Data for this paper are derived from baseline measurements collected for two longitudinal cohort studies designed to assess the etiology of youth obesity using an ecological model. The Identifying Determinants of Easting and Activity (IDEA) is one of several projects funded through the Transdisciplinary Research on Energetics and Cancer (TREC) initiative of the National Cancer Institute . TREC-IDEA is a three-year longitudinal study to identify determinants of adolescent (11-17 years old) obesity based on the multiple levels of the social-ecological model [1, 15]. The second study, Etiology of Childhood Obesity (ECHO), is also a three year longitudinal cohort study (age range 10-16 years old). Both studies have similar conceptual frameworks, use the same measurement instruments and protocols and both samples were drawn from the Minneapolis/St. Paul, Minnesota metro area. Based on these similarities, the samples were combined for these analyses.
The University of Minnesota's Institutional Review Board, Human Subjects Committee approved all study protocols. Participants for the TREC-IDEA study were recruited from: 1) an existing cohort of youth participating in the Minnesota Adolescent Community Cohort Tobacco Study, 2) a Minnesota Department of Motor Vehicle list restricted to the Twin Cities metro area, and 3) a convenience sample drawn from local communities using social networking, city parks and recreation bulletin boards, and fliers distributed at several school-based functions. For TREC-IDEA, a total of 349 parent/student pairs were measured at baseline (2006-2007). Additional details on recruitment procedures and study design have been reported elsewhere .
For ECHO, parent/child dyads were recruited from the membership of HealthPartners®, a large health management organization, within the seven-county metropolitan area of Minneapolis/St. Paul, Minnesota. To obtain a more diverse sample of youth, recruitment was targeted to enroll both healthy weight and overweight youth and parents and to oversample minorities. To be eligible for enrollment, youth were required to be current Health Partners members, in grades 6th through 11th in the fall of 2007, residing in one of the randomly selected middle or high-school districts included in the sample and have a parent willing to participate in a set of parental measures including a parent survey and the completion of the PAMI instrument.
For both studies, to obtain additional information about the home environment, youth were required to participate with one adult with whom he/she spends a significant amount of time (e.g., a parent/guardian, other relative or adult that cares for him/her on a regular basis). Due to the longitudinal design of these studies and the developmental time period covered (early to late adolescence), parent/student pairs were excluded from participating if they planned to move from the area in the next three years, had a medical condition that affects their growth (e.g., hypo/hyperthyroidism), were non-English speaking, and/or had any other physical or emotional condition that would affect their diet/activity levels or make it difficult to complete surveys or measurements. Only one parent/adolescent dyad per family was included in these analyses.
Physical Activity and Media Inventory (PAMI)
The PAMI is a self-report inventory of the availability and accessibility of equipment and other resources that may support household members' participation in activity and sedentary behaviors . Parent participants were instructed to go through each room/location in their home, and inventory the equipment present. For each room/location (including all rooms in the home, storage spaces, yard/garden, and garage), the PAMI included a list of response options, including 42 physical activity equipment items presented in alphabetical order. There also were five media equipment items listed: (1) television, (2) video cassette recorder (VCR) and/or DVD, (3) digital video recorder (DVR) and/or TiVo, (4) video game system, and (5) computer (desktop or laptop). Participants recorded the quantity of each type of item in the room/location and rated the accessibility of that item on a 4-point scale, with "put away and difficult to get to" coded as 1 and "in plain view and easy to get to" coded as a 4. A higher accessibility score indicated greater accessibility of the item. Active video games (e.g., Wii, DDR) were not included as an item on the original PAMI but during data collection several families indicated having these games through an option to write in a response. Active video games were coded as fitness equipment, similar to workout videos. The full PAMI instrument has been published previously .
The PAMI data were reduced to the following primary variables calculated separately for physical activity and media equipment: total number of items and the density of items in the home (total number of items divided by the total number of rooms/locations). In addition, we created variables to determine the number of televisions in the home and the number of televisions in children's bedrooms. We also created and assessed two summary scores. First, we calculated separate summary scores that accounted for availability and accessibility of the physical activity equipment (Physical activity Availability and Accessibility Summary Score (PAASS)) and media equipment (Media Availability and Accessibility Summary Score (MAASS)). A higher score reflects a greater overall presence in the home (both availability and accessibility) . To provide more detail, we calculated the PAASS for specific categories of items, including; sports equipment, fitness equipment, transportation equipment, water sports equipment, and outdoor/yard equipment. To rank the overall quality of the home, an overall home environment score was also calculated as the ratio of the PAASS to the MAASS (referred to as the Activity:Media Ratio Score). A higher overall Activity:Media Ratio Score would reflect a home more conducive for being physically active and less sedentary.
