The intervention phase of the study was implemented in two communities on each of Oahu and the Big Island of Hawaii with populations of 10,506 and 5,748, respectively. Comparison communities on each island had populations of 3,664 and 2,297, respectively. These four communities are home to more Native Hawaiian and Pacific Islanders (Oahu 27%, Big Island of Hawaii 10%) than the State of Hawaii as a whole (9% Native Hawaiian and Pacific Islanders). Income levels are low, with >75% of the population below the poverty level (US Census 2000).
Description of the CIQ instrument
The CIQ included sociodemographic characteristics (birth date, sex, marital status, and years of education). A “food getting frequency” recorded the number of times over the past 30 days that the respondent “got” various food items for the entire household. Thirty-nine different foods were included, focusing on foods to be promoted by the HFH intervention, and their less healthy alternatives.
The CIQ then asked the caregiver respondent to describe household meal patterns and child feeding. Questions addressed caregiver's psychosocial factors, including food-related self-efficacy, intentions, and knowledge. The self-efficacy section asked the respondent to describe their confidence in making healthy food selections, using preparation methods promoted by the intervention, and using food labels to make healthy choices when selecting foods. The food intentions section addressed the respondent's intention to purchase, consume, and prepare foods promoted by the intervention. The food-related knowledge section asked questions that related to nutrition information provided by the intervention. A section on household socioeconomic status assessed employment, participation in food assistance programs, food security, and ownership of material goods. The final section assessed general patterns of food consumption through a qualitative food frequency instrument of 39 foods.
Scale and score construction
A series of additive scores and scales were developed to measure the main psychosocial and behavioral constructs in the conceptual framework, based on Social Cognitive Theory (SCT) and the Theory of Planned Behavior (28). All scales were assessed for internal reliability using the Cronbach's α statistic. Similar scales have been developed for our other food store intervention trials and have been shown to be significantly related to study outcomes (15,23,24,25,26,27).
Caregiver food knowledge (behavioral capability, SCT) scale is the sum of scores from six multiple-choice questions that asked respondents to identify a food lowest in fat, highest in fiber, or lowest in total calories, or the cooking method that would result in the least the amount of fat. Scores had a possible and actual range from 0 to 6 with a mean of 4.3 (s.d. = 1.4, α = 0.58).
Caregiver food self-efficacy (SCT) scale is based on nine statements about healthy food purchasing, preparation, and consumption. Respondents were asked how easily they could do a particular behavior regularly. Scores had a possible range of 9–36, with an actual range from 16 to 36, and a mean of 30.3 (s.d. = 4.6, α = 0.65).
Caregiver food intentions (Theory of Planned Behavior) scale is based on seven questions where the respondent was asked to state their intention for future food choices. A higher score was given for the choice that reflected the lowest fat or lowest sugar choice. Scores had a possible range of 0–18 with an actual range of 7–18 and a mean of 10.8 (s.d. = 3.5, α = 0.61).
Healthy food getting frequency scale includes 16 different foods (and food groups) that were promoted as part of the HFH program (e.g., baked chips, water). Adult caregivers were asked to recall the number of times they got each food in the previous 30 days. Scores had a possible range of 0–480, with an actual range of 5–377, and a mean of 44.2 (s.d. = 45.9, α = 0.91).
Unhealthy food getting frequency scale includes nine different foods that were demarketed as part of the HFH program (e.g., regular chips). Caregiver respondents were asked to recall the number of times they got each food in the previous 30 days. Total unhealthy food getting frequency scores had a possible range of 0–270, with an actual range of 0–260, and a mean of 16.7 (s.d. = 23.3, α = 0.90).
Healthy food convenience perception scale includes three questions on the perceived convenience of selecting, preparing, and serving healthy food options. Caregivers were asked to provide their level of agreement with a series of statements (e.g., healthy food is inconvenient to prepare), with a higher score indicating greater perceived convenience of healthy foods. Scores had a possible and actual range of −3 to 3, and a mean of +1.9 (s.d. = 1.7, α = 0.75).
Caregiver healthy food consumption scale includes previous 30-day consumption frequency for 13 different low-sugar and/or high-fiber foods that were promoted as part of the HFH program. Respondents were provided with eight frequency categories ranging from “never in past 30 days” to “two times a day or more,” which were converted into number of times a month. Scores had a possible range of 0–780, with an actual range of 0–390, and a mean of 64.6 (s.d. = 67.6, α = 0.77).
