig. 1A) and the other of the

TABLE 1. Correlation among energy values calculatedfrom scale weights and weights estimatedby three observers for n 5 38 food items

Scale Observer 1 Observer 2 Observer 3

Scale 1.00Observer 1 .856* 1.00Observer 2 .854* .934* 1.00Observer 3 .894* .956* .897 1.00

* P , .008.


energy values of the same food items (Fig.1B). For the weights of food items, overallagreement between methods is good. For thefood items (n 5 8) with serving sizes greaterthan 250 g, there is more scatter, suggestinglower observer accuracy (Fig. 1A). The threefood items with the lowest accuracy weresoup (n 5 1) and stew (n 5 2). Energy valuescalculated from actual and estimated foodweights show a similar pattern of distribu-tion (Fig. 1B). The regression of weightdifferences (scale weight minus estimatedweight) on mean weight (Fig. 1A) has a slopewhich is not significantly different from zero,indicating that the size of the serving has noeffect on the difference between the twomethods. Similarly, when the values areexpressed in terms of energy, the slope of theregression is not significant. The mean differ-ence in energy between the two methods(scale weight minus estimated weight) was6.94 6 204.33 kJ (1.66 6 48.84 kcal), whichis equal to .03% of the total energy value ofthe weighed foods.

Anthropometric and sociodemographiccharacteristics of subjects

Mean stature and body weight of thesubjects were 155.6 6 6.1 cm and 53.9 6 7.3kg, respectively. Mean BMI was within thenormal range, 22.2 6 2.6 kg/m2. The sub-jects ranged in age from 19–43 years, with amean of 29.0 6 6.2 years. Formal educationaveraged 6.3 6 2.7 years (Table 2).

Comparison of records and recalls

Energy and macronutrient intakes ob-tained using records and recalls on week-days are shown in Table 3. In the records,mean intakes of energy, carbohydrate, pro-tein, and fat were 8,814 6 1,742 kJ (2,106.6 6316.4 kcal), 372.2 6 74.32 g, 61.0 6 14.01 g,and 45.4 6 13.72g per day, respectively.These values were significantly greater thanthose obtained using 24 h recalls (two-wayANOVA with repeat measures, ( 5 0.80,F(1.84) 5 20.49, P , 0.001). Absolute valuesobtained by records were higher by 11–13%than those obtained by recall. The timeeffect in the ANOVA model was not signifi-cant.

The average number of food items/women/day was significantly higher in the recordsthan in the recalls, 15.4 6 4.1 as opposed to13.1 (6 4.1 food items (t 5 -8.142, P , 0.001).

Fig. 1. A: Scatterplot of differences between scale weight and estimated weight and mean of twomeasurements following method of Bland and Altman (1986). B: Scatterplot of differences in energyvalues calculated from scale weight and estimated weight and energy values calculated from the mean ofthe scale and estimated weights following method of Bland and Altman (1986).

TABLE 2. Sociodemographic characteristicsof Cali women

Group age(years) n

Education completed(years)

Mean 6 SD Range

19 3 7.3 6 0.8 7–820–24 23 7.3 6 2.5 3–1325–29 17 6.5 6 2.3 2–1030–34 21 6.6 6 3.4 0–1435–39 19 4.8 6 1.9 2–940–43 2 4.5 6 2.1 3–6

Total 85 6.3 6 2.7 0–14


The discrepancy between records and re-calls was not correlated with either age (r 5-.008, P 5 0.467) or years of education (r 5-0.19, P 5 0.08). The linear relationshipbetween energy intake (records) and num-ber of food items food items/woman/day isshown in the scatterplot in Figure 2. Theregression equation for predicting energyintake from food items/woman/day is energyintake, MJ 5 0.386 · food items 1 2.88. Thenumber of food items/woman/day accountsfor approximately 41% of the variance inenergy intake. The correlation between en-ergy intake and items/woman/day is 0.64(P , 0.001).

