Cognitive processes in children’s dietary recalls: Insight from methodological studies

Abstract

Objective

This article summarises 12 dietary-reporting methodological studies with children (six validation studies, one non-validation study, five secondary analyses studies of data from one or more of the six validation studies), identifies research gaps, and provides recommendations for (a) improving children’s recall accuracy and (b) details to specify in publications of studies that utilise children’s dietary recalls.

Subjects/Methods

Randomly selected children (ages nine to ten) were observed eating school breakfast and school lunch, and interviewed to obtain dietary recalls.

Results

Children’s recall accuracy improved slightly between the first and third recalls, but individual children’s accuracy was inconsistent from one interview to the next. Although accuracy was poor overall, it was better for boys with reverse-order (evening-to-morning) prompts, but for girls with forward-order (morning-to-evening) prompts. Children recalled breakfast intake less accurately than lunch intake. Children’s accuracy did not depend on whether recalls were obtained in-person or by telephone, but was better for recalls obtained with open than meal format. Retention interval was crucial as children’s accuracy was better for prior-24-hour recalls (about the 24 hours immediately preceding the interview) than previous-day recalls (about midnight to midnight of the day before the interview). Observations of school meals did not affect children's recalls. Children’s recall accuracy was related to their age/sex body mass index percentile. Conventional report rates (which disregard accuracy for items and amounts) overestimated accuracy for energy and macronutrients, and masked complexities of recall error.

Conclusions

Research concerning errors in Children’s dietary recalls provides insight for improving children’s recall accuracy.

Introduction

In dietary-reporting validation studies, reported information from a method such as a 24-hour dietary recall (24hDR) is compared to reference information from a gold standard method such as direct observation which is assumed to be the truth, collected independent of the subject’s memory, and concerns the same meal(s) as reported information. Comparison of reported information to reference information allows identification of matches (referenced [eaten] items that are reported), omissions (referenced items that are unreported), and intrusions (reported items that are unreferenced [uneaten]) (Smith, 1991; Smith, Jobe & Mingay, 1991). Such comparisons are more appropriately called “relative validation studies” when both reported information and reference information are reported by subjects, for example, when 24hDRs are compared to food records.

Results from validation studies (Byers et al., 1993; Eck, Klesges & Hanson, 1989; Emmons & Hayes, 1973) underscore concerns that it is unrealistic to expect parents to accurately report their elementary-school children’s dietary intake, especially for meals eaten in locations at which parents are not present, such as at school (Livingstone & Robson, 2000). Thus, it is necessary to rely on elementary-school children to self-report their own dietary intake information.

Dietary self-report methods are generally used with children over age nine years, or third grade (Frank, 1991). Several methods have been used with varying levels of suitability. Food records create considerable subject burden (Dwyer, 1999) and often are completed later from memory instead of at the time of intake (Gillman, Hood, Moore & Singer, 1994; Vereecken & Maes, 2003). Another concern is that the process of completing food records may change eating behavior (Buzzard, 1998; Rockett, Berkey & Colditz, 2003; Smith, 1999). Food frequency questionnaires (FFQs) require cognitive skills (e.g., ability to average consumption) that many elementary-school children lack (e.g., Baranowski et al., 1997; Domel et al., 1994; Field et al., 1999; Rockett et al., 2003). Twenty-four hour dietary recalls, or recalls of intake at school, have been provided by elementary-school children without parental assistance for national surveys (e.g., Burghardt, Ensor, Hutchinson, Weiss & Spencer, 1993; US Department of Agriculture, Food and Nutrition Service, Office of Research, Nutrition & Analysis, 2007), for research studies (e.g., Moore et al., 2008; Todd & Kretsch, 1986), for evaluation of nutrition education interventions (e.g., Baranowski et al., 2003; Luepker et al., 1996; Lytle et al., 2002; Perry et al., 1998; Receveur, Morou, Gray-Donald & Macaulay, 2008), and for assessment of the relative validity of children’s FFQs (e.g., Field et al., 1999; Rockett et al., 1997; Wong, Boushey, Novotny & Gustafson, 2008). A 24hDR does not require that subjects be literate, and it is unlikely to alter intake (Buzzard, 1998); in addition, recalls can be conducted without advance notice.

Dietary recalls by children have been the focus of methodological research by Baxter and colleagues. This article summarises key findings from 12 methodological studies conducted with children by Baxter and colleagues and published between 2002 and 2007 — six dietary-reporting validation studies (Studies 1, 3 through 6, and 8), one non-validation study (Study 7), and five secondary analyses studies (Studies 2 and 9 through 12) that utilised data from one or more of the six validation studies. The article identifies research gaps and concludes with recommendations for (a) improving children’s recall accuracy and (b) details to specify in publications of studies that utilise children’s dietary recalls.

Materials / subjects and methods

Subjects and design

Before data collection for each study, approval was obtained from the appropriate institutional review boards for research involving human subjects, and child assent and parental consent to participate were obtained in writing. This section summarises details of data collection that were common to many of the studies; complete details are found in the publications for each study. Table 1 illustrates and defines terms. Table 2 provides an overview of each study including the purpose, sample, school year and number of schools, target period of recall and interview time, prompts or interview format, design, and key results.

Table 1.

