The Treasury

Global Navigation

Personal tools

Treasury
Publication

Characteristics of Children at Greater Risk of Poor Outcomes as Adults

Appendices

Appendix 1:Modelling poor outcomes for those aged 0-14 years

Predictive modelling was used to investigate the extent to which various characteristics (observed at age 0 through to age 14) were associated with poor outcomes as young adults.

In Treasury’s Analytical Paper 15/01 regression models were estimated for selected outcomes based on characteristics observed at age 5 and age 13 in the integrated data. This analysis identified that three particular characteristics were strongly associated with poor school attainment, long-term benefit receipt by age 21, and receiving a correction sentence before age 21:

  • being known to Child, Youth and Family (the child was the subject of a notification or report of concern)
  • having spent more than three-quarters of their lifetime supported by welfare benefits (ie, their parents or caregivers received benefits over this period)
  • having a parent with a corrections history (including both community and custodial sentences), where parents and caregivers were identified through the benefit data only.

These three characteristics were used to identify groups of children who were at risk of poor outcomes.

In Treasury’s Analytical Paper 15/02 a somewhat more systematic approach (for youth aged 15-24) was used where a large number of potential indicators derived from the integrated data were included in a large modelling exercise.

For youth a set of measures was developed based on the Youth Outcomes Framework[16], across the domains of:

  • Enjoying Economic Opportunity
  • Engaging & Achieving in Education
  • Maintaining Good Health, and
  • Enjoying Safety & Security.

Four outcomes measures were selected and defined as follows:

  • not achieving at least a Level 2 education qualification by age 19
  • use of mental health or addiction services whilst aged between 18 and 20
  • receiving a custodial or community sentence before age 21
  • being on benefit for 2 years or more before age 21.

Logistic regression models were run at each year of age for females and males separately for four outcome measures. Forward selection was used to select the model. This process allowed us to identify the key indicators for each age/gender combination and outcome measure, and calculate a predicted risk score for each outcome for each individual in the population. The average predicted risk score across the 4 outcomes was used to identify the 15% (and 10% and 5%) of youth with the highest average predicted score.

In this project a very similar modelling exercise was undertaken for children aged 0 to 14 years. The same outcome measures were used. This analysis helped us understand whether taking a more formal modelling approach would lead to a similar groups of children being identified.

Appendix 1 Table 1 and 2 shows the variables that were the most important based on the order they were included in the model.

Note that for the 1993 birth cohort, information about parents/caregivers corrections history and mother/female caregiver's educational attainment is coming from the welfare system, and hence is only known for children who have been supported by benefit at some point. (For children born before 2000 information on parent's date of birth in the birth register is incomplete and wasn't used in the cohort analysis.)

“Benefit caregiver has a community or custodial sentence history” took the values - child never supported by benefit, child supported by benefit and caregiver /parent did not have a sentence history, child supported by benefit and caregiver /parent had a sentence history.

“Benefit female caregiver has no formal qualifications” took the values - child never supported by benefit, child supported by benefit and female caregiver did not have a qualification, child supported by benefit and female caregiver had a qualification, child supported by benefit and did not have a female caregiver.

This regression analysis showed that for children aged 0-5 years being known to CYF (ie, the broader CYF contact measure), the proportion of time supported by welfare benefits, having a parent with a corrections sentence history, ethnicity, and gender were the characteristics most strongly associated with poorer outcomes. Having a “mother having no formal qualifications” was also associated with poorer outcomes, particularly qualification attainment, but less so than the other 3 indicators. This result likely reflects the partial nature of the mother's qualification measure that can be derived from the information available in IDI (this is discussed further in Section 2). The analysis of international longitudinal survey data shows that mother's educational attainment is correlated with children's outcomes.

