Measuring case complexity

We completed these reports as part of an initiative to improve how we assess and compare case workloads. The reports surveyed care and protection social workers working with children to inform design of a model that could predict case workloads based on the complexity of their needs.

We plan to use report findings to develop an improved, data-informed approach to social worker caseload management.

Report 1: Social workers survey

The first report (Part 1) presents the findings from a survey run with our care and protection social workers. The survey aimed to explore and measure the complexity of the cases on their workload.

Survey responses were structured into the 3 main phases of social work:

• intake/investigation/partnered response
• intervention, and
• placement phases.

The responses to the survey were used in the development of a model designed to predict case complexity using operational data.

Key findings

We found 3 elements contributed to case complexity for social workers according to the survey:

1. Tamaiti needs complexity. This increased with:
   • the level of involvement in the care and protection system
   • experiences of abuse and/or neglect,
   • a history of reports of concern, and
   • other factors.
2. Whānau needs and dynamics. This increased:
   • at the Intervention and Placement phases
   • for whānau with four or more tamariki
   • with any history of family harm/violence, mental health or substance use issues
   • with previous care and protection experience, and
   • when tamaiti themselves were complex.
3. Organisational complexity. The top 3 issues that affected our social workers ability to work with their caseload were:
   • a manageable workload
   • access to resources, and
   • supportive management. 

Report 2: Development of a predictive model

The second report (Part 2) outlines the development of the model. It uses structured data from our case management system to predict case complexity, as rated by social workers.

Key findings

We developed a predictive analytical model for case complexity, using survey ratings. The model is sufficiently accurate to predict either:

• ‘Low/Medium Complexity’ or
• ‘High Complexity’.

Prediction of high complexity was most accurate for tamariki with higher involvement in the system, such as those in placement phases. It was less accurate for lower phases.

Predicting unmanageable caseloads

We found that our model should only be used at a site, regional or national level due to its limited accuracy.

However, it helped to identify patterns of potentially unmanageable caseloads in sites and regions. For example, social worker caseloads with both above average:

• numbers of tamariki, and
• proportion of highly complex cases.

About 17% of social workers nationally fell into this category. Wellington had the highest proportion of social workers with potentially unmanageable caseloads at 18.5%, followed by Canterbury (17.3%), and South Auckland (16.9%).

Improving performance and reliability

We can improve the performance and reliability of the predictive analytical model with:

• regular adjustment using updated surveys and data from other organisational systems, for example HR
• new structured datasets in the existing case management system, and
• relating information from other data sources for example, Health, Education.

There’s also an opportunity to analyse the unstructured data in the case management system. We can do this using new Large Language Model machine learning techniques to add to the predictive power of the model.