FOURTH GRADERS’ REPORTS OF FRUIT AND VEGETABLE INTAKE AT SCHOOL LUNCH: DOES TREATMENT ASSIGNMENT AFFECT ACCURACY?

Abstract

Objective

Dietary interventions with children often use self-reported data to assess efficacy despite that objective methods rarely support self-report findings in validation studies. This study compared fourth graders’ self-reported to observed lunch fruit and vegetable intake to determine if the accuracy of self-reported intake varied by treatment condition

Design

Matched randomized follow-up design examined three treatment groups (high and low intensity interventions and control) post-intervention.

Subjects/Setting

379 middle-school children participating in a randomized controlled trial of a school-based fruit and vegetable intervention were observed during school lunch one day and asked to recall intake the following day.

Main Outcome Measures

Food items were coded as: “match,” “omission,” or “intrusion.” Students were classified as “accurate” if all food items matched, otherwise “inaccurate.” Matched foods’ portions were compared for accuracy. Servings were computed for total fruit and vegetable intake.

Analyses

Accuracy for fruits and vegetables were compared in separate analyses and tested for multiple potential associates: treatment condition, gender, race, BMI, subsidized meal eligibility, school district, fruit/vegetable availability, age and test scores. Fitted multivariable regression models included variables found to be significant in univariate or chi square analyses.

Results

Variables found to be significant for fruit item accuracy were availability at lunch, BMI, and subsidized lunch eligibility. For vegetable item accuracy, availability at lunch was significant. No differences were found for food portions or for efficacy of the intervention between the two methods of dietary data collection: observation and self-report.

Conclusion

Condition assignment did not bias recalled fruit and vegetable intakes among fourth graders.

Introduction

One goal of health promotion is to identify efficacious methods to advance healthy lifestyles through behavior change. Precursors to this activity are the identification of behaviors that improve health and the populations that would most benefit from them. One behavior that has been identified is eating five or more servings of fruits and vegetables daily (1, 2). As American children’s eating habits fall well below this level (3, 4) and these habits usually track into adulthood (5), this population has been targeted for dietary intervention.

Accurate assessment of dietary intake is important for understanding diet-disease relationships (6), policy decisions, and tracking progress towards dietary goals (7); it is also essential for the proper evaluation of dietary change interventions (8). Many methods are available to measure dietary intake but none are without error (9). Although objective methods, such as biological measures and observation, are available, they are costly and difficult to implement. This has resulted in dietary data being most frequently collected via self report. This method is relatively inexpensive, easy, and can gather information otherwise unavailable; however, it is particularly susceptible to social desirability (10) and memory fault (11) effects. These measurement issues are more problematic with children given the added variability in cognitive development and possible limited food experience and vocabulary (11).

In a study of intervention effectiveness, randomly distributed error would occur equally across treatment conditions, making comparisons across conditions or for longitudinal change reasonable for examining outcomes.. However, should one of these sources of error be systematic due to an effect related to treatment, the validity of findings would be compromised. In particular, knowledge about an intervention may boost social desirability responses in the treatment condition more than in the control group, biasing the study’s results toward finding the intervention efficacious. Of eleven studies reporting on interventions with middle school children that examined fruit and vegetable intake (12–22), only two used an objective measure to validate self-reported findings (13, 23) and neither found support for the efficacy that was based on self-reported intake. While limited, these findings suggest that the accuracy of self-reported intake may be different by intervention condition, perhaps as a result of social desirability, social approval, or recall bias (10, 24, 25).

Certain factors have been found more prevalent among participants who were less accurate in reporting dietary intake. Inaccurate reporting by gender (24) Body Mass Index (BMI) (26–29), race (30), age (31), social desirability (25, 31, 32) or type of food (33, 34) has been found in some studies, but not all (35, 36). Under-reporting of intake has been found among children with higher BMIs (37–39). Females have been found prone to socially desirable responses more than males (25, 40). Black students were found less accurate in reporting intake than White students (30).

