©   Access Economics Pty Limited

This work is copyright.  The Copyright Act 1968 permits fair dealing for study, research, news reporting, criticism or review.  Selected passages, tables or diagrams may be reproduced for such purposes provided acknowledgment of the source is included.  Permission for any more extensive reproduction must be obtained from Access Economics Pty Limited through the contact officer listed for this report.

Disclaimer

While every effort has been made to ensure the accuracy of this document and any attachments, the uncertain nature of economic data, forecasting and analysis means that Access Economics Pty Limited is unable to make any warranties in relation to the information contained herein.  Access Economics Pty Limited, its employees and agents disclaim liability for any loss or damage which may arise as a consequence of any person relying on the information contained in this document and any attachments.

Access Economics Pty Limited

ABN  82 113 621 361

www.AccessEconomics.com.au

Contents

Acknowledgements.i

Glossary.ii

Executive Summary.i

1          Introduction.5

2          Methodological overview..6

2.1      Definition of family functioning and the outcomes of interest.6

2.2      Data review..7

2.3      Literature review..10

2.4      Concepts and data underlying lifetime costing.10

2.5      Model construction..11

2.6      Cost benefit/cost effectiveness analysis (CBA/CEA) and the process for selecting interventions for analysis.17

3          Findings from the LSAC data investigation.20

3.1      LSAC variables.20

3.2      LSAC analysis.23

4          Findings from the ATP data investigation.30

4.1      ATP variables.30

4.2      ATP analysis.33

5          The costing model 37

5.1      Health outcomes.37

5.2      Productivity outcomes.43

5.3      Criminality outcomes.44

5.4      Lifetime costing framework.48

6          Cost benefit analysis.51

6.1      Communities for children (CfC).51

6.2      Positive Parenting Program (PPP).55

6.3      Reconnect.61

6.4      Shocking all variables – the value of PFF.68

7          Conclusions.70

References.75

Appendix A :  LSAC children and future addictive behaviours.83

Appendix B :  Literature review sources, ATP.86

Appendix C :  LSAC regression outcomes.89

Appendix D :  LSAC variable specification.102

Appendix E :  Mean and standard deviation of LSAC variables.106

Appendix F : ATP variable interpretation.109

Appendix G : ATP Regression outcomes.116

Appendix H : Mean and standard deviation of ATP variables.126

Appendix I :  Detailed costing methodology and tables.131

Charts

Chart 3.1 :  Explanatory power of regressions, by age.25

Chart A.1 :  Differences in cognitive development by socio-economic status.84

Tables

Table 2.1 :  Characteristics of family functioning domains.6

Table 2.2 : FF variables used for each intervention.15

Table 3.1 :  Regression constructs in LSAC.22

Table 4.1 : Regression constructs.31

Table 5.1 :  Summary of total costs of obesity (2010).38

Table 5.2 :  Summary of total costs of anxiety and depression (2010).40

Table 5.3 :  Summary of total costs of daily smoking (2010).41

Table 5.4 :  Summary of total costs of alcohol abuse (2010).42

Table 5.5 :  Summary of total costs of illicit drug use (2010).43

Table 5.6 :  Estimated effects* (%) of year 12 and undergraduate completion* on probability of participation and average earnings.44

Table 5.7 :  Social costs of crime by cost type (2010).45

Table 5.8 :  Net recurrent expenditure on criminal courts (2008-09) and % criminal court finalisations by court type  46

Table 5.9 :  Total real net operating expenditure on prisons and community corrections in Australia in 2008-09 47

Table 5.10 :  Discounted lifetime costs of adverse health outcomes (a) (2010 dollars).48

Table 5.11 :  Discounted lifetime costs of adverse productivity outcomes (a) (2010 dollars).49

Table 5.12 :  Discounted lifetime costs of criminality outcomes (a) (2010 dollars).50

Table 6.1 : Communities for Children target areas and related LSAC variables.53

Table 6.2 : Outcomes of Communities for Children variables used in this report.54

Table 6.3 : Outcomes of CfC as delivered over 2004-05 to 2007-08.55

Table 6.4 : PPP target areas and LSAC variables.57

Table 6.5 : Impact of large scale Queensland PPP trial 57

Table 6.6 : Impact of large scale Western Australian PPP trial 58

Table 6.7 :  Improvement in means scores of European PPP trial 59

Table 6.8 : Outcomes of PPP variables used in this analysis.59

Table 6.9 : PPP costs ($US).60

Table 6.10 : Outcomes of PPP as delivered over 2004-05 to 2007-08 in Queensland.60

Table 6.11 : ATP variables associated with the Reconnect Program..65

Table 6.12 : Estimate of effect size — young person’s perception of their ability to manage family conflict before Reconnect and now..66

Table 6.13 : Estimate of effect size — young person’s perception of their family’s ability to manage family conflict before Reconnect and now..67

Table 6.14 : Outcomes of the Reconnect Program, as delivered over 2004-05 to 2008-09.68

Table 6.15 : The value of PFF.68

Table 7.1 : Summary of costs and benefits of modelled interventions.71

Table B.1 :  Antisocial behaviour.86

Table B.2 :  Anxiety and depression.86

Table B.3 :  Smoking.87

Table B.4 :  Alcohol 87

Table B.5 :  Illicit drug use.87

Table B.6 :  Productivity.88

Table B.7 :  Overweight/obesity.88

Table C.1 :  Obesity B1 (ahs23c2).89

Table C.2 :  Obesity B2 (bcbmi).90

Table C.3 :  Obesity B3 (ccbmi).90

Table C.4 :  Obesity K2 (dcbmi).91

Table C.5 :  Obesity K3 (ecbmi).91

Table C.6 :  Productivity B1 (awlrnoi).92

Table C.7 :  Productivity B2 (bwlrnoi).92

Table C.8 :  Productivity B3 (cwlrnoi).93

Table C.9 :  Productivity K2 (dwlrnoi).93

Table C.10 :  Productivity K3 (ewlrnoi).94

Table C.11 :  Anxiety and depression B1 (apedsgc).94

Table C.12 :  Anxiety and depression B2 (bpedsef).95

Table C.13 :  Anxiety and depression B3 (cpedsef).95

Table C.14 :  Anxiety and depression K2 (daemot).96

Table C.15 :  Anxiety and depression K3 (eaemot).96

Table C.16 :  Antisocial B1 (apedsgc).97

Table C.17 :  Antisocial B2 (babitp).97

Table C.18 :  Antisocial B3 (caconda).98

Table C.19 :  Antisocial K2 (daconda).98

Table C.20 :  Antisocial K3 (eaconda).99

Table C.21 :  Addictions B1 (apedsgc).99

Table C.22 :  Addictions B2 (babitp).100

Table C.23 :  Addictions B3 (casdqta).100

Table C.24 :  Addictions K2 (dasdqta).101

Table C.25 :  Addictions K3 (easdqta).101

Table D.1 :  Standard LSAC variables.102

Table D.2 :  Combinations of LSAC variables.104

Illustrative distribution of categorical variables.107

Table G.1 : Underengagement predictor variables.116

Table G.2 : Completion of high school(a)(b).116

Table G.3 : Completion of high school (only) v University degree(a)(b).117

Table G.4 : Body mass index at 23-24 years(a)(b).118

Table G.5 : ATP anxiety/depression 2002 crosstabulation.120

Table G.6 : Logistic regression results — child age 19-20 years (year 2002)(a)(b).120

Table G.7 : Logistic regression results — child age 19-20 years (year 2002)(a)(b).122

Table I.1 :  Obesity prevalence rates and estimated obese people (number) in 2010.134

Table I.2 :  Anxiety and depression prevalence rates and estimated people with anxiety and depression (number) in 2010 134

Table I.3 :  Smoking prevalence rates and estimated current daily smokers in 2010.135

Table I.4 :  Prevalence rates for tobacco-caused diseases and conditions(a).136

Table I.5 :  Prevalence rates - drinking at risky-high risk(a) levels of long term health harm and estimated risky-high risk drinkers in 2010.136

Table I.6 :  Prevalence rates for recent use(a) of illicit drugs and estimated recent users in 2010 137

Table I.7 :  Offender age-gender prevalence profile and estimated offenders in 2008-09.137

Table I.8 :  Prisoner age-gender prevalence rates and estimated prisoners in 2009.138

Table I.9 :  Annual per-person costs of obesity - males (in 2010 dollars).138

Table I.10 :  Annual per-person costs of obesity - females (in 2010 dollars).139

Table I.11 :  Annual per-person costs of anxiety and depression - males (in 2010 dollars).140

Table I.12 :  Annual per-person costs of anxiety and depression - females (in 2010 dollars).140

Table I.13 :  Annual per-person costs of current daily smoking - males (in 2010 dollars).141

Table I.14 :  Annual per-person costs of current daily smoking - females (in 2010 dollars).141

Table I.15 :  Annual per-person costs of alcohol abuse - males (in 2010 dollars).142

Table I.16 :  Annual per-person costs of alcohol abuse - females (in 2010 dollars).142

Table I.17 :  Annual per-person costs of illicit drug abuse - males (in 2010 dollars).143

Table I.18 :  Annual per-person costs of illicit drug abuse - females (in 2010 dollars).143

Table I.19 :  Age-gender employment in the general population.144

Table I.20 :  Age-gender average weekly earnings(a) for the general population ($).144

Table I.21 :  Annual costs of year 12 non-completion ($ 2010).145

Table I.22 :  Annual costs of undergraduate degree non-completion ($ 2010).145

Table I.23 :  Age-specific undergraduate non-completion rates in 2007.146

Table I.24 :  Annual policing cost per offender ($).147

Table I.25 :  Annual court system cost per offender ($).147

Table I.26 :  Annual prison system cost per prisoner ($).147

Table I.27 :  Annual per-person societal costs of crime for males ($).148

Table I.28 :  Annual per-person societal costs of crime for females ($).148

Table I.29 :  Crime under-reporting multipliers and derived probabilities.149

Table I.30 :  Probabilities of court action on reported crimes.150

Table I.31 :  Court finalisation outcome probabilities.151

Table I.32 :  Custodial sentence probabilities in guilty verdict court cases.151

Figures

Figure 2.1 :  Approximate age of study cohorts and bridging the current information gap.8

Figure 2.2 :  Diagram of data map.9

Figure 2.3 :  Conceptual map for valuing costs of NFF.10

Figure 2.4 : Incidence versus prevalence approach.10

Figure 2.5 : Cost effectiveness analysis model map.16

Figure 2.6 :  CEA model pathway for interventions.17

Figure 6.1 : Value of PFF by benefit type, 2010 (total $5.4 billion), $bn and % total 69

Acknowledgements

Access Economics would like to acknowledge with gratitude the expert knowledge and inputs provided by members of the Expert Reference Group for this project.

Professor Ann Sanson

Department of Paediatrics, University of Melbourne, ARC/NHMRC Research Network Coordinator, Australian Research Alliance for Children and Youth

Brian Babington

Chief Executive Officer, Families Australia

Carol Ey

Branch Manager, Research and Analysis Branch, Department of Families, Housing, Community Services and Indigenous Affairs

Dr Lance Emerson

Chief Executive Officer, Australian Research Alliance for Children and Youth (ARACY)

Dr Marian Esler

Section Manager, Research Section, Family and Child Support Policy Branch, Department of Families, Housing, Community Services and Indigenous Affairs

Dr Matthew Gray

Deputy Director, Australian Institute of Family Studies

Megan Shipley

Research Section, Family and Child Support Policy Branch, Department of Families, Housing, Community Services and Indigenous Affairs

Paula Mance

Research Projects and Publications Section, Department of Families, Housing, Community Services and Indigenous Affairs

Rachel Henry

Research Section, Family and Child Support Policy Branch, Department of Families, Housing, Community Services and Indigenous Affairs

Professor Stephen Zubrick

Co-Director for Developmental Health, Curtin University of Technology

Access Economics would like to acknowledge in particular the staff of the Australian Institute of Family Studies (AIFS), who analysed the Australian Temperament Project (ATP) data and modelled those regressions.  Apart from Dr Gray, particular thanks go to Dr Ben Edwards and Dr Diana Smart.

Glossary

 Executive Summary

Access Economics was commissioned by the Australian Government Department of Families, Housing, Community Services and Indigenous Affairs (FaHCSIA) to quantify, in economic terms, the value of ‘goods and services’ provided by positive family functioning (PFF) and to conduct a cost benefit analysis (CBA) to establish the returns to government and society for investments made in supporting family functioning (FF).  This report follows a scoping study, also conducted by Access Economics, to establish the methodology for the project. The scoping study explains the equivalence of measuring the value of PFF as the costs of negative family functioning (NFF).  This study was overseen by a panel of experts.

