The methods section gives an overview of the trial. For a detailed description, please see the study protocol [24].
Trial design
The ABC program was evaluated in a multicenter randomized waitlist-controlled trial, where parents were randomized to either (a) receive the ABC program directly or (b) join a waitlist-control group receiving the intervention after approximately six months. All parents completed a pre-measurement questionnaire (at baseline), a questionnaire two weeks after the intervention (post-measurement), and a questionnaire six months post-baseline (follow-up measurement). Ethical approval for the trial was obtained from the Regional Ethical Review Board in Stockholm (Dnr: 2012/93-31/5).
Participants, randomization, and setting
Parents with children aged 3–12 were recruited to the trial during two waves, spring (February–March) and fall (September–October), in 2012. During the recruitment phase, all parents were invited to an informational meeting on local premises. Parents were informed orally about the trial by research staff, while trained group leaders informed them about the ABC program. They received written information about the trial, and those interested in participating signed an informed consent. All parents who agreed to participate completed the pre-measure assessment (a questionnaire). The intention was to include 300 parents each in the intervention and control groups, according to a sample-size calculation. Parents were randomized by the researchers at the individual level, at a ratio of 1:1. Randomization was performed for each municipality/city district using the random-sampling function in IBM SPSS Statistics for Windows, Version 20 (IBM Corp, Armonk, NY). Couples were randomized as a single unit. In total, 621 parents were recruited to the trial. See the flowchart (Figure 1) for details on the enrollment of the parents.
Recruitment of parents and implementation of parent groups were conducted locally in 11 municipalities and city districts in the County of Stockholm, Sweden. The settings for recruitment were maternity health services, child health services, preschools, and schools. Strategies for recruiting parents included: advertising in the local press and on websites, contacting parents personally, sending letters to parents, and showing a specially produced ABC promotional video at local supermarkets. The most common settings for hosting the ABC groups were schools and preschools; however, family centers and other community facilities were also frequently used.
Measures
Parental questionnaires were used to assess parental self-efficacy, child health and development, and the baseline characteristics of the study population.
Parental Self-Efficacy (PSE) was measured using a 48-item questionnaire. The development of this questionnaire was influenced by another measure, Tool to Measure Parental Self-Efficacy (TOPSE) [25]. The measure comprises eight subscales: positive emotion, being with your child, empathy, guiding, rules, pressures, acceptance, and experience. Parents rated statements such as, “I can show my child affection” and “I keep calm when my child misbehaves” on an 11-point Likert scale, ranging from 0 (completely disagree) to 10 (totally agree). This resulted in a total score between 0 and 480, where a higher score was equivalent to a higher level of PSE. The construct validity of PSE was explored via confirmatory factor analysis. The analysis showed that the fit of the model to the data was acceptable (RMSEA = .072). The internal reliabilities (Cronbach’s alpha) for the full PSE scale in the present trial were .94 at baseline, .93 at post-measurement, and .94 at follow-up measurement.
Child Health and Development (CHD) was measured using a 35-item questionnaire. The development of CHD was based on an established health-related quality of life instrument [26] and was tested on data from previous pilot studies of the ABC program (n =405). The questionnaire measures parents’ perceptions of their child’s physical and mental health, emotional development, independence, family relations, and social competence. Parents rated questions such as, “How would your child describe (s)he is feeling in general?” on a 5-point scale, yielding a total score between 35 and 175, where a higher score was equivalent to better child health status. The CHD was validated using pilot-study data and baseline data from the present trial. The confirmatory factor analysis showed that the fit of the model was acceptable (RMSEA = .074). Internal reliabilities in this trial were .93 for the complete scale at baseline, .92 at post-measurement, and .92 at follow-up measurement.
The questionnaire also included questions about the age and gender of the focus child, as well as the parent’s country of birth (i.e., born in Sweden/not born in Sweden), educational level (i.e., university-level/not university level), positive mental health, and number of children in the family (i.e., one child/several children). These variables were included in the study to test for moderators. Parent’s mental health was measured using the General Health Questionnaire (GHQ) [27], where the six positively phrased questions were used to measure positive mental health [28]. An example of a positively phrased question was, “Have you been feeling reasonable happy, all things considered?” [28]. Each of the six items was rated on a four-point Likert scale (0–3), resulting in a total score ranging from 0–18. A higher score was equivalent to higher positive mental health.