Although no true “gold standard” for assessing children’s intake has been established (41), direct observation of eating has been referenced as such (42–44). Observation is good for assessing food intake because it assesses foods rather than nutrients, which is particularly important when appraising health promotion messages that are food specific. Further, observation is relatively valid (43) and does not rely on the participant’s memory (45). It is not without error potential, however, as participants may react to being observed by changing behavior (46, 47), and observers can make errors; however, observer validity(48) and reliability can be checked and confirmed (49).

The primary purpose of this study was to use observed intake as a “gold standard” to examine if fourth graders’ accuracy in self-reported fruit and vegetable intake at lunch varied by treatment condition, after a fruit and vegetable promotion intervention. Additional goals were to examine if reporting bias occurred among sub-groups within the sample and to compare the study results regarding efficacy of the intervention based on lunch data from each of the two dietary data collection methods. It was hypothesized that children with more intervention exposure would be more prone to social desirability and therefore, less accurate in their reports of fruit and vegetable intake, and likely to over-estimate amounts eaten, than those with less exposure. This was expected to result in differences in findings for program efficacy between methods.

Methods

Sample

This study used a sample of 396 fourth-grade students randomly selected from the 1785 students participating in Hi5+, a fruit and vegetable promotion program implemented in 2002–2004 as part of the National Cancer Institute’s 5-A-Day for Better Health initiative (50). A total of 33 schools participated, drawn from four school districts characterized as urban, suburban, rural and parochial. Schools were matched into groups of three based on SES variables and then randomized so that each group had one school assigned to each of three treatment conditions: Control (C), School only (S), and School plus Home (S+). The primary aim of the Hi5+ intervention was to increase fruit and vegetable consumption of children enrolled in both the low (S) and high intensity (S+) intervention conditions. Both interventions included 14 twice-weekly half hour lessons; the S+ group received an additional seven family sessions at home. All lessons were based on Social Cognitive Theory (51) constructs and included taste-testing of fruits and vegetables. Details of the intervention have been reported (52, 53). Students in this post-intervention study were distributed across conditions with 121 in the Control, 118 in the School intervention, and, 140 in the School plus home intervention. This sample had dietary intake data measured by observation and recall for the same day. Figure 1 represents this study. This study was approved by the University of Alabama at Birmingham’s Institutional Review Board for Human Use.

Figure 1.

Overview of the Hi+ Study Design

Data Collection

Dietary data were collected by two methods, observation and self-report, to provide lunch fruit and vegetable intake information (both food items and portions) in this study. The “gold standard” observation data were used to assess how accurate self-reported intakes were for each treatment condition and how the conditions differed in their reporting accuracy. Three discrete study teams collected data. Team One collected demographic data and taught the children how to describe food portions using a 2-dimensional poster illustrating varying food shapes and sizes. Two weeks later, Team Two observed children eating lunch in the school cafeteria, following standardized “plate-waste” protocols. The day after the lunch observation, Team Three collected height, weight, and 24-hour recall data from children in individual face-to-face interviews.

The cafeteria observation team members trained on observation and estimation skills for 20 hours prior to data collection and bi-weekly skills practice and monitoring during the study. This maximized observer validity and reliability. Overall, validity for food items was 100%, while visual estimations of portions averaged 85% and inter-rater agreement averaged 83%, rates which are considered acceptable (49). Immediately prior to a lunch observation, foods being served were measured for average portion sizes and purchased to facilitate accurate descriptions during data entry. Students to be observed were randomly selected based on a seating chart. All students wore colorful wrist bands with specific colors identifying participating students to the staff. Each staff observed up to 3 students from 7 to 12 feet away, as unobtrusively as possible. Data forms and purchased food trays were returned to the study office for data entry into the Nutritional Data System for Research ([NDS-R®] version 2.4; Nutrition Coordinating Center, University of Minnesota, Minneapolis, MN) software system.