Methods

FF is defined through a variety of domains – emotional, governance, cognitive, physical, intra-familial and social (Table 2.1).  Literature review revealed three broad areas of outcomes associated with FF.

■  Health outcomes were observed through the occurrence of anxiety and depression, obesity and substance abuse (smoking, alcohol and drug abuse) later in life.  These are associated with health expenditures, productivity losses (through lower workforce participation and premature death), other financial costs, and loss of quality of life (QoL) (measured in disability adjusted life years or DALYs).

■  Productivity outcomes were reflected in secondary and tertiary educational achievement completion, flowing on to impact lifetime earnings.

■  Social outcomes were primarily measured through negative manifestations – antisocial behaviour such as delinquency and crime, resulting in criminal justice system costs.

Two longitudinal studies — the Longitudinal study of Australian Children (LSAC) aged up to 9 years and the Australian Temperament Project (ATP) for older children were selected to analyse the relationship between FF and child outcomes.  Regression analysis was conducted to establish relationships between ‘transition’ health, productivity and social outcome variables in LSAC and FF variables.  The latter were selected on the basis of literature evidence, after controlling for other factors such as socioeconomic status (SES).  Transition variables were carefully selected to match similar or identical ATP variables. Further regression analysis was undertaken using the transition variables, together with ATP FF and control independent variables, to establish relationships with ATP ‘interim’ health, productivity and social outcomes in early adulthood.  The interim outcomes were then used to predict lifetime health, productivity and social costs, based on an extensive costing process utilising multiple data sources (Chapter 5).

Findings

The net present value (NPV) of benefits from intervening in childhood and adolescence to prevent poor outcomes later in life are substantial, despite the fact that such intervention incurs costs today but discounted benefits are realised a long time into the future. 

■  In total, the potential NPV of benefits to be realised is in the order of $5.4 billion per annum in 2010 dollars.  This can be considered the cost of NFF currently, or the value of PFF gains possible.  Over half these gains (53% or $2.9 billion) are productivity gains, with a further 22% ($1.2 billion) of the benefits deriving from savings from fewer addictions.  Fewer cases of anxiety and depression would save $0.6 billion (11%), while lower rates of criminality and antisocial behaviour would accrue $0.5 billion (10%).  A reduction in obesity would save $0.3 billion per annum (5% of the total) - Figure i.

Figure i: Value of PFF by benefit type, 2010 (total $5.4 billion), $bn and % total

Source: Access Economics calculations.  Note: Shares may not sum to 100% due to rounding.

There are also marked social and economic benefits if cost effective prevention programs can be identified and implemented.  This analysis has focused on three interventions selected on the bases of a range of criteria (section 2.6)

■  The Communities for Children program, targeting pre-school and primary school aged children, is one of the major Australian Government investments in families.  The program improves outcomes in various FF areas including hostile parenting, parenting self-efficacy, parent mental health, quality of the home learning environment, parental relationship conflict, child total emotional and behavioural problems, childhood overweight, receptive vocabulary achievement and verbal ability.  The benefit:cost ratio for this program was estimated as 4.8:1, a 377% return on investment.

■  The Positive Parenting Program is one of the best evaluated FF programs for younger children.  The program improves FF outcomes in parental sense of competency, the dyadic adjustment scale, the Strengths and Difficulties Questionnaire (SDQ) emotional and conduct scales, the Eyberg Child Behaviour Intensity score, parental depression, parental laxness, parental over-reactivity, and parental verbosity.  The benefit:cost ratio for this program was estimated as 13.8:1, a substantial 1,283% return on investment.

■  The Reconnect program targets an older cohort of children and was found to improve outcomes in school bonding and conflictual relationships, with proxied effect sizes estimated for attachment to parents and harsh parenting.  The benefit:cost ratio for this program was estimated as 1.8:1, an 81% return on investment.

Costs and benefits are summarised in Table ii.

Table ii: Summary of costs and benefits of modelled interventions

Source: Access Economics calculations.  * Costs estimated over 2004-05 to 2007-08 except for Reconnect which extends to 2008-09.

Many of the family ‘inputs’ incorporated in the analysis were found to be statistically significant explanators of child outcomes with the relationshipconsistent with that predicted by the literature. 

■  Obesity was explained by key drivers such as previous obesity, parental obesity, lack of child persistence, and parent-child conflict.

■  Anxiety and depression were dependent on previous emotional problems, difficult temperament, lower socioeconomic status (SES), harsh discipline, parental anxiety/depression, alienation from parents and lack of child persistence.

■  Smoking in young adulthood (19-20 years) was determined by previous smoking in adolescence, parental permission to smoke at home and a conflictual parent-teenager relationship.  Alcohol abuse (binge drinking) in young adulthood was dependent on teen bingeing, lack of parental monitoring, father drinking and initiating drinking at an older age (over 15 compared to 14 or younger).  Illicit drug use in 23-24 year olds was dependent on the child’s temperament, lack of parental monitoring, and mother smoking.

■  Predisposition to smoking, alcohol abuse and illicit drug use was established in early years by parental smoking, temperament, harsh and/or inconsistent discipline, poor family cohesion and parental anxiety depression.

■  Productivity was driven by previous learning outcomes, consistent discipline, temperament, socioeconomic status, parent education and, in adolescence, persistence, relationship quality/warmth, parental monitoring and a positive attitude to school.

■  Antisocial behaviour and outcomes were determined by child lack of persistence, previous social/conduct problems and, importantly, were largely influenced by early life FF variables such as poor family cohesion, harsh discipline, parental smoking and low SES, along with parental anxiety/depression and the child’s temperament.

The greatest value in this project has been primarily to showcase how a broad, quantitative approach to social policy evaluation can work.  With better quality data in the future, there is scope to refine and continue to develop the modelling and elaborate on findings further.  The scope of this project has been both ambitious and challenging but, we believe, the methods developed and many findings and insights are of global significance.  The novelty of the research inspires further work in this field that we hope can be used to triangulate these findings internationally as well as continue to enhance the evidence base in Australia.

Access Economics

1              Introduction

Access Economics was commissioned by the Australian Government Department of Families, Housing, Community Services and Indigenous Affairs (FaHCSIA) to quantify, in economic terms, the value of ‘goods and services’ provided by positive family functioning (PFF) and to conduct a cost benefit analysis (CBA) and/or cost effectiveness analysis (CEA) to establish the returns to government and society for investments made in supporting family functioning. 

This report follows a scoping study, also conducted by Access Economics.  The scoping study in 2009 determined:

■  the feasibility of quantifying, in economic terms, the value of ‘goods and services’ provided by PFF;

■  a method for measuring the benefits of PFF; and

■  a method to conduct CBA and CEA of interventions to improve FF.

The methods developed in the scoping study form the basis for the analysis in this report.  The scoping study explains the equivalence of measuring the value of PFF as the costs of negative family functioning (NFF).

Both the scoping study and this full study were overseen by a panel of experts, as listed in the acknowledgements section of each report.  In addition, this study was undertaken with assistance from the Australian Institute of Family Studies (AIFS), which analysed the Australian Temperament Project (ATP) data.

The structure of this report is as follows.

■  The methodology is explained in chapter 2.

■  The findings from the investigation of the Longitudinal Survey of Australian Children (LSAC) are outlined in chapter 3.

■  The findings from the investigation of the ATP are detailed in chapter 4.

■  The costing model is described in 5.

■  In chapter 6, the interventions analysed and the CEA results are explained.

■  Conclusions are elaborated in chapter 7.

■  Detailed information is provided in the Appendices in relation to the methods, LSAC and ATP investigations, and the costing model.

2              Methodological overview

2.1       Definition of family functioning and the outcomes of interest

The focus of the scoping study was on the outcomes of family functioning for the child, without pursuing any family ‘ideal’ or promoting any specific type of family structure.  Children are not able to explicitly control their family environment and they are often viewed as the main victims of NFF. 

The scoping study identified that while no simple definition of PFF exists, consistent themes (or ‘domains’) of FF emerged from the literature.  These were developed and agreed, in consultation with the Expert Reference Group for the scoping study and with FaHCSIA.  These domains provide an overarching definition of the FF environment.  A summary is provided below.

Family functioning (FF) – positive and negative – is defined through a variety of emotional attributes, family governance frameworks, cognitive engagement and development characteristics, physical health habits, intra-familial relationships and social connectedness.  PFF is characterised by emotional closeness, warmth, support and security; well-communicated and consistently applied age-appropriate expectations; stimulating and educational interactions; the cultivation and modelling of physical health promotion strategies; high quality relationships between all family members; and involvement of family members in community activities.

The domains of FF are not mutually exclusive, but interact, complement each other and co-exist. 

Table 2.1:  Characteristics of family functioning domains

Source:  Access Economics in consultation with the Expert Reference Group and FaHCSIA.

The literature review for the scoping study revealed three broad areas of outcomes associated with FF.

■  Health outcomes were mostly observed through the occurrence of mental illnesses such as anxiety and depression later in life, but also included eating disorders, health behaviours (e.g. unsafe sex, physical inactivity, overweight and obesity) and substance abuse (e.g. smoking, alcohol and drug abuse), with the consequent physical impacts of these risk factors on morbidity and mortality outcomes.

■  Productivity outcomes were reflected in rates of labour force participation, employment and hourly wage rates, with a number of intermediate measures reported in the literature, such as reduced levels of literacy and numeracy and other measures of educational achievement.

■  Social outcomes were measured primarily through their negative manifestations –involvement in antisocial behaviour such as delinquency, and the probability of criminal behaviour during youth and later in life.  In contrast it was harder to quantitatively associate PFF with positive manifestations such as the quality of inter-personal relationships and future community contributions.

The criteria used to select the specific outcomes for analysis in this report were:

■  each outcome domain was covered (health, productivity and social);

■  outcomes were associated with high economic and social costs, including burden of disease (BoD), based on prevalence and/or existing studies of costs and BoD;

■  the outcomes were able to be measured by the data sets used as the basis for analysis (see below).

In this report, the correlation between FF is thus examined for the following outcomes:

■  productivity (school completion and completion of an undergraduate university degree);

■  health (anxiety and/or depression, obesity, smoking, drinking, illicit drug use); and

■  social (antisocial behaviour).

While the literature provided valuable insights into potential linkages between FF and child outcomes, the following cautions apply.

■  Many of the concepts are difficult to capture and measure and there is a mosaic of different and overlapping instruments and metrics.

■  Statistical techniques can be used to determine correlation rather than causation.

■  While measures of intermediate outcomes are available, it is difficult to convert these to final outcomes without lengthy study timeframes.  For example, intermediate outcomes such as numeracy and literacy cannot easily be converted into specific types of jobs and lifetime earnings streams.

2.2       Data review

Two longitudinal studies — the Longitudinal study of Australian Children (LSAC) and the Australian Temperament Project (ATP) were selected to analyse the correlation between FF and child outcomes.  The Longitudinal Survey of Australian Youth (LSAY) and the Household, Income and Labour Dynamics in Australia (HILDA) were also investigated but the LSAY did not collect information on FF variables and HILDA is limited to relationships between the family environment and parent’s participation in the labour force. 

■  LSAC has the advantage of containing a breadth of data, valuable for testing confounding factors.  However, a disadvantage is the relatively short timeframe of data collection as the eldest participants from the child cohorts are currently 10‑11 years of age.

■  ATP is currently the only study in Australia that allows the determination of long term impacts of FF on health, economic and social outcomes as the most recent data for participants in this study were collected at the age of 23-24 years.  However, measures of FF and parenting have only been recorded since the participants were in their early teens, with no measures during infancy or early childhood.

Variables from the ATP and LSAC within each of FF domains were mapped to the specific outcomes selected for analysis so that the likelihood of one of the events of interest occurring could be established across different age groups (Figure 2.1) and linked with outcomes.  The ability to join information from LSAC and ATP is limited by slightly different methods in each data set of measuring FF and differences in the generations (ATP children were teenagers during the 1990s, whereas LSAC children are growing up during the new century).  There are, however, conceptual consistencies in measurement of FF between the two data sets. 

Figure 2.1:  Approximate age of study cohorts and bridging the current information gap

LSAC (1) is the Birth cohort and LSAC (2) is the Kindy cohort.

Ideally, one or two more waves of information from the LSAC is required to successfully bridge the gap in ages between the two studies, to enable mapping outcomes for children aged 9 years to those aged 13 years.  This approach is considered acceptable, given the alternative is to wait some years for additional study waves to be completed.

Figure 2.2 shows how LSAC provides the dependent (end-period ‘transition’) variables and independent (early and mid-period family functioning and control) variables in regression analysis of children aged up to 9 years. 

