The 3 ^rd and 5 ^th grades were chosen where reading level would be sufficient to comprehend the questionnaire and allowing enough separation of age. Of the 180 students recruited from the third (n=77) and fifth (n=103) grades, 130 (72.2%) elementary school students at this medium-sized elementary school in the southeast participated. Permissions to conduct the study were obtained from the Research Committee at the County Public Schools and the Institutional Review Board (IRB) of the particular school. The designated teachers distributed consent forms to all third- and fifth-grade students. On the first day of the study, the purpose of the study was explained to the students. After all questions were answered satisfactorily, students of consenting parents/guardians were asked to sign the assent form. The researcher provided instructions on the correct way to complete the questionnaire and answered any questions. Questionnaires were collected and scanned for obvious errors such as a complete page of missing entries. Due to the constraints of the public school environment the researcher was not allowed to insist on completion of missing responses. One week later, the identical instrument was administered to the fifth-grade participants, thereby assessing the test-retest reliability of the instrument. As an incentive, each participant received a package containing a MyPyramid magnet, a Color Way Produce wheel and a pencil.

The Pro Children Project developed a cross-sectional questionnaire that not only assessed FVC intake, but also evaluated the determinants associated with these consumption patterns in children.⁶ Sections of the Pro Children Questionnaire included: (1) demographics of the students, (2) usual intake of FVC, and (3) environmental determinants affecting FVC consumption patterns. Haraldsdottir et al.¹¹ determined that the questionnaire was a valid and reliable instrument to determine FVC intake in European children. De Bourdeaudhuij et al.¹² concluded that the questionnaire was a valid and reliable tool for evaluating personal, social, environmental and cultural factors that influence FVC consumption in European children.¹³ The Pro Children Questionnaire was developed as the first phase of the Pro Children Project which aimed to promote fruit and vegetable intake in European school children. A detailed discussion of the development of this questionnaire is discussed elsewhere.¹³ Briefly, elements from the Attitudes-Social Influence-Self-efficacy (ASE), and the Ecological models were used, as the theoretical framework, to develop categories of influence based on a literature search. The specific determinants were chosen based on outcomes from qualitative studies (interviews and focus groups comprised of school aged children) and other self-administered food questionnaires administered in classrooms.

Determinants of FVC were measured as personal, social and environmental scales using constructs based on those used by De Bourdeauhuij et al.¹³, and Kristjansdottir et al. ¹⁵and modified based on Kendall’s tau b correlations of this study. Separate scales were used for fruit and vegetables for validation in the European populations and for this study. The choice for separate scales was in agreement with several other studies and was due to the factor analysis and distinctions in fruit and vegetable related behaviors.^16-18 The collinearity of two self-efficacy measured necessitated the removal of one. We retained “If I decide to eat vegetables every day, I can do it” and removed “It is difficult for me to eat vegetables everyday” since the former was specifically personal and the latter could have been due to environmental factors. Eight logical factors, designated at F for fruit and V for vegetables, were identified analogous for fruits (F1-4) and vegetables (V1-4). Positive attitudes, liking and self-efficacy (F1, V1) and barriers F2, V2) were measured for personal determinants. Perceived active encouragement from family, family rules, and peer and parental modeling were social determinants (F3, V3). Accessibility and availability were factors for the physical environment (F4, V4). Knowledge of the recommendations FVC (separate questions for fruit and vegetable intake) was considered a separate category.

Analyses were performed using all data without replacement of missing values. The factors were combined into scales and the internal consistency between each factor was measured by Cronbach’s alpha coefficient. Multiple regression assumptions regarding normality, linearity and homoscedasticity were acceptable. Test-retest reliability of the multi-dimensional scales was assessed by repeated measure multivariate analysis of variance as opposed to multiple univariate tests (e.g., Bland/Alterman) of agreement. Fruit and vegetables consumed per day were totaled separately and together. Mean fruit and vegetable intake per day was compared by gender and grade using the Student t-test. The General Linear Model was used to assess discriminative validity. Demographics were not found to be associated with FVC. Separate regression analyses were conducted for fruit and vegetables to test an ordered set of a priori predictions to explain the variance of FVC based on the factors. Step one included demographics and knowledge, step two personal, step three, social, step 4, encouragement, and step 5, physical environment (availability and accessibility). The dependent variable for fruit consumption was ‘how often do you eat fresh fruit’. Vegetable intake combined three questions asking ‘how often do you eat salad, other raw vegetables and cooked vegetables’. Knowledge was considered a separate construct and possible confounder when assessing personal, social and environmental determinants. The main effects of race, gender and grade were not associated with knowledge of FVC recommendations. Three-way interactions show differences; however, the number in each category was insufficient (< 30 and < 10 for African-Americans) to determine differences. Analyses were conducted using Statistical Package for Social Science (SPSS) version 18. A p-value of < .05 was considered significant.

