Non-linear incentives, plan design, and flood mitigation: the case of the Federal Emergency Management Agency's community rating system

Abstract

A basic proposition of ‘agency theory’ is that output-based performance incentives encourage greater effort. However, studies find that incentive schemes can distort effort if rewards for performance are discrete or non-linear. The Federal Emergency Management Agency's (FEMA) Community Rating System (CRS) is a flood mitigation programme with a non-linear incentive design. Under this programme, localities are incentivised to implement a mix of 18 flood mitigation activities. Each activity is performance scored, with accumulated scores corresponding to a percent discount on flood insurance premiums for residents that hold National Flood Insurance policies. Discounts range from 0 to 45% and increase discretely in increments of 5%. With multivariate statistical and Geographic Information Systems analytic techniques, tests are made to find whether observed changes in annual CRS scores for participating localities in Florida are explained by non-linear incentives, adjusting for hydrologic conditions, flood disaster histories, socio-economic and human capital controls that can plausibly account for local mitigation activity scores over time. Results indicate that local jurisdictions are discount-seeking, with mitigation efforts partially driven by the non-linear incentive design of the CRS programme. The paper ends with recommendations to improve the operation FEMA's flood mitigation programme.

Keywords:

• Florida
• flood mitigation
• community rating system
• non-linear incentives

Acknowledgements

This paper is based on research supported in part by the US National Science Foundation Grant No. CMS-0346673. The findings and opinions reported are those of the authors and are not necessarily endorsed by the funding organisations or those who provided assistance with various aspects of the study.

Notes

 1. Floods are the most lethal natural disaster in the United States (US). According to data from the Spatial Hazard Events and Losses Database for the United States (SHELDUS), 2353 persons were killed and 17,129 people injured by flood events from 1960 to 2005. Over this time period, fewer people were killed by hurricanes, tropical storms and tornados combined. The Centers for Disease Control mortality database corroborates SHELDUS estimates, showing 2164 persons killed by floods and cataclysmic storms from 1979 to 1998 (Zahran et al. 2008b). The economic impact from flood events is substantial. According to data from the Spatial Hazard Events and Losses Database for the United States (SHELDUS) from 1960 to 2005 flood events caused, on average, $1.16 billion dollars of property damage, and $335 million dollars of crop loss every year (figures expressed in 2000 inflation adjusted dollars).

 2. Series 300 (public information) activities measure the extent to which a locality informs local populace about flood hazards, insurance and protection measures. Series 400 activities (maps and regulation) measure regulatory enactment and enforcement activities that exceed the NFIP minimum standards. Series 500 activities (damage reduction) involve damage reduction measures such as acquiring, relocating or retrofitting existing buildings and maintaining drainage and retention basins. Series 600 activities (flood preparedness) co-ordinate local managerial efforts to minimise the effects of a flood on people, property and building contents.

 3. To provide readers with more detail on how public outreach activities are scored, the paper excerpts from the CRS website ( http://training.fema.gov/EMIWeb/CRS/m5s4main.htm). Outreach projects to the community are scored as “the sum of the points for each topic covered in written information sent to all properties in the community through a newsletter, utility bill, telephone book, or other document sent to everyone. A newspaper may be used if the information is not in a legal notice, small classified ad, or similar obscure location. The project must cover one or more of the 10 topics at least once a year to at least 90% of the properties in the community. Full coverage of each topic is worth six points. All the topics do not have to be covered in the same distribution, but the distribution must ensure that the topics credited are covered at least once each year. For example, a community with a quarterly newsletter may cover two topics in each edition and be credited for covering eight each year. There is no impact adjustment for this activity. For this credit, the outreach project must be sent to at least 90% of the properties in the community. ‘Properties’ can be counted as utility customers, tax parcels, or other measure that approximates all of the addresses in the community. Vacant lots need not be counted. Generally a distribution to all taxpayers, water customers, or property owners is considered 100% distribution. A commercial newspaper can only be counted if the community can document that it reaches 90% of the properties in the community”. CRS managers are trained in the assignment of points for mitigation activities undertaken. Training workshops are routinely held to educate managers in the CRS grading scheme. These workshops are designed to improve inter-coder reliability in scoring.

