Study design and sample
This study was part of a United States Department of Agriculture (USDA) multi-state project involving institutions from 11 states in the U.S: Arizona (AZ), California (CA), Colorado (CO), Hawaii (HI), Indiana (IN), Kentucky (KY), Michigan (MI), Minnesota (MN), Oregon (OR), Utah (UT), and Washington (WA). The study used a cross-sectional design and a purposive sampling scheme to recruit children aged 10–13 years old and the primary person responsible for food acquisition and preparation in the child’s household. Results from the National Osteoporosis Risk Assessment  identified Asian, Hispanic, and non-Hispanic white (NHW) women as being at highest risk for osteoporosis; therefore, these ethnic groups were targeted for this study. Other inclusion criteria for both parents and their children were: 1) the ability to read and speak English and 2) having lived in the U.S for at least 12 months.
The demographic profiles of 9 of the states, i.e., AZ, CA, CO, HI, MI, MN, OR, UT and WA were used to recruit approximately equal proportions of parent-child dyads from each ethnic group. Recruitment techniques designated as appropriate for this study were fliers, verbal announcements, personal contacts, written announcements in bulletins and newsletters, and presentations to groups. In order to ensure diversity of the study sample in terms of level of education and socio-economic status (SES), a wide range of organizations and groups were approached for this study. These included the Expanded Food and Nutrition Education Program (EFNEP), Supplemental Nutrition Assistance Program Education (SNAP-Ed/formerly Food Stamp Nutrition Education Program), faith-based groups, after-school programs, scouting groups, non-elite competitive age-group sports teams, and the Special Supplemental Nutrition Program for Women, Infants, and Children (WIC).
Data were collected between April 2006 and September 2008. A standardized data collection protocol was developed and used to administer the parent questionnaires and the child questionnaires in a consistent manner across all sites. The questionnaires were designed to be independently self-administered by parents and their children. Data were collected in community settings, such as community centers, libraries, and athletic facilities, and respondents’ homes. Researchers in several states also asked parents to return completed questionnaires with a stamped return envelope after obtaining consent in person (14–50% of questionnaires for these states). Parents took approximately 20–45 min, and children approximately 20–30 min to complete the questionnaires. All questionnaires were completed in English. Based on each research institution’s appropriate level of remuneration, parents and children were each given $5–$20 in cash or as gift certificates in return for their participation. All phases of this study were approved by the Institutional Review Board of each respective study site.
A previous study  reported a significant correlation in frequency of milk consumption between mother-child dyads based on a study sample of 180 pairs. Thus, it was assumed that a similar sample size was required for the current study in order to detect a significant association between parental psychosocial factors and their children’s calcium intakes. A total of 673 parent questionnaires and 680 child questionnaires were sent to a central location for scanning. In some instances, the adult respondent was a relative or caregiver other than the parent. For simplicity, the adult respondent is referred to as “parent” in this report.
Questionnaires and variables assessed
Parent and child characteristics
The adult questionnaire included questions regarding age, sex, education, and employment status of both the participant and his/her spouse. In addition, information on household characteristics, such as the number of adults and children living in the home, age and sex of the adolescent participant and his/her siblings, and participation in public assistance programs was collected. The child questionnaire requested information on age, sex, grade, and family members living in the home. Both the parent and child, in their respective questionnaires, were asked to select their ethnicity (Hispanic/Latino or Not Hispanic/Latino) followed by the race groups with which he/she most identified, as directed by the U.S Office of Management and Budget .
Dietary calcium intakes of early adolescents were estimated (mg/day) through the use of a 79-item calcium-specific semi-quantitative food frequency questionnaire (FFQ), which was previously developed and evaluated for use with Asian, Hispanic, and NHW adolescents . The FFQ asked participants to recall food and beverage consumption of the past month. Further details on the development of the FFQ have been published elsewhere . Although developed for multiethnic youth, this questionnaire has been evaluated for use with adult women . Therefore, the same FFQ was used with parents to allow for direct comparison of calcium intakes (mg) between parents and their children. The questionnaire also included questions regarding eating patterns and current supplement use.
Estimated calcium intakes (mg) were calculated following the procedures described previously . Estimated daily calcium intakes < 100 mg or > 2500 mg were considered implausible  and individuals with such values were excluded from analysis involving calcium intakes. Fifteen parents (2%) and 25 children (4%) had implausible calcium intakes. Therefore, 618 parents and 608 children representing 597 parent/child pairs were retained for statistical analyses of calcium intakes. All parents with implausible calcium intakes were Hispanic. Among children, 12 were Hispanic, 11 NHW, and 2 Asian (chi-square = 2.05, df = 2, P = 0.360). Dietary calcium intakes (mg) were classified into 2 categories previously used : Total dairy calcium (mg) represented the sum of calcium from dairy foods and calcium from mixed foods. An example of a mixed food containing both dairy and non-dairy calcium sources is pizza where calcium from the cheese is from a dairy source plus a non-dairy calcium source from the crust. Total calcium from food (mg) was the sum of total dairy calcium, calcium from non-dairy foods (e.g., broccoli, calcium-fortified orange juice), and mixed foods. Both calcium intakes were summarized as mg/day. Fluid milk consumption was defined using the food grouping, “milk to drink, white or chocolate.” The frequency responses were recoded to cups of milk per day (e.g., 1–3 cups per month would be 0.0667 cup/day; 1 cup per week would be 0.1429 cups per day and so forth). Similarly, soda pop consumption was re-coded to cans per day.