Material style of life (environment, SCT) scale was developed as a proxy for socioeconomic status. Respondents were asked to list the number of seven household items they owned in working condition (e.g., computer). Scores had a possible and actual range from 1 to 7 with a mean of 4.7 (s.d. = 1.1, α = 0.51).
Child healthy food knowledge identification scale assessed the focal child's ability to identify which of two foods is healthier (e.g., frosted flakes vs. Cheerios). Seven questions were included in the final scale, with correct answers each given one point. Scores had an possible and actual range from 0 to 7 with a mean of 5.0 (s.d. = 1.7, α = 0.61).
Child healthy food intentions assessed the focal child's predicted choice among three different foods. If the lower fat, lower sugar, and/or higher fiber option was chosen, the child received a point. Scores had a possible and actual range from 0 to 12 with a mean of 7.5 (s.d. = 3.1, α = 0.60).
Child healthy food consumption scale includes previous 7-day consumption frequency responses for eight different low-sugar, low-fat, and/or high-fiber foods promoted as part of the HFH program. Respondents were provided with four frequency categories ranging from “never” to “every day” which were converted into number of times a week. Scores had a possible range from 0 to 56, with an actual range from 4 to 51, and a mean of 27.4 (s.d. = 9.5, α = 0.66).
Child unhealthy food consumption scale includes previous 7-day consumption frequency responses for eight different foods that were high in sugar, fat, and/or lower in fiber, which were demarketed by the HFH program. Scores had a possible range from 0 to 56, with an actual range from 0 to 26, and a mean of 10.2 (s.d. = 6.1, α = 0.55).
Healthy Eating Index (HEI) component scores were calculated, each of which could range from 0 to 9 (29). A total HEI score was determined as the sum of the nine components; the dietary variety component was excluded because an appropriate database was not available to analyze the data. Although the Food Guide Pyramid food groups have now been revised to correspond with MyPyramid (30), and a new method of calculating the HEI has been released (31), these databases were not available for this study.
Prior to initiating the trial, we calculated the sample size needed for detecting meaningful differences in four variables: HEI score, percent of energy from fat, number of servings of vegetables, and number of servings of fruit. Using a two-sided paired t-test, a significance level of 5% and a power of 90%, we found that a sample size of 130 child–caregiver pairs would be required. Of the original 184 households sampled, 10 were eliminated due to missing data, yielding a final sample of 174 caregiver–child pairs at baseline. Postintervention, we attempted to interview as many of the baseline respondents as possible, and were able to interview 116, a 67% retention rate. The sample for whom we had pre- and postintervention data differed significantly from those for whom we were unable to get postintervention data, in terms of percent female (99.2 and 89.2%, respectively) and proportion employed (52 and 49%, respectively). No other differences in pre- and postintervention sociodemographic data were observed.
Multivariable linear regressions were conducted for each of the psychosocial and behavioral outcome, including HEI indicators and gram consumption of promoted foods to assess the programs' impact. We calculated the difference between baseline and postintervention scores for the caregiver and child psychosocial and behavioral variables, and then regressed the result on a series of independent variables, including intervention group assignment, caregiver age, sex, years of education, employment status, time period between baseline and postintervention measurements, caregiver ethnicity, and material style of life score.
We examined change in intake of the key targeted foods (i.e., water, 100% juice, diet soda, 2% milk, 1% milk, low-sugar cereals, local fruits, and vegetables) from pre- to postintervention. Due to low consumption of these foods (≥75% of respondents never consumed a particular targeted food, with water the exception), we treated consumption of the targeted foods as dichotomous (any nonzero value = 1, else 0). McNemar's paired tests were performed where each participant's final response was compared to their initial response to assess change in consumption of targeted foods. Data were analyzed using SAS 9.1 (SAS Institute, Cary, NC). We checked the distribution of the dependent variables and log transformation was conducted for variables that were not normally distributed. The variety of outcome variables being used increases the probability of type 1 errors. To obtain a global assessment of the impact (if any) of the intervention, multivariate tests for the effect of intervention on the entire set of change in quantitative outcome variables, adjusted for covariates, were used for the set of 20 caregiver variables and the set of 16 child variables using the “mtest” option in SAS.
Human subjects approval for this study was given by the University of Hawaii Committee on the Use of Human Subjects. Signed informed consent was obtained from all adult and child respondents.