One hundred and thirty third- and fifth-grade students participated in the study. Demographic characteristics (Table 1) showed that there were no statistically significance in the proportions of those within each gender and each ethnicity grouping between 3 ^rd and 5 ^th graders. The ethnic group categories were collapse to three groups, based on the estimated marginal means (African American, Hispanic and White non-Hispanic).

Table 2 presents Cronbach’s Alpha for the fruit and vegetable scales. A value of 0.70 to 0.79 is considered to be acceptable, .80 to .89 to be very good and .90 and higher to be excellent.¹⁹ The developers of the Pro Children questionnaire used .50 as a cut-off for a satisfactory reliability coefficient for self-administered questionnaire for children 10-11 years of age.¹³ Cronbach’s alpha values greater than .70 were found for all the scales except for the fruit ‘availability’ scale (F4) which was still in the acceptable range as stated by the developers of the questionnaire¹³ and near the cut-off point (.70) recommended by Cohen.¹⁹ Cronbach’s alpha values were higher for all vegetable factors except for barriers.

No differences were found between the test and the retest for the individual questions, factors, and the total means of fruits and vegetables with the exception of the question for mean perceived vegetable intake. The mean vegetable intake was lower at the retest (t = 2.29, p = .023).

Discriminative validity was tested via the general linear model, multivariate analysis of variance, to ascertain whether the instrument could demonstrate statistical significance differences between grade (third and fifth graders), and gender. The size of the effect was measured by the partial eta square and presented in Table 3 for each fixed effect. The discriminatory validity indicated the Pro Children Questionnaire can show differences across grade and gender levels for barriers of fruit (F2) and vegetables (V2), but not for other factors. Grade together with gender explained barriers to eating fruit and vegetables (F2).

The means for FVC per day are presented by gender and grade in Table 4. Fifth-graders compared to 3 ^rd graders and girls compared to boys reported a higher total fruit and vegetable consumption.

Table 5 presents the ordered set of regression analyses based upon prior assumptions for fruit and vegetable determinants in the study population. Separate but parallel models were conducted for reported fruit and vegetable consumption. Knowledge of recommended FVC was a determinant of FVC, adjusting for race, gender and grade (step 1). ‘Knowledge’ remained significant when ‘attitudes and likes’ (F1 or V1) was added (step 2). ‘Attitudes and likes’ (F1, V1) were significant with knowledge in the model. ‘Barriers’ were not associated with FVC and ‘knowledge’ was no longer significant with the addition of ‘barriers’ (F2, V2, step 3). ‘Attitudes and likes’ remained significant in the fruit model but not in the vegetable model with the addition of ‘barriers’. ‘Encouragement’ (F3, V3, step 4) was not associated with FVC. ‘Attitudes and likes’ became significant for the fruit model with the addition of F3. Third graders were associated with lower fruit consumption as compared to 5 ^th graders, adjusting for race, ‘knowledge’, ‘attitudes and likes’, ‘barriers’ and ‘encouragement’. ‘Availability’ (F4) and ‘attitudes and likes’ (F3) were determinants for fruit intake, adjusting for race, gender, grade, ‘knowledge’, and ‘encouragement’.

Figure 1

Table 1. Demographics of the Study Population

Figure 2

Table 2. Reliabilities for Fruit and Vegetable Factors

Figure 3

Table 3. Discrimination validity of fruit and vegetable factors by grade and gender

Figure 4

Table 4. Mean fruit and vegetable consumption by gender and grade

Figure 5

Table 5. Regression models for factors of fruit and vegetable intake adjusting for knowledge of fruit and vegetable recommendations

Convenience samples, and not representative samples, were used. All data were self reported and thereby subject to possible comprehension, memory and recording bias. The questionnaires were completed in classroom settings and the students may have been distracted by their peers. There were missing values for each determinant and this may introduce non-response bias. For the hierarchical regressions, two- and three-way interactions could not be tested due to insufficient power to determine a difference (less than 30 participants per category).The selected elementary school offered nutrition education to all of its students. Students from Time 1 to Time 2 were not paired for comparisons; however, reliability was established by the mean responses. This study did not adjust for socioeconomic factors. Children in single-parent households were less likely to consume FVC as compared to children in two-parent households (53% vs. 61%).³² Educational level and employment status were also identified as determinants of FVC consumption. ³⁶