Parental psychosocial factors
Based on SCT  and findings from previous studies [15, 22, 23], a questionnaire was developed to assess parental PSF hypothesized to influence children’s calcium intakes (Additional file 1). Three major constructs were defined: 1) social and environmental factors, 2) parent’s attitudes and preferences, and 3) parental knowledge of calcium intake during adolescence. These constructs were measured with 17 subscales. Briefly, for the social and environmental construct, 9 subscales emerged: culture/tradition for parent (calcium rich foods parent grew up with or considered culturally relevant), parenting styles (require or encourage child to eat food), parental role modeling of intake of CRF (CRF served at eating occasions throughout the day, as well as soda), parental concern about cost of CRF, availability of CRF in the home, parental encouragement of healthy eating practices (including talking to the child about eating healthfully), reasons for eating out, interruption of family meals, and parental perception of child’s access to CRF outside the home. Eight subscales were identified for the attitudes and preferences construct: parental taste preferences for CRF, parent’s perceived intolerance to milk and dairy foods, parental perception of health benefits of CRF for self, parental perception of the importance of calcium food sources for children, parental concern about dairy foods negatively influencing body weight, parental concern over adequacy of own calcium intake, parental expectations for child’s intake of beverages, and parental perception of convenience of CRF. In a separate study, these subscales met standards for psychometric properties with modest to acceptable Pearson correlation test-retest reliability coefficients for some subscales (0.68–0.85) and Cronbach α coefficients for internal consistency (0.50–0.79) [33, 34]. The scale composition, scale items and response categories have been published previously .
The raw data responses to the statements and questions representing each individual subscale were summed and then an average response was calculated for each subscale. In general, each response was coded as 1 to 5 or 1 to 6. The final score for a subscale of 5 statements would be the sum of the responses to the 5 statements divided by 5. As appropriate, the values associated with some responses were reverse coded for accurate directionality. For subscales containing 3 or more statements, a response was needed for at least 66% of the statements to calculate a score.
All questionnaires were reviewed and corrected for stray marks and multiple responses prior to sending to a central location for optical scanner reading. Scanning was completed in batches and separate raw data files were sent to a second central location for data cleaning and preparation for analysis. Of the parent questionnaires scanned, 4 were blank, 1 had a data scanning error, and 27 self-reported information outside of the inclusion criteria (i.e. missing ethnicity; ethnicity other than Asian, Hispanic, or NHW; or child’s age) resulting in a total of 641 acceptable questionnaires. Of the scanned child questionnaires, 5 were blank, 2 had data scanning errors, and 5 self-reported ages outside of the inclusion criterion resulting in 668 acceptable questionnaires. Only paired children and parents were retained for statistical analysis (n = 633). A flow diagram showing the details of the final sample size is provided in Additional file 2.
When discrepancies existed between the parent and the child responses regarding age, sex, and ethnicity of the child, the parent response was considered accurate and used in all analyses. Participants who selected multiple ethnic groups were classified as Asian, Hispanic, or NHW if one of these groups was also selected. Those who chose Asian and NHW, or Hispanic and NHW, were classified as Asian or Hispanic, respectively. In instances where Asian and Hispanic were both selected (n = 3), the predominant ethnic group of the sampling location was used for classification as either Asian or Hispanic.
Data were assessed for normality using normal probability plots and no variable was found to need a transformation. Season was defined as winter (November to January), spring (February to April), summer (May to July), and fall (August to October); employment status of parent was collapsed into no formal employment (student, homemaker/househusband, not employed and retired as they appeared on the questionnaire), part-time employment, and full-time employment; employment status of the household was defined as no formal employment and at least half-time employment of one individual; education level of parent was defined as high school diploma or less, some college, and 4-year college or more; parents’ ages were collapsed into 18–40 years and 41+ years; number of adults living in the home was defined as one, two, and three or more; and number of children in the household was collapsed to one, and two or more. Participation in federally funded programs was positive if the parent reported participation in WIC, SNAP, and/or free/reduced priced school lunch. The child’s report of breakfast location was collapsed to at home, at school, skips breakfast, and other (i.e. fast food restaurant, convenience store, other).
Analysis of variance was used to identify differences in quantitative variables among ethnic groups. Post-hoc Bonferroni analyses were performed on significant P-value results to identify which ethnic groups were significantly different. Differences between categorical variables were assessed using chi-squared analysis. Pearson’s correlations and linear regression accounting for sex were completed to examine the relationship between parent-child calcium intakes.
Multivariate linear regression analysis was conducted to quantify the predictive value of significant parental PSF towards calcium intakes (mg) of their early adolescent children while accounting for ethnicity. Parental psychosocial factors and parental calcium intake (mg) were modeled as predictors and children’s total calcium intake (mg) from food was the dependent variable. Potential confounding variables examined were age (of child and parent), soda intakes (of child and parent), and the indicator variables of season, sex (of child and parent), employment status (of parent and household), education level, number of adults in home, number of children in home, single household (yes/no), and participation in federally funded programs. Only those confounding variables significantly altering the association between parental PSF and children’s calcium intakes (mg) were retained in the models. Interaction terms were then examined and only significant interaction terms were retained in the models. Since parental role modeling and parent’s calcium intake (mg) were highly correlated (r = 0.491, P < 0.001) these were interchanged in separate models. The same statistical models were used with total dairy calcium (mg) as the dependent variable. To account for the complex sampling design, indicator variables for the states were examined with all final models. The results did not differ. All analyses were performed using Statistical Package for the Social Sciences (SPSS) software 17.0 for Windows (2007, Chicago, IL, USA). P ≤ 0.05 was used as a measure of statistical significance in all analyses.