■  End-period transition variables are those that relate to observations in the oldest groups i.e. the third wave of the Kindy cohort in LSAC.   Early and mid-period FF and control variables relate to observations in all the other age groups (the first and second wave observations and the third wave for the birth cohort).

ATP data then provide the dependent (end-period ‘interim outcome’) variables and independent (early and mid-period family functioning and control) variables in regression analysis of people aged up to 25 years. 

■  End-period interim outcome variables are those that relate to observations in the oldest groups of ATP relevant to the cost data and the circumstance.  For example, highest level of education completed, illicit drug use and body mass index were assessed at age 23-24 years, while other variables (completing year 12, smoking, binge drinking, anxiety/depression, and anti-social behaviour) were assessed at adulthood (18-19 years).  Early and mid-period FF and control variables relate to younger age observations.

The interim outcomes can then be used to predict, in an Excel model, lifetime health costs, lifetime productivity losses, social costs of crime and disability adjusted life years (DALYs).  The impact of childhood programs can be modelled by ‘shocking’ the LSAC variables, while adolescent interventions[1] can be modelled by ‘shocking’ the ATP variables.  ‘Shocking’ the model refers to changing input parameters from one level to another, simulating the effects of an intervention, and then observing the consequent change in model outputs (health, productivity and social outcomes).

Figure 2.2:  Diagram of data map

In laymen’s terms, LSAC was used to establish the impact of FF on children up to the age of 9 years, at which age transitional outcomes from childhood were spliced into the ATP analysis by matching this set of transitional outcomes across the two datasets.  Health, productivity and social interim outcomes in early adulthood thus depended on FF during adolescence as well as the transitional outcomes from childhood.  From the interim outcomes in early adulthood, lifetime cost impacts were predicted using a variety of other datasets.

Given the identification of the ATP and LSAC databases as providing adequate data to establish the relationship between family functioning and outcomes, collaborative relationships were established with AIFS researchers and FaHCSIA data specialists. Both were important participants in the project.  Furthermore, AIFS currently manages the ATP data set and access arrangements require that AIFS undertake any in depth analysis.

2.3       Literature review

Evidence from the literature was used to edify the selection of family functioning inputs from the ATP and LSAC that would be likely to be correlated with outcomes.  The findings are briefly outlined in chapter 3, and chapter 4, and evidence is summarised in Appendix A and Appendix B. 

2.4       Concepts and data underlying lifetime costing

Costs were attached to each outcome as depicted in Figure 2.3.

Figure 2.3:  Conceptual map for valuing costs of NFF

 Source: Access Economics.  Blue – health impacts. Red – productivity impacts. Green – Social/criminality impacts.

An incidence or ‘life time’ costing approach was adopted in line with the aetiology of lagged outcomes – i.e. ‘lifetime’ costs (hazard model).  An incidence approach is distinguished from a prevalence approach in Figure 2.4.

Figure 2.4: Incidence versus prevalence approach

Note: The years are illustrative and do not relate to this analysis.

The lifetime costs associated with each outcome include the standard cost categories for disease cost burden analysis from the health economics literature: 

■  Direct health costs – estimated with cost data sourced primarily from the Australian Institute of Health and Welfare (AIHW) data, National Hospital Cost Data Collection (Department of Health and Ageing), the Pharmaceutical Benefits Scheme (PBS), the Medicare Benefits Schedule (MBS) and epidemiological data sourced from the Australian Bureau of Statistics (ABS) National Health Survey (NHS), AIHW and other specific epidemiological studies reported in the peer reviewed literature. 

■  Costs of crime – estimated with data primarily from the Steering Committee for the Review of Government Service Provision (SCRGSP) Report on Government Services , and reports from the Australian Institute of Criminology and ABS. .

■  Productivity costs – are estimated using the human capital approach and reflect reduced labour force participation and absenteeism due to the outcomes selected.  Parameters and labour force data were drawn primarily from the ABS and reports by the Productivity Commission (PC), as well as peer reviewed literature.

■  Burden of disease (BoD) – was estimated using DALYs and determined using the same disability weights and methodology used by the AIHW (Begg et al, 2007).  Monetary values were estimated for the BoD using the value of a statistical life year from DOFD (2009).

■  Other financial costs – include costs associated with the provision of informal care (unpaid care provided by family and friends), health aids and appliances, deadweight losses (DWLs) (efficiency losses which arise due to transfer payments).  The main sources of these estimates are previous studies by Access Economics, and the ABS Survey of Disability, Ageing and Carers (SDAC) (ABS, 2004), and Lattimore et al (1997).

Further detail on costing methods, cost categories and data sources is provided in Appendix I.

2.5       Model construction

This section provides an outline of the model developed to investigate the benefits of PFF and the costs and benefits of the three family functioning programs selected for analysis.  The net benefits for each program were derived under a scenario with the intervention compared to a ‘base case’ without the intervention.

■  The costs for the CfC program, PPP and Reconnect programs are reported in chapter 6 and represent a direct input into the cost benefit analysis.

■  The benefits are based on the extent to which each intervention improves FF and reduces its associated costs.  The effectiveness of each intervention is reported in chapter 6 while the associated costs (health, productivity and social) are detailed in chapter 5.

The underlying principle of the model (developed in Microsoft Excel 2007) is that outcomes (dependent variables in the regression analyses) change as a result of the intervention affecting early and mid-period family functioning (independent) variables in the regression. The size of the intervention is a direct function of the effectiveness of the program on impacted functioning variables (chapter 6) and the size of the coefficients derived in the regressions (Appendix C and Appendix G).

A simple example illustrates the model construction using the impact of a 'shock' to consistent discipline on anxiety in children aged 4-5 (so the analysis starts with the B3 cohort of this age – incidentally the same age as the K1 cohort).  The ‘shock’ in this example was derived from PPP improvements in ‘parental laxness’ and mapped directly to parental consistent discipline as both in practice are aspects of parental consistency (Table 6.4).

The 'shock' can be viewed as changing the input parameters (e.g. parental consistent discipline) from one level to another, simulated by the effects of an intervention.  Before the shock, the multivariate regressions for anxiety are given below for age group 4-5 (B3), 6-7 (K2) and 8-9 (K3), respectively.  In each case, i represents the observational child in the dataset (so anxiety is that of the child, and discipline is towards that child).

As a result of the intervention, the new level of childhood anxiety for children aged 4-5 includes changes in parental consistent discipline. The change in parental consistent discipline from the shock was calculated by multiplying the effectiveness of the program by the average consistent discipline value for children aged 4-5.

As a result of the shock, the consequent change in the model output (Anxiety') is given by a percentage change.

Consequently, the direct percentage change in anxiety levels for children in the 4-5 age group is captured through to the next LSAC age group (6-7 year group) through the use of the lagged dependent variable changing by the intervention effectiveness. 

Following this pattern, Anxiety’’_K2,i and Anxiety’_K3,i are similarly calculated.  The change in the final period of LSAC compared with the actual outcome is then included in the ATP multinomial logit regression i.e. the new anxiety levels for K3 children aged 9 (the final regression in the LSAC analysis) is fed directly into the ATP logistic regression for children aged 13 years. 

This effectively assumes that no change occurs in outcomes between the ages of 10 to 12.  As mentioned before, this approach is considered acceptable, given the alternative is to wait some years for additional study waves to be completed. 

The multinomial logit regression models were used to analyse the impact of an intervention on the probability of each family functioning outcome for children aged 13-23 years.  Unlike the multivariate linear regressions, the results presented in the logit regression represent the probability of an outcome for a person with ‘average’ attributes.  Like the LSAC multivariate linear regressions, the interventions were modelled as a deviation from the mean of a particular explanatory variable (e.g. anxiety).  The baseline logistic regression is given below, where e is a mathematical constant, i is again each child observed (in ATP this time), βs are again coefficients, and εs are again error terms.  X is the sample mean.

After the percentage change from anxiety in the 8-9 year old age group, the new logistic regression is given, with changes to the lagged dependent variable.

In using the multinomial logit model, coefficient estimates are not directly interpretable so do not provide the same type of information as coefficients from an Ordinary Least Squares (OLS) model.  A more natural way of interpreting results from a multinomial logit model is to determine the impact on the probability of an outcome by changing the variables that would be impacted by the intervention while holding all others constant.  The impact of the intervention on the probability can therefore be represented by:

The impact of the intervention was therefore measured as the difference between the probability of an outcome for an average person with and without the intervention.  As such, the model projects the probabilistic change in the outcome as a result of the intervention and calculates the economic benefits attached (detailed in chapter 5).  The net economic gains between the scenario and the ‘base case’ (no intervention) can therefore be evaluated to determine the intervention’s overall return on investment using a dollar value. 

Table 2.2 summarises all the FF variables in the model for each intervention modelled (i.e. found to be significant in the regression analysis reported later on).  A conceptual map of the model is provided in Figure 2.5.

Table 2.2: FF variables used for each intervention

Source: Access Economics (2010).  Note: See Chapters 3, 4 and 6 for derivation.  * These variables are in the Reconnect model but the effect size was estimated as zero.

Figure 2.5 : Cost effectiveness analysis model map

Source: Access Economics (2010)

2.6       Cost benefit/cost effectiveness analysis (CBA/CEA) and the process for selecting interventions for analysis

The concept of CEA modelling of FF interventions is outlined in Figure 2.6.  All analyses compare the outcomes for children with the interventions, against outcomes for children without the intervention. The efficacy of a selected intervention in improving FF is derived from previous evaluations of the programs.  The Excel model is then used to explore how this improvement in FF (the ‘shock’) reduces lifetime costs (in dollars and DALYs).  These lifetime benefits can then be compared with the intervention costs, in net present value (NPV) terms.

Figure 2.6:  CEA model pathway for interventions

As noted in Section 2.2, ‘shocking’ the model for CBA and CEA involves comparing what happens in the absence of an intervention (the status quo), with what would happen if a particular target population received an intervention.  The intervention improves FF based on evaluated effectiveness of the program, which in turn improves transition and/or interim outcomes (based on the coefficients derived from the modelling).  Better outcomes are associated with lower costs, so the NPV of the benefits (lower costs of NFF) can then be compared with the costs of the intervention.  Benefits minus costs provide the ‘net benefit’ in dollars, while benefits divided by costs provide the ‘benefit:cost ratio’.

The process of selecting appropriate FF interventions commenced in the scoping study, when a preliminary assessment was undertaken of 12 types of interventions:

1.  family assistance and income support payments;

2.  family relationship services (FRS);

3.  Stronger Families and Communities Strategy, including Communities for Children and Invest to Grow;

4.  Positive Parenting Program (PPP);

5.  Early childhood education;

6.  Peel Child Health Survey;

7.  Responding Early Assisting Children (REACh);

8.  Reconnect;

9.  Youthlinx;

10.  Transition to Independent Living Allowance (TILA);

11.  SureStart;  and

12.  Parents under Pressure (PuP).

For this full study, criteria were established in the first Reference Group meeting for assessing appropriate interventions for CEA.  These four criteria were:

1. Model data : the LSAC and ATP datasets will be able to accommodate ‘shocks’ to the intervention;

2. Target age : the interventions will target different age groups (e.g. pre-school, primary school, youth)

3. Reach : the interventions have ‘reach’ i.e. they effectively target relevant (disadvantaged) groups; and

4. Efficacy data : adequate information is available from Australian (preferably) or international sources in order to provide an indication of the efficacy of interventions.

A follow-up meeting with Steve Zubrick identified three further criteria.

5. Specificity : Interventions specifically target family functioning, rather than indirectly affect it (e.g. income supplementation can assist with FF, but is in essence a poverty alleviation method).

6. Sustainability : Interventions are current, and likely to continue into the future.

7. Relevance : Interventions have strong connections to or relevance for FaHCSIA.

We also reviewed literature provided by the Reference Group – Karoly et al (2007) and Wise et al (2005).  The interventions selected from this process are summarised below.

2.6.2                      Communities for children (CfC)

CfC meets all criteria for the CEA.

1. Model data : LSAC FF data can be linked with CfC interventions.

2. Target age :  CfC is targeted to pre-school and primary school age children.

3. Reach : CfC targets all Australian population sub-groups.

4. Efficacy data : CfC has been evaluated with efficacy outcomes that can be imputed to the Excel model.

5. Specificity : CfC aims to improve family functioning and outcomes for children.

6. Sustainability :  The CfC program has forward funding.

7. Relevance : CfC is a FaHCSIA program.

2.6.3                      Positive Parenting Program

PPP meets all criteria for the CEA.