The 24-hour recall team members participated in a three-day training and certification process to acquire skills for interviewing and data entry into NDS-R® on laptop computers. The protocol used was similar to the one recommended by the Nutrition Coordinating Center at the University of Minnesota, creators of NDS-R®. This recall method requires three steps: 1) obtain a quick list of foods eaten and when; 2) obtain details about foods; and, 3) review all information. To maximize the validity of how foods were entered into the data system, interviewers had a list of “easy codes” for foods available in the cafeteria the previous day; this list was prepared by the observation data entry staff. In NDS-R®, any one food may have multiple codes based on preparation method and how it is served, and children would not be likely to know preparation details. As foods were the variables of interest, it was important that observed and reported foods were entered into and coded by NDS-R® in the same manner. Prior to the interview, in an attempt to minimize social-desirability bias, students were told that our purpose was to learn what typical fourth graders ate and to be honest and careful in reporting intake. As a memory prompt, when a child indicated no or very limited memory of foods (e.g., drink or dessert only), the interviewer would read the entire list of the foods served at the previous day’s lunch.

Barefoot heights, assessed to 1/8″, and weights, assessed to .2 pounds, were measured once by the interview team using a stadiometer (Seca Road Rod 214, Seca Corporation, Hanover, Maryland), and digital scale (Health-o-meter, Inc., Bridgeview, Illinois), respectively. BMI was calculated using the standard formula and rankings calculated by EpiInfo (Version 3.4, Centers for Disease Control, Atlanta, GA) software.

Data quality was monitored throughout the study. Data entered into NDS-R® were reviewed by the study’s registered dietitian daily. As well, recall interviews were periodically tape-recorded and reviewed with data re-entered for comparison by the study dietitian.

Additional data regarding possible influences on dietary intake were collected. Group level standardized test scores and subsidized lunch eligibility were provided by school administration. Study staff made seven unannounced lunchtime visits to each school cafeteria to collect fruit and vegetable availability data.

NDS-R® software provides nutrient and gram weight data rather than foods as output. Therefore, it was necessary to identify fruit and vegetable food codes to extract information on foods rather than nutrients. While 5-A-Day criteria were used to identify what counted as fruits and vegetables, serving determinations were based on the Food and Drug Administration criteria (54) used in NDS-R®. Observation data were compared to recall data to determine foods that matched (observed and eaten, or none observed and none eaten), intruded (not observed, but recalled) or were omitted (observed, but not recalled). Children were classified as “accurate” if all foods matched or “inaccurate” if any foods did not match. For foods that matched, serving portions were then compared for match, over-report or under-report. Borrowing from a previous definition of “precise estimate” as being within 10% of the true estimate 95% of the time (55), portions were accepted as matched if the recalled portion was within 10% of the observed portion.

Analyses

Post hoc power was calculated using the sample size and differences found among groups assuming a Type I error rate of ≤0.05. Power was found to be 0.96 for fruit accuracy; however, for vegetable accuracy power was only 0.1 due to the small differences in means among conditions. Outcomes analyses were performed using SAS (version 9.0, SAS Institute, Inc., Cary, NC). A p-value of 0.05 or less was considered significant. Descriptive and potentially confounding variables were examined for differences among conditions using chi square tests for categorical variables, the Wilcoxon test for variables with restricted variation, and ANOVA for continuous variables. Logistic regression was used to compare accuracy of food items among conditions. Analyses of fruit accuracy and vegetable accuracy were run separately. Univariate analyses were used to screen variables for inclusion in the multivariable models. Goodness-of-fit statistics determined model fit. Kruskal-Wallis tests determined intervention efficacy for combined fruit and vegetable servings using observation data, recall data, and the discrepancies between the two. Additionally, fruit servings and vegetable servings were examined separately.

