Whenever we apply computer technology, operations research and statistics to interpret meaningful patterns in data, we are using analytics. Predictive analytics takes that process a step further.
The process mines historical and current data and then uses modeling in order to predict future events. For example, an analytical model can draw on past events and behavior patterns to predict the probability that child support payments will be honored in specific cases. Cases can then be scored to allow child support workers to determine their work priorities.1
In other words, predictive analytics facilitates data-driven decision making. Most governments already have information they could use to identify people in need of assistance. It’s what they do with the data that is changing.
The potential of predictive analytics is coming to the fore in the way that governments deliver human services. Greater interagency coordination of services such as housing, mental health, child protection and addiction treatment becomes possible, bringing with it new insights and opportunities for innovation.
In the human and social services field, predictive modeling is highly effective in assessing various risks by scoring their probability. With that information, investigators can focus valuable resources where adverse events are most likely to occur.
Research linking birth data to later risk of maltreatment has been influential in improving child protection practice. Researchers at the Center for Social Services Research at the University of California at Berkeley, for example, released a study in 2011 that tracked over two million children. They were able to identify specific factors at birth that were linked to higher rates of reported or substantiated abuse by the time a child reached age five.2
Similarly, the California Child Welfare Indicators Project (CCWIP) provides an open-source database of customizable information on the state’s entire child welfare system. Data can be filtered by year, county, age, ethnicity, gender, and placement type, among other categories. Child welfare agencies are thus increasingly able to identify in real time the children most at risk from being harmed – and to target their interventions accordingly.
New Zealand is launching a predictive tool focused on those who need services, rather than those who provide them.3 SmartStart gives parents a customized timeline based on their personal profile. It provides tailored information about early childhood services, and establishes a newborn’s future digital relationship with government. The service can also be used by professionals working with expectant mothers or new parents, but it’s the parents who control which agencies have access to their information.
Big data and predictive analytics are also supporting efforts to combat homelessness. CAMBA, a non-profit agency in New York, offers a prime example.4
The agency takes an integrated approach to programs in education, youth development, family support, health, housing and legal services. In partnership with the City of New York, it has developed the innovative HOMEBASE program, which is premised on the concept that preventing homelessness is more effective – and less costly – than reversing it.
That means using data to identify those at relatively high risk of becoming homeless, and simultaneously to direct resources most efficiently. Among key risk factors are prior use of a shelter, moving more than four times within a 12-month period, being under age 28, or being a young single mother.
Proprietary software is used to map these factors and related data onto the widely available Google Maps interface, producing color-coded dots to show concentrations of people at the greatest risk for homelessness. Different colors represent factors such as a prior visit to housing court or to a shelter. Field workers can click on a dot to see family composition and other details, and then organize their caseload accordingly.
Predictive analytics offers significant benefits to those who provide human services. Not least are vastly improved targeting of services and benefits – often resulting in greater job satisfaction for service providers – and significant savings to the public purse.
As an example, the United States Social Security Administration (SSA) has established a fraud prevention unit dedicated to building data analytics to help workers detect and avert scams. The approach involves applying analytics to determine common characteristics and meaningful patterns of fraud, based on data from past allegations and known cases of fraud. The predictive tools increase the ability of the SSA to identify suspicious patterns of activity in disability claims and to prevent fraudulent applications from being processed.
Similarly, the Massachusetts health and human services department, MassHealth, is using predictive modeling to combat Medicaid fraud. In the first six months after the system was launched in May 2013, investigators recovered US$2 million in improper payments and avoided paying hundreds of thousands of dollars in fraudulent claims.5
Predicting service demand
These types of tools can also help agencies to identify where their resources will be needed in future. The Administration for Children’s Services (ACS), for example, administers child welfare in New York City, but it had no method of developing a profile of the future foster care population or predicting the facilities and resources this population would need. With KPMG, it built a predictive model to support forecasting, planning and budgeting.
The modeling approach not only accounted for events while an individual was within the child welfare program but also developed child demographic characteristics, and then evaluated the extent to which these events and demographics could determine the probability of a child staying within or leaving the program.
The model can be re-run with updated data to refresh forecasts for the number and types of services and facility places needed over any given period. That means the ACS is better able to plan and budget, and therefore to negotiate with vendors and facilities. In other words, improved service planning leads to improved service responsiveness and efficiency.
Predictive analytics is an exciting field, but it is not without challenges and potential limitations. It is important to be aware that predictive models generate probabilities, not facts. The undoubted power of such analytical tools may encourage heightened responsiveness from service providers, when appropriately targeted responsiveness is the goal.
The sensitivity of the data to be harnessed can be an issue as well. Interagency cooperation is essential to overcome restrictions to data access that could hinder the successful application of analytics.
What comes first? Robust data and analytics (D&A) requires cooperation among involved parties across departments and at every stage of the process. But recent research by KPMG has shown that organizations around the world still need to build trust in their D&A – 60 percent of them say they are not very confident in their D&A insights.6 Decisions that are based on inaccurate predictions will quickly erode the confidence not only of frontline workers but also of the decision makers themselves.
Because predictive models are built on historical events (such as known fraud cases), they forecast accurately only when scoring new data that contains similar relationships to those in previous data. As patterns of behavior change, predictive models must be reprogrammed with examples of the new behavior, or else they quickly become outdated and can no longer make accurate predictions.
Data collection, interpretation and sharing clearly stands or falls on the quality of the information gathered – and on determining the most appropriate response to the results. It’s evident that a genuinely transformative use of predictive analytics depends on the people making data-driven decisions, but most individuals find the operation of algorithms and models too opaque to verify – we don’t know how they work – and that complexity can create a trust gap.
KPMG defines four ‘anchors of trust’ that must underpin the successful application of analytics: quality, effectiveness, integrity and resilience. To strengthen those anchors, decision makers must take a systematic approach to managing the D&A lifecycle: from data gathering, through analysis and modeling, and ultimately to service delivery and performance measurement. Trusted, high-quality processes and timely, effective responses are key to using data in the service of the citizen.
1Dr Steven Golightly, Los Angeles County Child Support Services, Predictive Analytics: Los Angeles County CSSD (PDF 336 KB), presentation to Western Interstate Child Support Enforcement Council (WICSEC) 30th Annual Training Conference, 20–24 October 2013, Kansas City, MO.
2See E. Putnam-Hornstein and B Needell, “Predictors of child welfare contact between birth and age five: An examination of California’s 2002 birth cohort” (PDF 236 KB), Children & Youth Services Review 33, no. 11 (2011): 2400–07.
3New Zealand, Department of Internal Affairs, “SmartStart: The best start for parents and babies”.
4The discussion of CAMBA is based on information from Forbes Insight, Digitizing Human Services: Field notes and forecasts from the front lines of government’s technological transformation (2015), 27.
5Stephanie Kanowitz, “Social Security to step up fraud detection with predictive analytics”, GCN magazine, 22 April 2014.
6KPMG International Cooperative, Building Trust in Analytics: Breaking the Cycle of Mistrust in D&A, 2016.