1. Model data : LSAC FF data can be linked with PPP interventions.

2. Target age :  PPP is targeted to pre-school and primary school age children.

3. Reach : PPP targets all Australian population sub-groups.

4. Efficacy data : PPP has been evaluated with efficacy outcomes that can be imputed to the Excel model.

5. Specificity : PPP aims to improve family functioning and outcomes for children.

6. Sustainability :  The PPP program is successful and growing.

7. Relevance : PPP is a program relevant to FaHCSIA core business.

2.6.4                      Reconnect

In the Scoping Study Reference Group and in the 13 January Reference Group meeting the Reconnect program was identified as being a good candidate for CEA.  Again it meets all criteria for the CEA.

1. Model data : ATP FF data can be linked with Reconnect interventions.

2. Target age :  Reconnect targets youth aged 12-18 (who are homeless or at risk of homelessness) and their families.

3. Reach : Reconnect targets all Australian population sub-groups. [2]

4. Efficacy data : Reconnect has been evaluated with efficacy outcomes that can be imputed to the Excel model (based on two longitudinal studies).

5. Specificity : Reconnect aims to improve family functioning and outcomes for high school aged children.

6. Sustainability :  The Reconnect program is funded into the future.

7. Relevance : Reconnect is a FaHCSIA program.

3              Findings from the LSAC data investigation

LSAC consists of two cohorts ‘B’ (for Baby) and ‘K’ (for Kindergarten).  Each of these cohorts has three sets of time data ‑‘waves’ of survey data taken at two yearly intervals.  Information is collected through self-reporting and also observational measures (such as parent-child interactions). Child outcomes are measured as: behavioural and emotional adjustment; language and cognitive development; and social competence.

LSAC contains specific research questions.  One question focuses particularly on the impacts of family relationships, composition and dynamics on child outcomes, and changes to these over time.  The question includes the analysis of:

■  the size and make-up of family;

■  the involvement of extended family;

■  roles of family members;

■  character of parental relationships and level of conflict in the family;

■  parenting practices;

■  child’s temperament;

■  impact of family break-up and re-formation; and

■  family coping strategies, particularly in times of stress.

3.1       LSAC variables

Evidence from the literature was used to edify the selection of family functioning inputs from the ATP and LSAC that would be likely to be correlated with outcomes.  The evidence is summarised in Appendix B.  In most cases, the choice of relevant LSAC variables to represent these general literature categories in the model is fairly straightforward. 

■  Appendix C provides a complete list of LSAC variables included in the model.

■  Appendix D provides a detailed description of each individual variable.

■  Appendix E provides a distribution of LSAC variables.

Cases that involved more consideration are discussed below.

■  In some cases, concepts linked in the literature to negative outcomes for children — either as control variables or family functioning variables — were not available from the datasets.  In the LSAC, no variables were identified for child abuse, school bonding, enmeshment, or negative conflictual relationships between children and parents.  In addition, some literature concepts were not relevant to LSAC children such as study child’s alcohol consumption or smoking.

■  For the obesity regression, the category ‘Sedentary activities’ is not derived directly from the literature discussed in Appendix B.  The FF variable used here (e-tainment) measures total hours per week of electronic entertainment: using computers, playing video games and watching television.  The concept is that too much sedentary activity contributes to obesity.

■  Also, under obesity, the control variable ‘needs extra medical care’, while a direct LSAC variable, is not derived directly from the literature in Appendix B.  The rationale here is that some children may be physically unable to exercise, which could in turn contribute to weight gains.

■  Under productivity, academic competency (learning outcomes) is derived from the literature but, given the young age of the children involved, academic competency could vary considerably within the two-year span of each wave.  Accordingly, the LSAC variable ‘study child age in months’ was added as a control variable.

■  For the very young cohorts (particularly B1 and sometimes B2 and older), there were no distinct measures of addictions, antisocial behaviour, or anxiety and depression.  In some cases the best proximate dependant variable was the same (such as apedsgc for the B1 cohort across all three regressions).  Since the only impact of these early cohorts is to feed into older age groups as a general indicator of previous problems, this was considered sensible, since the same predisposing factor can be a risk for two different outcomes later.  Note that for B2, babitp is used as the dependent variable with identical independent variables except that parental overprotection is used to predict addiction but not antisocial behaviour, based on the literature (Emmelkamp and Heeres, 1988), which reported that parental overprotection was a significant factor in adult drug addiction, hence its inclusion there.

■  Parental anxiety and depression has been recoded from an average to a sum of component measures.

■  The literature category ‘socio-economic status’ is proxied by the LSAC variables household income[3] and highest level of parent education[4]. 

■  Alcohol consumption has been recoded from six categories down to three categories – abstain, safe and unsafe.

■  In an effort to obtain a ‘cradle to grave’ coverage for the full model, where the same dependant variable for the regressions is not available for all age groups, close substitutes are used.

There were also issues around sample size.  The initial sample for both cohorts was around 5,000 children.  After excluding children who have not participated in all three waves, the sample size is reduced to between 4,000 and 4,500.  Using standard regressions, for any given variable, participants with nil responses are excluded.  That is, the variable with the most missing observations determines the sample size for the regression.  Thus, there is a trade off between more variables (greater explanatory power) and fewer variables (larger sample size).  Preference has been given here to greater explanatory power, but this does limit sample sizes to between 2,000 and 3,000 observations.

■  Certain participants are more likely to drop out as the study proceeds or to be non-responders (which will affect sample size and bias in the analysis sample) – e.g. people with low socioeconomic status (SES), Indigenous Australians and those from non-English speaking backgrounds.  To account for this, sample weightings were employed.

■  The model did not adjust for complex survey design, as this is not usually an issue unless location-specific factors are important.

■  Also, to maintain sample size, the primary carer was used to represent parents, rather than having separate (and thus smaller) equations for mothers and fathers.  This is also consistent with the ATP, which generally does not differentiate parents by gender.  However, to allow for gender differences, sex of parent was included as a control variable.

■  The LSAC variable ‘family cohesion’ was used as a proxy for the literature category ‘marital conflict’.  LSAC also measures ‘argumentative relationships’, although this variable is not asked of sole parent families, who are an important subsample to retain in the modelling.

■  Scales from the Strengths and Difficulties Questionnaire (SDQ) were utilised from LSAC in relation to anxiety and depression, anti-social behaviour and addictions (Table 3.1).

Table 3.3:  Regression constructs in LSAC

Notes: In the left column, variables in normal font are universal, variables in italics are particular to the individual regressions in brackets.  In the right column, variables outside brackets are those used mostly or exclusively and variables in brackets are those used when the main variable was not available. Appendix D provides greater description of these variables.
# Previous wave of the dependent variable is also used as a control variable.
† = LSAC variables that have been combined to form new variables.
‡ = variable not directly suggested by literature search.
* Where a variable code contains a wildcard (*) it covers all age groups (represented by first letters a,b,c,d,e) unless another age-specific variable code is included next to the variable name.
1  caemot would have been preferred, but yielded nonsensical results (R-squared of 0 in Stata and 1 in Excel).

2  Household annual income was categorical, converted to dollars per year.

3.2       LSAC analysis

The LSAC presents a number of methodological issues for time series analysis.  As noted in the preamble to chapter 3, the Baby and Kindergarten cohorts each have three waves of time data taken at two yearly intervals.

■  While this should yield a total of six time series observations, the two cohorts overlap at ages 4-5 years (intentionally), leaving only five time series observations.

■  Moreover, as each cohort of children are entirely distinct populations, this creates conceptual difficulties in splicing the two together to create one ‘continuous’ time series.

The situation is further complicated by the fact that, despite having over 16,000 variables, the great majority of them are not common across all wave/cohort groups.  This is to some extent unavoidable given the great differences between newborns and nine year olds, but it still makes it difficult to track variables continuously over time periods.

Accordingly, a number of approaches were investigated for modelling LSAC data.  A hypothetical obesity regression is used here to illustrate the differences between these approaches. 

■  1) a straightforward longitudinal analysis.  This derives averages across each total age group (e.g. average obesity for 4-5 year olds) and uses them as regression variables.   However, this approach was not satisfactory for a number of reasons, including the lack of suitable variables which covered all age groups.

■  2) a semi-panel approach, combining elements of longitudinal and cross sectional.  Illustratively, obesity in wave (W) 3 is a function of obesity in W2, obesity in W1, diet in W3, diet in W2, diet in W1.  This had the advantage that if the same variable cannot be found for diet across all groups, it does not matter, as different diet variables can be used for each age group.  On the other hand, the number of variables with significant explanatory power was not high, and most of them belonged to W3.  Also, because effective sample size for regression analysis was limited to the smallest variable among any of the constituent age groups, the number of observations was usually around 1,500, which was low.

■  3) a stepped time series model.  Illustratively, obesity in W2 is a function of W2 diet, other W2 control and FF variables, and previous period (W1) obesity.  While explanatory power was good for the older groups, and average for intermediate ages, it was very poor for the B1 infants group (Chart 3.1).  A further issue with this approach is that the K1 group could not be utilised, as they lacked a link to previous periods. [5]  Fortunately, this age group (four to five year olds) is the only one covered with two sets of survey data, so B3 could be used instead to represent this age group.

While none of the methods were optimal, method three was adopted as it had the least shortcomings.  Outcomes for individual regressions are discussed below.

Chart 3.1 :  Explanatory power of regressions, by age

3.2.2                      Obesity

Overall, obesity had the best explanatory power (R²) among the regressions, ranging from .08 for B1, up to 0.77 for K3.  As in all regressions, B1 was lowest and K3 highest in terms of explanatory power.

As with all regressions, the lagged dependant variable (in this case, last period’s obesity) was significant for every age group.  Previous obesity also had the largest coefficients.  Parental obesity was also always significant.

Female gender for the child was a risk for the B1 cohort, while male gender was a risk for K2 and K3.  Male gender of the primary carer was a risk factor in two cohorts. 

Interestingly, fatty foods, electronic entertainment and exercise tended not to be significant, and when they were, occasionally had an unexpected sign in the younger cohorts (when the children were aged 2 and 4).  When children are young, body mass index is a less reliable indicator and this may explain why some of the findings are not intuitive.  Moreover, children in the younger cohorts may be less likely to use electronic entertainment and eat fatty foods.  It should also be noted that the model captures children’s preference to being active or not, not to actual measures of physical activity.

3.2.3                      Productivity

The explanatory power (R²) for productivity regressions ranged from .07 to .51.

Previous learning index outcomes were always significant, as were consistent discipline, temperament (easy temperament effect), SES and parent education[6].  SES, however, had an unexpected sign for the B1 cohort.  This may reflect that higher income is associated with returning to work which may, in turn, be associated with less ability to spend time the baby and poorer early learning outcomes as a consequence.

Both temperament and child gender had consistently strong coefficients.

Being a girl was an advantage up to the K2 age group, when being a boy was significant (neither gender was significant for K3).

Other variables also sometimes had unexpected signs – such as parental warmth (B2, B3, K2), parental monitoring (K2), and parental involvement in education (K3).  This is discussed further in Section 3.2.7.

3.2.4                      Anxiety and depression

Anxiety and depression, as measured by the LSAC’s Strengths and Difficulties Questionnaire (SDQ) emotional symptoms scale had explanatory power ranging from 0.01 (B1) to 0.36 (K3).

Previous emotional problems was always significant with the expected sign, while difficult temperament also had a strong impact and SES had a significant impact in some age groups, mostly with the expected sign.  Harsh discipline/hostile parenting and parental anxiety/depression were significant.  Family cohesion and over-protection also sometimes had unexpected signs.

Note that some outcome measures used are coded in less than intuitive directions. For example, while in the PEDS global concerns item (apedsgc and *pedsef), lower scores represent concern, but for the SDQ high scores equate with more problems (i.e. poor functioning).  This is important to understand when interpreting the coefficients in Appendix C.

3.2.5                      Anti-social behaviour

Antisocial behaviour, as measured by the LSAC’s SDQ conduct problems scale, had explanatory power ranging from 0.01 (B1) to 0.50 (K3).

Previous conduct problems, family cohesion, harsh discipline, parental smoking, low SES and the child’s temperament were always significant with the expected signs.  Difficult temperament had consistently large coefficients, along with parental anxiety/depression in the B2 cohort.

Parental efficacy, parental anxiety/depression, gender (with risk for boys), and household income were mostly significant.  Consistent discipline was mostly significant but with an unexpected sign in the K2 cohort. 

Inductive reasoning was significant for children aged 4 years and older, but with an unexpected sign.  It may be that parents are more likely to use inductive reasoning with children who have conduct issues (i.e. reverse causation).  Parental alcohol consumption had an unexpected sign in the two younger cohorts cases where it was significant, possibly due to the same issue⁷.

3.2.6                      Addictions

‘Addictions’ is a composite measure covering the literature concepts of smoking, drinking and drug use.  It is rare for children in the LSAC age groups to smoke, drink or take other drugs.  Thus, some LSAC variable needed to be found as an intermediate outcome that could act as a pointer to future addictive problems when the children reach ATP age and later adulthood.

