A variety of community outreach methods (as above in section 3.3.1) were conducted to disseminate advertising material to potential participants. Parents who were interested in taking part in the study were asked to contact the PhD candidate. Upon contact, parents were fully informed of the research protocol and, if still interested in participating, parents were screened for eligibility (further details are in section 3.3.2). Eligible participants were enrolled in the study and sent a copy of the participant information sheet (PIS), the consent form (CF), and the pre-intervention measures (either online or in hard copy depending on the parent’s preference). The PIS and CF are displayed in Appendix E. After pre-intervention measures were completed participants were randomly assigned to one of the two intervention conditions (for further details see section 3.6.3). For two-parent families, if both parents consented to participating, this occurred when the measures from both parents were received. An allocation call or email was made to inform each family which condition they were allocated to. For families who were allocated to the multiple exemplar condition, parents were asked to nominate the two additional groups they would like to attend from the three options (Fighting and Aggression, Doing Chores, and Building Self-esteem).Post-intervention measures were administered following the intervention. For families allocated to the multiple exemplar condition, post-intervention measures were administered immediately after the end of the intervention. For those in the single exemplar condition, postintervention measures were administered at the equivalent time of post-intervention for the multiple exemplar condition (approximately 4 weeks after the session). The 6-month followup measures were administered approximately 6-months after post-intervention.
The study protocol was approved by The University of Auckland’s Human Participants Ethics Committee (Reference 2011/360). An extension to the ethics approval was granted on 12/02/2014 (Reference 7431, expiry date 20/07/2017) to allow for further data collection with the aim of increasing the sample size.
A simple randomisation procedure was used. Simple randomisation involves allocating to condition using a computer generated list of random numbers on a random number table (Torgerson & Torgerson, 2008). Simple randomisation was used in the current study as it is easy to undertake (Torgerson & Torgerson, 2008) and was deemed the best design for comparing the effects of a two-armed trial. Randomisation occurred at the level of individual target children and allocation to condition was in sequence of completion of pre-intervention measures. Randomisation to condition was conducted after pre-intervention measures had been completed to reduce bias that can occur if randomisation happens prior to administration of pre-intervention measures, such as discontinuation if a desired intervention condition was not allocated (Torgerson & Torgerson, 2008). A key limitation of simple randomisation is that an imbalance in the number of participants allocated to groups may occur. In order to reduce the impact of any potential imbalances, allocation to condition was stratified by the area of Auckland in which the participants resided (Central or West), and household configuration(one-parent vs. two-parent household) and randomisation occurred with each strata. The allocation was stratified in this way to ensure that there would be a balance of participants in both conditions at each site and to reduce the possibility of single parents being overrepresented in one condition.To conduct randomisation, a spreadsheet was set up which was separated into two columns: Central and West Auckland. Within each area, the spreadsheet was further divided into two columns based on household configuration. Computer generated lists of random numbers were downloaded from random.org and a list of numbers was entered into each column (e.g., Central one-parent household families, Central two-parent household families). As described above, randomised occurred in sequence of completion of pre-intervention measures and participating families were allocated to condition within each strata. Allocation to condition was conducted by individuals independent of the study to ensure there was no bias in allocation.