Illustration (using items and energy [in kilojoules (kJ)] for an interview for a given child) and definitions of terms

	Observed		Reported		Meal	Weightb	Total	Match	Omission	Intrusion
	servingsa	kj	servingsa	kj	component		inaccuracy
Breakfast
Milk, white	1.00	335	0.75	250	Beverage	1.00	0.25	1.00
Sausage	1.00	420	--	--	Breakfast meat	1.00	1.00		1.00
Biscuit	1.00	525	1.00	525	Bread/grain	1.00	0.00	1.00
Cereal, Corn Flakes	--	--	1.00	335	Bread/grain	1.00	1.00			1.00
Peaches	0.50	170	1.00	335	Fruit	1.00	0.50	1.00
Lunch
Hamburger on bun	1.00	965	--	--	Combination entrée	2.00	2.00		2.00
Mustard	1.00	40	--	--	Condiment	0.33	0.33		0.33
Carrots	0.50	105	1.00	210	Vegetable	1.00	0.50	1.00
Milk, strawberry	1.00	630	0.75	475	Beverage	1.00	0.25	1.00
Ice cream, chocolate	1.00	585	1.00	585	Dessert	1.00	0.00	1.00
Sausage pizza	--	--	0.75	1050	Combination entrée	2.00	1.50			2.00
Correspondence rate: For energy or any nutrient, this is calculated as (sum of corresponding amounts from matches/total observed amount) × 100%. It is a genuine measure of accuracy that is sensitive to reporting errors. It has a lower bound of 0% (indicating that nothing observed eaten was reported eaten) and an upper bound of 100% (indicating that all items observed eaten and all amounts observed eaten were reported correctly). Higher rates reflect better accuracy. In the illustration, for energy, the correspondence rate for breakfast and lunch is (2110 / 3775) × 100% = 56%.
Corresponding amount from a match: For energy or any nutrient, this is the smaller of the reported and observed amounts of a match (or the reported amount if it equals the observed amount of a match). In the illustration, for energy, the sum of the corresponding amounts from matches for breakfast and lunch is 250 (from 0.75 serving of white milk) + 525 (from 1.00 serving of biscuit) + 170 (from 0.50 serving of peaches) + 105 (from 0.50 serving of carrots) + 475 (from 0.75 serving of strawberry milk) + 585 (from 1.00 serving of chocolate ice cream) = 2110 kilojoules.
Inflation ratio: For energy or any nutrient, this is calculated as ([sum of over-reported amounts from intrusions + sum of over-reported amounts from matches] / total observed amount) × 100%. It is a measure of reporting error. It has a lower bound of 0% (indicating no over-reported amounts of matches and no over-reporting from intrusions), but no upper bound because there is no limit on what an individual can report. Lower inflation ratios reflect better accuracy. In the illustration, for energy, the inflation ratio for breakfast and lunch is ([1385 + 270] / 3775) × 100% = 44%.
Intruded kilojoules: This is calculated as (sum of kilojoules from reported amounts of intrusions) + (sum of kilojoules from parts of matches for which reported amounts exceeded observed amounts). For energy, this is the same as (sum of over-reported amounts from intrusions) + (sum of over-reported amounts from matches).
Intrusion: This is a food item that was not observed eaten but was reported eaten for that meal. In the illustration, for breakfast and lunch, intrusions are Corn Flakes cereal and sausage pizza, and the weightedb number of intrusions is 1.00 + 2.00 = 3.00.
Intrusion rate:b This food-item rate is calculated as (sum of weighted intrusions/[sum of weighted intrusions + sum of weighted matches]) × 100%. Values are undefined if children were observed to eat something but reported eating nothing. Defined values may range from 0% (indicating no intrusions) to 100% (indicating that no items reported eaten were observed eaten); thus, lower intrusion rates indicate better accuracy. In the illustration, the intrusion rate for breakfast and lunch is (3.00 / [3.00 + 6.00]) × 100% = 33%.
Match: This is a food item that was observed eaten and reported eaten for that meal. In the illustration, for breakfast and lunch, matches are white milk, biscuit, peaches, carrots, strawberry milk, and chocolate ice cream, and the weightedb number of matches is 1.00 + 1.00 + 1.00 + 1.00 + 1.00 + 1.00 = 6.00.
Matched kilojoules: For an interview, this variable is the sum of kilojoules from amounts of matches that overlapped between reported and observed amounts. For energy, this is the same as (sum of corresponding amounts from matches).
Number of items observed eaten:b This is the sum of the weighted number of items observed eaten in any non-zero amount. In the illustration, the weighted number of items observed eaten for breakfast and lunch is 1.00 + 1.00 + 1.00 + 1.00 + 2.00 + 0.33 + 1.00 + 1.00 + 1.00 = 9.33.
Number of items reported eaten:b This is the sum of the weighted number of items reported eaten in any non-zero amount. In the illustration, the weighted number of items reported eaten for breakfast and lunch is 1.00 + 1.00 + 1.00 + 1.00 + 1.00 + 1.00 + 1.00 + 2.00 = 9.00.
Observed kilojoules: This is the sum of kilojoules from amounts of items observed eaten. For energy, this is the same as the total observed amount.
Omission: This is a food item that was observed eaten but was not reported eaten for that meal. In the illustration, for breakfast and lunch, omissions are sausage, hamburger on bun, and mustard, and the weightedb number of omissions is 1.00 + 2.00 + 0.33 = 3.33.