Types of foods in records and recalls

The average number of food items/woman/3 intake days in each of 16 foodgroups is compared for records and recalls inTable 4. The four most frequently consumedfoods in both the records and recalls werecoffee, rice, bread, and meat. These four foodgroups accounted for over 50% of the totaldietary energy intake in the records. The

average number of food items in each foodgroup tended to be higher in the records andwas significantly higher in seven of 16 foodgroups. In only one food group, soft drinks,were the recall values higher than the rec-ords. The food groups showing the greatestdiscrepancies between the two methods werecoffee, bread, meat, fruit, and salad. Foodgroups showing relatively good agreementbetween the records and recalls were rice,roots, fruit juices, legumes, candy, mixedvegetable dishes, milk, and other.

Is underreporting on recalls relatedto average energy intake?

The difference between mean energy in-take in records and recalls for each subject isplotted by tertile of intake in Figure 3. Forsubjects in the middle tertile, the averageintake group, most of the differences be-tween methods (records minus recalls) werepositive, indicating a tendency to underre-porting. Subjects in the high intake tertileshow a similar pattern. The mean differ-ences between methods are positive, 1.1 61.37 and 1.4 6 1.89 in the middle and highenergy intake tertiles, respectively. For sub-jects in the low-intake tertile, the meandifference is also positive but absolutelysmaller, 0.12 6 1.92 MJ, indicating that onaverage there is less discrepancy betweenrecalls and records. There is, however, morevariability in the low-intake tertile. Themeans of the low-intake and high-intaketertiles are not significantly different fromthat of the middle tertile, but they aresignificantly different from each other (one-way ANOVA, Bonferroni post hoc test, P 50.017). A similar analysis of the number offood items in the records and recalls for eachtertile gave similar results.

Fig. 2. Scatterplot of energy intake (MJ/day) againstnumber of food items/woman/day in diet records for 85women for 3 days each.

TABLE 3. Energy and macronutrient intake of Cali women (n 5 85) from diet records (3 nonconsecutive days) and24 h recalls (3 nonconsecutive days)

NutrientRecords

(mean 6 SD)24 h recalls

(mean 6 SD) Significance

Energy (kJ) 8,814 6 1,742 7,923 6 2,024 ,.001Energy (kcal) 2,107 6 316.4 1,894 6 483.9Carbohydrate (g) 372.2 6 74.32 334.2 6 82.00 ,.001Protein (g) 61.0 6 14.01 54.0 6 15.79 ,.001Fat (g) 45.4 6 13.72 40.6 6 15.82 .006


DISCUSSIONObserver accuracy

The accuracy of observers in estimatingweights of common food items was good, andenergy values calculated with estimatedweights were not significantly different thanthose calculated from scale weights. Bothover- and underestimation occurred buttended to cancel one another out such thatthe mean difference between actual andestimated weights was low. The greatest

discrepancies between actual and estimatedweights were for food items (soups andstews) with total portion sizes greater than250 g. The lower accuracy of observers inestimating the weight of food served inmultiple spoonfuls results probably becauseeach spoonful may have a different level offullness, and the repeat spoonfuls typicallyoccur in rapid succession. The greater vari-ability in the estimated weights of thesetypes of foods probably had a minimal im-pact on the calculation of daily energy in-take because food items with weights of 250g or greater occurred only in 20% of all fooditems in the diet records.

A recent study by Romieu et al. (1997) alsoreported good agreement between actualfood weights and weights estimated bytrained observers. We concur with Romieuand colleagues that adequate training re-quires a substantial time investment. Otherstudies using trained observers have re-ported lower levels of agreement betweenactual and estimated weights, with CVs of16–53% (Rutishauser, 1982; cited in Bing-ham and Nelson, 1991) and 21–70% (Bollardet al., 1988). Studies using untrained sub-jects have reported higher CVs (between 1and 96%) (Young et al., 1952), as would beexpected.