The Triple P Discussion Groups were used in the current study and are an example of a low-intensity topic-specific group parenting programme (Sanders, 2012). The Triple P Discussion Groups are a level 3 Triple P programme (see Appendix A for a detailed description of the Triple P multilevel system of support) and are a recent addition to the Triple P system (Sanders, 2012). The groups are two-hour interventions for parents looking for specific advice about a common child behaviour or developmental issue (e.g., disobedience, fighting and aggression). During the session, parents are taught the principles of positive parenting, including the use of positive encouragement and attention to motivate children to learn new skills and show desirable behaviour, and the use of consistent, assertive discipline techniques to manage misbehaviour and promote self-control. These strategies are alternatives to coercive and ineffective discipline strategies and are tailored to focus on the specific topic that is being addressed. The information is presented in a variety of ways: parents watch video-modelling of behaviour management strategies, complete a variety of exercises, and receive assistance in developing a plan to prevent and effectively manage difficult child behaviours related to a specific topic. Parents are also given the opportunity to practice their skills in session, and discuss their experiences and implementation plans with other group members. A group workbook is given to attendees to reinforce material presented in the session.Parents from Central Auckland attended sessions held at the Faculty of Education campus in Epsom. For parents who resided in West Auckland, sessions were held at a community based setting in Titirangi. Among two-parent families, both mothers and fathers were encouraged to attend. Sessions were held on weeknight evenings from 7-9pm. In the current study, all sessions were delivered by the PhD candidate who is a trained accredited Triple P Discussion Group practitioner. Sessions were delivered according to the standard manual (Sanders & Turner, 2011).
Single exemplar condition
Parents in the single exemplar condition attended one two-hour parenting group on Dealing with Disobedience (see Table 3.4 for further details).
Multiple exemplar condition
Parents allocated to the multiple exemplar condition attended four two-hour group sessions. There were two compulsory topics (Dealing with Disobedience and Being a Positive Parent; see Table 3.4) that all parents were asked to attend. Families were then asked to attend two additional sessions and could choose from three options targeting other specific behaviours and concerns (Fighting and Aggression, Doing Chores, and Building Self-esteem). Parents allocated to the multiple exemplar condition completed the Dealing with Disobedience and Being a Positive Parent sessions before attending sessions on the additional topics. For parents allocated to the multiple exemplar condition, the groups were held weekly at the same time; thus, attendance at four sessions occurred over a four or five week period depending on which additional sessions they chose.
Craig et al. (2008) suggest that the fidelity of the programme and quality of implementation is an important area to study and Flay et al. (2005) also state that, when evaluating an intervention, it is desirable to measure programme implementation. Intervention fidelity for the Triple P Discussion Groups was measured using session checklists that were developed by the programme developers for the purpose of assessing intervention fidelity (Sanders & Turner, 2011). In the current study, the sessions were delivered by the PhD candidate, who is a trained accredited Triple P practitioner. Budget constraints did not allow for a Triple P practitioner independent of the study to deliver the sessions. However, as parents completed outcome measures in their homes and the DGSQs were anonymous, it is unlikely that this influenced their responses. Furthermore, it was not possible for the PhD candidate to be blind to participant condition and therefore potential bias cannot be eliminated. The session checklists were completed at the end of each Triple P Discussion Group and the proportion of the content covered was calculated to determine if each group session was delivered according to the standard manualised protocol. In order to determine the reliability of the practitionerrated intervention fidelity, the content covered in the sessions was also rated independently by a second PhD student who was a trained accredited Triple P practitioner using the same session checklist. All Triple P Discussion Groups were video or audio recorded and approximately 30% of the recorded sessions were checked at random by the second rater. The independent second rater was blind to intervention condition (e.g., single vs multiple exemplar). The proportion of content covered was calculated. In addition, agreement between the practitioner completed and second-rater completed checklists were examined to determine the extent of inter-observer agreement on intervention fidelity.The PhD candidate delivered 21 sessions using the standard manualised protocol (Sanders & Turner, 2011). Frequent clinical supervision was provided to the PhD candidate to ensure that the intervention was being delivered to a high level of quality and to discuss any process issues. Adherence to the intervention protocol was high. Practitioner completed ratings of the proportion of content covered in the sessions ranged from 84.6% to 100.0% (M = 92.5%, SD = 0.04) for the 21 sessions (seven Dealing with Disobedience groups, four Being a Positive Parent groups, three Fighting and Aggression groups, two Doing Chores groups, and four Building Self-esteem groups). Inter-rater reliability was obtained for eight randomly selected group sessions. The ratings of the proportion of the intervention protocol covered in the sessions ranged from 84.1% to 100.0% (M = 91.5%, SD = 0.07) according to the second independent rater. Inter-rater agreement between the adherence ratings provided by the practitioner and those provided by the independent rater was high (ranged from 87.5% to 97.4%, M = 92.0%, SD = 0.04).