 4. It is important to note that official CRS literature de-emphasises the discount incentive. The benefits of participation in the CRS programme emphasised include free technical assistance, access to Federal assistance programmes, quality of life gains by protection of the environment, insulating residents from casualty risks and protection of private property. With regard to discount incentives, the literature states: “Few, if any, of the CRS activities will produce premium reductions equal to or in excess of their implementation costs. In considering whether to undertake a new floodplain management activity, a community must consider all of the benefits the activity will provide, not just insurance premium reductions in order to determine whether it is worth implementing” (italics our emphasis, http://training.fema.gov/EMIWeb/CRS/m1s3main.htm).

 5. For example, during a two-day period in early October of 2000 Broward, Collier, Miami-Dade and Monroe Counties collectively received over a foot of rain, causing over $450 million dollars of flood damage (NCDC 2005), prompting over 51,000 individuals to request financial assistance from the Federal Emergency Management Agency.

 6. The list of participating county governments includes: Alachua, Bay, Brevard, Broward, Charlotte, Clay, Collier, Columbia, Dade, Escambia, Franklin, Gulf, Hardee, Hendry, Hernando, Highlands, Hillsborough, Indian River, Jefferson, Lee, Levy, Madison, Manatee, Marion, Martin, Okaloosa, Orange, Osceola, Palm Beach, Pasco, Pinellas, Polk, Santa Rosa, Sarasota, St. Johns, St. Lucie, Sumter, Suwannee, Taylor, Wakulla, Volusia.

 7. A 10-year rolling average is both theoretically and practically reasonable. From theory and empirical evidence, it is assumed that flood mitigation planning is incremental. On the question of institutional change, Nobel laureate Douglass C. North (1995, pp. 19–20) maintains that, “the overwhelming majority of change is simply incremental and gradual”. A single flood event may not be enough to induce local institutional change as much depends on prior probability distributions. It is presumed that a locality is more likely to do costly mitigation if flood events are recurrent and cumulate in institutional memory. Practically, a 10-year rolling average sufficiently smoothes the noise in flood frequency and intensity measures.

 8. The SHELDUS consists of a county-level inventory of 18 natural hazard types, including hurricanes, floods, wildfires and drought. Hazard event records include a start and end date, estimated property damage and crop loss, as well as the number of human injuries and deaths. SHELDUS data are derived from public sources such as National Climatic Data Center monthly publications and NGDC's Tsunami Event Database. The data are limited to disaster events that cause more than $50,000 in crop loss or property damage.

 9. It is important to note that local government participation in the CRS programme is not synonymous with policy holder savings – in many localities lower than expected counts of policy holders are observed, and relatively low dollar savings. Indeed, in an interview conducted by the lead author with members of the CRS Task Force, officials noted that in at least three CRS localities zero NFIP policy holders are observed.

10. As noted by a thoughtful anonymous reviewer, there is no upper bound on ‘between interval growth’ while ‘within interval growth’ is capped at 499 points by interval widths. To eliminate the analytic risk that observed differences in interval condition are a function of a selection mechanism that advantages non-bounded ‘between interval growth’, the study truncated 17 place/year between interval growth observations that exceeded 499 points. Comparisons reported in Table 3 reflect this statistical adjustment. Even when the 17 between interval high growth place/year observations are dropped from the analysis (biasing down the expected result), strong evidence is found of strategic behaviour (0.99% vs. 33.25%, t = −15.4865).

11. Recall, the classification scheme moves in a counter-intuitive direction – a class rating of 9 in worst than a class rating 8, with a class rating of 1 constituting superior flood mitigation behaviour. The direction of class scores were inverted for a more intuitive interpretation. Therefore, a unit increase in CRS class rating increases the odds of within-place stalled growth by a multiplicative factor of 56.85 (95% CI, 19.2–168.02).

12. Indeed, a related analysis of the CRS programme found that Florida localities disproportionately pursue class 300 (public information) and class 400 (mapping and regulation) activities while at the same time under-pursue class 500 and 600 activities. For example, in 2005, almost 74% of total points earned (on average) came from class 300 and 400 interventions, a figure 28% higher than the proportional weight assigned to these activities. Florida localities (on average and over the time period assessed) deviate below the proportional weight of class 500 scores by 23.49% (Brody et al. 2009).