Physical Activity, Sedentary Time and Screen Time
Physical activity and sedentary time were assessed with the ActiGraph (Model 7164; Pensacola, FL) accelerometer. The ActiGraph has been previously validated for use with youth in laboratory and field settings [16–18]. It is a small (5.1 × 3.8 × 1.5 cm), lightweight (42.6 g), single plane (vertical) accelerometer that collects and stores accelerations from 0.05-2.00 G with a frequency response of 0.25-2.50 Hz. These settings capture normal human motion but will filter out high frequency vibrations from mechanical sources (e.g., operating a lawn mower).  The analog acceleration is filtered and converted to a digital signal and this value (count) is stored in user-specified time intervals. Thirty-second intervals were used for this study. ActiGraph monitors were initialized to begin collecting data at 5:00 am the day following the clinic visit where surveys and physical measures were completed . Adolescents were instructed to wear the accelerometer on their right hip for seven days (except for sleeping, swimming or bathing) and to return the unit in a pre-paid courier service envelope following the data collection period. Upon return, each monitor was downloaded to a computer for subsequent data reduction and analysis.
ActiGraph data were reduced using a custom-developed software program [20, 21]. All data contained within the time frame from when the monitor was initialized until the same time the following week was processed. For example, if students received their monitors at 9:00 AM on Friday the data from Friday at 9:00 until the next Friday at 9:00 would be processed through the program. For days 2-7, all data from 00:01 until midnight was reduced to summary variables. Day one and day eight were combined to form a composite seventh day of data.
Data files were scanned for data points ≥7,500 counts/30 seconds to identify implausible bodily movement or ActiGraph malfunction; no data points met this criterion. Daily inclusion criteria were established to determine days and times with acceptable accelerometer data. Blocks of time incorporating at least 30 continuous minutes of "0" output from the ActiGraph were considered to be times when the subject was not wearing the monitor . These data points were eliminated and not used in any calculations. Following these deletions, days with less than 10 hours of data were eliminated from data reduction to account for unrepresentative days of physical activity. Lastly, students with at least four out of seven days of ActiGraph data were retained for the analysis sample (n = 613, 86%) .
After processing the data through the exclusion criteria, summary variables were calculated. Time spent in moderate to vigorous physical activity (MVPA) and sedentary times (SED) were calculated as the average number of minutes per day spent in MVPA and SED, respectively. Due to the wide age range of the participants, recently developed age-adjusted cutoffs (using single year increments) were used to classify accelerometer data into intensity categories .
Measuring sedentary time via accelerometry captures all sedentary activities including sitting at school, reading/doing homework, crafts and other developmentally appropriate activities but does not specifically identify screen time. Screen time is one very prevalent sedentary behavior, and one particularly associated with obesity . In order to specifically measure screen media behaviors (or screen time), adolescents completed a self-administered survey in which they reported usual weekday and weekend time spent in the following activities: watching TV, watching DVDs or videos, Nintendo/Play Station/computer games, internet/computers. Response options ranged from "None" to "6+ hours per day" [25–28]. A weighted average of weekday and weekend screen media use was calculated to estimate mean hours per day spent using screen media. To be consistent with accelerometer-derived variables, screen time was converted to mean minutes per day.
Heights and weights of parents and their children were obtained during a 2-hour clinic visit. Measurements were made by trained staff using a direct read portable stadiometer (Shorr Productions, Olney, MD) for height and an electronic scale/body composition analyzer (Tanita TBF-200A; Tanita Corporation of America, Inc., Arlington Heights, IL) for weight. For the adolescents, age and gender specific BMI% were calculated using the 2000 CDC growth charts http://www.cdc.gov/growthcharts/percentile_data_files.htm.
Parents of adolescent participants completed a survey that included questions regarding the highest education level attained for all adults in the home. In addition, parents reported whether or not the child in the study received free or reduced price school lunch (yes/no), an indicator of family income. Parents also reported the number of adults (≥18 years old) and children in the home. Adolescents reported their age (in years) and their race/ethnicity.
Baseline data from the IDEA (2006-2007) and ECHO (2007-2008) studies were included in these analyses. All analyses were conducted with SAS version 9.1 (Cary, NC). Since boys are typically more active than similarly-aged girls and since the associations between the home environment and activity may be gender specific, the sample was stratified by gender and analyses were performed separately. Means (standard deviations) and percents were calculated to describe the sample and the distribution of values for the dependent and independent variables. Pearson correlation coefficients were calculated to determine bivariate associations between home environment variables (from the PAMI) and MVPA, SED, and screen time. Non-normally distributed variables were log transformed to more closely represent a normal distribution. If the distribution remained non-normal after the transformation, Spearman correlation coefficients were calculated using the non-transformed values. Home environment variables that were associated with MVPA, SED, or screen time behavior at the P ≤ 0.05 level were included in the multivariable general estimating equation (GEE) models. Participants sampled from the same school may appear to be more similar than those from other schools. GEE regression models were used to account for these contextual effects and clustering within our sample, although the interpretation of these coefficients is the same as for standard linear regression coefficients. These models included age, race/ethnicity, highest level of parent education, FRPL status, number of people in the home, parent BMI, month of data collection (to control for seasonal effects on activity and accessibility of some equipment items), and gender, and also examined possible gender interactions with PAMI variables. The GEE models also accounted for clustering at the school level since a number of students shared similar school environments. Finally, we also included the study (TREC IDEA or ECHO) as a fixed effect to account for possible sample differences due to different recruitment methods.