We used baseline data obtained from September 2007 through November 2009 from KAN-DO, a randomized, controlled intervention trial of 400 overweight postpartum mothers and their preschool-aged children. Details of the KAN-DO Study are published elsewhere (12). Briefly, inclusion criteria for the KAN-DO Study included a postpartum maternal body mass index (BMI) of at least 25 kg/m², delivery of an infant 1 to 7 months before randomization, and a child aged 2 to 5 years in the household. Exclusion criteria included mother less than 18 years old, mother unable to speak and read English, no regular access to a telephone, lack of a permanent mailing address, and presence of a health condition precluding daily physical activity in the mother or child. If multiple children in the home were aged 2 to 5 years, we chose the child born in the earliest month of the year for study participation.

We recruited study participants from 14 counties in the Triangle and Triad areas of North Carolina by using a combination of birth certificate records, publicly available mailing lists, and flyers posted in clinic and community locations. Of more than 40,000 women who received information about the KAN-DO Study, 4,445 were screened; and of those, 1,617 refused and 2,180 were ineligible, leaving 648 who were eligible and interested. Of those, 152 did not attend their first scheduled appointment, 80 refused or were deemed ineligible at their first study appointment, and 16 did not complete all components of the baseline assessment. This left 400 eligible mothers from whom we obtained written informed consent. The participants were randomized based on 4 strata (child's age, mother's race, site, and days from birth of recent child). The institutional review boards of Duke University Medical Center and the University of North Carolina at Greensboro approved this study.

The outcomes were measured objectively by using an accelerometer. Accelerometers are small instruments worn on the body that measure accelerations that can be converted to intensity of physical activity. The measurements are averaged over prespecified time periods called epochs (13). In the KAN-DO Study at baseline, children were instructed to wear an Actical omnidirectional accelerometer (Mini Mitter Co, Inc, Bend, Oregon) placed at the hip for 7 days. They were instructed to remove the monitor only for bathing and nighttime sleeping. The accelerometers were water-resistant, and children were instructed to also wear them for water-based activity. The accelerometers were set to record activity in 15-second epochs. The 2 outcomes of interest were child's minutes per day of sedentary time and child's minutes per day of MVPA. Sedentary time was defined by using a cutpoint of less than 12 counts per 15 seconds (14). MVPA was defined by using a cutpoint of 715 counts or more per 15 seconds (15). To be included in the analysis, children needed at least 3 valid days of wear (2 weekdays and 1 weekend day with at least 6 h/d of wear). Accelerometer data were available for 392 children; 55 of these children had fewer than 3 valid days of wear, resulting in 337 children available for analysis.

The correlates (Table 1) were divided into demographic, biological, behavioral attributes and skills, social and cultural, and physical environment domains adapting a framework used by Sallis et al (8). The mothers were asked to wear the same type of accelerometer as the child, but set to 60-second epochs, at the hip for 7 days. They were instructed to remove them only for bathing or nighttime sleeping and to wear them for water-based activity. Maternal physical activity was examined as total counts per day. For the maternal accelerometer data to be included in the analysis, mothers also had to have at least 3 valid days (2 weekdays and 1 weekend day with at least 6 h/d) of wear. For both the mother and child, we measured height by using the Seca 214 portable stadiometer (Seca, Hamburg, Germany) and measured weight by using the Tanita BWB-800s digital scale (Tanita Corp of America, Inc, Tokyo, Japan). We categorized children as underweight, healthy weight, overweight, or obese by using American Academy of Pediatrics recommendations (16). All other potential correlates were from the baseline questionnaires completed by the mother. To screen for postpartum depression, we used the Edinburgh Postnatal Depression Scale (EPDS), a 10-item questionnaire with a maximum score of 30, and we considered scores of 13 or higher as a positive screen for depression (17). To assess the presence of a chronic health problem among the mothers, we asked, "Do you have any longstanding illness, disability, or medical condition? That is, anything that affects your work or other regular daily activities such as type 2 diabetes, cancer, and heart disease?"