Omission rate:b This food-item rate is calculated as (sum of weighted omissions/[sum of weighted omissions + sum of weighted matches]) × 100%. Values are undefined if children were observed to have eaten nothing regardless of what was reported eaten. Defined values may range from 0% (indicating no omissions) to 100% (indicating that no items actually eaten were reported eaten); thus, lower omission rates indicate better accuracy. In the illustration, the omission rate for breakfast and lunch is (3.33 / [3.33 + 6.00]) × 100% = 36%.
Omitted kilojoules: This is calculated as (sum of kilojoules from entire unreported amounts of omissions) + (sum of kilojoules from parts of matches for which reported amounts were smaller than observed amounts). For energy, this is the same as (sum of unreported amounts from omissions) + (sum of under-reported amounts from matches).
Over-reported amount from an intrusion: This is the entire reported amount of an intrusion. In the illustration, for energy for breakfast and lunch, the sum of the over-reported amounts from intrusions is 335 (from 1.00 serving of Corn Flakes cereal) + 1050 (from 0.75 serving of sausage pizza) = 1385 kilojoules.
Over-reported amount from a match: This is the part of the reported amount that exceeded the observed amount of a match (or zero if the reported amount was less than the observed amount of a match). In the illustration, for energy for breakfast and lunch, the sum of the over-reported amounts from matches is 165 (from 0.50 serving of peaches) + 105 (from 0.50 serving of carrots) = 270 kilojoules.
Previous-day recall: This recall concerns midnight to midnight of the day before the interview. For example, for an interview conducted on a Tuesday at 1:30 p.m., the subject would be asked to report meals/snacks eaten on Monday between midnight and midnight.
Prior-24-hour recall: This recall concerns the 24 hours immediately preceding the interview. For example, for an interview conducted on a Tuesday at 1:30 p.m., the subject would be asked to report meals/snacks eaten between 1:30 p.m. on Monday and 1:30 p.m. on Tuesday.
Report rate: For energy or any nutrient, this is calculated as (total reported amount/total observed amount) × 100%. It is a conventional measure of accuracy that is indifferent to reporting errors. It has a lower bound of 0% (indicating that nothing was reported), but no upper bound because there is no limit on what an individual can report. Report rates with values close to 100%, greater than 100%, and less than 100% have typically been interpreted as indicating better accuracy, over-reporting, and under-reporting, respectively. However, research shows that report rates overestimate accuracy because the report rate = correspondence rate + inflation ratio. In the illustration, for energy, the report rate for breakfast and lunch is (3765 / 3775) × 100% = 100%.
Reported kilojoules: This is the sum of kilojoules from amounts of items reported eaten. For energy, this is the same as the total reported amount.
Total inaccuracy:b This is the sum of three components for food items and amounts: (a) the sum, over matches, of the absolute difference between amounts observed and reported for each match times the weight; (b) the sum, over intrusions, of each intruded amount times the weight; and (c) the sum, over omissions, of each omitted amount times the weight. This measure cumulates errors (in servings) for all items. Values close to zero indicate better accuracy; greater values indicate worse accuracy. In the illustration, total inaccuracy for breakfast and lunch is 0.25 + 1.00 + 0.00 + 1.00 + 0.50 + 2.00 + 0.33 + 0.50 + 0.25 + 0.00 + 1.50 = 7.33 servings.
Total observed amount: For energy or any nutrient, this is the sum of amounts observed eaten; this is the same as (sum of corresponding amounts from matches) + (sum of under-reported amounts from matches) + (sum of unreported amounts from omissions). In the illustration, for energy, the total observed amount for breakfast and lunch is 335 + 420 + 525 + 170 + 965 + 40 + 105 + 630 + 585 = 3775 kilojoules.
Total reported amount: For energy or any nutrient, this is the sum of amounts reported eaten; this is the same as (sum of corresponding amounts from matches) + (sum of over-reported amounts from matches) + (sum of over-reported amounts from intrusions). In the illustration, for energy, the total reported amount for breakfast and lunch is 250 + 525 + 335 + 335 + 210 + 475 + 585 + 1050 = 3765 kilojoules.
Under-reported amount from a match: This is the part of the observed amount that exceeded the reported amount of a match (or zero if the observed amount was less than the reported amount of a match). In the illustration, for energy, the sum of the under-reported amounts from matches for breakfast and lunch is 85 (from 0.25 servings of white milk) + 155 (from 0.25 serving of strawberry milk) = 240 kilojoules.
Unreported amount from an omission: This is the entire observed amount of an omission. In the illustration, for energy, the sum of the unreported amounts from omissions for breakfast and lunch is 420 (from 1.00 serving of sausage) + 965 (from 1.00 serving of hamburger on bun) + 40 (from 1.00 serving of mustard) = 1425 kilojoules.

