ERIC ED610285: Four Principles to Make Advanced Data Analytics Work for Children and Families

2020·0.42 MB·English

by  ERIC

four principles to make
ADVANCED DATA ANALYTICS
work for children and families
The Annie E. Casey Foundation is a private philanthropy that creates a brighter future
for the nation’s children by developing solutions to strengthen families, build paths to
economic opportunity and transform struggling communities into safer and healthier
places to live, work and grow. For more information, visit www.aecf.org.
© 2020, The Annie E. Casey Foundation, Baltimore, Maryland
introduction
ADVANCES IN DATA SCIENCE
and computing power are rapidly
shifting the opportunities available to citizens, changing how systems interact with
almost every aspect of our lives: as students, job seekers, patients, citizens and consumers.
T his document presents the l Expand opportunity. Advanced to achieve, more equitable
Annie E. Casey Foundation’s analytics should open doors outcomes for historically
point of view on the use to greater opportunity for disadvantaged groups.
of advanced analytics in social children and families, not merely
Taken together, these principles
programs and policy, based on incrementally improving the
direct a spotlight toward areas that
an exploration of the promise status quo.
need a great deal more attention,
and consequences of advanced
analytics for youth, families and l Provide transparency and investment, innovation and
evidence. Agencies and businesses consultation with families and
communities. It also outlines a
providing services fundamental affected communities. At a time
set of principles that distinguish
to the welfare of families and when the COVID-19 pandemic
between the useful, acceptable
communities owe the public has laid bare the inequities in many
and harmful applications of new
evidence about the use and public systems and prompted
tools. Developed through broad
effects of their decision-making reimagination of more effective
consultation with data scientists,
analytics. approaches, organizations and
civil rights groups, public leaders
advocates who care about children
and family advocates convened by
l Empower communities. These and families must not only
the Foundation, as well as extensive
data tools should empower
seize the opportunity to prevent
consultation with Foundation staff
communities to hold systems
emerging technology from causing
and leadership, these principles
accountable.
harm but also work affirmatively
represent points of agreement
to support and promote the
between those who are optimistic l Promote equitable outcomes.
constructive use of these tools for
about the good these tools can do New tools and models should
equity and progress.
for families and those who remain explicitly promote, and be
extremely concerned about their judged against their ability
abuse:
four principles to make advanced data analytics work for children and families / 1
THE ESCALATING ROLE
of advanced analytics and big data
Advanced data analytics are deeply embedded in how public and private systems make decisions, already
shaping the opportunities available to youth and families. Algorithms guide decisions on everything from which
neighborhoods police patrol to whether loan officers extend credit and which job applicants a hiring manager
selects to interview. (See Table 1.) With the volume, variety and velocity of information available on all of us (also
known as “Big Data”) growing dramatically in the past decade, these analytical tools will play an increasingly
central role in translating that information into decisions about families and communities in the coming decade.
BIG DATA is an umbrella term VVOOLLUUMMEE THE EMERGENCE OF
referring to the dramatic change in MASSIVE NEW DATASETS
the volume, variety and velocity of
information available to — and about
— all of us.
VVAARRIIEETTYY THE EXPANSION OF
DATA SOURCES
l VOLUME: The digitization of
all aspects of our lives and
environments has created massive
new datasets to mine for insight.
VVEELLOOCCIITTYY THE ACCELERATED PACE
OF INTELLIGENCE
l VARIETY: In addition to
administrative data intentionally
collected by programs, companies
can scrape social media sites and The new tools computer programmers and statisticians
handwritten case notes to create have developed to interrogate these large and varied
unanticipated composite records. datasets go by many names — such as machine learning and
artificial intelligence — and they introduce both additional
l VELOCITY: The speed of collecting opportunities and challenges compared to more familiar
and analyzing information
statistical modeling methods, as discussed below. But
is constantly increasing, as
whatever technique is used, the core questions raised by the
processors, bandwidth and data-
use of advanced analytics for prediction, risk assessment
science techniques improve.
and decision making are similar.
the annie e. casey foundation/www.aecf.org / 2
Table 1
APPLICATIONS OF ADVANCED ANALYTICS
l Affordable housing protection l Job awareness and employment l Pretrial risk assessment
decisions
l Criminal record expungement l School assignment and transportation
l Learning analytics in higher education
l Early warning systems in K–12 l Screening calls to child protective services
l Policing
l Evidence-based sentencing l Tenant screening
l Predicting risk of injury to youth
l Insurance pricing in foster care
Many uses of advanced analytics officers. The problem for civil have learned about adolescent
are controversial. Civil rights rights advocates isn’t chiefly bias development in the past decade:
advocates worry that more or naivety on the part of those Youth are exceptionally resilient
algorithmic and automated building these models — it’s and able to recover from past
decision-making systems will the racism built into the policy adversities but also vulnerable
worsen existing inequities across and historical practice of many to the harm caused by deep
lines of race, age, income and institutions. Data science can be engagement with child welfare and
gender. Organizers, academics used to combat these problems justice systems.2
and journalists have increasingly rather than worsen them, but
Such structural biases in decision
documented these problems in doing so requires putting these
making are not unique to risk
many advanced analytical tools in tools in the hands of different
assessments built on advanced
common use. (See the “Evidence- people with explicitly different
analytics. “Differentiation” and
Based Sentencing” example in intention.