Data from the participants were collected in two ways: via online and hardcopy versions of the measures. Data collected through hardcopy versions of the measures were entered into SPSS. The entry of approximately 30% of data collected through hardcopy versions was checked by an individual independent from the research. Only one error in the data entry was found, indicating high confidence in the accuracy of data entered. Data collected online were downloaded and merged with the data collected through hardcopies.Prior to conducting the main analyses, preliminary analyses were conducted. First, analyses were conducted to identify any differences between those who were randomised (n = 77) and those who dropped out prior to randomisation (n = 12) on disruptive child behaviour, child age, and single parent status reported at screening. There were no significant differences between those who were randomised and those who dropped out before randomisation on measures of disruptive child behaviour, child age, and single parent status at screening. This indicates that families who continued with the study were not demographically different from those who dropped out prior to randomisation.An intention-to-treat (ITT) analysis was used in the current study to examine the effects of the two intervention conditions. An intention-to-treat analysis includes all participants who were randomised in the analysis, regardless of whether the intervention was received and whether withdrawal occurred (Gupta, 2011). Therefore, in an ITT analysis all randomised participants are analysed. An ITT analysis preserves sample size and statistical power, is seen to reflect the real-world situations where withdrawal and protocol deviation is common, and allows for greater generalisability (Gupta, 2011). Furthermore, ITT analyses prevent bias that may be associated with withdrawal and protocol deviation, such as those who withdraw may report poorer outcomes than those who adhere to protocol (Montori & Guyatt, 2001). An ITT has also been recommended for RCT designs in the CONSORT statement (Moher et al., 2010). However, one limitation of an ITT analysis is that intervention effects may be diluted by including those who did not comply with the protocol (Gupta, 2011). In the current study, all randomised participants were included in the ITT analysis which consisted of 75 mothers (single exemplar condition: n = 34, multiple exemplar condition: n = 41) and 58 fathers (single exemplar condition: n = 27, multiple exemplar condition: n = 31). Per protocol analyses were also conducted to examine the effects among those who completed measures at more than one time point.First, missing variable analyses and Little’s Missing Completely at Random (MCAR) tests were used to identify the extent and pattern of missing data. Missing data can be classified in three ways: Missing Completely at Random (MCAR), Missing at Random (MAR), and Missing Not at Random (MNAR, van Buuren, 2012). If data are MCAR, the likelihood of a value being missing is the same for all cases (van Buuren, 2012). MAR is more general than MCAR where the probability of missingness is the same within defined groups. If data are neither MCAR nor MAR, then MNAR holds where the probability data are missing varies for unknown reasons. Acock (2005) explained that data MCAR are rare in family studies and usually data MAR are expected. The findings from missing variable analyses and Little’s MCAR tests are reported in section 4.2.Multiple imputation (MI) was then used to impute all missing data. Therefore, for those who did not complete post-intervention and 6-month follow-up measures, their data were imputed using MI. MI is a technique for dealing with missing data that differs from single imputation by imputing more than one possible value in a distribution (Rubin, 1987). By pooling a number of plausible values, rather than a single imputed value, MI results in an improved parameter estimate (Acock, 2005). MI is currently considered to be the best method for dealing with missing data (Tabachnick & Fidell, 2013; van Buuren, 2012). SPSS can conduct MI using two methods: linear regression and predictive mean matching. In the current study, the predictive mean