There is abundant literature showing clear links between poor FF inputs in childhood and crime and addiction outcomes in adulthood.  However, because such literature is frequently based on contiguous longitudinal surveys, there is no focus on transitional outcomes in nine year old children, as is needed here to bridge between the LSAC and the ATP.

There is evidence from paediatric and neurological studies (Shonkoff et al, 2010) to suggest that anti-social behaviour in primary school children may be linked to both addictions and anti-social behaviour in adulthood, with stress as the common factor.  Continual environmental stress – such as that occasioned by NFF – can cause structural changes in the brain that strengthen negative emotions such as fear and aggression while simultaneously diminishing the forward-thinking and reasoning capacities which normally moderate reactions to such emotions.  While longitudinal studies on the influence of high stress have only been covered in animals so far, retrospective studies show that people who experienced trauma as children are much more likely as adults to smoke, drink and use drugs (Felitti et al, 1998) and become criminals (Currie and Tekin, 2006). 

High levels of childhood stress can manifest as psychological and emotional dysregulation in childhood (Evans and Kim, 2007).  Thus, for the purposes of regression modelling, ‘addictions’ are represented by the LSAC variable ‘total SDQ score’.  A detailed rationale for this choice can be found in Appendix A.

The addictions regression, thus defined, had explanatory power ranging from 0.01 (B1) to 0.62 (K3).  No variables were significant in the youngest B1 cohort.

In the four older cohorts, previous total SDQ problems, harsh discipline, inconsistent discipline, family cohesion, smoking, temperament, gender (risk for boys), and parental anxiety/depression were always significant with the expected signs.  Temperament and parental anxiety/depression consistently had the largest coefficients.

Parental warmth and overprotection were mostly significant with expected signs, while SES was sometimes significant with the expected sign.  Inductive reasoning was sometimes significant, but with unexpected signs – possibly for the same reasons as for antisocial behaviour (e.g. reverse causation).  Parental alcohol consumption had an unexpected sing in the B2 and B3 cohorts also. [7]

3.2.7                      General observations

A number of variables in the LSAC analysis had unexpected signs. Family cohesion and parental warmth may be in this category because of their skewed distributions; LSAC data indicates that a large proportion of people report that they are warm parents and their family has good cohesion.

Inductive reasoning usually has an unexpected sign.  In the LSAC, ‘inductive reasoning’ asks parents if they ‘explain to children why they are being disciplined / talk it over’.  Note that most studies of the PPP include a variable to measure parental verbosity, which has a consistent negative effect on behaviour.  It is possible that ‘inductive reasoning’ is in fact capturing ‘verbosity’ instead.  In other words, the parent may perceive a detailed explanation, while the child may perceive ‘going on about it’ (at worst, nagging or harassment).  It may also be that inductive reasoning is positively associated with conduct and total emotional problems because parents are more likely to use this parenting technique with children with difficult behaviour.

Parental involvement in education at home was negatively associated for the K3 cohort but positively correlated for the children aged 2 and 4 as expected.  It should be noted that the measure changes (due to developmental differences) between the cohorts.  When the children are younger, the measure includes activities such as reading stories, drawing, music and singing.  These activities are likely to have a link to educational outcomes.  In the K3 cohort, there are only three home activities questions – everyday activities (e.g. cooking, caring for pets), outdoor activities (e.g. swimming) and an adult in the family reading to the child.  The link between caring for pets, swimming and learning outcomes may not be as clear as activities such as drawing and music.  By the age of 8-9 a child may spend more time reading to themselves and/or their parents than vice versa, particularly higher performing children.  Another possible explanation of the unexpected sign in the older cohort is reverse causation.  That is, if a child has learning difficulties, a parent may have to spend more time helping them.

■  Indeed, in general it should be noted that these analyses are largely cross-sectional so direction/causation cannot be perfectly tested, although to some extent this is overcome by including lagged dependent variables in subsequent stages of the modelling.  However, in some cases (notably this variable and inductive reasoning), rather than the parents’ behaviour causing the child’s problem, it may be that the parents’ behaviour is in reaction to the child.

Parental alcohol consumption often has an unexpected sign, being significantly negatively associated with anxiety and depression (K2, K3), anti-social behaviour (B2, B3) and addictions (B2, B3).  This might possibly be due to reverse causation together with other factors in these toddler and pre-school cohorts.

Finally, income sometimes has unexpected signs.  This may be improved by standardising for household size.  That is, a household with higher income may not have better outcomes than another household with lower total income, if income per child is in the first house is actually lower than income per child in the second house.  In two wave 1 regression variables, income coefficients had an unexpected sign – for B1 productivity and obesity.  We think this can be explained as follows.  For the babies, the obesity metric is in fact weight – so in this case the result is as expected – higher income is associated with heavier babies (low SES with underweight babies).  As income goes up, learning in babies may be negatively impacted by mothers returning to work earlier (Ruhm, 2000), which could explain the productivity finding.

Readers should note that some variables are reverse coded in the LSAC database, including parental depression, parental highest education achieved and family cohesion — and at first glance may appear to have unexpected signs.

4              Findings from the ATP data investigation

4.1       ATP variables

Access Economics and the Australian Institute of Family Studies (AIFS) worked to specify appropriate models of family functioning based on the literature and variables available from the Australian Temperament Project (ATP). 

ATP variables were mapped to family functioning domains (emotional, governance, engagement and cognitive development, physical health, intra-familial relationships and social connectivity — and to outcomes (in the domains of health, productivity and criminal behaviour).  A literature review was also conducted to inform the selection of family functioning and other variables from the ATP likely to be correlated with the selected outcomes.  Statistical analysis determined the final model specifications, and only variables found to be significantly correlated with outcomes were included in the CEA model.

In the literature, positive outcomes for young people were commonly associated with authoritative parenting models.  Based on the two dimensions of parenting in Baumrind’s categorisation (see Wise 2003) — authoritative parenting is characterised by high levels of both acceptance (which entails emphasis by parents on warmth and the expression of affection) and control (which can be divided into control based on the use of inductive reasoning and control based on parents’ use of power). 

Authoritative parenting is to be distinguished from alternatives often associated with poorer outcomes for children — authoritarian parenting (high on control and low on acceptance),  permissive parenting (high on warmth and low on control) and neglectful parenting (low on both control and acceptance) (see Wise 2003). 

Other features of parenting or family functioning commonly identified in the literature as influencing outcomes for children include:

■  family conflict and family cohesion;

■  marital relationships;

■  sibling relationships;

■  child abuse (including physical, sexual and emotional abuse) and neglect;  and

■  parental smoking and/or drinking.

The tables in Appendix B provide a summary of the literature reviewed.

Confounding variables available in the ATP and included in the analysis included:

■  the child’s innate temperament (persistence, reactivity and approach);

■  parents’ education (as a proxy for socioeconomic status);

■  family stress (e.g. death of a parent or divorce) — some of which may reflect family functioning (such as the circumstances surrounding marriage breakup).

In rare cases, concepts linked in the literature to outcomes for children — either as control variables or family functioning variables — were not available from the datasets.  For example, in the ATP we were unable to allow for parental mental health — mother’s post natal depression or parent’s anxiety/depression linked to child’s anxiety/depression (Rapee 1997, van Gastel et al, 2009) — or parent’s body mass index as a factor linked to child’s overweight or obesity outcomes (Wardle et al, 2001, Danielzik et al, 2004).  Family income was another variable not available from the ATP.  However, in some cases, proxies were found, for example, in place of family income, the ATP parameter: ‘each resident parent’s highest education level attained’.  Positive affect towards school was selected as an indicator of school bonding based on AIFS advice about the findings of previous research.

The initial regression constructs investigated are summarised in Table 4.1. 

It is important to note that ‘family functioning’ cannot always be isolated from other influences such as environmental factors or child temperament, and is difficult to measure.  The ATP variables selected inevitably reflect a mixture of family functioning and other factors and are at times difficult to interpret as either reflecting family functioning or some other factor external to the family.  In addition, there is the potential for omitted variable bias where variables do not completely capture the concept being measured or where there are missing variables within the ATP.  The statistical relationships reported here need to be viewed in this context.

Table 4.4: Regression constructs

4.2       ATP analysis

The statistical analysis was undertaken by AIFS using ordinary least squares, logistic and multinomial regressions.  The variables are described in  Appendix F and details of each of the regression outcomes are in Appendix G.  The means and standard deviations of the regression variables are in Appendix H.  In summary:

■  Variables were constructed at various child ages depending on the most appropriate sample size for that variable in any given year. 

■  In some instances, composite variables were constructed which included data for children at a range of different ages.  Inter-correlations were computed to assist decisions about the viability of using across-time variables.  Composite variables were only constructed for variables which were highly correlated across time.

■  Tests for multicollinearity suggest this was not a problem in any of the final equations.

■  The initial ATP sample comprised 2,443 families from urban and rural areas of Victoria in 1983, approximately two-thirds of whom are still participating after 24 years.  However, the sample sizes for the regression analyses were determined by the samples available for all of the variables included in the regression analysis — and were affected by missing data and drop outs over time.

The most recent wave of the ATP for which data are available (2006) reports on ATP children at the age of 23-24 years.  While regression analysis was conducted on the transitional outcomes for children aged 19-20 as well as 23-24, dependent variables for children aged 19-20 were generally preferred on the basis that family influences were likely to be stronger at this age.  Exceptions are obesity, completion of a university degree and illicit drug use. 

■  For the first two (obesity and completion of a university degree), dependent variables were not available for this age group.

■  For illicit drug use, the regression equation for the older group was preferred because one of the independent variables was of an unexpected sign (see Appendix G). 

4.2.1                      Productivity

A logistic regression was also conducted for under‑engagement (children not employed and not studying versus other children) but none of the predictor variables was significant.  The sample size for this dependent variable was a lot less — 574 children of whom only 58 children were not employed and not studying.

Completed high school versus did not complete

A logistic regression was conducted comparing children who completed secondary school (year 12) with those who did not.  The responses are based on young adult report at age 19-20 years.  There were 687 children in the sample, 65 of whom did not complete high school.  The Nagelkerke R² was 0.309.

The following predictor variables were significant:

■  Relationship warmth (inverse relationship with completion of high school);

■  Parental monitoring/supervision of teen’s activities/associates, teenager report 17-18 years (an increase in supervision was associated with an increased chance of completing school);

■  Mother’s education (higher educational achievement was associated with a greater chance of the child completing high school);

■  The child’s academic competence at 11-12 years based on teacher report (greater academic competence was associated with a greater likelihood of completing school);

■  School bonding[8] (better bonding was associated with a greater likelihood of completing school);  and

■  Persistent temperament style, composite mean score of parent report at 11-12, 13-14, & 15-16 years (persistence was associated with a greater likelihood of completing school).

Completed high school versus completed university degree

In the sample of 595 young adults at age 23-24 included in a logistic regression, 143 had completed high school only, and 277 had completed a university degree.  The Nagelkerke R² was 0.276.

The following predictor variables were significant:

■  Attachment to parents child age 17-18 (children with higher levels of attachment are more likely to complete a university degree than only completing high school);

■  Relationship warmth (greater warmth is associated with a greater likelihood of only completing high school);

■  Father’s education (higher educational achievement of the father was associated with a greater chance of the child completing a university degree)

■  School bonding[9] (better bonding was associated with a greater likelihood of completing a university degree).

4.2.2                      Obesity

In the ATP, body mass index is generated from self report (rather than actual measurements) at 15-16 years and 23-24 years.  The sample size was 513, with 44 obese, 128 overweight and the rest underweight or normal.

A logistic regression analysis comparing the chance of being overweight or obese at age 23-24 compared with ‘normal’ or underweight found the following significant variables.  The Nagelkerke R² was 0.518.

■  Overweight

¬  Body mass index in 1998 (child age 15-16) (a high index at age 15-16 is associated with a greater chance of being overweight at age 23-24);

¬  Father’s education (lower levels of education are associated with a greater chance of being overweight at age 23-24);  and

¬  A persistent temperament style (composite mean score of parent report at 11-12, 13-14, & 15-16 years) is associated with a lower chance of being overweight at age 23-24.

■  Obesity

¬  Body mass index in 1998 (child age 15-16) (a high index at age 15-16 is associated with a greater chance of being obese at age 23-24);

¬  A negative/conflictual parent-teenager relationship, (parent report 17-18 years), is associated with a greater chance of obesity at age 23-24;

¬  Father’s education (lower levels of education are associated with a greater chance of obesity at age 23-24); and

¬  A persistent temperament style (composite mean score of parent report at 11-12, 13-14, & 15-16 years) is associated with a lower chance of obesity at age 23-24.

