Study looks at ways FEMA can improve data-based decisions
There’s no uniform plan for disaster response because needs vary widely from one event to another. Even so, the Federal Emergency Management Agency (FEMA) is focused on improving services to disaster survivors, especially when it comes to making sound data-based decisions.
In particular, studies by the Government Accountability Office (GAO) have recommended that FEMA can do better at using data to plan for day-to-day operational decisions, such as locating and staffing the disaster recovery centers (DRCs) that serve survivors.
Urgency often privileges effectiveness above efficiency, and political dynamics can also factor into decision making. Any fresh approach would have to work within FEMA’s established policy and regulatory framework, be reasonably flexible to accommodate each unique situation, and prove both valuable and trustworthy.
In a recent article for Decision Sciences, researchers Julia Moline (now with FEMA), Jarrod Goentzel, and Erica Gralla presented two models for locating and resourcing DRCs, evaluated cost savings and service improvements compared to actual operations, and described the issues surrounding their ongoing implementation at FEMA. One model aligns with current decision-making constraints and culture, while the other suggests more sophisticated approaches.
DRCs are set up across a disaster-affected community, usually within a week after the event. These centers provide in-person access to specialists who help with applying for financial aid, temporary housing, funds to repair property, small-business support, crisis and legal counseling, and other assistance. Disaster “survivors” (FEMA’s term) interact first with a greeter, who helps identify needs, then proceed to relevant service stations and, finally, an exit interview. Visits can take minutes or hours, plus wait times.
DRCs are typically set up inside community centers, schools, or other public buildings, and can be open for a week to 18 months. They can only be opened in counties where the United States President has made declarations of disaster and individual assistance.
At the time of the research, FEMA did not follow a standard process for allocating DRCs and there was no written, standardized guidance for opening and closing the centers.
The decision-making process considered location; the anticipated need for DRC services; the need for public transport or survivors’ willingness to drive long distances; the post-disaster state of transportation infrastructure; equitable access for all populations; the perception of fairness across political and jurisdictional boundaries; the need for a visible response presence; and the availability of appropriate facilities.
Generally, FEMA wants a more efficient process for reducing costs and improving service.
DEVELOPING AND TESTING TWO APPROACHES
The researchers developed simple mathematical models to support two types of decisions: (1) siting and sizing DRCs and (2)adjusting staffing over time and eventually closing those DRCs. Dimensions for the service level are twofold: travel time for survivors to reach a DRC, as well as the number of people who cannot be served due to limited capacity. Costs are defined based on both fixed and variable expenses for opening and operating a center.
The researchers then developed two distinct approaches. The first, the jurisdiction model, operates within existing regulations that require separate decision making from county to county. The second, a travel time model, ignores county boundaries and locates DRCs within a specified travel time for all potential DRC visitors.
The outcomes of these two models were compared with each other, as well as against the actual decision-making processes in three past disasters. These events spanned different disaster types and areas (rural or urban), but all occurred in 2013: flash floods in Colorado (FEMA Disaster Number 4145), flash floods in Illinois (4116), and tornadoes in Illinois (4157).
THE TWO MODELS
The jurisdiction model follows current FEMA regulations under which individual counties decide on the site and size of DRCs. The model is a simple algorithm designed to lower costs and ensure adequate service by opening and staffing DRCs only when the expected demand in a county meets a minimum threshold.
The chief difficulty lies in determining that threshold, along with staffing requirements. For their research, each of these quantities was estimated based on FEMA policies and data from recent disasters. Generally, FEMA sets that threshold at one visitor per staff hour, or90 visitors per week.
To determine the number of DRCs required in a county and their staffing levels, it was assumed that initial DRC capacity must meet the expected peak number of weekly visitors, or the largest number of visitors expected in any single week over the projected duration of need.
It is less expensive to operate a DRC at maximum capacity than to open additional centers. (For simplicity, the model assumes all DRCs operate at the same capacity, which is reasonable because it is also assumed they operate at or near maximum capacity if more than one is needed).
The travel time model differs from the jurisdiction model in two key ways. First, researchers ignored jurisdictional lines and chose DRC sites for the entire disaster-affected area. Second, there is no minimum threshold for opening a center; instead, the model determines a set of locations that every expected visitor can reach within a set travel time. Cost is minimized and service is improved. The model also requires sufficient capacity to serve every expected visitor.
DRC visits typically peak within two weeks after a disaster, then decrease steadily. Therefore, staffing should be reduced and DRCs closed accordingly. Initial decisions created by the jurisdiction and travel time models should be revised weekly based on demand: for the jurisdiction model, decisions continue to be made county by county, whereas the travel time model considers the entire system.
