Challenges to Coordination: Understanding Intergovernmental Friction During Disasters

While idealized crisis response involves smooth coordination between relevant actors, friction between levels of government and between the state and civil society in responding to catastrophe may be more common. This article builds a theory of cross-level friction during and after crisis by analyzing the conditions when discord is most likely. With a medium-N dataset (N = 18) of disaster responses from, among other countries, Chile, Haiti, Japan, North America, the Philippines, and Somalia, I carry out quantitative and qualitative analysis of cases with a variety of levels of friction to investigate the conditions that lead to misalignment. Tobit regression, qualitative comparative analysis, and case studies that take into account levels of economic development, government structure, nongovernmental organization density, and levels of damage demonstrate that low levels of development, lower levels of economic costs from the crisis, and poor planning and logistical infrastructure correlate with a higher likelihood of friction between disaster response stakeholders. Although not definitive, these findings come with theoretical and practical implications as climate change makes extreme weather events and future disasters more likely and more powerful.


35
All societies hope to minimize the casualties, property 36 damage, and business interruption that come with disasters 37 that, because of human induced-climate change, are 38 becoming more costly (Below and Wallemacq 2018). A 39 number of policies reduce the human and economic costs 40 of crises, including preventing settlement in vulnerable 41 areas through zoning laws, wide scale evacuation before 42 the event, and rapid and organized response from local and 43 national authorities (Metaxa-Kakavouli et al. 2018). Suc-44 cessful post-disaster outcomes are not as common as 45 observers would like. Experts have pointed to a handful of 46 well-coordinated responses following disasters, including 47 the intensive search after the Columbia shuttle disaster that 48 resulted in locals and professional responders rapidly 49 locating the bodies of the astronauts and critical debris 50 from the shuttle. Similarly the largest boat evacuation in 51 history after the 9/11 terror attacks on the Twin Trade 52 Towers in New York City resulted in the successful 53 evacuation of 500,000 people in an afternoon thanks to 54 coordination between the Coast Guard and private boats 55 (Boin and Bynander 2014). Successful disaster responses 56 facilitate strong and rapid population return, generate 57 access to capital for affected businesses and entrepreneurs, 58 create stocks of permanent housing for the displaced, and 59 rebuild the social safety net of schools, hospitals, and social 60 services (BCLC 2010). 61 But failures in coordination and communication during 62 disaster response have become so common that they are 63 expected (Boin and Richardson 2015). Some USD 95 64 million of international aid intended for victims of the 65 1984-1985 famine in Ethiopia was diverted to local war-66 lords and corrupt aid officials, resulting in the deaths of 67 many from preventable starvation. Satellite images from 120 This article contributes to the growing literature on 121 disasters and resilience in several ways. First, rather than 122 merely describing cases of cooperation or discord, this 123 article uses mixed methods to qualitatively and quantita-124 tively evaluate the importance of various factors in coor-125 dinating disaster response. Where past investigations of 126 friction during crisis have often relied on a single 127 methodological approach-most commonly case studies-128 this attempt moves between regression, QCA, and case 129 studies. Rather than avoiding research topics that require a 130 resource-and time-intensive data collection process, this 131 research points to the potential for medium N investiga-132 tions for future research. 133 Next, where previous studies of disaster response have 134 often relied on the analysis of a single case or handful of 135 cases, this study uses data from 18 catastrophes around the 136 world to avoid making inference based on a narrow range 137 of cases from a single country or disaster (Johnson and 138 Olshansky 2017 (Girdwood 2013).