Results

For this study, data were collected from 379 fourth grade students in 33 schools. Individual participant attributes, specifically age, BMI, race, and gender, were similar across conditions. Rates of study participation, as measured by classroom consent rates also were similar among conditions. Among school/class level variables, differences were seen for subsidized lunch eligibility, standardized test scores, and the availability of fruits and vegetables during school lunch. These school/class level variables were highly correlated with each other and with school district, suggesting these differences were due to the unequal distribution of schools from each district in each condition. Table 1 provides details for these data.

TABLE 1.

Differences among Treatment Groups in Demographics, Consent Rate, SAT Scores, and Fruit and Vegetable Lunch Availability

	School plusHome Intervention	School only Intervention	Control	All participants	Test statistic	p
Number in sample	140	118	121	379
Race
% Black	36.4%	42.4%	40.5%	39.6%	χ2=1.684	0.794
% White	59.3%	55.1%	57.0%	57.3%
% Other	4.3%	2.5%	2.5%	3.2%
Gender
% Male	49.3%	50.8%	51.2%	50.4%	χ2=0.113	0.945
% Female	50.7%	49.2%	48.8%	49.6%
Body Mass Index
Mean Percentile rank	72.9	70.8	77.0	73.6	F=1.757	0.174
Age in months
Mean	121.5	120.9	121.4	121.3	F=0.572	0.565
School District
% Urban	20.0%	18.6%	29.8%	22.7%	χ2=38.79	<.001
% Parochial	15.7%	10.2%	10.7%	12.4%
% Suburban	52.1%	32.2%	25.6%	37.5%
% Rural	12.1%	39.0%	33.9%	27.4%
Subsidized Luncha
Mean % eligible	44.6%	43.8%	48.3%	45.5%	χ2=8.761	0.013
Stanford Achievement Test (SAT)a
Mean score 3rd gradeb	55.01	55.83	55.46	55.04	χ2=4.66	0.098
Consenta
Mean Consent Rate	68.9%	66.4%	67.7%	67.9%	F=2.30	0.102
Fruit and Vegetable availabilitya
Mean Fruits servedc	1.50	1.28	1.33	1.38	χ2=5.253	0.072
Mean Vegetables servedc	2.09	2.06	2.77	2.30	χ2=11.894	0.003

^abased on school or classroom clustered statistics

^bTest is given to all 3-graders in Alabama Schools; this is a nationally normed test with scores ranging from 1–99

^cservings based on 5-a-day inclusion criteria using FDA serving sizes; mean number of servings based on seven days offerings

Accuracy rates for fruit items and vegetable items were examined individually, with no differences in accuracy found among conditions. Factors found positively associated with fruit item accuracy were higher BMI, increased fruit availability at lunch, and subsidized lunch eligibility of less than 50%. The only factor associated with better vegetable item accuracy was fewer vegetables available at lunch. (see Table 2)

TABLE 2.

Factors Associated with Accuracy of Fruit Items and Vegetable Items based on Recall compared to Observation

	Fruit Item Accuracy		Vegetable Item Accuracy
	OR (CI)	p-value	OR (CI)	p-value
Conditions –
Control vs. School Plus Home	1.80 (0.94–3.45)	0.093	0.91 (0.55–1.52)	0.2777
School vs. School Plus Home	0.92 (0.50–1.67)		1.39 (0.62–0.86)
Race – Black vs. White		ns		na
Gender – Male vs. Female		na		ns
Body Mass Index percentile	1.01 (1.00–1.02)	0.036		ns
Age		na		ns
School District		na		ns
Subsidized lunch eligibility <50% vs. ≥50%	2.98 (1.79–4.99)	<0.0001		na
Stanford Achievement Test scores		ns		na
Fruit or Vegetable Availability at lunch	0.60 (0.41–0.88)	0.009	0.73 (0.62–0.86)	0.0002

^*variables entered into multivariable model if p ≤0.10 in univariate analysis; forward selection

na=not entered into multivariable model due to non-significance in univariate models

ns=non-significant in multivariable model

For fruit and vegetable food items that matched, portion accuracy was examined by chi square tests comparing “Accurate” vs. “Over-report” between Intervention (S+ and S) and Control (C) conditions. These tests indicated no differences between groups for fruit portion accuracy (Fisher’s Exact Test χ²=0.929; p=0.489) or vegetable portion accuracy (χ²=1.205; p=0.272). (see Table 3)

TABLE 3.