“discrimination” are two sides
Appendix A.) Some have gone
This challenge is of particular of the same coin. Agencies and
as far as to call for abolishing
concern to organizations businesses must make decisions
these tools, arguing that agencies’
advocating for youth and youth about people all the time, often
application of advanced analytics
adults, who are more likely to be in situations of great uncertainty.
to the increasing kinds and quality
assessed by tools that treat age as In every one of those situations,
of data available to them increases
a liability rather than an asset. For managers already have screening
opportunities for discrimination.
example, an analysis of the most rules in place — implicit or
Algorithms implemented by common pretrial risk assessment explicit, simple or complex.
systems (public or private) will models — Correctional Offender The lessons that civil rights
tend to automate decision-making Management Profiling for organizations and system reformers
criteria already in place, learning Alternative Sanctions (COMPAS) have learned about implicit and
from data that capture how these Violent Recidivism Risk Score — structural bias, organizational
systems have historically allocated estimated that fully 60% of its risk incentives and the way they pollute
resources or punishment and score was a function of a person’s and skew data apply with equal
amplifying the existing incentives age rather than any criminogenic force to this newest generation
for frontline workers, including factor or case history.1 This practice of analytics tools and predictive
case managers, judges and loan runs contrary to what researchers algorithms.
four principles to make advanced data analytics work for children and families / 3
What’s more, new data and data Many of these predictive models improving the accuracy of an
tools have real potential to do great can be thoroughly tested before assessment disproportionately
good. Predictive risk assessments being implemented. They can benefits groups subject to
can be fairer and more accurate help more accurately identify and discrimination.
than the actuarial tools already in measure sources of disadvantage,
It is true that some data science
wide use throughout government making the mental models used
techniques introduce genuinely
and business. Each time systems by judges and social workers
new challenges when applied to
introduce new technologies into more explicit and transparent.
decisions about people’s rights
familiar decision-making processes, As prominent social scientists
and well-being. Programmers
advocates get an additional chance and practitioners argue, “when
increasingly rely on algorithms to
to critique and improve how algorithms are involved, proving
parse and identify patterns in vast
decisions affecting children and discrimination will be easier —
reservoirs of data and then use
families are made — but only if or at least it should be, and can
those patterns to make predictions
agencies and regulators include be made to be.” 3 In many cases,
about people. Often called
them as critical partners. the existing frameworks used to
machine learning, this approach
make decisions about youth are
can create spurious or dangerous
so fundamentally biased that
correlations unless data scientists
work closely with practitioners
to test their conclusions. What’s
more, the methods involved can
BASELINES FOR COMPARISONS
generate findings that are difficult
to interpret. For example, it
can be difficult to explain what
In judging whether a new application of advanced analytics
combination of characteristics
is fair and worthwhile, the answer depends largely on how
a machine learning algorithm
the question is framed.
identified as predictive of heart
Is the new tool: failure. In some contexts, this is
less important — just knowing
l fair and accurate, or at least as close as possible given the a patient is likely to be at risk
available data? can prompt a doctor to provide
preventive care. But when making
l an improvement over the process in use now?
high-stakes decisions about
whether a consumer qualifies
l better than any other reform that could be applied to this
for a bank loan or a young man
system?
is eligible for parole, being able
Which combination of these elements constitutes the to explain and defend that risk
“right” standard for review is a matter for debate — and it assessment becomes a matter of
is common for data scientists, public administrators and civil rights.
civil rights advocates to disagree.4
the annie e. casey foundation/www.aecf.org / 4
Upturn, a Washington, D.C.– among disadvantaged youth, l a lack of transparency, not just
based organization that advocates put it: “The real world is often about how the models work but
for more equitable use of more complicated than one-way also the lack of any notification
technology, uses this illustration tests. ... If it is a combination that these algorithms and
of data mining creating a spurious of characteristics (like age, “automated decision systems” are
correlation: A data scientist might neighborhood and family income) even in use; and
use machine learning techniques that defines who is benefiting
to analyze millions of electronic from a program, we are never l fundamentally different visions
for reform that focus on, for
health records and observe going to find that with our
example, decriminalization rather
that patients with an asthma standard ways of testing.” 6 These
than fairer enforcement. Relying
diagnosis are, surprisingly, less tools can help to model large,
on advanced analytics to mitigate
likely to die when hospitalized multivariate problems — like how
near-term problems, reformers
with pneumonia. Does asthma to simultaneously create a more
worry, might inhibit or foreclose
protect against pneumonia? Just equitable bus schedule for affluent
longer-term structural solutions.
the opposite: What the data and low-income communities
reflect here is that hospitals rank and rebalance school bell times to
While calls to “abolish” this whole
any asthmatic patient admitted create a later start for older youth
category of technology are not likely
for pneumonia as “high risk” and — that previously could not be
to work, these concerns are well
treat them with corresponding solved together.