Comprehensive information on school enrolments is only available from 2006 onwards, which meant that when these children born in 1993 information on school characteristics and attendance was only available from age 13 onwards. Appendix 1 Table 2 shows that being stood down from school becomes an important variable, along with being known to CYF (ie, the broader CYF contact measure), the proportion of time supported by welfare benefits, ethnicity, and having a parent with a corrections sentence history.

Appendix 1 Table 1: The most important variables in the regression models
for children aged 0-5 years
Benefit 2+ years Corrections sentence No Level 2 qualification Mental health service use
Extent of time supported by benefit since birth Benefit caregiver has a community or custodial sentence history Extent of time supported by benefit since birth Known to CYF
Known to CYF Known to CYF Benefit female caregiver has no formal qualifications Ethnicity
Ethnicity Ethnicity Known to CYF Benefit caregiver has a community or custodial sentence history
Neighbourhood deprivation index Regional Council Ethnicity Type of benefit
Benefit caregiver has a community or custodial sentence history Extent of time supported by benefit since birth Benefit caregiver has a community or custodial sentence history Regional Council
Appendix 1 Table 2: The most important variables in the regression models
for children aged 13-14 years
Benefit 2+ years Corrections sentence No Level 2 qualification Mental health service use
Known to CYF Days stood down from school Extent of time supported by benefit since birth Known to CYF
Extent of time supported by benefit since birth In CYF care Days stood down from school Ethnicity
In CYF care Benefit caregiver has a community or custodial sentence history Known to CYF Days stood down from school
Ethnicity Known to CYF School type (private, public, integrated) In CYF care
Days stood down from school Ethnicity Days truant from school Territorial Authority

This regression analysis showed that being known to CYF (ie, the broader CYF contact measure), the proportion of time supported by welfare benefits, having a parent with a corrections history, ethnicity, and gender were the characteristics most strongly associated with poorer outcomes. Having a “mother having no formal qualifications” was also associated with poorer outcomes, particularly qualification attainment, but less so than the other 3 indicators. This result likely reflects the partial nature of the mother's qualification measure that can be derived from the information available in IDI. The analysis of international longitudinal survey data shows that mother's educational attainment is correlated with children's outcomes.

This regression analysis showed that being known to CYF (ie, the broader CYF contact measure which include children notified to CYF) was more strongly associated with poor outcomes than the narrower CYF contact measure (based on having a substantiated finding of abuse or neglect or a period in care). For children aged 0-14 years the proportion of children known to CYF was around twice the proportion that had a substantiated finding or had spent time in care (8% compared to 17%). The broader CYF contact measure leads to leads to around 17.2% of children aged 0-14 having two or more of the four indicators, compared to 13.9% based on the narrower measure.

Appendix 1 Table 3 shows the characteristics of children identified through the modelling exercise as being most at risk of poor outcomes, based on the average predicted score. The 15% of children identified as being at highest risk of a particular outcome were much more likely to have a finding of abuse or neglect, supported by benefits most of their lifetime, and to have a parent with a correction sentence history. They are also more likely to be male (59%) and Maori (67%).

