Recommendation 5: Develop and Implement Predictive Risk-Modeling
The final state-level solution involves expanding the role of the Bureau of Educator
Certification to include a Predictive Risk-Modeling department. Using organizational psychology tools
as well as business standards for predictive analytics, or risk-modeling, this department would be
responsible for developing a methodology to predict the turnover risk for each certified teacher or
principal in the state of Florida. These risk scores would be used to identify teachers and principals
at risk of exiting the profession and would be shared with districts on a regular basis. This risk score
could be contained as an element of the statewide Automated Staff Information System that each
district is required to maintain. Predictive Risk-Modeling algorithms currently used by the business
sector use employment data such as entry cohort, continuing education courses, outside job
prospects, and other variables to predict an employee’s likelihood to exit their role (Rosenbaum, 2019).
IBM estimates that their proprietary risk-prediction algorithm has a 95% accuracy rate and has saved
the company nearly $300 million dollars related to attrition costs (Rosenbaum, 2019).
This proposal is difficult to determine a cost for since it would require multiple years of
development and could potentially include a request for proposal process. The state would bear much
of the costs for implementing this program and any savings would likely be realized at the local
school district level rather than at the state level. However, given that much of the monies spent at the
local level are distributed at the state level, the state might be able to recoup a portion of any savings
that are realized. There are significant case studies from the business world that suggest attrition can
be substantially reduced, and that the reduced attrition in turn provides cost savings (Feffer, 2014).
There are relatively few examples of risk-modeling use in other areas of human services. One
notable exception is the Allegheny Family Screening Tool (AFST) used by Allegheny County,
Pennsylvania. This tool is “designed to improve child welfare call screening decisions” at the first
point of contact that a caller has with the local child abuse hotline (Vaithianathan et. al, 2017). Three
outside reviews of the methodology used to develop the tool as well as the tool itself were
completed. This process required a two-year evaluation and construction period before the AFST was
fully deployed.
This is a very politically ambitious proposal and there are several stakeholder groups who may
be concerned by this proposal. State legislators might be concerned with several facets of this
proposal, including the lengthy time between the initial legislative approval and the eventual
deployment of the tool, the uncertainty of the costs associated with this proposal, and the overall
ability of local school districts to use the information generated by this tool. Local school districts
may have similar concerns and may request additional training in order to understand and best use
the risk scores generated by FLDOE. Other stakeholders who may have significant concerns include
teacher unions. Teacher unions would likely suggest that risk scores could be improperly used by
principals or school districts, resulting in early terminations of teachers or principals with high risk
scores.
The researcher recognizes that there are significant political and practical hurdles to
implementing this recommendation. Though this area of analytical practice is valuable, it is extremely
unlikely to gain widespread adoption at this time. This recommendation could be considered by local
school districts, though it might be cost-prohibitive for a district to complete this level of
sophisticated analysis with their existing resources.
21