170 This is true even in policy fields that do not involve 171 potential loss of life, extensive scrutiny from media, and 172 high levels of regulatory oversight (if occasionally only ex 173 post) as we find in disasters. For example, ''[t]he challenge 174 for the Netherlands-as well as for other countries-is to 175 harmonize a national adaptation policy with its spatial 176 planning policy'' (De Bruin et al. 2009, p. 25). Government 177 experts and decision makers typically hold access to cut-178 ting edge technology and literature on best practices. 179 Nevertheless, local experts, because of their detailed 180 knowledge of hyperlocal conditions, have a better grasp on 181 how best to implement new plans. 182 So too in the field of energy efficiency, German cities 183 experimenting with energy efficiency, renewable energy, 184 and a smaller carbon footprint have not been able to find a 185 sweet spot in their interactions with the central govern-186 ment. These localities need to balance a bottom-up, 187 entrepreneurial culture of individual designers and archi-188 tects with standardized top-down planning from regional 189 and federal government experts who set laws and require 190 national compliance (Fastenrath and Braun 2017). The 191 Chinese central government must regularly renegotiate 192 national environmental enforcement protocols and laws 193 with thousands of regional and local governments with 194 varying degrees of economic development and environ-195 mental pollutants (Shi et al. 2013). 196 With these regular problems of friction even in none-197 mergency policy areas, the challenges of coordination and 198 communication during crisis have more severe and obvious 199 consequences: lives are lost and recovery retarded should 200 responders and decision makers fail to move resources, 201 personnel, and aid in time to disaster struck communities. 202 Nations have moved to create frameworks intended to 203 better handle catastrophe coordination, as ''[t]urning policy 204 into practice requires finding the appropriate balance 205 between top down and bottom up engagement'' (GNDR 206 2009a, b, p. vi). Analysts have argued that the ideal disaster 207 response involves simultaneous engagement from com-208 munity-and locality-based resources in coordination with 209 national authority plans and resources (Muir-Wood 2016). 210 Precisely what that balance is, or the conditions under 211 which it is likely to be reached, are not clear. North 212 American authorities, for example, seeking to coordinate 213 the actions of thousands of agencies, institutions, and 214 NGOs around the country, built the National Disaster 215 Recovery Framework (NDRF) in hopes of a harmonious 216 response structure (DHS 2016). The NDRF mentions 217 coordination more than 100 times in 53 pages. But few 218 would argue that the United States has mastered the art of 219 coordinated disaster response. 220 Scholars studying disaster response have sought to shed 221 light on this problem in a variety of ways. One researcher 222 created a framework for analyzing disaster coordination 223 based on an economic model of the labor market that 224 involves job seekers and employee seekers. Under such 225 assumptions, governments and relevant actors-labeled 226 external and internal agents in this framework-need 227 information and engage in transactions for stabilization and 228 recovery. In this model, communities hit by disasters serve 229 as the internal agents while national government agencies 230 like the Federal Emergency Management Agency (FEMA) 231 are external ones who provide resources (Siembieda 2012).
232 Other teams of researchers have argued that coordination in 233 disaster response works best in conditions of ''emergent 234 coordination,'' that is, through indirect, unplanned, and 235 spontaneous collaboration (Faraj andXiao 2006, p. 1166).
236 Other research has focused on the ways that government 237 agencies and decision makers take credit (or blame) for the 238 response pattern (Boin et al. 2008). Another investigation 239 of extreme weather events and adaptation in the premodern 240 world by Peregrine (2018) looked at greater local partici-241 pation in decision making and coordination in a balance 242 with the enforcement of norms, and described the core 243 balance as one involving flexibility to adapt to changing 244 circumstances and tight adherence to social norms.
245 Given this past research, a number of potential factors 246 stand out for further systematic investigation: level of 247 development, governance, impact of the disaster, civil 248 society capacity, goal sharing between levels of gover-249 nance, administrative capacity of regional and local gov-250 ernments, and level of logistics and planning. I investigate 251 each of these factors and how they may correlate with 252 disaster response coordination in turn below.
253 The level of development of a society or nation hit by 254 disaster could impact the likelihood of friction in a number 255 of ways (Manandhar and McEntire 2014). Developed 256 societies, with higher levels of professionalization in their 257 bureaucrats and higher levels of education, may have an 258 easier time coordinating local and national responses.
259 Alternatively developing societies, which more regularly 260 have authoritarian or at least less democratic governance, 261 may find that central authorities can steamroll or overcome 262 local resistance to central plans.