Comparison among Conditions of Accuracy of Fruit and Vegetable Food Portions where Observed and Recalled Foods Matched

	School plus Home Intervention	School Only Intervention	Control	All Participants	p-value
Fruits (n=92)
Under-report %	25.0	23.3	14.3	22.8
Accurate (Matcha) %	47.9	56.7	71.4	54.4	0.627b
Over-report %	27.1	20.0	14.3	22.8	0.489c
Vegetables (n=230)
Under-report %	26.5	19.0	21.4	22.6
Accurate (Matcha) %	43.4	41.4	37.1	40.4	0.547b
Over-report %	30.1	39.7	41.6	37.0	0.272d

^arecall within 10% of observed

^bCochran-Mantel-Haenszel General Association statistic for 3 conditions by 3accuracy categories

^cFisher’s exact test for S+ and S vs. C for over-reporting vs. accurate (n=71)

^dChi square S+ and S vs. C for over-reporting vs. accurate (n=178)

Differences in total servings of fruits and vegetables at lunch were analyzed among conditions using three different data: 1) using the observation data; 2) using the recall data; and, 3) using the discrepancy between the two measures (e.g., servings from recall data minus servings from observation data). Results indicated no difference between conditions for total fruit and vegetable servings, regardless of assessment method used. However, analyses of fruit servings and vegetable servings, separately, revealed statistically significant differences among conditions for fruit servings for both recall and observation data and significant differences for vegetables only in the recall data. In general, recalls produced larger serving intakes than observation, especially for vegetables. Table 4 displays these data.

TABLE 4.

Efficacy of Intervention to Promote Fruit and Vegetable Consumption based on Servings Eaten at School Lunch—as determined by two methods of data collection: Observation and Recall

	School plus Home Intervention	School only Intervention	Control	χ2a	p-value
Mean (SD)
Fruit Servingsb
Observation	0.29 (.45)	0.24 (.36)	0.13 (.27)	12.39	0.0020
Recall	0.29 (.42)	0.26 (.37)	0.09 (.23)	31.29	<0.0001
Vegetable Servingsb
Observation	0.48 (.63)	0.46 (.69)	0.54 (.54)	5.85	0.0535
Recall	0.55 (.62)	0.60 (.83)	0.71 (.76)	6.31	0.0427
Fruit & Vegetable Servingsb
Observation	0.76 (.75)	0.70 (.74)	0.67 (.63)	0.81	0.6654
Recall	0.84 (.71)	0.86 (.86)	0.80 (.82)	0.85	0.6539

^aKruskal-Wallis test

^bservings are mean for group based on 5-a-day inclusion criteria using FDA serving sizes

Discussion

This study is one of the few that have examined outcomes of a children’s dietary intervention with two measures (56), objective and self-report, and the only, to our knowledge, in which the objective data supported the self-report data for program efficacy. As with other studies examining the validity of self-reported intake, misreporting was common. In this study, this was particularly true for vegetable items and portions, possibly reflecting characteristics related to this food group or to the cognitive development of the age group examined.