The jurisdiction model does not define specific locations for DRCs within each county, but rather assumes the existing decision process will be followed. Therefore, researchers used the DRC locations that were opened in reality, and when the jurisdiction model called for fewer centers in a given county, they selected the largest cities and/or most central locations.
Also, the actual number of visitors for each week was used to project the next week’s visits as the basis for updating staff hours and closing DRCs. Because some actual locations were not opened by using the researchers’ models, those visitors were assigned to open DRCs. In the jurisdiction model, DRC visitors were assumed to visit the open DRC nearest to the one they actually visited.
In the travel time model, actual visitors were distributed among DRCs using the proportion of each town’s visitors assigned to each DRC by the mixed-integer program, which took into account their travel time.
In the first week of operation, the capacity is set by the jurisdiction and travel time models. Thereafter, it is adjusted based on the actual number of visitors from the past week. In each of the three disasters used for this study, the actual capacity far exceeded the total number of visitors (providing between 318% and 1,206% of the required capacity in the first week).
In every case, both models made an initial decision that resulted in a capacity much closer to the actual demand (78% to 257% for the jurisdiction model; 94% to 300% for the travel time model).
These figures underscore the importance of combining models for the initial DRC allocation decision with those for ongoing operations. Because conditions and data quality change rapidly, it is critical to be able to revise decisions quickly.
The jurisdiction model occasionally provides less capacity than required (4-6%) because its county-by-county decisions do not account for visitors from counties without a DRC. In contrast, the travel time model considers the expected visitors across all affected counties and plans accordingly.
The second aspect of service level is travel time, which was significantly reduced by the travel time model in all three sample disasters. The model did not necessarily reduce average travel times, but it did reduce maximum travel times. This was a result of openingfewer DRCs – at a large cost savings to FEMA – while still serving the affected population. The jurisdiction model’s performance varied widely across the three disasters, mostly due to differences in city and rural-area traffic congestion.
Finally, both models were able to reduce costs in the three sample events by 55-85% in the first week and by 66-88% over the entire 15-week study period. The savings are largely explained by the number of DRCs opened, although staff-hour reductions also contributed.
Comparing the two models, the jurisdiction model appears to lead to less expensive DRC deployments (but also to less capacity overall).
DISCUSSION AND IMPLEMENTATION
Both models offer major improvements over the current method of DRC allocation. Cost savings are significant with both – $158,000 to $1.5 million in the first week alone – while service is sufficient – at worst, 78% of required capacity in the first week and 94% in the first four weeks. The travel time model additionally guarantees shorter maximum travel times.
FEMA’s implementation of this study thus far has included weekly visitor forecasts, staff throughput capacities, and thresholds for closure.
Among the barriers to implementing the full process is that FEMA shares decision-making duties with other entities that vary from region to region and disaster to disaster. State officials may have different goals, approaches, and considerations. In addition, despite FEMA’s growing interest in data-driven decision making, a tendency remains to rely on experience.
Still, successful partial implementation of these models shows that DRC demand can be usefully predicted, and that models can provide cost-saving and service-enhancing solutions. Significantly, the success to date belongs to the jurisdiction model, which aligns with FEMA’s current decision-making culture and practices, requires no specialized software, and is transparent and easily understood. However, the travel time model remains available as a further step. A combined approach could be used: the jurisdiction model could determine the number of DRCs for each county, and a modified travel time model could be used to guide locations. This sort of hybrid approach might be adopted later or in stages, as the trust of disaster-response officials grows.
Citation: Moline, J., Goentzel, J., and Gralla, E. (2019). Approaches for locating and staffing
FEMA’s disaster recovery centers. Decision Sciences 50(5), 917-947. https://doi.org/10.1111/deci.12359
Julia Moline is Deputy Chief, Program Delivery Branch, Public Assistance Division, at the Federal Emergency Management Agency (FEMA). She completed her master’s degree in MIT’s Technology and Policy Program.
Jarrod Goentzel is founder and director of the MIT Humanitarian Supply Chain Lab in the MIT Center for Transportation & Logistics. His research focuses on meeting human needs in resource-constrained settings through better supply chain management, information systems, and decision support technology. He received a Ph.D. from the School of Industrial and Systems Engineering at the Georgia Institute of Technology.
Erica L. Gralla is an assistant professor in the Department of Engineering Management and Systems Engineering at the George Washington University. Her research focuses on decision-making in real-world contexts. She seeks to combine the strengths of human intuition and mathematical models to create better decision-making approaches. She completed her Ph.D. degree in the Engineering Systems Division at the Massachusetts Institute of Technology.