Last week The Economist claimed data is the world’s most valuable resource — the oil of the 21st century. This enthusiasm is well placed. Phone companies predict customer churn based on social network data. Spending patterns help banks predict credit worthiness and segment their customers. Sommeliers even use rainfall data to predict wine quality. Yet use of data to improve government service delivery remains stuck in the 20th century.
We’re all better off when government spending is effective. Which students should receive mentoring and tutoring to prevent dropout? Which job-search services are best suited for a former factory worker recently made redundant? Should a defendant arrested for a low-level drug offence await trial in jail, or at home?
These are decisions that allocate scarce government resources with human and financial consequences. Predictive analytics — techniques that use data analysis to make predictions — can help.
Like on trading floors, decisions in schools should rest on a foundation of predictive analytics. There has been considerable progress in recent years. Wisconsin, among other states, has developed a Dropout Early Warning System (DEWS). Each year, 225,000 Wisconsin middle school students are assessed for their likelihood of dropout or late graduation. Early results from a randomized controlled trial for 73 Midwest schools are promising — decreased failure and absenteeism after only one year using a DEWS.
More could be done. Many existing early warning systems are not arduous, using measures such as achievement, attendance, behaviors, and mobility. Yet only half of all U.S. high schools have them.
For schools that already have early warning systems, emerging research suggests student social-emotional well-being and school culture are important lead indicators. Examples of social-emotional measures include emotion regulation, perspective taking, and adoption of a growth rather than a fixed mindset to one’s learning. Collecting and acting on this data is in line with the recently passed Every Student Succeeds Act. School culture, meanwhile, includes a sense of belonging, safety, and perceived support for academic learning. Incorporating these measures in early warning systems presents exciting opportunities for student specific and system-wide improvement.
Preventing long-term unemployment
More than 2,500 job centers support many of the roughly 25 million Americans who experience unemployment each year. These centers offer workforce development services ranging from job search, career counseling, and subsidies that supplement wages. Job seekers are a diverse group. The central challenge is targeting the right services to the right person at the right time.
New Zealand, through its “investment approach,” uses job seeker characteristics to estimate unemployment duration. A young adult who dropped out of high school with a patchy job history is at high risk of long-term unemployment. In this case, early investment in intensive job services seeks to prevent entrenched welfare dependence. Meanwhile, someone who holds a bachelor’s degree with extensive work experience will likely find employment quickly, so would only need light touch services.
While it is early days, welfare dependency in New Zealand is decreasing. Critics have rightly raised concern that people should not be punitively pushed off welfare. Instead, implemented effectively, the investment approach targets limited resources at individuals who will benefit most to sustainably reduce long-term unemployment.
Every day judges decide whether defendants should be detained in jail on remand or released while they await trial. At $85 per person per day, the cumulative cost of holding defendants in jail is substantial. A recent study proposes using defendant characteristics like their criminal record, age, and arrest location to predict reoffending risk. This information is used to identify low-risk defendants who should be released. The author's predictive analytics model is highly effective with the potential for significant declines in both jail populations and crime rates.
Moreover, using the model, currently over-represented groups — African Americans and Hispanics — would constitute a lower proportion of total prisoners. Judges, while doing their best, make mistakes. Supported by data, they can achieve much better results.
Limitations and conclusion
Predictive analytics are not a panacea. First, government is trying to address problems that are complex — many of which cannot be solved by using better data alone. Second, prediction is no substitute for human judgment. For instance, algorithms should complement, not replace the deliberations of a judge. Third, models can be wrong and data inaccurate. Finally, if misinterpreted, models risk becoming deterministic. If dropout risk, for example, is seen as a diagnosis rather than a prediction, teachers may be inclined to give up on students who need the most help.
There are a myriad of predictive analytics applications that mean your tax dollars can be spent more effectively. Private sector companies have already learned that data-driven decision-making means smarter, more efficient decisions. Taxpayers ought to demand that government follow suit.
Matt Tyler is an economist who works to improve government effectiveness with a particular focus on social services. Tyler is a former management consultant, where he supported executives in developing and implementing strategy across financial services, telecommunications, manufacturing, postal services, and retail. He worked as an economist for Australia’s foreign service and as a policy adviser to the Federal Australian Labor Party on economic and social policy. He has also worked for Third Sector Capital Partners where he assisted with the construction of two Social Impact Bonds in Salt Lake City. He is currently completing a Master of Public Policy at Harvard’s Kennedy School of Government. He tweets as @matt_b_tyler. To read more of his reports — Click Here Now.
© 2021 Newsmax. All rights reserved.