^aAmounts observed eaten and/or reported eaten of standardised school-meal portions were recorded using a qualitative scale and then assigned numeric values as none = 0.00, taste = 0.10, little bit = 0.25, half = 0.50, most = 0.75, all = 1.00, or as the actual number of servings if more than one was observed eaten and/or reported eaten.

^bFor some variables, a weight was assigned to each match, omission, and intrusion according to meal component (e.g., beverage, bread/grain, breakfast meat, combination entrée, condiment, dessert, entrée, fruit, miscellaneous, vegetable) with combination entrée (e.g., hamburger on bun) = 2, condiment (e.g., mustard, syrup) = 0.33, and remaining meal components = 1.00, so that errors for combination entrées counted more than errors for condiments and remaining meal components.

Table 2.

Overview by study for each of 12 methodological studies with fourth-grade children summarised in this article

Study and reference	Purpose or focus of studya	Sampleb	School year and number of schools	Target period of recall and interview time	Prompts or interview format	Design	Key results
Study 1 – Baxter et al., 2002	DRVS to evaluate the accuracy and consistency of children's recalls of school meals obtained during 24- hour dietary recalls (24hDRs)	104 total: 24 AA m 27 AA f 25 W m 28 W f	1999 – 2000 school year; 6 schools	Previous-day recall obtained in the morning (after school breakfast)	Forward- order prompts	104 children were each observed and interviewed once;92 (of the 104) were each observed and interviewed a 2nd time; 79 (of the 92) were each observed and interviewed a 3rd time. There were 25 to 99 days between any 2 consecutive interviews for a child. Interviews were conducted in person.	Across 275 recalls, mean omission rate was 51%, mean intrusion rate was 39%, and mean total inaccuracy was 7.1 servings. Total inaccuracy decreased from the 1st to 3rd recall (P = 0.006; mixed- model analysis of variance [ANOVA]). The intraclass correlation coefficient for omission rate was 0.15 (P < 0.04; mixed-model ANOVA) and for total inaccuracy was 0.29 (P < 0.0003; mixed- model ANOVA).
Study 2 – Baxter et al., 2006b	SAS to investigate the influences of body mass index (BMI) and sex on children's accuracy for reporting school meals during multiple 24hDRs	Secondary analyses of data from Study 1 (for 79 children interviewed 3 times each [on non- consecutive days] about the previous-day's intake).				Age/sex BMI percentiles of 79 children: 48 healthy weight (≥ 5th and < 85th percentiles), 14 at risk of overweight (≥ 85th and < 95th percentiles), and 17 overweight (≥ 95th percentile). For each of 5 outcome measures, a full general-linear- model repeated- measures analysis was conducted.	For number of items observed eaten, the BMI- category-x-trial interaction was marginally significant (P = 0.079); over 3 trials, this outcome was stable for healthy-weight children, decreased and then stabilised for children at risk of overweight, and was stable and then decreased for overweight children. Also, this outcome was greatest for overweight children and least for healthy-weight children (P = 0.015). For number of items reported eaten, no significant effects were found. For omission rate (P = 0.028) and intrusion rate (P = 0.083), the BMI- category-x-trial interaction was significant and marginally significant, respectively; over 3 trials, both outcomes decreased for healthy-weight children, decreased and then stabilised for children at risk of overweight, and increased and then stabilised for overweight children. Also, both omission rate (P = 0.006) and intrusion rate (P = 0.025) decreased and then stabilised over 3 trials. Total inaccuracy decreased slightly over trials(P = 0.076); also, this outcome was greater for boys than girls (P = 0.049).
Study 3 – Baxter et al., 2003c	DRVS to evaluate whether children would report school meals during 24hDRs more accurately with reverse- order (evening-to- morning) or forward-order (morning-to- evening) prompts	121 total: 31 AA m 29 AA f 32 W m 29 W f	2000 – 01 school year; 11 schools	Previous-day recall obtained in the morning (after school breakfast)	Each child had 2 interviews (1 with forward- order prompts and 1 with reverse- order prompts)	Each child's 2 interviews were separated by ≥ 28 days; a random half of each race/sex group was assigned for prompting in forward order for their 1st interview and reverse order for their 2nd interview, while the other half had the complementary assignment. Children did not know that order prompts would vary between the 2 interviews. Interviews were conducted in person.	Across 242 recalls, mean omission rate was 57%, mean intrusion rate was 36%, and mean total inaccuracy was 6.7 servings. A significant order-x-sex interaction was found for omission rates, which were lower (better) for boys with reverse-order prompts (53%) than forward-order prompts (62%), but for girls with forward-order prompts (53%) than reverse-order prompts (61%; P < 0.008, mixed-model ANOVA). Intrusion rates were acceptable (defined as ≤ 30%) for slightly more boys with reverse- than forward- order prompts (P = 0.095, McNemar's test).
Study 4 – Baxter et al., 2003b	DRVS to evaluate whether in- person interviews would yield more accurate recalls from children than telephone interviews	69 total: 18 AA m 19 AA f 16 W m 16 W f	2001 – 02 school year; 10 schools	Same-day recall obtained in the evening (average beginning time was between 7:15 p.m. and 7:28 p.m.)	Forward- order prompts	Each child was interviewed once either in person (n = 33) in a research van parked outside of their homes or by telephone (n = 36). Children (and parents) did not know in advance when or how children would be interviewed.	Accuracy did not differ significantly between in- person and telephone recalls (ANOVAs). For in- person recalls and telephone recalls, mean omission rates were 34% and 32%, respectively; mean intrusion rates were 19% and 16%, respectively; and mean total inaccuracy was 4.6 servings and 4.3 servings, respectively.
Study 5 – Baxter et al., 2003a	DRVS to evaluate the effect of interview format (i.e., reporting instructions) on children's dietary recall accuracy	23 total: 10 m 13 f 15 AA 8 W (number for race x sex not mentioned in publication)	2001 – 02 school year; 8 schools	Same-day recall obtained in the evening (after 6:00 p.m.)	Either open format (no prompts) or meal format (meal name prompts)	Each child was interviewed once using either open format (n = 12) or meal format (n = 11). Children did not know that the interview format would vary, or which interview format would be used. Interviews were conducted by telephone.	The number of items observed eaten did not differ by format, but greater numbers of items were reported eaten with meal format than open format (P = 0.029, t test). However, accuracy was better with open format than meal format: Omission rates did not differ between the formats, but intrusion rates (P = 0.036, t test) and total inaccuracy (P = 0.050, t test) were higher (worse) with meal format than open format.
Study 6 – Baxter et al., 2004	DRVS to evaluate the effect of retention interval (i.e., elapsed time between when the meal[s] happen[s] and the recall occurs) on children's accuracy for school meals obtained during 24hDRs	60 total: 30 m 30 f 42 AA 16 W 2 O (number for race x sex not mentioned in publication)	2002 – 03 school year; 6 schools	Prior-24-hour recall or previous-day recall obtained in the morning (after school breakfast), afternoon (after school lunch), or evening (after 6:30 p.m.)	Forward- order prompts	Each child was interviewed once using 1 of the 6 conditions created by crossing 2 target periods with 3 interview times; each condition had 10 children (5 per sex). Morning and afternoon interviews were conducted in person; evening interviews were conducted by telephone. Children did not know that the target period would vary, or which target period would be used.	There was an effect of target period on omission rates (P = 0.006, ANOVA), intrusion rates (P < 0.001, ANOVA), and total inaccuracy (P < 0.001, ANOVA); each was lower (better) by one-third to one- half for prior-24-hour recalls than previous-day recalls (mean omission rates: 47%, 67%; mean intrusion rates: 29%, 54%; mean total inaccuracy: 5.5 servings, 8.6 servings). There was a marginally significant interaction of target period x interview time on omission rates (P = 0.08, ANOVA) which were lowest (best) for prior-24-hour recalls obtained in the afternoon (30%) and highest (worst) for previous-day recalls obtained in the afternoon (73%); although the pattern was similar for intrusion rates, significance was not detected (P = 0.15, ANOVA).