263 The level of governance in a nation could be correlated 264 with development and therefore have an impact on coor-265 dination ability afterwards a shock or disaster (Gall et al. 266 2014). More democratic and transparent societies might 267 find collaboration easier as they would seek to consult with, 268 rather than overrule, local disaster managers, town mayors, 269 and regional governors. Democratic countries may envi-270 sion civil society as a desired and active participant in the 271 disaster response stage, helping to organize volunteers into 272 organizations such as voluntary organizations active in 273 disasters (VOADs), community organizations active in 274 disasters (COADs), and other trained community-based aid 275 givers.
The cost of the disaster could potentially reduce or 277 increase the transactions costs associated with coordination 278 (Benson and Clay 2003). I envision costs in two ways: the 279 economic costs of the crisis, captured here in terms of the 280 disaster's costs as a percentage of overall GDP, and also 281 the human cost, measured as a percentage of the population 282 killed. When a society faces a larger disaster-with either 283 higher human or economic costs-it may be more incum-284 bent on the national government to show leadership by 285 leading local governments and ignoring dissent. This was 286 the case in China's 2008 Wenchuan Earthquake, the Chi-287 lean 2010 earthquake and tsunami, and Japan's 3/11 triple 288 disasters. However, it may also be possible that, in a 289 massive disaster that creates chaos and reduces the effec-290 tiveness of standard service providers in the area, the 291 central government should back away from seeking to 292 standardize and coordinate responses and allow local res-293 idents and governments to lead (Blakeman 2017). Or, as 294 others have argued, ''because recovery is a process rather 295 than an outcome, it is best accomplished at local levels of 296 government'' (Johnson and Olshansky 2017, p. 9). 297 The ability of the civil society to coordinate and 298 respond, measured here in terms of NGOs per capita, can 299 also alter whether outcomes move smoothly (Shaw and 300 Godao 2004). As with previous variables, the impact of 301 higher levels of NGOs per capita is unclear. It may be that 302 having more NGOs provides little space for top-down 303 decision making and opens the way for conflict and dis-304 connection post crisis. Alternatively societies with more 305 NGOs may be more democratic and therefore more likely 306 to see NGOs as an ally of local, regional, and national 307 governments during crisis. 308 How well the local authorities and the central govern-309 ment harmonize their visions of a response-a variable I 310 label ''goal sharing''-may alter the trajectory of the 311 response (Edgington 2010). If the national government 312 uses the disaster response as a chance to clear residents 313 from vulnerable areas, for example, that may disconnect 314 with local plans of increasing livable and affordable 315 housing in the area. If the central government regularly 316 coordinated before a disaster shock with local authorities 317 on extreme weather, resilience, and disaster management 318 plans-as has been the case in the Netherlands, for 319 example-the arrival of a different major catastrophe, such 320 as a flood, may provide another scenario where cooperation 321 is likely. 322 Whereas NGO capacity captures the ability of the third 323 sector to respond to needs, local capacity instead seeks to 324 capture how well the local government is equipped to 325 respond in a crisis. Some local governments find them-326 selves underfunded through local tax revenue and heavily 327 dependent on the central government for administrative 328 and financial resources-as in Japan. In other cases local 329 cities and regions may themselves have tremendous 330 resources-such as New York City-and be able to coor-331 dinate their own well-oiled responses to crises. 332 The ability of the society to get personnel, material, and 333 information where it needs to go during a crisis can be 334 categorized in terms of the strength of logistical infras-335 tructure. Even before the Haiti Earthquake, for example, 336 that nation struggled to provide potable drinking water, 337 education, and health care to its citizens. After the earth-338 quake killed many members of the armed forces, police 339 and fire teams, and government officials, it became even 340 more of a struggle for the government to assist its citizenry.
341 In contrast Japan's air, rail, road, and shipping infrastruc-342 ture made it possible to deliver needed relief supplies 343 within days even to the most damaged areas of Tōhoku 344 following its 3/11 triple disasters.
