All analyses were conducted using IBM SPSS Statistics version 23. Outliers for the pedometer data were identified as days with <1,000 or >30,000 steps/day for children and <1,000 or >25,000 for adults were set as missing . Outliers for the child PA questionnaire (>6 h/day), and for the other continuous variables (? ± 3.29 SD) were truncated .
Pearson unit–moment correlations were go to test new unadjusted bivariate dating between parents’ and you can children’s PA dating as the measured from the pedometers
Of the 28 variables included in this study, 83 % were missing on at least one value (number of non-missing values for each variable is available in Table 1). Across cases/participants, 55 % were missing on at least one variable, and across the entire dataset, 13 % of the values were missing. Missing and non-missing cases were compared for variables with >10 % missing data. Significant (p < .05) or marginally significant (p < .10) differences existed on parental BMI for parents' and children's steps/day. Importantly, families who participated in the initial assessment and those that returned the pedometers did not differ on parent self-reported leisure time MVPA (t = ?.67, p = .50) or children's parental-proxy reported PA (t = ?.38, p = .38). We therefore assumed at least a partial missing at random mechanism and imputed all of the missing data (including all covariates, predictor variables, criterion variables) using multiple imputation in SPSS. This procedure uses the fully conditional specification method and imputes data using linear regression for continuous variables and logistic regression for binary variables. We used 100 iterations, which resulted in 100 separate datasets . Relevant variables from the wider dataset (i.e., screen time, aerobic fitness, grip strength, dog ownership, walkability of neighborhood) were included as auxiliary variables.
I along with checked out which relationships separately by-child and you can mother intercourse, man and you may mother lbs standing, sex homogeneity, lbs position homogeneity, mother training, house earnings, and you can city-peak SES. Linear regressions were utilized to check the research questions and you will partial r shown effect size. Cohen’s required impact brands out of brief = .10, typical = .29, high = .fifty were utilized so you can translate how big is outcomes. New covariates for everyone analyzes were guy age, sex, and weight updates; father or mother gender, lbs condition, and you may education; domestic income; area-top SES; and you will 12 months. Per research incorporated between eleven and you can 13 details. With regards to the IBM SPSS Statistics SamplePower 3, with eleven covariates (typical joint impact dimensions), that predictor variable (typical impact proportions) and you can a communicating term (quick impression dimensions), 413 players was in fact expected to choose effects during the power = .80 getting ? = .01. Hence we had been good enough pushed for everyone analyses.
To address research questions 1 (whether parents’ steps/day was related to children’s steps/day), a linear regression was run with children’s steps/day as the criterion variable and parents’ steps/day and covariates as predictor variables. Coefficients were deemed significant at p < .05. To address research question 2 (potential moderators of the parent–child step/day relationship), children's steps/day was entered as the criterion variable and parent's steps/day and covariates as predictor variables. One by one we tested potential interactions including parent steps*child gender, parent steps*parent gender, parent steps*gender homogeneity, parent steps*child weight status, parent steps*parent weight status, parent steps*weight status homogeneity, parent steps*parent education, parent steps*household income, parent steps*area SES. In the models where the gender homogeneity and weight status homogeneity interactions were tested, these variables were also included as main effects. Before creating the interaction terms, the continuous variables (i.e., parent steps, area-level SES) were centered on their mean . To control for the increased probability of finding a significant result due to running multiple tests, a more stringent significance level was applied (p < .01) to the interactions. For significant or near significant interactions, a simple slopes analysis was performed to determine the beta coefficients and p-values for each group. Beta coefficients for the simple slopes were calculated by hand using the pooled results . The pooled results did not provide sufficient information to calculate the significance of the slopes by hand so the p-value (set at p < .05) was computed using the initial dataset (i.e., before the multiple imputation). To address research question 3 (parent–child PA relationship as measured by questionnaires), children's proxy-reported PA was entered as the criterion variable and parent self-reported leisure time MVPA and the covariates were entered as predictor variables. Coefficients were deemed significant at p < .05.