Study 7 – Smith et al., 2007a	NVS to investigate whether being observed eating school breakfast and school lunch influenced children's 24hDRs	120 total:* 39 AA m 45 AA f 18 W m 13 W f 3 O m 2 O f * 60 of these 120 children were from Study 6 because Studies 6 and 7 were conducted at the same time	2002 – 03 school year; 6 schools	Prior-24-hour recall or previous-day recall obtained in the morning (after school breakfast), afternoon (after school lunch), or evening (after 6:30 p.m.)	Forward- order prompts	Each child was interviewed once after being randomly assigned to 1 of the 12 conditions created by crossing 2 target periods, 3 interview times, and 2 observation-status groups (observed or not observed eating school meals in the target period); each condition had 10 children (5 per sex). Interviewers were blind to children's observation status. Morning and afternoon interviews were conducted in person; evening interviews were conducted by telephone. Five response variables were calculated and compared for children observed versus not observed — interview length, meals/snacks reported in the target period, meal components reported for school meals, items reported for school meals, and kilojoules reported for school meals.	The 5 variables were transformed to principal components; 2 were retained (1 – school meal variables; 2 – interview length and number of meals/snacks reported in the target period). Observation status did not affect scores on either retained principal component (P values > 0.6, ANOVAs). Results suggested that observations of school breakfast and school lunch did not affect 4th-grade children's recalls, which in turn suggested, but did not guarantee, that conclusions about recalls by 4th-grade children observed eating school breakfast and school lunch may be generalised to recalls by comparable children not observed.
Study 8 – Baxter et al., 2006a	DRVS to investigate the effects of BMI, sex, and interview protocol on children's accuracy for reporting kilojoules at school meals during 24hDRs	40 total:** 20 low-BMI (≥ 5th and < 50th percentiles; 10 m, 15 AA) 20 high-BMI (≥ 85th percentile; 10 m, 15 AA) **These 40 children had each been interviewed once (about three to four months earlier) for Study 7.	2002 – 03 school year; 6 schools	Prior-24-hour recall obtained in the morning (after school breakfast) or previous-day recall obtained in the evening (between 6:30 p.m. and 9:00 p.m.)	Forward- order prompts	Each child was interviewed once either about the prior 24 hours in the evening, or about the previous day in the morning with 10 low- BMI (5 per sex) and 10 high-BMI(5 per sex) children in each of the 2 protocols. Evening interviews were conducted by telephone and morning interviews were conducted in person. Five kilojoule variables were analysed using separate four-factor (BMI group, sex, race, interview protocol) ANOVAs.	No effects were found for reported kilojoules or matched kilojoules. There were more observed kilojoules (P < 0.02) and omitted kilojoules (P < 0.05) for high-BMI than low-BMI children. For intruded kilojoules, means were smaller (better) for high-BMI girls than high-BMI boys, but larger for low-BMI girls than low-BMI boys (interaction P < 0.04); high-BMI girls intruded the fewest kilojoules while low-BMI girls intruded the most kilojoules. For interview protocol, omitted kilojoules and intruded kilojoules were slightly lower (better), although not significantly so (P values < 0.11), for prior- 24-hour recalls obtained in the evening than previous- day recalls obtained in the morning.
Study 9 – Baxter et al., 2007a	SAS to compare children's recall accuracy for school breakfast versus school lunch obtained during 24hDRs	Secondary analyses of data from Study 1 (for 104 children interviewed 1 to 3 times each [on non- consecutive days] about the previous-day's intake) and data from Study 3 (for 121 children interviewed twice each [once per reporting-order prompt condition; on non-consecutive days] about the previous-day's intake).				For each study, a general linear mixed- model analysis was conducted for each of 9 variables.	For each study, 5 measures were greater for school lunch than school breakfast – number of items observed eaten, observed kilojoules, number of items reported eaten, reported kilojoules, and correspondence rate for kilojoules (all 10 P values < 0.001) – and 3 measures of reporting error were greater for school breakfast than school lunch for each study – omission rate, intrusion rate, and inflation ratio for kilojoules (all 6 P values < 0.005). Total inaccuracy (a measure of reporting error) was greater for lunch than breakfast (P values < 0.001), but this could be due to the significant differences in numbers of observed items and reported items for breakfast and lunch.
Study 10 – Baxter et al., 2007b	SAS to illustrate that conclusions about children's recall accuracy for energy and macro- nutrients over multiple interviews depend on the analytic approach for comparing reported information and reference information	Secondary analyses of data from Study 1 (for 104 children interviewed 1 to 3 times each [on non- consecutive days] about the previous-day's intake).				Analyses compared 2 approaches for analyzing energy and macronutrients in validation studies: conventional (which disregards accuracy of reported items and amounts) and reporting-error-sensitive (which classifies reported items as matches [eaten] or intrusions [not eaten], and amounts as corresponding or over-reported).	With the conventional approach for energy and each macronutrient, report rates did not vary systematically over interviews (all 4 P values > 0.61, mixed-model analysis). However, with the reporting-error-sensitive approach for energy and each macronutrient, correspondence rates increased over interviews (all 4 P values < 0.04, mixed-model analysis), indicating that accuracy improved over time. Furthermore, over interviews, inflation ratios decreased for energy and each macronutrient, although not significantly so, suggesting that accuracy improved over time. Correspondence rates were lower than report rates for energy and each macronutrient.
Study 11 – Baxter et al., 2007c	SAS to use data from a validation study (concerning the effect of order prompts on recall accuracy) to illustrate that conventional energy and macro- nutrient variables misrepresent recall accuracy	Secondary analyses of data from Study 3 (for 121 children interviewed twice each [once per reporting- order prompt condition; on non-consecutive days] about the previous-day's intake).				Conventional approach and reporting-error- sensitive approach as described for Study 10	With the conventional approach, report rates were higher for the 1st than 2nd interview for energy, protein, and carbohydrate (P values ≤ 0.049, mixed models). However, with the reporting- error sensitive approach, correspondence rates were higher for boys with reverse- order prompts but for girls with forward-order prompts (P values ≤ 0.041, mixed models); furthermore, inflation ratios were lower with reverse-order prompts than forward-order prompts for energy, carbohydrate, and fat (P values ≤ 0.045, mixed models). Correspondence rates were lower than report rates for energy and each macronutrient.
Study 12 – Smith et al., 2007b	SAS to use data from a validation study (concerning the effect of interview modality [in person; by telephone] on recall accuracy) to illustrate that conventional energy and macro- nutrient variables mask the complexity of dietary reporting error and overestimate recall accuracy	Secondary analyses of data from Study 4 (for 69 children interviewed in person or by telephone in the evening about that day's intake).				Conventional approach and reporting-error- sensitive approach as described for Study 10	There was no significant effect of interview modality on report rates or correspondence rates for energy or any macronutrient (P values > 0.14, Wilcoxon rank-sum tests), so data from the two modalities were combined. With the conventional approach, median report rates ranged from 76% to 95% for energy and each macronutrient. However, with the reporting- error-sensitive approach, median correspondence rates ranged from 67% to 79%; also, median inflation ratios, which ranged from 7% to 17%, differed significantly from zero (P values < 0.0001, sign tests).