345 The degree to which all authorities engaged in realistic 346 disaster planning may serve as a final factor that can 347 influence coordination or discord after disaster. While 348 colleagues have referred to such plans as ''fantasy docu-349 ments'' (Clarke 1999) development, from developing, to mid-range developed, to 362 advanced development. Next, these disasters bring with 363 them a wide range of government capacity and democracy. 364 These disasters draw from a number of regions around the 365 world, including Africa, the Americas, and Asia. Finally, 366 these disasters have a broad range of economic and human 367 costs, from very few deaths to 0.05% of the overall nation's 368 population. Most of the cases come after 2001 because of 369 the availability of consistent data on these events. Using a 370 team of coders I sought to capture the variables outlined 371 above from secondary and tertiary reports in relatively 372 simple terms, often using a binary outcome but at other 373 times using a continuous scale if it was feasible across all 374 observations. The unit of analysis is therefore the disaster 375 response itself.   Table 2 provides the estimated 389 Tobit regression coefficients. 390 Using this pilot quantitative analysis of the dataset, and 391 holding other factors constant, three factors stand out as 392 statistically significant, defined here as those Tobit coeffi-393 cients with p values of equal to or less than 0.05: the level 394 of development, the economic cost of disaster, and logis-395 tical infrastructure. Level of development, disaster cost, 396 and logistical capacity come with a negative coefficient 397 indicating that as these factors rise, friction is less likely.
398 These estimates would indicate that low levels of devel-399 opment, low economic costs, and poor logistics make 400 friction more likely. These preliminary findings based on a 401 quantitative analysis of a small dataset require further 402 investigation and I use QCA to do so.

404
The underlying principles of qualitative comparative 405 analysis (QCA) rest on the belief that measurable outcomes 406 in empirical processes come from multiple paths (Thomas 407 et al. 2014). That is, friction during a crisis may not solely 408 be a function of governance, but rather governance, high 409 economic costs, and poor logistics interacting with each 410 other. By themselves, any one factor may be necessary but 411 insufficient to trigger friction. As quantitative regression 412 models are hard pressed to test various combinations of 413 factors (and instead test each in isolation of the others), 414 QCA builds on Boolean logic and truth tables to probe 415 combinations and recipes involving various levels of fac-416 tors (Ragin 2000). Qualitative comparative analysis bridges 417 qualitative and quantitative methodologies, allowing for 418 process tracing through verifiable and testable evaluation.
419 For this study, rather than using purely binary outcomes 420 (often labeled crisp sets) as with earlier research involving 421 truth tables, I use fuzzy set logic to allow partial mem-422 bership in a category. Hence values can fall within the 423 range of 0 to 1 rather than exclusively at 1 or 0 (in or out of 424 a category) (Table 3).
This study has sought to probe the role of nine factors 426 (level of development, governance, and so on), and the 427 number of possible combinations of these factors is quite 428 large (2 to the 9th power, or 512). These are far too many 429 to test by hand or visual observation alone. Software can 430 simplify and reduce the large number of potential combi-431 nations of cases so that in this case two combinations of 432 characteristics stand out as significant as seen in Table 4, 433 which provides details on the combination of factors that 434 lead to friction along with their coverage and consistency 435 levels.
436 The first pathway (which some scholars have called a 437 ''recipe'') involves low levels of development, high gov-438 ernance, low economic costs, fewer NGOs per capita, low 439 goal sharing, higher local capacity, poor logistics, and poor 440 planning. This combination of factors has the best raw and  Author Proof 441 unique coverage, meaning these conditions are likely to 442 lead to friction. The second potential combination is mostly 443 similar to the first recipe except it has more people killed 444 (as a percentage of population) rather than low economic 445 costs. However, this second configuration of factors is far 446 rarer among the 18 cases than the first set as seen in its raw 447 coverage (0.003 as compared to 0.213). Raw coverage 448 explains the ''extent to which each recipe can explain the 449 outcome,'' while unique coverage is the proportion of cases 450 that can be explained exclusively by that set (Legewie 451 2013, p. 19). While the second combination of factors has 452 higher solution consistency, its unique and raw coverages 453 are very low in comparison, so the first set has more pro-454 mise. The Tobit results from the previous section overlap 455 strongly with these from QCA, namely both indicate that 456 development, economic costs, and logistics seem to be 457 critical factors in friction outcomes. Having seen strong 458 overlap between these methodologies, I turn to cases on 459 and off of the predicted outcomes to better ground these 460 claims.  474 It also revealed the consequences of social and economic 475 inequalities in the nation that resulted in disparate out-476 comes for residents (Walch 2018). Despite the activation of 477 evacuation and aid deliveries, this crisis response encoun-478 tered a high level of friction (1 out of a maximum of 1) due 479 to a lack of coordination between local, regional, and 480 national agencies. To begin, ''the Philippine government 481 failed to anticipate the extent of the damage and miscal-482 culated its own capacity to respond'' (Yamada 2017, p. 77), 483 likely because it lacked ''an emergency plan that coordi-484 nated risk assessment, preparation, and resourcing'' (Al-485 cantra 2014, p. 342). With local governments overwhelmed 486 by scale of the disaster (only 60 out of 2250 employees in 487 Tacoban reported for duty, for example), the President 488 himself took control of relief and response operations. But 489 public feuding-what others have called the blame game 490 (Boin et al. 2008)-between the central and regional 491 governments further slowed the overall response and 492 politicized it.