matching method for MI was used. Predictive mean matching is an easy to use method (van Buuren, 2012), preserves the observed distribution of variables (Barnes, Lindborg, & Seaman, 2006), and provides some protection against violations to the assumptions of normality (Barnes et al., 2006). A further advantage of predictive mean matching is that only possible values within the observed data range are imputed (Barnes et al., 2006; van Buuren, 2012). Acock (2005) reported that from Schafer’s research five imputations have been found to be adequate when up to 30% of the values are missing at random. Five imputations were used in the current study resulting in five complete data sets. As the patterns of data were likely to differ according to intervention condition and parent gender, missing data were imputed separately for mothers and fathers in each condition and then merged together.After MI was completed, the total scale scores were then calculated and the internal reliability of each variable was examined using Cronbach’s alpha (the relevant statistics are presented for each measure in the Measures section above). Each variable was then screened for outliers and for violations to the assumptions of normality (skewness and kurtosis). Univariate outliers were examined by converting each participant’s score for each variable into z-scores (Field, 2013). Field’s (2013) guidelines for identifying outliers using z-scores were used: if z > 1.96 = potential outlier (about 5% of the scores in a normal distribution would have a z-score greater than 1.96), if z > 2.58 = probable outlier (about 1% of scores in a normal distribution would have a z-score greater than 2.58), and if z > 3.29 = extreme outlier (no scores in a normal distribution would have a z-score greater than 3.29). Field (2013) recommends reducing outliers that have z-scores above 300 to three standard deviations from the mean. Details on univariate outliers found in the data are reported in section 4.2.The assumption of normality was checked by examining the skewness and kurtosis statistics of each measure at each time point. Skewness and kurtosis statistics were obtained through SPSS and converted into z-scores by dividing the test statistic by the standard error(Field, 2013). The obtained z-scores were compared to Field’s guidelines described previously to identify variables with significant non-normality problems. Variables that violated the assumption of normality and resulting transformations and decisions are described in section Multivariate normality was then checked for variables used in the multivariate analyses described below. Multivariate outliers were examined by comparing the Mahalanobis distance values to a critical value (Pallant, 2010). Pallant (2010) specified that the critical value for two dependent variables is 13.82 and Tabachnick and Fidell (2013) recommend removing multivariate outliers from the data set. Details on multivariate outliers found in the data are reported in 4.2.Preliminary analyses were then conducted using independent t tests for continuous variables and chi-squared tests of independence for categorical variables. When cell sizes were less than five, Fisher’s exact tests were used to calculate the exact probability that the statistic is accurate (Field, 2013). First, preliminary analyses were conducted to identify whether there were any significant differences between those who were randomised and those who dropped out before randomisation on disruptive child behaviour, child age, and single parent status reported at screening. Preliminary analyses were then conducted to determine whether there were any differences by condition in demographic variables and pre-intervention measures for families, mothers, and fathers. Independent t tests and chi-squared tests were also used to compare demographic variables and pre-intervention measures for those families, mothers, and fathers who completed post-intervention and 6-month follow-up measures and those who did not. In addition, paired t-tests were used to compare pre-intervention and post-intervention scores for mother-father pairs (n = 55) to explore whether mothers and fathers in the same family were reporting similar problems.
