This research protocol was approved by the Research Ethics Board of the University of British Columbia and received Institutional Review Board approval from the Baylor College of Medicine.
Sample
The sample consisted of parents of 5-12 year old children who were living in Canada or the USA. Parents were recruited by an internet research polling firm (YouGovPolimetrix, US) from their web-based panel membership. Recruitment from a polling firm represented a cost-effective approach of obtaining a large representative sample over other sampling approaches. Recruitment occurred between November 2013 and February 2014. All panel members provided consent to be part of the panel and to participate in the survey. To be eligible, participants had to be the primary guardian of a 5 to 12 year old child. Participants were excluded if the child had a physical or learning disability that limited their child’s physically activity. Panel members were sampled to reflect the socio-economic and ethnic diversity of the two countries based on the 2012 US and 2011 Canadian census estimates. To reflect socio-economic diversity, country-specific household income cut-points (< 40th percentile, ≥40th to ≤80th percentile, and >80th percentile) were created and a corresponding percentage of participants (40%; 40%; 20%) were recruited into each group. To reflect ethnic diversity, participants were recruited based on the percentages of the largest ethnic groups for each country (White, Hispanic, Black, and other in the US and White, East/Southeast Asian, South/West Asian, and other in Canada). In addition, the sample was balanced between parents with younger (5-8 years of age) and older (9-12 years of age) children. In total, 134 parents (73 US and 61 Canadian) provided valid responses. Participants received 2000 points for their participation in this survey which could be redeemed as cash or gift cards valued at about $5 USD/Cdn.
Questionnaire
Parents were asked to respond to a series of screening and socio-demographic questions to ensure eligibility. Those who met the eligibility criteria were asked to respond to the following four questions based on their youngest or oldest child within the age criteria (selected by the data collection program): 1) What sorts of things do you do to encourage your child to be physically active?; 2) What rules or guidelines do you have that may encourage your child to be physically active?; 3) What sorts of things might you do that may unintentionally affect your child from being physically active?; 4) Thinking about other parents with children of the same age, what things do they do that may discourage their children from being physically active? The fourth question asked parents to respond generally about other parents to avoid socially desirable responses. Each of the four questions was open-ended and parents could provide up to ten 160 character responses per question. A character limit was set to limit each response to one practice and to encourage parents to provide multiple responses (i.e., identify several parenting practices). Parents were prompted to expand on their responses if they provided short answers (<50 characters). As the questions were open-ended, parents were blocked from completing the questionnaire on their mobile devices which may limit their ability to type more detailed responses. The online survey was piloted among 25 Canadian parents using cognitive interviewing techniques [14]. Responses from questions 1 and 2 were grouped and categorized as encouraging parenting practices and responses from questions 3 and 4 were grouped and categorized as discouraging parenting practices.
Coding of responses
Detailed information about the development of the coding scheme used in this study can be found elsewhere [6]. Briefly, the coding scheme was initially developed to code 74 published questionnaires/instruments designed to assess physical activity parenting practices, which included 608 items [6]. The coding scheme was developed based on a review of these published constructs, associated items and was informed by conceptual frameworks of parenting practices for physical activity and nutrition [15, 16]. The coding scheme consisted of 6 broad domains, 14 dimensions, and 1 to 5 sub-dimensions [6]. The domains (dimensions in parenthesis) were structuring of the activity environment (monitoring, structure of the environment); emotional support (expressing positive emotions, parental encouragement); parental control (expressing negative emotions, lack of parental control, pressure to be active, restriction, rewards and discipline); informational support (modeling, teach/reason); autonomy promotion (autonomy support, co-participation); and tangible support (logistic support/facilitation) (see our previous study for the full coding scheme and detailed definitions) [6]. Each item from the 74 instruments was assigned to a dimension and sub-dimension in order to group similar items. The list of items was further reduced using a winnowing process that grouped similar items into a statement that best captured the parenting practice (608 items winnowed to 100 parenting practice statements). For example, three published items asking about parent co-participation in physical activity (“During a typical week, how often has a female adult done physical activity with you?”; “How often do you exercise with one or both of your parents?”; and “How often do you do the following activities together as a family with at least one adult family member – Play sports?”) were reduced to the following generic statement “Participate in [physical activity, sports, exercise] or play active games with my child.”
Each parent response to the four questions on the online questionnaire was coded to a parenting practice from the consolidated list of 100 parenting practice statements. For example, when parents were asked what they did to encourage their child to be physically active, one parent responded “Ask if she wants to join sports at school, usually she does not” which was coded to the item “I allow my child to choose whether s/he participates in sports or vigorous physical activity in his/her free time.” As the item was meant to describe a concept, responses in a negative direction could have been coded to an item in the opposite direction, if that already existed in the coding scheme. For example, one parent’s response to a question asking what the other parents did to discourage their child from being physically active responded “Not participating with their kids at all” which was coded to the item “I participate in [activity type] with my child.” If the parent response had elements of two distinct items, then the response was coded to both items. Coding was conducted independently by two researchers, discrepancies were discussed, and a consensus was reached. A third researcher reviewed the coding and discussed discrepancies with the other two researchers. A new code was created for any unique parenting item that did not appear in the literature. If more than one response from the same parent was coded to the same item, the additional responses were removed to avoid repetition.
Analysis
Coded responses to the encouraging and discouraging questions were ranked by domain and dimension. To assess whether there were differences in responses by country, parental sex, age of child, or household income, a log-linear analysis with iterative proportional fitting was conducted. Log-linear analysis is an iterative process which can test for higher order (e.g., three-way) associations among categorical variables. Analyses were conducted on the above four sets of variables with each set including the coded domain (6 categories), a dichotomous variable indicating whether the response was encouraging or discouraging (2 categories), and one of country (USA, Canada), parental sex (male, female), age of child (5-8, 9-12), or household income (below median, above median). All two- and three-way interactions were assessed within each set. For each set, a saturated model was formed and each higher order term was sequentially removed to examine the goodness-of-fit. Model fit was based on Pearson’s Chi-square statistic (χ2) and the deviance statistic (G2). Models with significant goodness-of-fit statistics suggest poor fit. The most parsimonious model (model with least number of interaction terms) was chosen as the best fitting model for each set. A p-value of <0.05 was considered significant. Stata (version 13.1, College Station, Texas) was used for all analyses.