^aDRVS = dietary-reporting validation study

SAS = secondary analyses study

NVS = non-validation study

^bAA = African American

W = White

m = male

f = female

Children (ages nine to ten) were recruited from fourth-grade classes in public elementary schools in one district in a southern state in the USA; schools were selected based on high participation in school breakfast and school lunch. For each school year of data collection, the sex/race composition of children who provided written child assent and parental consent to participate was similar to that of children invited to participate. More children were recruited each school year than required so that final random selection of children from the recruited population could be stratified by race and sex.

An individual child was not interviewed for more than one study with one exception: The 40 children interviewed for Study 8 were a subset of the 120 children who were each interviewed once, approximately three to four months earlier, for Study 7.

School-meal observations

To obtain reference information, dietitians observed randomly selected children eating two consecutive school meals (breakfast and then lunch, or lunch and then breakfast). Because it can be difficult to unobtrusively identify contents of meals brought from home (Simons-Morton et al., 1992), only children who obtained meals provided by school foodservice were observed. Most foods were served to children because “offer-versus-serve” (which allows children to refuse some foods) was not implemented in the district’s elementary schools (US Department of Agriculture & Food and Nutrition Service, 2007). Observations were conducted during usual school meal periods in cafeterias with children seated according to their school’s typical arrangement; at most schools, children sat as they arrived in the cafeteria for breakfast, but with their classes for lunch. Children were observed for their entire meal periods to identify food trades (Baxter, Thompson & Davis, 2001). An observer simultaneously observed one to three children and recorded, for each child, items and amounts eaten in servings of standardised school-meal portions. An observer stood by tables where groups of children sat and appeared to watch the entire group or class; thus, children could see that an observer was present, but children did not know specifically who was being observed or would be interviewed later. Each school year, practice observations were conducted to familiarise children with an observer’s presence (Simons-Morton & Baranowski, 1991). Observers were trained using modeling, practice, and assessment of interobserver reliability. Throughout data collection for Studies 1 and 3 through 8, interobserver reliability was assessed regularly (e.g., weekly) to ensure that information collected did not depend on who conducted observations (Baglio et al., 2004).

Dietary recall interviews

To obtain reported information, each child was interviewed individually (without parental assistance) by a dietitian who usually had not observed that child eating the school meals covered by the recall. As shown in Table 2, children were interviewed in the morning, afternoon, or evening about intake for a specified target period that was the previous day, that same day, or the prior 24 hours. Interviews were conducted in various private locations, audio-recorded, and transcribed. Interviewers followed written, multiple-pass protocols usually patterned after that of the Nutrition Data System for Research (NDSR, Nutrition Coordinating Center, University of Minnesota, Minneapolis, MN, USA) with the following four exceptions or modifications. First, during all interviews, instead of using computerized software, interviewers wrote information reported by children onto paper forms. Second, Study 3 utilised the NDSR protocol with forward-order (morning-to-evening) prompts (Regents of the University of Minnesota) and a protocol that we created based on NDSR but modified to contain reverse-order (evening-to-morning) prompts (Baxter et al., 2003c). Third, Study 5 utilised a protocol with an open format modeled after the automated multiple-pass method of the United States Department of Agriculture (US Department of Agriculture & Agricultural Research Center) and a protocol that we created based on meal name prompts (Baxter et al., 2003a). Fourth, for Studies 6, 7, and 8, children who were interviewed about the prior 24 hours were asked to report intake for the interview day first, and then intake for the previous day, to complete the 24 hours. Interviewers were trained using modeling, practice, and assessment of quality control for interviews. Throughout data collection for Studies 1 and 3 through 8, quality control for interviews was assessed regularly by having at least one interview per interviewer each week, or day, randomly selected by another interviewer and checked for adherence to protocol (Shaffer et al., 2004).

Weight and height measurements

Dietitians used established procedures to measure children’s weight and height (without shoes) on digital scales and portable stadiometers at school on days when no observations or interviews were conducted at that particular school (Lohman, Roche & Martorell, 1988; Maternal and Child Health Bureau). For each child, weight and height were measured twice (back-to-back) by a dietitian; if the two weight or height measurements were not within a tenth of a pound (0.045 kg), or a quarter of an inch (0.64 cm), respectively, then a third weight or height was measured. If three weight and/or height measurements were obtained, then the average of the closest two was used for the child’s weight and/or height. Inter-rater reliability was assessed daily across pairs of dietitians on a random 10% of children; the intraclass correlation reliability was > 0.99 for weight and > 0.99 for height. Each child’s age, sex, height, and weight were used to determine his or her age/sex body mass index (BMI) percentile (Centers for Disease Control and Prevention & US Department of Health and Human Services). For Studies 1 and 3 through 7, children were observed and interviewed irrespective of age/sex BMI percentile. For Study 8, according to the design, only children with an age/sex BMI percentile that we defined either as low BMI (≥ 5^th and < 50^th percentiles) or as high BMI (≥ 85^th percentile) were observed and interviewed (Baxter et al., 2006a).

Classification of observed and/or reported information

Children were to report all meals/snacks eaten during the specified target period, but accuracy was assessed for only the school-meal parts because only those meals were observed. Meals in children’s recalls were treated as referring to school meals if children identified school as the location where meals were eaten, referred to breakfast as breakfast or school breakfast, referred to lunch as lunch or school lunch, and reported mealtimes to within an hour of observed mealtimes. Each item observed eaten for a school meal was classified as a match if it was reported eaten (in any non-zero amount) by the child for that school meal; otherwise, it was classified as an omission. Each item reported eaten for a school meal was classified as a match if it had been observed eaten (in any non-zero amount) by the child at that school meal; otherwise, it was classified as an intrusion. Because children can report foods many ways, items reported eaten were classified as matches unless it was clear that children’s reports did not describe items observed eaten. For example, matches included all kinds of white milk (e.g., whole milk observed, non-fat milk reported) and all types of pizza (e.g., pepperoni pizza observed, cheese pizza reported). Intrusions included fruit juices (e.g., grape juice observed, apple juice reported), milk flavours (e.g., strawberry milk observed, chocolate milk reported), ready-to-eat cereal (e.g., doughnut-shaped cereal observed, flake-shaped cereal reported), and vegetables (e.g., carrots observed, spinach reported).