founded, and organizations and
preventive protocols. A data
advocates that care about children
Despite the potential to use
scientist not familiar with hospital
and families have a vital role to
advanced analytics to improve
protocol would draw exactly
play. Advanced analytical tools
decision making for youth and
the wrong conclusion about the
are already embedded in business
families, many prominent civil
relationship between asthma and
and government operations, and
rights organizations in the country
pneumonia from the data mining
increasingly in state and federal
have aligned in opposition to them
alone.5
policy. Whether they benefit or
— and a new generation of data-
harm communities will depend
Advanced analytics also empower driven civil rights organizations
on their design, use and oversight;
agencies and child advocates to such as Data for Black Lives have
the technologies themselves are
create new, never-before-possible grown up specifically to guard
neither “good” nor “bad.” It’s
solutions. For example, machine against their misuse. Among the
critical that communities and
learning tools are uniquely helpful most frequently cited reasons
advocates contribute to a policy
for studying how youth move for curbing or ending the use of
framework and set of professional
through complex social systems algorithms in decision making are:
standards that can protect children
and refining evidence on which
programs work for whom, under l distrust in the capacity, incentives from negligent or harmful uses of
and historical practices of systems advanced analytics and highlight
what circumstances. As Sarah
that control the underlying data where these tools can be used
Heller, a leading researcher on
and decision-making processes; to transform ineffective and
strategies for reducing violence
inequitable systems.
four principles to make advanced data analytics work for children and families / 5
The social sector cannot stay on departments to end predictive benchmark. This challenge can —
the sidelines of this debate. Public policing (as happened recently in and should — surface fierce debate
agencies, philanthropies and both Los Angeles and Chicago); about the risks and opportunities
policymakers will have to make and presented by new kinds of
judgment calls about uses of algorithmic decision making.
advanced analytics. Situations that l a need to modernize our existing
approaches to risk assessment, But surprisingly broad agreement
can bring about such judgment
most of which have well- exists about the values that should
calls include:
established flaws and biases. guide how these tools are used and
l the growing number of regulated. The implications of
The challenge facing the field is
deliberative city and state task these values on policy and practice,
how to distinguish among the
forces on public algorithms; explored in the recommendations
useful, acceptable and harmful
below, outline a vision for fair and
l calls from citizens and activist applications of advanced data effective use of advanced analytics
groups, such as those organized analytics, and to help practitioners
that looks very different from the
to “Stop the Cradle to Prison and advocates use the consequences
investments being made today.
Algorithm” in St. Paul Public of these tools for youth,
Schools or to force police families and communities as the
the annie e. casey foundation/www.aecf.org / 6
4 PRINCIPLES TO USE ADVANCED ANALYTICS FOR GOOD
expand
OPPORTUNITY
Most of the established uses of advanced analytics
in education, social services and criminal justice
direct the attention of counselors and social workers to potential
problems facing youth and families — for example, dropouts, abuses
and infractions. Interventions are frequently intended to prevent worse
harm to the individual, family or community, but very often they result
in punishments that disproportionately track along familiar lines of age,
gender and race. An alternative and encouraging use of advanced analytics
focuses not only on risk but also on identifying so-called odds beaters and
new opportunities for youth. Systems also might use advanced analytics
to identify untapped talent, sequence more successful pathways into the
workplace and — as Los Angeles recently did — expunge old charges
from criminal records.
four principles to make advanced data analytics work for children and families / 7
An example of the benefits: Good intentions, however, can cause harm if the
analytics effort is not well designed.
The Children’s Data Network at the University of
Southern California is helping the state’s departments Chicago police used advanced analytics to develop
of education and social services move beyond a “heat list” of individuals more than 500 times as
documenting their poor results with high-needs likely as the average city resident to be involved in
students (especially youth who have been in foster gun violence, as either a victim or shooter. Despite a
care, experienced homelessness, run away or lived in commitment to follow up with these mostly young
migrant families) to begin investing in doing better. men with prevention services, the police “failed to
Their project, Exploring Resilience Among Vulnerable provide any services or programming” and “instead
Students in California, explores why some students increased surveillance and arrests.” 8 A 2016 study
succeed, despite these negative experiences, and will by the RAND Corporation found no perceptible
identify protective factors — individual, family and change in gun violence. As the author wrote, “to make
school experience — that merit more investment.7 a long story short: it didn’t work,” and the project
demonstrated how initiatives aimed at personalizing
support for young people who need it can shift
on the ground to targeting “chronic offenders” for
incarceration.9
Recommendation: This is a promising but underexplored area for research and investment. Government and its
philanthropic partners, in particular, will need to test whether novel applications of data science (like so-called
“precision analytics”) can create new insights for the field and under what circumstances it is responsible to apply
them. These advanced analytics are likely to be particularly helpful in understanding and tailoring opportunities
for complex populations — for example, youth involved in several systems and youth aging out of foster care.
the annie e. casey foundation/www.aecf.org / 8