Appendix 1 Table 3: Characteristics of children aged 0-14
by predicted risk score
Total Level of predicted risk
Highest
5% average
risk
Highest
10% average
risk
Highest
15% average
risk
Lowest
85% average
risk
Number of children 873,180 43,653 87,306 131,025 742,155
Percentage of children 100.0 5.0 10.0 15.0 85.0
Percentage                                       
Gender
Male 51.3 63.9 60.8 59.3 49.8
Female 48.7 36.1 39.2 40.7 50.2
Ethnicity
Asian 10.4 0.2 0.3 0.4 12.2
NZ European 50.0 19.1 22.9 25.5 54.3
Māori 27.0 76.3 70.7 66.5 20.1
Other 10.2 3.9 5.3 6.7 10.9
Pacific 2.0 0.6 0.7 0.9 2.2
Indicators
Substantiated finding of abuse of neglect 8.2 62.3 50.1 40.8 2.5
Mainly supported by benefits since birth 15.2 69.8 63.3 57.9 7.6
Parent has a community or custodial sentence history 17.0 78.5 69.2 61.5 9.2
Mother has no formal qualifications 10.4 31.7 28.3 25.9 7.6
Number of risk factors
0 69.0 2.2 6.2 11.0 79.3
1 17.1 13.7 21.1 26.2 15.5
2 8.9 35.4 36.1 34.4 4.4
3 4.1 37.0 28.7 22.8 0.8
4 0.9 11.7 7.8 5.7 0.1
2+ 13.9 84.1 72.7 62.9 5.3
3+ 5.0 48.7 36.6 28.5 0.8
Combination of 2+ risk factors
2 factors: NO QUAL and one of: CYF, BEN, CORR 2.7 4.9 6.4 7.2 2.0
2 factors: CYF, BEN or CORR 6.2 30.5 29.7 27.2 2.4
3 factors: NO QUAL and two of: CYF, BEN, CORR 2.1 14.4 12.4 10.6 0.6
3 factors: CYF, BEN, CORR or 4 factors: CYF, BEN, CORR, NO QUAL 2.9 34.3 24.2 17.8 0.3

There is a moderate degree of overlap between the high risk 15% identified through the modelling exercise and the priority (or target) population with 63% of those identified having 2 or more of the four indicators. Hence a more formal modelling approach leads to somewhat different groups of children being identified, compared to using a small number of key indicators.

Appendix 1 Table 4 shows that the expected outcomes for the two groups of children are very similar, and a simplified approach based on using a small number of key indicators identifies a similarly high risk group.

Appendix 1 Table 4: Projected outcomes for children aged 0-14 years
by predicted risk score and number of indicators present
Total Level of
predicted risk
Number of 4
indicators present
Highest
15% average
risk
Lowest
85% average
risk
Two or
more
None or
one
Number of children 873,180 131,025 742,155 121,377 751,803
Percentage of children 100.0 15.0 85.0 13.9 86.1

Projected outcomes before age 21 (%)

Further contact with CYF 18.6 67.5 10.0 59.5 12.0
Contact with Youth Justice 4.4 17.5 2.1 15.9 2.6
Did not achieve any school qualifications 17.7 40.7 13.6 38.7 14.3
Did not achieve a level 2 qualification 27.0 54.7 22.2 52.2 23.0
Received a benefit for more than 2 years 7.1 22.4 4.4 22.6 4.6
Received a community or custodial sentence 5.9 20.2 3.4 18.6 3.9
Received a custodial sentence 1.7 7.5 0.7 6.9 0.9

Projected outcomes when aged 25 to 34 (%)

Received a community or custodial sentence 8.5 23.9 5.8 21.9 6.3
Received a custodial sentence 3.2 10.7 1.9 9.7 2.2
Spent more than 5 years receiving benefit 8.6 21.9 6.3 21.9 6.5

Projected average costs before age 21

Average total cost of CYF, YJ, benefits as a child 21,600 75,500 12,100 77,400 12,600
Average benefit costs as an adult 5,700 15,400 3,900 15,500 4,100
Average corrections costs 1,200 5,600 500 5,200 600
Average total cost* 28,500 96,600 16,500 98,100 17,300

Projected average costs before age 35

Average benefit costs 37,700 88,600 28,700 90,000 29,300
Average corrections costs 7,700 31,100 3,500 28,600 4,300
Average total costs* 67,000 195,100 44,400 196,000 46,200

* includes CYF, YJ, child benefit costs

Notes

  • [16]This framework includes five domain areas (the four listed above, as well as Social Participation, for which an appropriate measure proved difficult to identify in the IDI). It was derived from the Global Youth Wellbeing Index, which set out six domains by which youth wellbeing could be defined; equivalents to the five used in the Youth Outcomes Framework, as well as Information Communication and Technology. Information about the Global Youth Wellbeing Index can be accessed at www.youthindex.org
Page top