Amounts observed eaten and amounts reported eaten of standardised school-meal portions were recorded using a qualitative scale and then assigned numeric values as explained in Table 1, footnote a. For each item observed eaten and/or reported eaten, standardised school-meal portions were used to obtain per-serving information about energy and macronutrients (protein, carbohydrate, fat) from the NDSR database; for items not in the NDSR database, energy and macronutrient information was obtained from the school district's nutrition program. Although the portion-size estimates may have been imprecise, the same approach was used to estimate energy and macronutrients for observed items and for reported items.

Analytic variables

Food-item variables (calculated per interview, except for Study 9 when breakfast and lunch were calculated separately) included number of items observed eaten, number of items reported eaten, omission rate, intrusion rate, and total inaccuracy, as illustrated and defined in Table 1. These food-item variables were calculated after assigning a weight to each item according to meal component, with combination entrée = 2.0, condiment = 0.33, and each remaining meal component = 1.0, as explained in Table 1, footnote b. Lower values for omission rate, intrusion rate, and total inaccuracy indicate better accuracy.

For Study 8 only, kilojoule variables (calculated per interview) included observed kilojoules, reported kilojoules, matched kilojoules, omitted kilojoules, and intruded kilojoules, as illustrated and defined in Table 1. Higher values for matched kilojoules, and lower values for omitted kilojoules and intruded kilojoules, indicate better accuracy.

For Studies 9 through 12, variables for energy and each macronutrient (calculated per interview, except for Study 9 when breakfast and lunch were calculated separately) included observed amount, reported amount, report rate, correspondence rate, and inflation ratio, as illustrated and defined in Table 1. Higher values for correspondence rate, and lower values for inflation ratio, indicate better accuracy. Conventional interpretation of report rates is that values close to 100%, greater than 100%, and less than 100% indicate better accuracy, over-reporting, and under-reporting, respectively; however, as results for Studies 10 through 12 show, conventional report rates overestimate reporting accuracy.

Summary of Results with Discussion

Consistency of children’s recall accuracy

Children’s accuracy improved slightly from the first to the third recall (P = 0.006) in Study 1 (Baxter, Thompson, Litaker, Frye & Guinn, 2002). However, intraclass correlation coefficients were low, indicating that individual children’s recall accuracy was inconsistent from one interview to the next.

Forward- versus reverse-order prompts

Boys recalled items more accurately when prompted to report meals/snacks in reverse order in Study 3 (Baxter et al., 2003c), but girls recalled items more accurately when prompted to report meals/snacks in forward order. However, overall accuracy was poor.

Interview modality (in person versus by telephone)

Children’s recall accuracy did not differ significantly between in-person versus telephone recalls in Study 4 (Baxter et al., 2003b). This is important because with telephone interviews, there are potential savings in travel time (for subjects and investigators) and transportation costs. Also, this result allows for both interview modalities to be utilised in a single study without compromising the ability to compare children’s recall accuracy (e.g., compare in-person recalls conducted at school in the morning or afternoon with telephone recalls conducted in the evening).

Meal name prompts

Although more items were reported eaten with meal format (i.e., meal name prompts) than open format (i.e., no prompts) in Study 5 (Baxter et al., 2003a), accuracy was better with open format than meal format for two measures — intrusion rates and total inaccuracy. At least one item was intruded by one-third of the children interviewed with open format, but by more than four-fifths of the children interviewed with meal format.

Retention interval

The retention interval (i.e., elapsed time between when the meal[s] happen[s] and the recall occurs) is defined by the combination of target period and interview time. Two possible target periods for a 24hDR are the prior 24 hours (24 hours immediately preceding the interview) and the previous day (midnight to midnight of the day before the interview). The interview time may be any time of day (e.g., morning, afternoon, evening) for each of these target periods.

The importance of retention interval on children’s dietary recall accuracy is evident across and within studies summarised in this article. Accuracy was better for same-day recalls obtained in the evening in Study 4 (Baxter et al., 2003b) than for previous-day recalls obtained in the morning in Studies 1 and 3 (Baxter et al., 2002, 2003c). Accuracy was better for prior-24-hour recalls than previous-day recalls in Study 6 (Baxter et al., 2004). Accuracy was slightly better for prior-24-hour recalls obtained in the evening than previous-day recalls obtained in the morning in Study 8 (Baxter et al., 2006a). Results concerning retention interval will soon be available from a large dietary-reporting validation study with children (similar in design to Study 6).

Recall accuracy for school breakfast versus school lunch

Accuracy was worse for school breakfast intake than school lunch intake in secondary analyses in Study 9 (Baxter, Royer, Hardin, Guinn & Smith, 2007a). These results have implications when dietary recalls are obtained from children in studies to determine the effects of school breakfast on children’s nutritional adequacy, body weight, and cognitive and academic performance.

Children’s BMI and dietary recall accuracy

Children’s dietary recall accuracy was related to their age/sex BMI percentile in Studies 2 and 8. High-BMI children (≥ 85^th percentile) omitted more kilojoules than low-BMI children (≥ 5^th and < 50^th percentiles) in Study 8 (Baxter et al., 2006a). Also, high-BMI girls intruded fewer kilojoules than high-BMI boys, but low-BMI girls intruded more kilojoules than low-BMI boys.