493 The central government overlooked the potential roles 494 for regional and provincial level agencies along with 495 international aid groups when it directed the response and 496 recovery processes. This may be because nongovernmental 497 agencies and civil society organizations lacked partner-498 ships with Philippine government organizations at all 499 levels (Dy and Stephens 2016, pp. iii-iv). During this 500 major event the local and central governments ended up at 501 odds with each other, openly politicizing the event and 502 failing to respond harmoniously. In terms of the factors 503 under study here, the Philippines has low levels of devel-504 opment, a relatively democratic governance system, a 505 disaster with moderate levels of economic impact (0.008% 506 of the GDP), higher levels of its population killed 507 (0.00006% of overall population), modest numbers of 508 NGOs per capita, low capacity, poor logistics, and poor 509 planning. This disaster in the Philippines sits squarely on 510 the predicted pattern created by the QCA truth tables. It 511 also embodies the findings from the Tobit regression, 512 namely that low levels of development, poor logistics, and 513 moderate levels of economic damage are highly correlated 514 with poor coordination in disaster response. On 11 March 2011 Japan experienced a massive triple 517 disaster: a 9.0 magnitude earthquake followed by a series 518 of tsunami (some as high as 20 meters, or 60 feet) and 519 nuclear meltdowns at the Fukushima Dai-ichi nuclear 520 power plants. With more than 18,400 people killed, 521 125,000 homes destroyed, and a cost of more than USD 522 235 billion in damages, these crises were accompanied by a 523 moderate level of friction (0.5 out of 1). Japan's central 524 government undertook top-down response and recovery 525 plans, sending in private engineering consultants to affec-526 ted cities for over a year. Locals argued that Tokyo often 527 ignored their wishes and that localities relied on cookie 528 cutter plans in a rush to move forward (Dimmer 2017), a 529 long recognized tension in the recovery process (Johnson 530 and Olshansky 2017). Tokyo based bureaucrats pushed for 531 14.5 m seawalls against future tsunami despite little evi-532 dence that physical infrastructure had saved lives during 533 the tsunami (Aldrich and Sawada 2015) and resistance 534 from local communities hoping to maintain tourism and 535 fishing industries through lower walls. Local governments 536 in the affected Tōhoku region had some plans in place to 537 receive aid from localities further from ground zero 538 through ''sister city'' agreements, such as Hyogo Prefecture 539 bureaucrats assisting Sendai's local government, which had 540 mixed results on the ground (Watarai 2012). 541 In responding to the ongoing nuclear disaster, the central 542 government failed to provide guidance to mayors and 543 governors on how to react to the radioactive contamination 544 so that many communities faced additional danger by not 545 evacuating to safe areas (Aldrich 2019). As a result of the 546 poor coordination and information from the government, 547 many residents in Japan became skeptical of the govern-548 ment's claims about the disasters and moved away from 549 interactions with it to instead engage in citizen science, 550 especially in the field of radiation exposure (Suzuki and 551 Kaneko 2013). Japan's 2011 disasters involved high levels 552 of development, a strong democratic governance system, 553 high economic costs (0.03% of the GDP), moderate levels 554 of fatalities, high capacity, good planning, and poorly 555 shared goals. This case sits slightly off the predictions of 556 both the QCA and Tobit analyses as the models would 557 predict that this disaster's characteristics would lead to 558 lower levels of friction. With USD 10 billion in infrastructure damage, the 566 crises remain the largest attack on American soil for the 567 past two centuries. Despite the chaos of the day, the joint 568 government and civil society response to evacuate civilians 569 following the 9/11 terror attacks in Manhattan had a low 570 level of friction (classified as a 0 out of 1). In the largest 571 boatlift in history-even larger than the popularized 572 account of the 1940 Dunkirk withdrawal of the UK and 573 French troops during World War II-government and 574 private sector vessels worked to evacuate all people south 575 of the Twin Trade Towers within nine hours.
