Schools
Two comparable elementary schools in the same school district in a small New England town were invited to participate in the present study and were randomly assigned to condition. Both schools have predominantly Caucasian students (89% intervention, 90% control), very few students who qualify for free or reduced lunch (fewer than 10% in each school), and very high rates of 4th graders scoring at or above grade level on the Connecticut Mastery Tests (91% for the intervention school, 87% for the control school). The enrollment is 309 students at the intervention school and 337 at the control school. Both schools participate in the same district-wide food service program, therefore, have the same foods available each day. The food service director reported that on average, 50% of children buy lunch at each of the two schools.
Intervention
The idea for the intervention emerged from the school district's Health Advisory Committee and the author was asked to help the district assess the impact of the intervention. Prior to the beginning of the school year, the district superintendent, food service director, and school principals agreed to the protocol for the study and the protocol was approved by the Yale University Human Subjects Review Board.
The food service regularly offered children a choice of at least two types of fresh or canned fruit, and one or two types of 100% juice, each day. In the intervention school, the cafeteria workers were instructed by the food service director to provide the following verbal prompt to each child while he or she was standing front of the fruit servings: "Would you like fruit or juice?" The implication of this statement was that children were expected to take a fruit serving, however, if a child indicated that he or she did not want fruit or juice, the food service worker did not prompt further. In the control school, no changes were made; the same fruit and juice options were available each day in the cafeteria line, but no verbal prompt was given.
A few weeks into the school year, the food service director reported that the intervention had been successfully implemented. The author visited the intervention school and spoke with the cafeteria workers to verify this. The present study did not influence the fruit choices provided in the schools, as the cafeteria workers and food service providers did not know which days would be used for data collection.
Observational data collection
The first observation day was in January 2005 and the second was in March 2005. The intervention and methods of observation on the two days were identical. The district superintendent and principals of each school gave permission for parent volunteers to observe fruit consumption in the cafeteria for two days. The volunteers were all parents of children in the school. They were instructed to casually observe what children were eating without initiating conversation with the children. If asked what they were doing, the observers were instructed to say that they were "just interested in seeing how the children were enjoying their lunch." During each lunch period, the observers recorded the following information for each child who purchased lunch: Did the child take fruit or juice? Did the child consume the fruit or juice? If the child consumed the fruit or juice, the observer rated whether it was partially (i.e., less than half) or fully consumed (i.e., equal or greater than half). There were four lunch periods in each school, corresponding to grades 1 through 4.
Analyses
The data were first analyzed for the two days separately, and then averaged. Fewer than 3% of the observations indicated "partial consumption," so this variable was collapsed into two levels (i.e., yes or no) for the analyses. Frequency analyses were used to determine: (a) the number of students who purchased lunch in each school, (b) the percentage who took fruit and juice, and within that (c) the percentage who ate the fruit and drank the juice. To calculate the odds ratios for taking the fruit serving, and eating the fruit serving, the data were analyzed using the formula to calculate the odds ratios and confidence intervals in an unmatched case control study [9].