The outcomes were modeled by using a natural log transformation to meet the assumptions of linear regression. Since the study was conducted at 2 study sites (Durham and Greensboro), study site was included in all models to account for potential site difference. To account for differences in wear time, we also included mean hours per day of monitoring for the children in all models. We first conducted linear regression between each correlate and the transformed outcome to produce minimally adjusted β coefficients, adjusted only for site and wear time. We then conducted multivariable analysis to produce adjusted β coefficients. Because the outcome was log transformed, the β coefficients were retransformed ([exponentiated β coefficient − 1] × 100) to represent the percentage of change in the outcome per unit change in each independent variable. We explored collinearity between correlates using criteria of a condition index of 30 or higher to consider the presence of multicollinearity (18). Linearity with the outcome for each potential correlate was explored and, if necessary, correlates were either transformed or modeled by using indicator variables. Maternal physical activity was modeled by using quartiles, with the highest quartile representing the most total counts per day of maternal activity.

We created separate models for each of the domains of the potential correlates (demographic, biological, behavioral attributes and skills, social and cultural, and physical environment). For each of these models, we conducted a partial F test for the potential correlates only because site and wear time were kept in all models. If the P value was <.20, we used backward selection with partial F tests to remove variables in the domain with P values of ≥.20. Variables retained in these separate models were combined in the full model. To create the final model, we removed each domain from the full model with partial F tests using a P value ≥.20 as the criteria for removal. We considered variables with a P value of ≤.05 to be significant. We ensured that the final models met the assumptions of linear regression. For the model examining the outcome of sedentary time, we removed 8 observations from the multivariable model because they led to violations of the assumptions of linear regression in the final model. For the model examining MVPA, we removed 1 observation from the multivariable model for the same reason. These observations were removed because they had studentized residuals of greater than 3 or less than −3 in the final models. For consistency, these observations were also removed from the unadjusted analyses. We performed all analyses with Stata 11.0 (StataCorp LP, College Station, Texas).

The 337 children in the analysis sample (Table 1) had a mean age of 3.5 years (standard deviation [SD], 1.1); 58% were boys. The children were monitored for a mean (SD) of 6.3 (1.4) days and had 6.1 (1.3) hours per day of sedentary time (Figure 1), and 14.9 (9.5) minutes per day of MVPA (Figure 2). Of the participants, 61% of children were seen at the Durham study site and 39% at the Greensboro study site. The variable with the most missing information was presence of a chronic health problem in the mother (12% missing). All other variables had less than 3% of observations missing.

The 337 children in the analysis sample were not different from the 63 children enrolled in the KAN-DO Study without sufficient accelerometry data for inclusion in our analysis with respect to maternal age, maternal marital status, maternal education, and maternal ethnicity. The 337 children in the analysis sample were on average 0.5 years older than the 63 children not included (P < .001), more likely to be boys (58% included vs 44% not included; P < .05), and more likely to have a household income greater than $60,000 per year (59% included vs 45% not included, P < .05). Maternal race differed among the children in the analysis sample (77% white, 23% nonwhite), compared with those not included (63% white, 37% nonwhite; P = .02).

In minimally adjusted analyses (adjusting only for site and monitoring time), correlates of the child's amount of sedentary time were child's sex, child's time spent outdoors, and maternal activity (Table 2). In multivariable analyses, the final model contained only demographic, biological, and social and cultural variables. The only correlate of sedentary time in the final model was child's sex. Boys had less sedentary time than girls (P < .001).

In minimally adjusted analyses, correlates of the child's amount of MVPA were child's age, child's sex, household income, maternal education, and mother's report of child's time spent outdoors. The final model contained only demographic and physical environment variables. Correlates of MVPA were child's sex, child's age, household income, and child's time spent outdoors. Both child's age and child's time spent outdoors were modeled with squared terms because they exhibited quadratic relationships with the outcome. Older children engaged in more MVPA than younger children (P = <.001), and boys had more MVPA than girls (P < .001). Children who spent more time outdoors had more MVPA (P = .001). Children in households with a household income of at least $60,000 per year had more MVPA than children in households with a lower income (P = .005).