Over three recalls per child, accuracy for items varied according to children’s age/sex BMI category in secondary analyses in Study 2 (Baxter, Smith, Nichols, Guinn & Hardin, 2006b) – accuracy improved for healthy-weight children, improved and then stabilised for children at risk of overweight, but deteriorated and then stabilised for overweight children. Considering the current attention given to the increased prevalence of childhood obesity, these results are especially pertinent. Also, many studies require individual children to complete multiple 24hDRs (e.g. to assess the relative validity of FFQs or to evaluate the effectiveness of nutrition interventions); results from Study 2 suggest that for studies in which multiple 24hDRs are obtained from children, it should not be assumed that recall accuracy is invariant over trials and independent of BMI category.

Conventional versus reporting-error-sensitive analytic methods

Studies 10, 11, and 12 (Baxter, Smith, Hardin & Nichols, 2007b, 2007c; Smith, Baxter, Hardin & Nichols, 2007b) consisted of secondary analyses to compare two approaches of analysing energy and macronutrients in dietary-reporting validation studies. The conventional approach disregards accuracy of reported items and reported amounts by transforming reference information, and reported information, to kilojoules and macronutrients for each subject, and then calculating report rate for energy and each macronutrient. The reporting-error-sensitive approach classifies reported items as matches or intrusions, and reported amounts as corresponding, over-reported, or under-reported, before calculating correspondence rate and inflation ratio for energy and each macronutrient. As shown in Table 1, the conventional report rate is a sum of the correspondence rate (a genuine measure of accuracy) and the inflation ratio (a measure of error).

In secondary analyses in Studies 10, 11, and 12 (Baxter et al., 2007b, 2007c; Smith et al., 2007b), conventional report rates for energy and each macronutrient overestimated reporting accuracy and masked the complexity of reporting errors. Specifically, report rates were higher than correspondence rates, indicating that reporting accuracy was overestimated by conventional report rates. Furthermore, conventional report rates did not detect improvement over multiple recalls that were evident with the reporting-error-sensitive variables of correspondence rates and inflation ratios in Study 10 (Baxter et al., 2007b), nor did conventional report rates detect sex differences that were evident with correspondence rates and inflation ratios in Study 11 (Baxter et al., 2007c).

Using school-meal observations to validate children’s dietary recall accuracy

Throughout data collection for each validation study summarised in this article, assessment of interobserver reliability demonstrated adequate agreement between observers. Thus, research has established the use of school-meal observations to validate children’s dietary recall accuracy.

Furthermore, Study 7 (Smith et al., 2007a) supports the generalisability of using school-meal observations as the validation method to assess children’s dietary recall accuracy. Specifically, results from Study 7 suggested that school-meal observations did not affect children’s dietary recalls. These results suggest, but do not guarantee (because the small sample prohibited equivalence testing), that conclusions about dietary recalls by children observed eating school meals in validation studies may be generalised to dietary recalls by comparable but unobserved children in non-validation studies. This is important because the goal of validation studies is to generalise conclusions to subjects in non-validation studies (such as national surveys, epidemiologic studies, and nutrition interventions) for whom reference information is not collected. Results will soon be available from equivalence testing in a large trial (similar in design to Study 7) that investigated whether school-meal observations influenced children’s 24hDRs.

Recommendations

Methodological validation studies summarised in this article provide insight into errors in children’s dietary recalls. However, research gaps remain.

First, validation studies are needed concerning the accuracy of information about children’s dietary intake when information is obtained from child-only recalls, parent-only recalls, and joint parent-child recalls. A 1989 study by Eck and colleagues (Eck et al., 1989) found that consensus recalls provided by mother, father, and child yielded better estimates of observed intake of a single cafeteria meal by children ages four to ten years than did recalls from either the mother or father alone. Unfortunately, the study by Eck and colleagues is often incorrectly cited as a rationale for obtaining joint 24hDRs (of children’s intake) from parents and children because for that study, children by themselves did not provide recalls, so no comparison could be made of the accuracy of child-only recalls, parent-only recalls, and joint parent-child recalls. Furthermore, because consensus recalls were always obtained after the mother and father had each provided separate recalls, the back-to-back recalls could have altered accuracy during the second recall, which was always the consensus parent-child recall.

Second, validation studies are needed concerning the combined effect on children’s dietary recall accuracy of retention interval, reporting-order prompts (reverse, forward), and interview format (e.g., meal, open). An appropriate validation method, such as direct observation, should be used to provide information about actual items and amounts eaten. Relative validation studies may be misleading because they compare information from two methods that both rely on self-reports provided by subjects; thus, actual intake is unknown.

Third, validation studies are needed concerning the consistency of children’s dietary recall accuracy. Despite the fact that multiple 24hDRs often are obtained from individual subjects, Study 1 (Baxter et al., 2002) appears to be the only study published to date that has investigated the consistency of children’s dietary recall accuracy.

Fourth, methodological validation studies are needed concerning potential correlates of children’s dietary recall accuracy. Such correlates include children’s social desirability, self-esteem, body image, BMI, sex, race/ethnicity, age, and memory/cognitive ability.

Fifth, validation studies should include both omission rates and intrusion rates when assessing accuracy for recalling food items because these rates characterise different aspects of reporting accuracy (Smith, 1991; Smith et al., 1991). Furthermore, when assessing accuracy for energy and nutrients, a reporting-error-sensitive analytic approach is recommended because conventional report rates fail to provide insight concerning whether reporting errors in dietary recalls are due to items that are intruded or omitted, or to amounts of matches that are under-reported or over-reported (Baxter et al., 2007b, 2007c; Smith et al., 2007b).

In conclusion, when designing studies that will utilise dietary recalls obtained from children, to improve recall accuracy, investigators should make deliberate decisions to minimise the retention interval between intake and report. For prior-24-hour recalls, the end of the 24 hours coincides with the beginning of the interview, and no meals intervene between the to-be-reported meals and the interview; however, for previous-day recalls, as the interview is held later in the day, the length of the retention interval increases, as does the number of intervening meals. Furthermore, because of the profound influence of retention interval on recall accuracy, publications of studies (validation and non-validation) that utilise 24hDRs should specify details concerning target period and interview time. Finally, publications of studies (validation and non-validation) that utilise children’s 24hDRs should clearly indicate whether parents helped children during recalls because when parents help children with recalls, it is impossible to determine whether recall errors are related to children’s characteristics versus parents’ characteristics.