Intervention
ABC consisted of four 2.5-hour structured sessions given to parents every other week. Components included in the sessions were: positive attention and warmth, parent–child time and child-directed play, positive parenting strategies, and consistent parenting. The sessions consisted of discussions and short films, while role-plays exercises were used to facilitate the discussions. Each ABC group was run by two trained group leaders, and groups within the trial contained seven parents on average. After approximately 2–3 months, a booster session was offered to parents; during the trial, this was offered after the six-month post-baseline measurement. The booster session included review from the four previous sessions, plus an introduction to one of three new topics (e.g., “boys and girls,” “sibling relations,” “teens”). The program targeted an important protective factor for children, namely, the parent–child relationship [1], and aimed to promote children’s development. The ABC program has been described thoroughly in other sources [24].
Statistical analyses
Chi-square tests and t-tests for independent groups were performed to examine baseline differences between the intervention and control groups. The same tests were also used to examine differences between parents who did not complete the post- and follow-up measurements and parents who did.
Multilevel linear modeling (MLM) with a repeated-measures design was used (SPSS, mixed models) to evaluate the effectiveness of the program. There are several advantages to MLM, compared with other analyses, such as the lack of requirement for complete data across time points and the possibility to test individual differences in growth curves [29]. In a repeated-measures study design, individual scores are nested within individuals and might in turn be nested within, for example, households [30]. Nested data are more likely to correlate within the group. For example, responses from parents in the same family unit are more likely to correlate highly, compared with responses by parents in general.
A three-level model was run to account for the 621 individuals being nested within three measurements, and that 220 of the included individuals were nested in 110 households. Time-related variables were constructed to manage the nonlinear growth trajectories in both primary outcome measures. There are different ways to code the time-related variable when a growth trajectory is found to be nonlinear. One way is to use a quadratic time variable (where the three measurement points, i.e. baseline-, post-, and follow-up measurement, are coded as 0, 1, 4), which captures any fluctuations in the rate of change across time points [31]. Another way is to code the first measurement occasion (the baseline) as 0 and the third measurement occasion (the follow-up) as 1. The growth that occurs over the entire trend is captured when using this approach [31]. The code for the second measurement point (the post-measurement), is in this approach, found by generating a variety of specifications and is determined by the best model fit [31]. The second measurement point is an intermediate point of the growth trajectory, and the code shows if the growth trajectory is close to being linear (i.e., if it is close to .5). Both the quadratic time-variable approach and 0–1 approach were tested on our data. The approach of coding the time variables as 0–1 gave the best model fit and was therefore selected. The three measurement points were coded as 0, .95, and 1 for PSE, while CHD was coded as 0, .85, and 1. The model included an interaction between time and condition. The intercept and time-related variables were used as random effects in the models, and best model fit was achieved with unstructured covariance type.
Effect sizes (η2) were calculated by subtracting the residual variance of the larger model from the residual variance of the intercept model, and then dividing the sum with the residual variance of the intercept model (using models that did not include random effects of time) [29]. Adopting the guidelines of Cohen [32], .02 represents a small effect size, .13 a medium effect size, and .26 a large effect size.
Another advantage of MLM is that the model can be adjusted simultaneously for the effects of numerous covariates. Several MLM analyses were run to evaluate the potential moderating effects, including interactions between condition, time, and the potential moderator (e.g., child age and gender, parent’s country of birth, educational level, positive mental health, number of children in the family). As a second step, all significant interactions were included in a final model, with scaled identity used as covariance type. For the moderation analyses, the variables of child age and positive mental health were centered on the grand-mean.
Intention-to-treat analyses were conducted, and IBM SPSS Statistics for Windows, Version 22 (IBM Corp, Armonk, NY) was used for all statistical analyses. The alpha level was set to < .05.