Records vs. recalls

For energy and macronutrient intake, thevalues obtained using 24 h recalls were

Fig. 3. Scatterplot of differences between recordsand recalls plotted by tertile of mean energy intake inrecords (1 5 low-intake, 3 5 high-intake). Values aremeans of records (n 5 85) and means of recalls (n 5 85).Mean 6 SD are shown for each tertile. *Significantlydifferent than tertile 3 (one-way ANOVA, Bonferronipost hoc comparisons, P 5 0.017)

TABLE 4. Comparison of mean number food items per woman (n 5 85) for 3 days of intake by food groupin diet records and 24 h recalls1

Food group

Items/woman/3 intake days

Record Recall Significance2

1. Coffee, chocolate, aguapanela 9.0 6 5.12 7.1 6 3.86 ,.0012. Rice, rice dishes, pasta, pasta dishes 6.0 6 1.87 5.5 6 1.64 .0143. Bread, pastry, cookies 5.7 6 2.24 4.7 6 2.47 ,.0014. Meat, fish and mixed dishes 5.7 6 2.1 4.9 6 2.00 ,.0015. Roots, tubers, plantains 3.0 6 2.14 2.4 6 1.61 .0286. Soft drinks, other drinks 2.9 6 2.00 4.9 6 1.93 ,.0017. Fruit juices 2.4 6 1.93 2.1 6 2.00 .2048. Fruits 2.0 6 2.44 0.9 6 1.29 ,.0019. Soup, stew 1.9 6 1.44 1.4 6 1.43 .003

10. Legumes 1.8 6 1.66 1.5 6 1.44 .16611. Salads, vegetables 1.2 6 1.34 0.6 6 0.98 .00112. Other 1.2 6 1.48 0.7 6 0.89 .00413. Candy 1.1 6 1.40 0.9 6 1.34 .32014. Mixed vegetable dishes 0.7 6 1.00 0.4 6 0.75 .06815. Milk and milk products 0.7 6 1.07 0.5 6 1.03 .24916. Alcohol 0.1 6 0.40 0.1 6 0.38 .4361 Food groups listed by frequency of consumption in the diet records.2 P , .0045 considered significant.


11–13% lower than the values obtained us-ing records. Although an error of that magni-tude might be acceptable in some studies, itwas not acceptable in this one because one ofour goals was to measure the components ofenergy balance at the group level.

The lower mean energy intake calculatedfrom the recalls can be almost entirely ex-plained by the lower number of food items inthe recalls. The differences between recordsand recalls were 891 kJ for energy intakeand 2.3 food items. Based on the regressionequation in Figure 2, a difference of 2.3 fooditems is equal to 888 kJ. Only a few otherstudies have reported data on the extent towhich the number of food items consumed isin error (Bingham, 1987). Acheson et al.(1980) found that men who recorded theirown food intake and were then asked to fillout a questionnaire reporting what they hadeaten in the past 24 h usually left out atleast one food item. The incorrect number offood items, together with a tendency tounderestimate portion size, resulted in anunderestimation of energy intake by 21%.Thomson (1958) found that pregnant womenin Scotland who kept weighed food recordsand were then asked to recall food intake fora 24 h period covered by the records alsoomitted food items. The 24 h recalls underes-timated energy intake by 17%, approxi-mately 5% of which could be accounted forby missing food items. Linusson et al. (1974)kept weighed diet records for lactatingwomen (n 5 88) in a hospital setting andfound that women omitted food items in 24 hrecalls, although which or how many is notclear. The women also underestimated foodquantities in eight of 14 food groups.

In the present study, virtually all of thedifference between energy intake in the rec-ords and recalls can be accounted for by alower number of food items in the recalls,whereas in the other studies portion sizeestimation appeared to be more of a prob-lem. Important differences between ourstudy and the others might help explainthese discrepancies. First, in this studytrained observers did both the records andrecalls and had completed a calibration ofhousehold dishes and serving utensils be-fore administering the recall. This shouldhave improved the accuracy of serving size

estimation. Second, subjects in our studywere women living in poverty for whom theacquisition of food was constrained by lim-ited budgets, and high food intake had posi-tive social value. Under such circumstances,it is likely that subjects are more focused onfood quantities than are nondieting subjectsliving in food-rich environments. The womenin our study impressed us with their accu-racy in describing food quantities and theirability to divide large amounts of food intomultiple, equal-sized servings. Further, in acultural context in which being able to con-sume large amounts of food is valued, a biastoward underestimating food eaten is un-likely. Third, the observers in our study werewell trained and shared the language re-lated to food and food portions with thesubjects. This should have facilitated com-munication of portion size from the subjectand its correct coding by the observer.