4.2.3                      Anxiety and/or depression

Anxiety and depression were combined for this analysis.  The ATP scales for anxiety and depression are based on Lovibond and Lovibond (1995), which lead to similar prevalence results to the Diagnostic and Statistical Manual of Mental Disorders (DSM IV) used more broadly in Australia. 

A logistic regression comparing young adults aged 19-20 who were at risk of being in the clinical range for anxiety and/or depression with those who were not was conducted using a sample of 642 children, 166 of whom were anxious and/or depressed.  The Nagelkerke R² was 0.213.

Anxiety and/or depression at age 19-20 was significantly positively associated with:

■  alienation of the child from his or her parents;

■  not having a persistent temperament; and

■  previous experience of anxiety/depression.

4.2.4                      Drinking

A logistic regression was conducted using a sample of 580 young adults aged 19-20, divided into those who spent 1-4 days per month binge drinking (245 young adults), those who spent 5 or more days per month binge drinking (196 young adults) and those who did not binge drink (0 days per month).  The Nagelkerke R² was 0.121.

Binge drinking on 1-4 days per month at age 19-20 was significantly positively associated with drinking at age 13-14 years.

Binge drinking on 5 or more days per month at age 19-20 was significantly positively associated with:

■  drinking at age 13-14 years;

■  less parental monitoring at age 15-16;

■  father drinking, (composite mean score of parent report at 13-14, and 17-18 years); and

■  a higher age at which the teenager was first allowed to drink at home (this variable has two categories — the teen was allowed to drink at home at 15 years or older or the teen was allowed to drink at home at 14 years or younger — (based on parent report at 17-18 years).

4.2.5                      Smoking

A logistic regression was conducted on a sample of 830 young adults aged 19-20 years, 121 of whom smoked daily.  The Nagelkerke R² was 0.297.

Daily smoking in this group was significantly positively associated with:

■  Smoking at age 13-14 years;

■  Being allowed to smoke at home;

■  Negative conflictual parent-child relationship at age 17-18; and

■  Mother smoking.

4.2.6                      Illicit drugs

A regression was based on a sample of 664 young adults aged 23-24 years who reported the number of differing types of illicit drugs used on one or more day/s in past month (range 0-4).  The R² was 0.076.

Independent variables which were significantly positively associated with the dependent variable were:

■  Less parental supervision/monitoring at age 17-18;

■  A mother who smoked;  and

■  An approaching temperament.

4.2.7                      Antisocial behaviour

A logistic regression was conducted using 755 young adults aged 19-20, 112 of whom reported that they were involved in 3 or more differing types of antisocial activities in past 12 months (including illicit substance use).  Tne Nagelkerke R² was 0.145.

Antisocial behaviour in 1996 at age 13-14 years and not having a persistent temperament were the only independent variables significantly positively associated with antisocial behaviour.

5              The costing model

As explained in the introductory chapter, FF was defined as having six domains, and variables from the ATP and LSAC within each of these domains were mapped to specific outcomes.  The outcomes were drawn from three domains: health (anxiety and/or depression, obesity, smoking, alcohol abuse and illicit drug use), productivity (not completing year 12 and not completing a university undergraduate degree) and social/criminality (antisocial behaviour).  The cost of each adverse outcome was then estimated (recall Figure 2.3). 

This chapter describes the method for estimating the social and economic costs associated with adverse outcomes attributable to NFF. 

The costs of each outcome include: 

■  Health costs – of five main conditions: tobacco use, alcohol abuse, illicit drug use, obesity, and anxiety and depression.  These were sourced from the Australian Institute of Health and Welfare (AIHW) data and Australian Bureau of Statistics (ABS) National Health Survey (NHS) primarily. 

■  Costs of crime – using data sources from the  PC’s Report on Government Services for 2009, as well as reports from the Australian Institute of Criminology.

■  Productivity costs – estimates of reduced participation and absenteeism related to health impacts are based on Access Economics cost of illness methods, plus other sources for alcohol and illicit drugs.  Productivity costs associated with school completion, TER and post-secondary education outcomes are from the  PC and other sources.

■  Burden of disease (BoD)– was determined using the AIHW Begg et al (2007) BoD study for the year 2003 (with extrapolations modelled to future years) and the value of a statistical life year (based on DOFD, 2009).

■  Other financial costs – these include costs associated with the provision of informal care, aids and appliances, deadweight efficiency losses and various other costs for people with long term substance abuse and other health problems.  The main sources comprise previous studies by Access Economics and the Survey of Disability, Ageing and Carers (SDAC) 2003 (ABS, 2004).

Further detail on costing methods, cost categories and data sources is provided in Appendix I.

5.1       Health outcomes

Analysis of the LSAC and ATP data-sets led to the identification of the following main health conditions and behaviours associated with NFF:

■  anxiety and depression;

■  obesity;

■  current daily smoking;

■  alcohol abuse; and

■  illicit drug abuse.

In line with Access Economics’ Disease Cost Burden Analysis (DCBA) framework, health outcomes are associated with health system costs, productivity losses, DWLs, other financial costs and BoD.

For anxiety and depression, obesity and smoking, past Access Economics studies were used to collate total costs in 2010 and calculate per-person costs.  For alcohol abuse and drug abuse, Collins and Lapsley (2008) was used.  Total reported costs were disaggregated and matched, as closely as possible, with Access Economics’ DCBA categories.  It is noted that there are limitations matching costs from Collins and Lapsley (2008) with the DCBA categories, due to differences in costing methodologies.  However, Collins and Lapsley (2008) is the most recent and relevant source for costs of drug abuse in Australia and is thus employed.  This report can be referred to for details on the methodology it uses.

5.1.1                      Obesity

Obesity is the accumulation of excessive body fat, defined here as a body mass index of over 30 for adults. For children and adolescents aged 2 to 18 years, a set of age-gender specific thresholds was used.

The cost of obesity was estimated for this project using Access Economics (2008) — that report included the costs of type 2 diabetes, cardiovascular disease, osteoarthritis and cancer attributable to obesity.

The 2008 costs were inflated to 2010 using a health inflation rate of 3.5% per annum (based on historical averages reported by the AIHW).  and 3% per annum for non health expenditure (Table 5.1).

Table 5.5:  Summary of total costs of obesity (2010)

(a)‘Other financial’ category includes DWL from transfers, carer costs and other indirect costs.

Source: Access Economics (2008 and 2010 calculations)

Obesity prevalence rates were also drawn from Access Economics (2008).  The rates were applied to 2010 population projections from Access Economics Demographic Model (AE-Dem).  Total costs were divided by the estimated number of obese people in 2010 (around 4 million) to obtain a per-person annual cost for each cost type.  The age-gender prevalence profile for obesity was then applied to these per-person costs to further refine costs into per-person annual costs by age and gender.  To ensure per-person costs multiplied by estimated obese people summed to the correct total, a minor level adjustment downwards was applied while preserving the prevalence profile. 

5.1.2                      Anxiety and depression

Anxiety and depression are defined consistent with the clinical definitions used for prevalence studies as follows:

■  Anxiety disorders are those in which ‘anxiety is a predominant feature’ (American Psychiatric Association, 2009) such as panic disorder, obsessive-compulsive disorder, a phobia, or generalised anxiety disorder. 

■  Depression and related affective disorders are defined in the International Classification of Diseases Tenth Revision (ICD-10) by the World Health Organization (WHO, 2010) as a lowering of mood, reduction of energy, and decrease in activity.  Capacity for enjoyment, interest, and concentration is reduced, and marked tiredness after even minimum effort is common.  Sleep is usually disturbed and appetite diminished. Self-esteem and self-confidence are almost always reduced and, even in the mild form, some ideas of guilt or worthlessness are often present.  The lowered mood varies little from day to day, is unresponsive to circumstances and may be accompanied by so-called ‘somatic’ symptoms, such as loss of interest and pleasurable feelings, waking in the morning several hours before the usual time, depression worst in the morning, marked psychomotor retardation, agitation, loss of appetite, weight loss, and loss of libido.

Cost data were drawn from Access Economics (2009) — a study of the costs of six categories of mental illness in the 12 to 25 years age group including substance abuse disorders, anxiety disorders, affective disorders, bipolar disorder, schizophrenia and other mental illness.  The costs were adjusted so that they applied to all age groups by:

■  applying 2003 age-gender prevalence rates, relative risks of mortality attributable to anxiety and depression and applying the relevant disability weights for anxiety and depression from Begg et al (2007);

■  expanding productivity losses to incorporate all age groups using NHS data for 2004-05 (special request from the ABS for Access Economics, 2009) - parameters for employment/productivity effects of all mental illness were used to proxy anxiety and depression;

■  adjusting carer costs to apply to the prevalent group of those with anxiety and depression; and

■  applying average annual government welfare payment rates in 2009 (Centrelink, 2009) to total recipients with mental illness.

The prevalence profile for anxiety and/or depression was obtained from Begg et al (2007), which grouped anxiety and depression as one disease category due to a high degree of co-morbidity and similarity in psychological and drug treatment between these conditions.  The number of people with anxiety and/or depression was calculated by applying the prevalence rates to 2010 population projections from the AE-Dem.

The total aggregated cost of anxiety and depression in 2010 is presented in Table 5.2 by cost type.  The 2009 costs were inflated to 2010 using a health inflation rate of 3.5% per annum (based on historical averages reported by the AIHW)) and 3% per annum for non health expenditure.

Table 5.6 :  Summary of total costs of anxiety and depression (2010)

(a)‘Other financial’ category includes DWL from transfers, carer costs and other indirect costs.

Source: Access Economics (2008 and 2010 calculations)

Total costs were divided by the estimated number of people with anxiety and/or depression in 2010 (around 2 million) to obtain a per-person annual cost for each cost type.  The age-gender prevalence profile for anxiety and depression was then applied to these per-person costs.  Adjustments were made to ensure per-person costs summed to the relevant total.

5.1.3                      Smoking

A person is defined as being a smoker if he or she currently smokes on a daily basis.  Others e.g. ex-smokers or people smoking less frequently are excluded. 

The costs were estimating using the findings of Access Economics (2007), adjusted as follows:

■  removing the stipulation of having a mental illness and being a smoker from all relevant costs to obtain total costs of current daily smoking rather than ‘excess’ costs;

■  applying 2007 age-gender prevalence rates for current daily smoking for those without a mental illness from Access Economics (2007);

■  calculating relative risks of mortality attributable to current daily smoking for all age groups using Begg et al (2007);

■  adjusting BoD estimates to incorporate average disability weights for tobacco-caused diseases for males and females using unpublished data from the AIHW requested for preparation of Access Economics (2007);

■  calculating productivity losses using NHS 2004-05 data (special request from the ABS) for all current daily smokers; and

■  applying welfare payment rates (Centrelink, 2009) to estimated recipients with tobacco-caused cancer (Access Economics, 2006a) and tobacco-caused cardiovascular illness (Access Economics 2006b) in 2005.

Two separate prevalence profiles were applied to estimate per-person costs of daily smoking.  The first profile employed prevalence rates for current daily smoking for people aged 18 and over, from Access Economics (2007).  This prevalence profile displays falling prevalence rates with higher age groups and was applied to estimate per-person productivity costs, DWLs from transfers, carer costs and indirect costs.  This profile was also applied to AE-Dem 2010 population estimates to calculate estimated current daily smokers in 2010.

The second prevalence profile employed prevalence rates for tobacco-attributable diseases and conditions.  This profile displays higher prevalence rates for older age groups, and was applied to health system and BoD costs.  An increasing-prevalence profile was deemed to be more suitable for these categories due to the cumulative nature of health deficits from smoking with increasing age (Hubbard et al, 2009).  Thus, annual health expenditures and loss of wellbeing costs are likely to increase with age.

Prevalence data for this second profile was obtained from the AIHW via a special data request in 2004 for Access Economics’ past study (2007).  Prevalence rates were calculated by dividing the 2004 prevalent cases of total tobacco-caused diseases per age group by the 2004 population in that age group (from AE-Dem).

The total aggregated cost of current daily smoking in 2010 is presented in Table 5.3 by cost type.  The 2005 costs were inflated to 2010 using a health inflation rate of 3.5% per annum (based on historical averages reported by the AIHW ) and 3% per annum for non health expenditure.

Table 5.7:  Summary of total costs of daily smoking (2010)

(a)‘Other financial’ category includes DWL from transfers, carer costs and other indirect costs.

Source: Access Economics (2008 and 2010 calculations)

Total productivity and other financial costs were divided by the estimated number of smokers in 2010 (2.9 million) to obtain a per-person annual cost for these categories.  The smoker age-gender prevalence profile was then applied to these per-person costs to further refine costs into per-person annual costs by age and gender.