Short-Term and Long-Term Condition Effects
As SPSS is not currently able to automatically pool the results from the imputed data sets, the three step approach described in Acock (2005) was used. First, data were imputed five times. Second, the analyses were then run on each of the data sets separately. Finally, the results are pooled to give a single solution. Therefore, all analyses described below were conducted on each of the five imputed data sets and the results were pooled manually. The analyses examining the short-term and long-term condition effects were also conducted with the sample of mothers and fathers who completed post-intervention and 6-month follow-up measures and the results were compared to the ITT sample.To analyse change between pre- and post-intervention for the two conditions, a series of multivariate and univariate analysis of covariance (MANCOVA and ANCOVA) were used. A MANCOVA examines whether there are differences between groups when there is more than one conceptually related dependent variable whilst controlling for one or more covariates (Tabachnick & Fidell, 2013). An ANCOVA examines the difference between groups for one dependent variable whilst controlling for a covariate (Field, 2013; Pallant, 2010). In the current study, pre-intervention scores of the dependent variable were used as covariates (e.g., for the ANCOVA examining PS Total scores, post-intervention PS Total scores were the dependent variable and pre-intervention PS Total scores were used as the covariate). To examine the long-term effects of the two conditions, the series of MANCOVAs and ANCOVAs were repeated using 6-month outcome measures as the dependent variables. Preintervention scores of the dependent variable were again used as covariates. Prior to conducting the MANCOVAs and ANCOVAs the specific assumptions for these analyses were tested, e.g., multicollinearity, homogeneity of variance-covariance matrices, and linearity for MANCOVAs; homogeneity of regression slopes, the assumption of equality of variance, and linearity for ANCOVAs (Pallant, 2010).In the current study, MANCOVAs were used to examine the differences in postintervention scores for the two intervention conditions, after controlling for pre-intervention scores of the dependent variables. They were used for the following conceptually related dependent variables: disruptive child behaviour (ECBI Intensity Total and ECBI Problem Total), parenting self-efficacy (PTC Behavioural self-efficacy and PTC Setting self-efficacy), target and non-target negative child behaviours displayed on weekdays and weekend days (PDR Weekday Total and PDR Weekend Total), and inter-parental conflict (PPC Extent Total and PPC Problem Total). When significant multivariate effects were found, the univariate analyses were examined to determine which dependent variables contributed to the multivariate effect. For the MANCOVAs, multivariate tests of significance were reported using Pillai’s trace as it is more robust for small samples sizes, unequal n’s, and violations of assumptions (Pallant, 2010).ANCOVAs were used to determine differences in post-intervention scores for the two intervention conditions for the following unidimensional measures: parenting practices (PS Total score), parenting experiences (PES Parenting Experiences Total), child psychosocial problems (SDQ Total Difficulties), parental mental health (DASS-21 Total), partner support (PES Parent Support Total), and partner relationship quality (RQI Total).Effect sizes were calculated to report the pre- to post-intervention effect of the multiple exemplar condition over the single exemplar condition. This was done using the following equation: the difference in mean pre- to post-intervention scores for the multiple exemplar condition minus the difference in mean pre- to post-intervention scores for the single exemplar condition, divided by the pooled pre-intervention standard deviation (Morris, 2008). According to Cohen (1992), an effect is meaningful but small in size if d 0.20 ≤ 0.49, medium in size if d 0.50 ≤ 0.79, and large if d ≥ 0.80. A similar procedure was used to calculate effect sizes from pre-intervention to 6-month follow-up. The results of the analyses examining the short- and long-term condition effects are reported.
Change Over Time by Condition
In addition to the MANCOVAs and ANCOVAs, analyses were conducted to also examine change over time in outcome measures for each of the conditions separately as the comparison group in the current study was an active intervention. To examine change in outcome measures over the three time points (pre-intervention, post-intervention, and 6-month follow-up), a series of doubly multivariate repeated measures analyses (see Kerr, Hall, & Kozub, 2002) and one-way repeated measures univariate analysis of variance (ANOVAs) were used. These analyses were conducted separately for each condition. Doubly multivariate repeated measures analyses were used for conceptually-related dependent variables described above in section 3.7.2 and one-way repeated measures ANOVAs were used for unidimensional measures.If a significant main effect for time was found, follow-up Bonferroni pairwise comparisons were used to examine change from pre- to post-intervention, and pre- and post-intervention to 6-month follow-up. Pre- to post-intervention effect sizes were calculated to examine the change over time for each outcome measure. For this effect size, the following formula was used for each condition separately: the difference in mean pre- to post-intervention scores divided by the pooled pre- and post-intervention standard deviation (Cohen, 1992). The results of these analyses are reported in section 4.5.
