Study design and population
Project EAT-II is a longitudinal, follow-up study of Project EAT-I, which explored the socioenvironmental, personal, and behavioral determinants of dietary intake and weight-related variables among an ethnically diverse adolescent population. Project EAT-I enrolled 4746 participants including a younger cohort of middle-school students and an older cohort of high-school students recruited during the 1998–1999 academic year from 31 middle and high schools with ethnically and racially diverse student populations. Project EAT-II aimed to resurvey the original participants by mail to assess changes in their eating patterns and weight status 5 years later (2003–2004). Of the original study population, 22.5% (1074) were lost to follow-up for various reasons, primarily missing contact information at Project EAT-I (n = 411) and no address found at follow-up (n = 591). Of the remaining 3672 participants who were contacted by mail, 2516 competed surveys. These 2516 participants represent 53.0% of the original study sample and 68.4% of participants who could be contacted for Project EAT-II. One-third of the participants (32.0%) were in the younger cohort; at Time 1, their mean age was 12.8 years (SD = 0.8) and at Time 2 their mean age was 17.2 years (SD = 0.6). Two-thirds of the participants (68.0%) were in the older cohort; at Time 1, their mean age was 15.8 years (SD = 0.8) and at Time 2 their mean age was 20.4 years (SD = 0.8). Although the younger cohort is smaller than the older, there is sufficient power to precisely estimate the relationships between the parental variables and adolescents' outcomes in both cohorts.
Surveys were sent by mail to the address provided by the participant during Project EAT-I. Methods to locate Project EAT-I participants and increase the survey response rate have been described elsewhere. Data collection occurred between April 2003 and June 2004 and was conducted by the Data Collection and Support Services in the Division of Epidemiology and Community Health at the University of Minnesota. The University of Minnesota's Institutional Review Board Human Subjects Committee approved all study protocols.
Physical activity and sedentary behavior
Project EAT-I and EAT-II surveys included several questions to assess physical activity and sedentary behavior. Questions related to physical activity were modified from the Leisure Time Exercise Questionnaire[16, 17]. Two survey items individually assessed moderate and vigorous activity asking, "In a usual week, how many hours do you spend doing the following activities...." Vigorous activity was described as strenuous, during which the heart beats rapidly, and moderate activity was described as not exhausting. More than 10 examples of specific activities were given after each question. Possible responses ranged from 0 to ≥6 hours per week. A continuous variable representing total weekly hours of moderate and vigorous physical activity (MVPA) per week was created by summing the weekly hours of moderate and vigorous activity.
Items to measure time spent watching television and videos were adapted from Planet Health. Participants reported average hours per weekday and weekend day (Saturday or Sunday) "watching TV & videos." Possible categorical responses ranged from 0 to ≥5 hours per day. These items have been validated and show moderate to high test-retest correlations. A continuous variable, total hours of TV & videos per week, was created by summing the average weekday hours multiplied by five, and average weekend hours, multiplied by two.
Adolescents' perception of parents
The Project EAT-I survey included two items asking how much the participant's mother encouraged them to be physically active and how much she cared about staying fit and exercising. The same two items asked about the participant's father's encouragement and care about staying fit and exercising. Possible response to these questions were: 1 = "not at all," 2 = "a little bit," 3 = somewhat," and 4 = "very much." These items had test-retest correlations between 0.66 and 0.69.
Gender, age, ethnicity/race and socioeconomic status (SES) were based on adolescents' self-report in Project EAT-I. The primary determinant of SES was parental educational level, defined by the highest level of educational attainment of either parent. In addition, an algorithm was developed to take into account family eligibility for public assistance, eligibility for free or reduced-cost school meals, and employment status of the mother and the father.
Because attrition in the study population during the 5-year study did not occur completely at random, the data were weighted to adjust for differential response rates in Project EAT-II using a response propensity method. The use of this method with Project EAT data has been described in detail elsewhere, where it has been evaluated as a means of correcting potential response bias. At Time 1, non-responders to Project EAT-II reported participating in fewer MVPA hours per week (7.1 vs. 7.5, p < .01) and more TV/video hours per week (18.9 vs. 17.8, p < .01), compared to those who did participate in Project EAT-II. Additionally, there were some differences in the amount of parental encouragement and care for fitness reported by these two groups. After adjusting for sociodemographics and weighting, there were no significant differences found in MVPA nor TV/video hours per week at Time 1 between responders and nonresponders at Time 2. The weighted ethnic/racial and SES proportion are 48.5% white, 19.0% black, 19.2% Asian, 5.8% Hispanic, 3.5% Native American, and 3.9% mixed or other race. Thirty-seven percent of the sample were of low or low-middle SES.
ANOVA models were used to calculate cohort and gender-specific weighted means and standard errors of baseline parental factors, as measured during Project EAT-I (Time 1), and cohort and gender-specific weighted means and standard errors of the primary outcomes, hours of physical activity and sedentary behavior per week, as measuring during Project EAT-II (Time 2). The gender by cohort interaction was evaluated to determine if the pattern of means by gender of the parental variables or primary outcomes differed by cohort. If significant, post-hoc t-tests were used to test differences between the individual means.
Separate multiple linear regression models were developed to examine the association between maternal and paternal variables and the adolescents' behavioral outcomes of weekly hours of MVPA and TV/video watching five years later. All analyses were stratified by gender and cohort ("younger cohort" refers to those transitioning from junior high/middle school to high school and "older cohort" refers to those transitioning from high school to young adulthood) to assess the differential impact of mothers and fathers on children of the same or opposite sex, and potential differences in the relationship between parental influence and behavioral outcomes for the younger versus older cohort. All of the models were first adjusted for race and SES (Model 1), and then adjusted for baseline values of the primary outcome (i.e. adolescent physical activity or TV/video viewing) along with race and SES (Model 2).