Total health system costs and the value of the BoD were divided by estimated prevalent cases of tobacco-caused disease in 2010 (692,523) to obtain a per-person annual cost for these categories.  The tobacco-caused disease prevalence profile was then applied to these per-person costs to further refine costs into per-person annual costs by age and gender.

Adjustments were made to ensure per-person costs summed to the relevant total.

5.1.4                      Alcohol abuse

Collins and Lapsley (2008) note that the definition of ‘abuse’ in relation to alcohol is problematic, with the National Alcohol Strategy showing no preference or separation between the terms ‘misuse’ and ‘abuse’.

For the purposes of this report, alcohol abuse is defined as consumption of alcohol at a level that produces sufficient risk of long-term harm to health.  This definition was taken from the 2004 National Drug Strategy Household Survey (NDHS) (AIHW, 2005).  The categories of ‘risky’ drinking and ‘high risk’ drinking were summed.  ‘Risky’ drinking is defined by the AIHW as 29 to 42 drinks per week for males and 15 to 28 drinks per week for females.  ‘High risk’ drinking is defined as 43+ drinks per week for males and 29+ drinks per week for females.  Short term risk was not included as the data were insufficient in LSAC and ATP to identify short term risk (e.g. from bingeing).

The primary data source for costs was Collins and Lapsley (2008).  The prevalence profile for long-term risky drinking was obtained from the 2004 NDHS (AIHW, 2005).  These rates were applied to AE-Dem 2010 population projections to estimate the number of people affected in 2010. 

The total aggregated costs of alcohol abuse in 2010 are in Table 5.4.  The 2004-05 costs were inflated to 2010 using a health inflation rate of 3.5% per annum (based on historical averages reported by the AIHW) and 3% per annum for non health expenditure.

Notably, crime costs were excluded to avoid double-counting — the costs of crime related to substance abuse are included in section 5.3.

The estimates of the BoDfrom alcohol abuse appear small when compared to those for smoking, anxiety and depression, and obesity.  This is partly due to limitations in matching the Collins and Lapsley study categories with Access Economics disease cost burden analysis categories due to differences in study methodologies.

Table 5.8:  Summary of total costs of alcohol abuse (2010)

(a)‘Other financial’ category includes accidents not elsewhere included, fires not elsewhere included and abusive consumption costs.

Source: Collins and Lapsley (2008) – costs adapted to Access Economics DCBA categories 

Total costs were divided by the estimated number of people with the condition in 2010 (1.8 million) to obtain a per-person annual cost for each cost type.  The age-gender prevalence profile was then applied to these costs to develop a set of per-person annual costs by age and gender group.  Adjustments were made to ensure per-person costs summed to the relevant total.

5.1.5                      Illicit drug abuse

Illicit drug abuse in this report is defined to include all illicit drugs (e.g. marijuana), consistent with the source of the cost data (Collins and Lapsley, 2008). 

The prevalence profile for recent use of illicit drugs was obtained from the 2004 NDHS (AIHW 2005), with recent use defined as use within the last 12 months.  Recent drug users in 2010 were calculated by applying prevalence rates to the 2010 population estimates from the AE-Dem. 

The total aggregated cost of illicit drug abuse in 2010 is in Table 5.5.  The 2004-05 costs were inflated to 2010 using a health inflation rate of 3.5% per annum (based on historical averages reported by the AIHW)  and 3% per annum for non health expenditure.

As with alcohol abuse, the costs of crime associated with illicit drug abuse were excluded to avoid double-counting.

Table 5.9:  Summary of total costs of illicit drug use (2010)

(a)‘Other financial’ category includes accidents not elsewhere included, fires not elsewhere included and abusive consumption costs.

Source: Collins and Lapsley (2008) – costs adapted to Access Economics DCBA categories 

Total costs were divided by estimated people with the condition in 2010 (around 2.8 million) to obtain a per-person annual cost for each cost type.  The age-gender prevalence profile was then applied to these costs to develop a set of per-person annual costs by age and gender group.  Adjustments were made to ensure per-person costs summed to the relevant total.

5.2       Productivity outcomes

Based on analysis of the LSAC and ATP data-sets, and parameters available from the literature, the loss of earnings from not completing year 12 and not completing a tertiary (undergraduate) degree (if completed year 12) were able to be costed.

Human capital theory supports the idea that people with higher education levels display higher labour productivity, as proxied by their earnings (Forbes et al, 2010).  Additionally, those with higher education are more likely to participate in the workforce, due to higher labour demand for these workers, and labour supply factors such as expectations of higher wages and derivation of greater utility from the social and intellectual stimulation work provides attracting these workers to the labour force (La Plagne et al, 2007).

Consequently, there are costs (in the form of forgone income) associated with non-completion of year 12 and/or undergraduate studies.

A number cost components for productivity losses were estimated as follows:

■  Parameters for the effect of not completing year 12 and/or a university degree on employment and productivity were drawn from La Plagne et al (2007) and Forbes et al (2010).  La Plagne et al (2007), through econometric modelling, estimated the effects of different educational attainment levels on the probability of participation in the labour force, relative to the baseline of year 11 completion only.  A more recent study (Forbes et al 2010) estimated the impact of different education levels on average hourly wages (taken as an indicator of labour productivity), relative to the baseline of year 11 completion.  Parameters from these studies are presented in Table 5.6.

■  Differences in participation depending on educational attainment were used as a proxy for differences in employment.  This difference was combined with average weekly earnings (AWE) (ABS, 2009b) and employment rates (ABS, 2009a) for each age-gender group to estimate lost earnings.  The ABS AWE data were for 2008 and were inflated to the end of 2009 using labour price index data (ABS, 2010a). 

■  Wage differences were used as a proxy for the impact of educational attainment on labour productivity. These calculations were also based on ABS AWE (ABS, 2009b) and employment rate data (ABS, 2009a).

■  DWLs arising from foregone earnings were estimated by calculating taxation revenue foregone by applying the average personal income tax rate and average indirect taxation rate to foregone earnings.  The tax rates for 2010 (19.48% and 11.68% respectively) were drawn from Access Economics’ Macroeconomic Model (AEM).   DWLs were then calculated by applying a DWL rate of 28.75% of each tax dollar foregone (Lattimore, 1997; PC, 2003) and Access Economics estimates of Federal Government administration costs).

Table 5.10:  Estimated effects* (%) of year 12 and undergraduate completion* on probability of participation and average earnings

* Relative to the baseline of year 11 completion

Source: Productivity Commission - La Plagne et al (2007) and Forbes et al (2010)

The proportion of the population not completing an undergraduate degree was based on unpublished data on undergraduate award course completions for 2007 from the Department of Education, Employment and Workplace Relations (DEEWR) via a special data request in 2009.

5.3       Criminality outcomes

Four broad categories of costs were included:

■  societal costs - due to the impacts on victims and society;

■  policing costs;

■  court system costs; and

■  prison system costs.

The costs were matched with the relevant probability e.g. of the crime being prosecuted and resulting in a conviction and prison sentence.

5.3.1                      Societal costs

Societal costs can include:

■  medical costs for victims of violent interpersonal crimes;

■  productivity losses for hospitalised and non-hospitalised victims;

■  the costs of pain, suffering and lost quality of life (Rollings, 2008);

■  property-related costs incurred by individuals and businesses due to property damaged/destroyed and non-recovered stolen property; and

■  other costs such as cost of volunteer services and those costs non-separable into the above categories.

Estimates of societal costs were taken from an Australian Institute of Criminology report (Rollings, 2008).  Rollings (2008) quantified societal costs for the following types of crime:

■  homicide;

■  assault;

■  sexual assault;

■  robbery – including individual, commercial, burglary, thefts of vehicles, thefts from vehicles, shop theft and other theft;

■  criminal damage;

■  arson;

■  fraud; and

■  drug offences.

The total social cost in 2010 is in Table 5.7.  The 2005 costs from Rollings (2008) were inflated to 2010 using a health inflation rate of 3.5% per annum (based on historical averages reported by the AIHW) and 3% per annum for non health expenditure.  Societal costs were divided by the number of offenders in 2008-09 (344,274) to determine a set of per offender costs by cost type (Table 5.7).

Table 5.11 :  Social costs of crime by cost type (2010)

(a)Those costs non-separable into above categories

Source: Rollings (2008)

5.3.2                      Policing costs

Policing costs are defined here as recurrent government expenditures on crime-related policing services in a year.  Data on policing expenditures were obtained from SCRGSP (2010) and adjusted down by 30% consistent with Rollings (2008) to account for the fact that not all police time is spent on crime. 

A prevalence profile of offenders was developed using ABS data on offenders in Australia (ABS 2009c; 2010b).  There were approximately 344,274 total offenders in Australia for all crime types in 2008-09 (ABS, 2010b). An age-gender distribution of offenders based on 2007-08 data (ABS, 2009c) was applied to this figure to estimate offenders by age-gender group. 

A set of offender prevalence rates was developed by dividing these offender estimates by population estimates from the AE-Dem. 

5.3.3                      Court system costs

Court system costs are defined here as recurrent government expenditures for all levels of criminal court:

■  Supreme courts;

■  District/County courts;

■  Magistrate courts; and

■  Children’s courts.

Data on total recurrent expenditures on criminal courts were obtained from SCRGSP (2010).  This publication also contained estimates of recurrent expenditures per finalisation for each court level. 

Total recurrent expenditure on criminal courts in Australia in 2008-09 and recurrent expenditure per court finalisation is presented in Table 5.8, by court level.  A weighted average net recurrent expenditure per finalisation for all courts was calculated by applying weights based on the percentage of total criminal court finalisations at each court level.  These weights are presented in the last column of Table 5.8.

Table 5.12 :  Net recurrent expenditure on criminal courts (2008-09) and % criminal court finalisations by court type

Source: Access Economics calculations using data from (SCRGSP, 2010)

The offender prevalence profile was applied to the weighted average expenditure per finalisation estimate to develop per-offender annual court costs based on age-gender group.

5.3.4                      Prison costs

Prison costs are defined here as real net operating expenditures by the government on prisoners in a year.  Data on real net operating expenditure per prisoner were obtained from SCRGSP (2010).  This contained figures for real net operating expenditure on prisons and community corrections in Australia in 2008-09. 

An age-gender profile of prisoners was developed using data on prisoners in Australia from the ABS (2009d).  A set of prisoner prevalence rates was developed by dividing prisoner estimates in each age-gender group by age-gender population estimates from the AE-Dem. 

Total real net operating expenditure on prisons and community corrections in Australia in 2008-09 is presented in Table 5.9.  Overall, total real net operating expenditure on prisons and community corrections was approximately $2.5 billion in 2008-09.  This figure was divided by the number of prisoners in 2008-09 (29,319) to determine a cost per prisoner of approximately $85,064.

Table 5.13:  Total real net operating expenditure on prisons and community corrections in Australia in 2008-09

Source: (SCRGSP, 2010)

The prisoner prevalence profile was applied to the cost per prisoner to attain a set of per-prisoner annual prison system costs by age-gender group. 

5.3.5                      Probabilities of incurring costs

Since not all crimes incur all cost types (i.e. dependent on visibility of crime, severity), probabilities based on crime type were applied to costs within the final costing model.  All crimes were assumed to attract the societal costs mentioned in section 5.3.1, regardless of whether they were reported.  However a schematic set of probabilities was developed to determine whether a crime attracted policing, court and prison system costs.  This set included:

■  probability of a crime being reported to the police (thus attracting a policing system cost);

■  probability of court action on a reported crime (thus attracting a court system cost);

■  probability of a ‘guilty verdict’ in a court finalisation; and

■  probability of a guilty verdict involving a custodial sentence (thus attracting a prison system cost).

Crime types were based on ABS (2010b) for the application of probabilities as follows:

■  homicide and related offences;

■  acts intended to cause injury;

■  sexual assault and related offences;

■  dangerous or negligent acts endangering people;

■  abduction/harassment/other offences against the person;

■  robbery, extortion and related offences;

■  unlawful entry with intent;

■  theft and related offences;

■  fraud, deception and related offences;

■  illicit drug offences;

■  prohibited/regulated weapons and explosives offences;

■  property damage and environmental pollution;

■  public order offences; and

■  offences against justice.

5.4       Lifetime costing framework

The lifetime costing framework was based on the application of each set of per-head annual costs by age-gender, across an estimated lifetime.  Total lifetime costs were dependent on an assumed age of onset for the condition and assumed life-span over which costs would apply.

Summation of these costs across life-spans gave a total lifetime, undiscounted cost from the presence of that NFF outcome.  Application of a discount rate across the lifespan allowed the estimation of discounted lifetime costs (net present values) in 2010 dollars.

The assumed age of onset for all health, productivity and criminality outcomes, excluding undergraduate degree non-completion, was 19 years.  For undergraduate degree non-completion, the assumed age of onset was 21 years. 