Accuracy Findings Compared to Other Studies

Variations in the methods used to examine the accuracy of children’s reported intake, including the observational unit, weighting foods differently and statistical methods, hamper comparisons across studies. Nevertheless, some consistencies are found among validity studies of middle school-aged children’s self-reported diet. For the purpose of comparing this study’s findings to others, results will be discussed at the more commonly-used match, omission and intrusion level rather than accuracy category used in these analyses. Unlike accuracy categorization, a child may fall into more than one category (e.g., match and omission = two foods eaten, one reported and one omitted). Others have consistently found that children omit reporting foods they’ve eaten more often than they intrude foods they have not eaten (14, 23, 35, 36, 42, 57–59), with omissions for fruits and vegetables generally found to be higher than for all foods. Where rates were reported, omissions were 34% (35), 37% (36) and 67% (42) for fruits and 43% (35) 60% (36), and 69% (42) for vegetables. These rates of omissions are much higher than those found in this study (12.7% for fruit and 20.3% for vegetables). There are several possible reasons for higher rates in other studies: 1) other studies combined food item and portion (e.g., reporting ½ cup of grapes instead of 1 cup observed would be omitting 1 fruit); 2) different criteria were used to define fruits or vegetables (e.g., other studies included French fries); and, 3) differences in prompting methods used in the recall interviews.

The prompting method used in this study may have helped the child remember foods eaten, reducing omissions. Reading the list of all available foods, including all fruits and vegetables, when participants could not remember what they ate likely elicited memories of foods eaten more than the more general prompt, “What else did you have?” Baxter and colleagues suggest the type of prompting used can affect memory retrieval (35). While possibly reducing omissions, this method did not appear to affect intrusions as rates in this study were similar to those in other studies. Intrusions in this study were 15% for fruits and 32% for vegetables, whereas other studies have reported 14% (35), 20% (36) and 44% (42) for fruits, and 14% (35), 16% (36) and 47% (42) for vegetables.

This study required foods to match exactly and did not allow similar foods (e.g., peas and green beans) to be considered a match (57); therefore, it might be expected that lower rates of matching would be seen. However, the match rates (78% fruit and 73% vegetables) in this study were similar to, or higher than, those found in other studies for all foods. While other investigators have reported on matching of food items or food portions or some combination of item/portion, none have described including those not eating and not reporting any fruits or vegetables as a match (“zero matches”). These studies addressed food rather than child as the unit of accuracy; our interest was in the accuracy of the child as the source of error arises from the child and not the food. It is possible that the high match rates found in this study are due to the inclusion of “zero matches” which likely inflated the match rates as these accounted for between 30.6% and 77.7% of food matches. Removal of these from the overall matches would change both the denominator and the numerator, probably reducing the match percentages. Additionally, using pre-determined “easy codes” for entering foods the children reported probably facilitated the exact food code match requirement of this study.

Rates for portion matching were 54% for fruits and 40% for vegetables, which are very similar to the 49% (fruits) and 42% (vegetables) found in the Pathways study with 8–10 year olds (23). Remembering food items appears to be a lesser source of error for children this age than portion estimation, which others have reported (36). Our study found higher match rates on portions for fruits than vegetables, doubtless reflecting the frequent consumption of whole fruits which are easier to estimate. Vegetable portions, with their frequently amorphous shapes, may require a more advanced cognitive process to accurately describe them using the conservation of mass principle (60).

Accuracy among Conditions

Accuracy for food items and accuracy for food portions were not found to be different among conditions. This suggests that social desirability demand on participants was equally extant in all conditions, and assessments were not perceived as related to intervention messages as might be expected in the intervention conditions. Although not significant, a trend to over-reporting food portions was noted in the groups who were observed to eat more of those foods (fruits in S+ and vegetables in C) perhaps reflecting a tendency to over-report foods eaten more often. This would be consistent with dietary intrusions for adults which were found to be from the “usual diet” rather than from a specific day (61).