It is clear that the subjects in the presentstudy tended to recall some foods betterthan others (Table 3). One might assumethat the food items most frequently con-sumed would be recalled most accurately,but this was not the case. Three of the fourmost frequently consumed items (coffee,bread, meat) appeared less frequently in therecalls than in the records. On the otherhand, values for the four least frequentlyconsumed foods (candy, mixed vegetabledishes, milk and alcohol) were similar in therecall and records. Other foods that ap-peared with comparable frequencies in therecalls and records were rice, roots, fruitjuices, and legumes. These foods were typi-cal components of the midday meal, themain meal of the day and the culturallymost significant meal (Dufour et al., 1997).Meat and salad are also normally part of themidday meal, but they were relatively under-reported in the recalls. Fruit, a food eatenbetween meals, also showed significant un-derreporting in the recalls relative to therecords.

Other studies looking at the agreementbetween actual and recalled diet intake interms of foods have also found larger discrep-ancies for some foods than others (Thomaset al., 1954; Adelson, 1960; Bransby et al.,1948; Linusson et al., 1974; Hankin et al.,1975). The foods showing the largest discrep-


ancies vary from study to study, suggestingpopulation-specific factors are associatedwith the recall of specific foods. One popula-tion-specific factor is obviously the choice ofcore foods. For example, a study of adultmales in the USA by Hankin et al. (1975)found that items eaten regularly (i.e., thecore foods) tended to be recalled more accu-rately, but that generalization did not applyto all food items. It is clear, then, thatknowledge of the type of food items mostlikely to be omitted in the recalls can be usedto improve interviewer’s ability to probe inthe interview. In addition, knowledge of thetypes of food items most likely to be missedusing recall techniques in a given popula-tion would also be important in epidemiologi-cal studies examining the link between dis-ease prevalence and intake of specific foods.For example, a study of the relationshipbetween cancer and fruit and vegetable in-take in Cali women would be compromisedby the significant underreporting of thosefoods using 24 h recall.

When the Cali subjects were grouped ac-cording to tertile of energy intake, we founda tendency to underreport on the recallsrelative to the records in the middle andhigh energy intake tertiles. Subjects in thelow energy intake tertile showed the bestaverage agreement between records and re-calls but the greatest variability. In thistertile, the tendency of some women to under-report on the recalls was balanced by overre-porting by an approximately equal numberof women. Hence, in this group of subjectswe find no clear support for the idea thatunder- and overreporting on recalls is re-lated to average energy intake measuredusing food records.

In conclusion, we found that 24 h recallsunderestimated energy intake by 11% incomparison to estimated records kept bytrained observers. The underestimation canbe almost entirely explained by the fewernumber of food items in the recalls. Somefood items tended to be more accuratelyrecalled than other foods. Underreportingon recalls was a general tendency in thisgroup of subjects, and under- and overreport-ing on recalls was not clearly related toaverage energy intake.

ACKNOWLEDGMENTS

The authors express their gratitude toZoila de Maldonado, Betty de Orozco, andJairo Ardila for their excellent technicalsupport and also to Blanca Diaz, Clara InesGil Rivas, Mayerling Valencia Castro, RudtGarcia Ramirez, and Yasmin Villota for theirinvaluable assistance in data collection. Weare also grateful to Coldeportes del Valle forlogistical support, to the Instituto Nacionalde Nutricion for the analysis of food samples,and to Fundacion para la Educacion Supe-rior for their continued support. We thankthe women of Agua Blanca for their patienceand cooperation. Support for data analysisand manuscript preparation was providedby a CRCW faculty fellowship to D.L.D. fromthe University of Colorado.

LITERATURE CITED

Acheson KJ, Campbell IT, Edholm OG, Miller DS, StockMJ. 1980. The measurement of food and energy intakein man—an evaluation of some techniques. Am J ClinNutr 33:1147–1154.

Adelson SF. 1960. Some problems in collecting dietarydata from individuals. J Am Diet Assoc 36:453–461.