Lifetime costs were discounted back to 2010 at an annual real discount rate of 7%, consistent with the Australian Government’s choice of discount rate for assessing regulatory interventions (Australian Government, 2010). 

Assumed life spans were based on life expectancies for males and females taken from ABS life tables (2009e).

The costs are summarised in the tables below.

Table 5.14:  Discounted lifetime costs of adverse health outcomes (a) (2010 dollars)

(a) Assuming an age of onset of 19 years, life expectancies from ABS and 7% discount rate

(b) ‘Other financial’ category for obesity, anxiety and depression and smoking includes DWL from transfers, carer costs and other indirect costs.

(c) ‘Other financial’ category for alcohol abuse and drug abuse includes accidents not elsewhere included, fires not elsewhere included and abusive consumption costs.

Source: Access Economics calculations.

Table 5.15:  Discounted lifetime costs of adverse productivity outcomes (a) (2010 dollars)

(a) Including costs from lower employment/participation, productivity and DWL due to lower taxation revenue. 

(b) Assuming an age of onset of 19 years, life expectancies from ABS and 7% discount rate.

(c) Assuming an age of onset of 21 years, life expectancies from ABS and 7% discount rate.

Source: Access Economics calculations.

Table 5.16 :  Discounted lifetime costs of criminality outcomes (a) (2010 dollars)

(a) Assuming an age of onset of 19 years, life expectancies from ABS and 7% discount rate.

(b) ‘Intangible’ costs are those due to pain, suffering and lost QoL (Rollings, 2008).

(c) Those costs non-separable into above categories.

Source: Access Economics calculations.

6              Cost benefit analysis

The Communities for Children (CfC), Positive Parenting Program (PPP) and Reconnect programs were selected for analysis.  The reasons for selection of these is explained in section 2.6.

6.1       Communities for children (CfC)

This section is based on the evaluation of the Stronger Families and Communities Strategy report (FaHCSIA, 2009).  CfC is one of the major Australian Government investments in families.  It has already been shown to be efficacious, and the CEA evaluation can also determine at what cost its effective outcomes are achieved.

The CfC program was an initiative funded by the Australian Government from 2004–2009.  The program aimed to:

■  improve coordination of services for children 0 to 5 years and their families;

■  identify and provide services to address unmet needs;

■  build community capacity to engage in service delivery; and

■  improve the community context in which children grow up.

Under the CfC, FaHCSIA funded non-government organisations as ‘Facilitating Partners’ in 45 disadvantaged geographic areas around Australia to develop and implement a whole-of-community approach to enhancing early childhood development.  CfC was one of three models of service delivery funded under the Australian Government’s Stronger Families and Communities Strategy (SFCS) 2004–2009.

Most of the outcomes targeted by CfC are included as LSAC variables within Access Economics’ model (Table 6.1).  The main exception is building child-friendly communities, which is not directly related to family functioning.

FaHCSIA (2009) presented the results of the evaluation of the short-run impacts of the CfC initiative on child, family and community outcomes.  The evaluation study was based on a three-wave longitudinal study of 2,202 families living in 10 sites that had a CfC program and five sites that did not have a CfC program but were in other ways comparable to the CfC sites (contrast sites).  The study had a large sample, representing 42 per cent of the population of 2 year-old children in 10 CfC sites and five contrast sites at wave 1.  Follow up (wave 3) was conducted in 2008, approximately one year after CfC program activities were under way. 

The effects of the CfC initiative were estimated using statistical techniques that allowed child, family and community outcomes in the CfC sites to be compared to what they would have been in the absence of the CfC intervention (using outcomes in the contrast sites). The outcomes measures related to four priority areas:

■  healthy young families—child injuries requiring medical attention, child and parent physical health, children’s experiences of emotional and behavioural problems, children’s prosocial behaviour, children being overweight, and parents’ mental health;

■  supporting families and parents—harsh parenting, parenting self-efficacy or self-confidence, parent relationship conflict, and living in a jobless household;

■  early learning and care—children’s receptive vocabulary achievement and verbal ability, and the quality of the home learning environment; and

■  child-friendly communities—parents’ involvement in community service activities, the level of support parents receive from others to raise children, the quality of the neighbourhood as a place to raise children, parents’ sense of community social cohesion, their perception of the quality of facilities in the community, and the level of unmet service needs.

The overall conclusion is that, on balance, there is evidence that CfC had the following positive impacts:

■  fewer children were living in a jobless household;

■  parents reported less hostile or harsh parenting practices; and

■  parents felt more effective in their roles as parents.

At initial follow up, any impact of CfC was expected to be small, given that the evaluation was of short-run effects and the intervention was intended to have an impact on all families living in CfC sites and not just those who directly accessed services.  As expected, for many of the outcome measures, the estimated impact of CfC was not statistically significant, which is not surprising given that any effects were likely to be small in the short-run.  Depending on the statistical model estimated, between two-thirds and three-quarters of the outcome variables indicated a positive, although not necessarily statistically significant, effect.

■  Although non-significance means that it is not possible to say with a high level of confidence that the individual effect was not different from zero, the skewed pattern of results towards positive effects provides support for the conclusion that CfC has had some positive impacts in the short-run.

■  The effect sizes of the CfC impacts on all outcomes were small, but can be considered positive relative to what was observed in the early phase of Sure Start (a large-scale area-based initiative in the United Kingdom). The current results were also comparable in size to those found in the later impacts evaluation of Sure Start, where 3 year-old children were exposed to more developed programs from birth.

■  Also important is the extent to which CfC outcomes compare with alternative early childhood interventions that target specific client groups and seek to enhance child outcomes through other processes, such as centre-based programs, home visiting programs, case management interventions and parenting programs.  Wise et al (2005) found that most studies reported effect sizes on parenting and child outcomes that were negligible to small.  It should also be noted that most of these evaluations measured outcomes for children who were directly enrolled in the program, whereas CfC is aimed at improving outcomes for children in the whole community.

The fact that the effect sizes of CfC were comparable to, if not greater than, many alternative early childhood interventions, and that these effects were evident irrespective of whether parents and children in the CfC communities had actually received services, seems to point towards an additional effect over and above the provision of new, stand-alone services, possibly as the result of a better coordinated local system of early childhood services and/or other enhancements to the community context in which children develop.

6.1.1                      CfC outcomes

Three of the five LSAC categories used as interim variables in this report can be used as proxies for variables modelled in FaHCSIA (2009): emotional and social problems, overweight (obesity) and vocabulary and verbal ability (learning outcomes).  Four more CfC variables ‑ hostile parenting (harsh discipline), parental self-efficacy, parental mental health (depression) and parental relational conflict (family cohesion) are used as inputs to the other two interim outcomes.  The last CfC variable, quality of the home learning environment (parental involvement in education at home) is used as an input into the learning outcome interim variable (Table 6.1).

Table 6.17: Communities for Children target areas and related LSAC variables

Note: Variables in bold are mapped to LSAC dependent variables.  Variables in normal font (not bold) are explanatory variables.  Variables with dashes do not occur in our model as the literature review did not support them as being significant for our model. 
Source: FaHCSIA (2009).

All of the relevant outcomes of the CfC, as modelled in FaHCSIA (2009) had the expected signs (that is, they were all improvements.)  The LSAC sample includes 364 children who were also participants in the CfC program.  The data were drawn from the K1 cohort to match these children.

By comparing CfC effect sizes against LSAC averages for the same variables, it is possible to estimate the magnitude of the average improvements under the program (Table 6.2). 

Most of the substantial effects from CfC were improvements to parenting[10].  The smallest effects were on child outcomes – but they were still all positive.

Table 6.18: Outcomes of Communities for Children variables used in this report

Source: FaHCSIA (2009).  Note: Variables in bold are mapped to LSAC dependent variables.  Variables in normal font (not bold) are explanatory variables. 
1 Obtained from LSAC database observations on CfC children.

6.1.2                      CfC costs

The cost of funding CfC over the four years from 2004-05 to 2008-09 was $142 million.  The majority of this investment was spent on service delivery, with Community Partners (the local service provider that delivered the services/activities) receiving 60% of the funding, Facilitating Partners receiving 7% of the funding and 3% was used for local evaluation.  The remaining 30% was for community resource funding (development, implementation, project management and community development).

Based on the number of 0 to 5 year olds in each CfC site in 2006 (n=28,810), $840 was spent on each 0 to 5 year-old child living in the CfC communities between 2004–05 and 2007–08.  This four year period was thus selected for the costing – estimated as 4/5^ths of $142 million or $113.6 million.

6.1.3                      Cost benefit results

The difference between the economic benefits and costs under a scenario with the intervention and one without was estimated by using an incidence approach i.e. the difference in the number of people affected by each outcome.  Under the CFC, the resulting reductions in the incidence of negative outcomes and the increases and improvements in positive outcomes are shown in Table 6.3, along with the intervention effect in percentage terms.  The net economic gains between the scenario and the ‘base case’ (no intervention) were evaluated to determine the intervention’s overall return on investment. 

The total financial benefits represent discounted social and economic savings associated with improvements in outcomes attributable to CFC over the life of each individual.  Consequently, the benefits were estimated to be $541.4 million, or $135.4 million per annum (assessed over 2004-05 to 2007-2008).

Table 6.19: Outcomes of CfC as delivered over 2004-05 to 2007-08

Source: Access Economics (2010). * The intervention as modelled targets children aged 0-1 years.

6.2       Positive Parenting Program (PPP)

PPP is one of the best evaluated programs targeted at improving family functioning and outcomes for younger children.  While its efficacy is well-proven, there are fewer studies on its cost effectiveness and this CEA can also act as a tool to test/triangulate the power of the model.

Developed at the University of Queensland, the PPP (also referred to as ‘Triple-P’) is a multi-level system of parenting and family support.  It can be provided individually, in a group, or a self-directed format.  It incorporates five levels of intervention on a tiered continuum of increasing strength for parents of children and adolescents from birth to age 16.  The multi-disciplinary nature of the program allows utilisation of the existing professional workforce in the task of promoting competent parenting.  The program targets five different developmental periods from infancy to adolescence.  Within each developmental period, the reach of the intervention can vary from being very broad (targeting an entire population) to quite narrow (targeting only high-risk children).  The PPP enables practitioners to determine the scope of the intervention given their own service priorities and funding.

The aims of PPP[11] are:

■  To promote the development, growth, health and social competencies of children and young people.

■  To promote the development of non-violent, protective and nurturing environments for children.

■  To promote the independence and health of families by enhancing parents' knowledge, skills and confidence.

■  To enhance the competence, resourcefulness and self-sufficiency of parents in raising their children.

■  To reduce the incidence of child abuse, mental illness, behavioural problems, delinquency and homelessness.

PPP is prevention oriented, multi-disciplinary and has five levels. The program has been developed through over 20 years of clinical trials.  Five different developmental periods are targeted at each level – infants, toddlers, preschoolers, primary school aged children and teenagers.  The program aims to promote parental competence and enable parents to become independent problem solvers.  Five key principles of parenting – safe, engaging environment; positive learning environment; assertive discipline; reasonable expectations and taking care of self as parent.  The program has five levels, as outlined below.

■  Level 1 – provides universal access to parenting information through print and electronic media. This level aims to increase community awareness of parenting resources, encourage parent participation in PPP and create a sense of optimism.

■  Level 2 – One or two primary health care sessions that provides ‘anticipatory developmental guidance’ to parents who have children with mild behaviour problems. Parenting tip sheets and videotapes are commonly used at this level.

■  Level 3 – Four primary care intervention sessions for parents of children with mild to moderate behaviour problems. The sessions provide active skills training for parents.

■  Level 4 – Intensive program of eight to ten sessions that are either individual sessions, group based or self-directed. This level is for parents of children with more severe behaviour problems.

■  Level 5 – An enhanced behavioural family intervention program for parents whose difficulties are complicated by other issues such as relationship conflict and parental depression. This level therefore targets not only parenting skills but also distressing parental emotional reactions (including depression, anger and stress) via cognitive behavioural techniques.

6.2.1                      PPP outcomes

PPP has been the subject of dozens of peer-reviewed journal articles over the years.  The majority of the LSAC regression dependent variables in this analysis have been reported directly in Australian PPP studies (Table 6.4). 

■  PPP targets childhood problem solving skills (which is similar to the dependant variable in the productivity regression (learning outcomes).  While the literature search did not uncover any quantitative reporting for this outcome, several input variables (or common psychological variables that can be used as proxies for input variables) were reported.

■  PPP also targets health, of which obesity is a factor.  Again, no studies were found that reported on obesity directly, but a number of input variables have been studied.

Table 6.20: PPP target areas and LSAC variables