Statistically Reliable and Clinically Significant Change
The Reliable Change Index (RCI, Jacobson & Truax, 1991) and the clinical cut-offs were used to examine statistically reliable and clinically significant change from pre- to postintervention for the ECBI Intensity Total, the ECBI Problem Total, and the PS Total scores for each condition. The RCI provides an indication of whether the extent of change is statistically reliable or whether the extent of change is likely due to variation in inaccurate measurement (Jacobson & Truax, 1991). To determine whether reliable change has occurred, an index is calculated based on the difference between pre- and post-intervention scores and the standard error of that difference. If the index is larger than 1.96, the differences between scores is likely to reflect real change. Based on the index, each participant’s pre- and post-intervention difference in ECBI Intensity Total, ECBI Problem Total, and the PS Total scores were grouped into three categories: reliable improvement, reliable deterioration, or no reliable change.The clinical cut-offs were also used to determine if movement in and out of the clinical range had occurred between pre- and post-intervention for these measures among parents in each condition. Based on the scores on the ECBI Intensity Total, the ECBI Problem Total, and the PS Total, participants were grouped into four categories on the level of clinical change. The four categories were: clinically significant change (scores above the clinical cut-off at preintervention and below the clinical cut-off at post-intervention), did not achieve clinical change (scores above the clinical cut-off at both pre- and post-intervention), worsened (scores below the clinical cut-off at pre-intervention and above the clinical cut-off at postintervention), and not in clinical range (scores below the clinical cut-off at both pre- and postintervention).Chi-squared tests for independence were used to examine whether there were any significant differences in the distribution of statistically reliable and clinically significant change by intervention condition. Where there were fewer than five cases in a cell, Fisher’s exact tests were used. See section 4.6 for the results from the statistically reliable and clinically significant change analyses.
Chapter 1 Introduction
1.1 Overview of Thesis
Chapter 2. Literature review
2.1 Overview of Chapter
2.2 Conduct Problems
2.3 Intervention for Child Conduct Problems
2.4 The Issue of Poor Parental Mental Health
2.5 Summary of the Literature and Aims and Hypotheses of the Current Research
Study One: Enhancing Intervention Outcomes of Low-Intensity Parenting Groups for Parents of Primary School Aged Children Through Generalisation Promotion Strategies
Chapter 3. Method
3.1 Overview of Chapter
3.2 Trial Registration
3.7 Data Analysis
Chapter 4. Results
4.1 Overview of Chapter
4.2 Preliminary Analyses
4.3 Short-Term Condition Effects
4.4 Long-Term Condition Effects
4.5 Change Over Time by Condition
4.6 Statistically Reliable and Clinically Significant Change at Post-Intervention
4.7 Participant Attendance and Satisfaction
Chapter 5. Discussion
5.1 Overview of Chapter
5.2 Summary and Discussion of Key Findings
5.4 Future Research
5.5 Implications for Practice
5.6 Key Contributions of the Study
Study Two: Enhancing Intervention Outcomes of Low-Intensity Parenting Groups by Simultaneously Addressing Parenting and Parental Mental Health
Chapter 6. Method
6.1 Overview of Chapter
6.2 Study Registration
6.7 Data Analysis
Chapter 7. Results
7.1 Overview of Chapter
7.2 Preliminary Analyses
7.3 Change Over Time in Outcome Measures
7.4 Statistically Reliable and Clinically Significant Change at Post-Intervention
7.5 Analysis of Interviews
7.6 Participant Attendance and Satisfaction
Chapter 8. Discussion
8.1 Overview of Chapter
8.2 Summary and Discussion of Key Findings
8.4 Future Research
8.5 Implications for Practice
8.6 Key Contributions of the Study
Chapter 9. Conclusion
9.1 Overview of Chapter
9.2 Major Conclusions
9.3 Overall Future Directions
GET THE COMPLETE PROJECT
Low-intensity topic-specific group parenting programmes