Bingham SA. 1987. The dietary assessment of individu-als: methods, accuracy, new techniques and recommen-dations. Nutrition Abstracts and Reviews (Series A)57:705–742.

Bingham SA, Nelson M. 1991. Assessment of food con-sumption and nutrient intake. In: Margette BM,Nelson M, editors. Design concepts in nutritionalepidemiology. New York: Oxford. p 153–191.

Blake EC, Durnin JVGA. 1963. Dietary values for a 24hour recall compared to a 7-day survey on elderlypeople. Proc Nutr Soc 22:i.

Bland JM, Altman DG. 1986. Statistical methods forassessing agreement between two methods of clinicalmeasurement, Lancet 1:307–310.

Bollard JE, Yuhas JA, Bollard TW. 1988. Estimation offood portion sizes: effectiveness of training. J Am DietAssoc 88:817–820.

Bransby ER, Daubney CG, King J. 1948. Comparison ofresults obtained by different methods of individualdietary survey. Br J Nutr 2:89–110.

Dufour DL, Staten LK, Reina JC, Spurr GB. 1997.Living on the edge: dietary strategies of economicallyimpoverished women in Cali, Colombia. Am J PhysAnthropol 102:5–15.

Gersovitz M, Madden JP, Smiciklas-Wright H. 1978.Validity of the 24-hr. dietary recall and seven-dayrecord for group comparisons. J Am Diet Assoc 73:48–55.

Greger JL, Etnyre GM. 1978. Validity of 24-hour dietaryrecalls by adolescent females. Am J Public Health68:70–72.

Hankin JH, Rhoads GG, Gobler GA. 1975. A dietarymethod for an epidemiologic study of gastrointestinalcancer. Am J Clin Nutr 28:1055–1061.

Huss-Ashmore R. 1996. Issues in the measurement ofenergy intake for free-living human populations. Am JHum Biol 8:159–168.


Instituto Nacional de Nutricion. 1988. Tabla de composi-cion de alimentos Colombianos, 5th ed. Bogota, Colom-bia: Instituto Nacional de Nutricion.

Linusson EEI, Sanjur D, Erickson EC. 1974. Validatingthe 24-hour recall method as a dietary survey tool.Arch Latinoam Nutr 24:277–294.

Lohman TG, Roche AF, Martorell R, editors. 1988.Anthropometric standardization reference manual.Champaign, IL: Human Kinetics.

Madden PJ, Goodman SJ, Guthrie HA. 1976. Validity ofthe 24-hr. recall. J Am Diet Assoc 68:143–147.

Ohlson MA, Jackson L, Boek J, Cederquist DC, BrewerWD, Brown CG. 1950. Nutrition and dietary habits ofaging women. Am J Public Health 40:1101–1108.

Pennington JAT, Church HN. 1985. Food values ofportions commonly used, 14th ed. New York: Harper &Row Publishers.

Romieu I, Hernandez-Avila M, Rivera JA, Ruel MT,Parra S. 1997. Dietary studies in countries experienc-

ing a health transition: Mexico and Central America.Am J Clin Nutr 65:1159S–1165S.

Rutishauser IHE. 1982. Food models, photographs orhousehold measures? Proc Nutr Soc Aust 7:144.

Spurr GB, Dufour DL, Reina JC. 1996. Energy expendi-ture of urban Colombian women: a comparison of pat-terns and total daily energy expenditure by the heart rateand factorial methods.Am J Clin Nutr 63:870–878.

Thomas RU, Fox HM, Kelly HJ, Moyer EZ, Macy IG.1954. Rapid method for qualitative appraisal of foodintake of groups. J Am Diet Assoc 30:865–871.

Thomson AM. 1958. Diet in pregnancy. Br J Nutr12:446–461.

Thompson FE, Byers T. 1994. Dietary assessment re-source manual. J Nutr 124:2245S–2317S.

USDA. 1994. USDA nutrient data bank (online data base).Young CM, Hagan GC, Tucker RE, Foster WD. 1952. A

comparison of dietary study methods. II Dietary his-tory vs. seven-day record vs. 24hr recall. J Am DietAssoc 28:218–221.


Documents

Estimating energy intake of urban women in Colombia: Comparison of diet records and recalls