Factors Associated with Accuracy

It has been suggested that self-reported intake bias may be attributed to attributes of the reporter (62). Investigators have reported finding significant misreporting by females (24) and Blacks vs. Whites (30) while others report none (35, 36). This study did not find any patterns of misreporting by gender or race. An increased likelihood of under-reporting intake as weight increased has been reported (26–29). It has been thought that overweight children are particularly sensitive to social desirability (63). In this study, more accurate reporting rather than under-reporting was found. Possibly as fruits and vegetables are viewed as “good” foods (64), social desirability would dictate to report them. Further, the socially desirable behavior may have been conceived as “accurate reporting” which was stressed during the interview introduction.

The differences in fruit intake and vegetable intake among conditions discussed in the previous sections may be due to availability at lunch. Consistent with others’ reports (65), this study found that where more food items were available, more were eaten: S+ for fruit and C for vegetables. Interestingly, as the number of food items available increased, intake increased but accuracy decreased, as did the number of “zero matches.” For each additional food item that was available, the likelihood of accurate reporting was decreased by 40% for fruits and 27% for vegetables. It may be harder to accurately report what was eaten than what wasn’t. Remembering is a complicated process requiring several types of memory. For some children, perhaps having seen a greater variety of fruits or vegetables in the lunchroom resulted in a misinformation effect (66); that is, observing a food being served or on their lunch tray may have constructed a false memory of eating the food.

Eligibility for subsidized school meals was found to differ among conditions and was independently and negatively associated with fruit accuracy. In schools with less than half the students entitled to subsidized meals, students were three times more likely to be accurate in their reporting than those with over half of students eligible for this program. It is unclear how this factor affected accuracy, but this was a school-level variable and imprecise; a future study is needed to test this relationship where subsidized meal status is measured individually.

The efficacy of the intervention was examined by comparing differences among the three treatment conditions for overall fruit and vegetable servings. Although it was expected that study results for fruit and vegetable servings would differ by observation and recall, no significant differences were seen among conditions. Post hoc analyses were run separately for fruit servings and vegetable servings. Fruit servings differed statistically with both measures indicating that the two intervention conditions had more fruit servings than the control condition. For vegetable servings, the two measures produced different findings; the recall method produced a significant difference; whereas, the observation method found only a trend. Vegetables servings were greater in the control condition than either intervention condition. When combined, the opposite directions of intakes for fruits and vegetables cancelled out the differences seen among conditions when examined individually.

In summary, the findings from this study indicate that food availability and other factors are more important than intervention messages in influencing accurate fruit and vegetable recall from middle-school aged children. Influences were found to be food specific, as BMI, subsidized lunch eligibility, and availability were associated with fruit accuracy, while only availability was associated with vegetable accuracy. Children may think of fruits and vegetables differently and their ability to report them accurately in this study reflected this, with fruits being more accurately reported than vegetables. Finally, self-reported intake generally inflated observed intake consistently across conditions.

Limitations

There are several limitations to this study. Related to study design, pre-intervention comparability among treatment groups was not tested. Matching schools and random assignment to condition should have ameliorated the effect of this limitation. Also, some data were available only at the school/class level which reduced the power and precision to examine some associations sufficiently. Another limitation is the possible error in height and weight measurements as these measures were only taken once and instruments were portable. Limited statistical power for some measures was a result of this study being a post hoc examination of data collected for another purpose with a pre-determined sample size. Finally, the findings of this study may not be generalizable to other populations.

Conclusions

Four messages for practice arise from this research. First, methods for collecting dietary data from children should enhance memory, but limit cuing, to minimize recall and social desirability biases. Second, although only moderately representing “true” intake, self-reported fruits and vegetables intakes appear to be sufficiently representative for examining the efficacy of an intervention. Third, including a more objective measure to assess validity of self-reported data can reinforce confidence in a study’s findings, even if only collected from a representative sub-sample. Fourth, fruits and vegetables, although often promoted together, are different foods and probably conceptualized differently, which has implications for both intervention messages and data collection methods.

Some questions remain. Investigating how social desirability works within the context of assessments in an intervention study may elucidate how to minimize it. Also, further examination of food availability, in terms of consumption level and its affect on accuracy, is needed.