576 Following the attacks, half a million residents, office 577 workers, and others sought to leave Manhattan but were 578 stymied due to the shutdown of normal traffic out of many 579 of the typically accessible tunnels and bridges from the 580 island. Thanks to an ''ad hoc flotilla of ferries, tugs, 581 workboats, dinner cruise boats and other assorted harbor 582 craft'' some 500,000 people evacuated from Lower Man-583 hattan by the end of the day in a case of ''distributed 584 sensemaking'' (Kendra and Wachtendorf 2016, pp. 2, 30).
585 The coordination was possible because of the low level of 586 control exercised by the Coast Guard (which has jurisdic-587 tion over the area) and the emergent volunteers who came 588 on a variety of private craft to assist when the Coast Guard 589 called for all available boats. The 9/11 terror attacks 590 response involved high development, strong democracy, 591 low economic damage, low fatality rates, strong logistics, 592 good capacity, and highly shared goals. It sits among the 593 very few disaster responses that have achieved high ratings 594 of coordination.

596
One suggested solution to the friction found in many of the 597 cases here and in the broader dataset involves advancing 598 mitigation and response simultaneously in parallel at dif-599 ferent levels. Local residents, experts urge, need to live in 600 cultures of disaster while decision makers and national 601 leaders must be part of strong institutions (Muir-Wood 602 2016). Local disaster culture translates into homeowners, 603 families, and businesses being mindful of threats and risks 604 and creating and drilling crisis responses. Institutional 605 engagement means that decision makers and bureaucrats in 606 national governments create and maintain strong regula-607 tions and governance practices that minimize harm. One 608 example would be optimizing the bottom-up and top-down 609 responses in vulnerable, coastal regions of nations with a 610 high likelihood of tsunami occurrence, whether Talc-611 ahuano, Chile, Portland, Oregon, or Sendai, Japan. In an 612 ideal world, citizens living in such areas would evacuate 613 each time they felt or were warned of an earthquake and 614 school children and adults alike would regularly practice 615 walking, moving via wheelchairs, or running to higher 616 ground. At the same time central government officials 617 would put in physical infrastructure that effectively miti-618 gated threats while creating escape zones that minimize 619 dangers of inundation. Government agencies would also 6 Shortcomings and a Research Agenda 674 This study has used a medium sized dataset of disasters 675 around the world and as a result the dataset cannot be 676 thought of as a representative sample. Because of data 677 availability, most of these events occurred in the early 678 twenty-first century. Further, while we used coders when 679 categorizing our nine variables of interest, future teams 680 should consider machine coding to provide stronger relia-681 bility in coding decisions. This is especially true for the 682 outcome of interest, namely the level of coordination (or 683 friction) in the response, which scholars have argued 684 involves subjective judgment (Boin and Richardson 2015).
685 Future studies should simultaneously build up a broader 686 dataset of disasters-perhaps along the structure of EM-687 DAT (CRED 2019)-while employing teams of simulta-688 neous coders who can seek to accurately capture more 689 ambiguous variables of interest.

691
This article has sought to use qualitative comparative 692 analysis, Tobit regression analysis, and mini-case studies to 693 illuminate the factors that lead to friction-and not coor-694 dination-during disaster response. Across these methods 695 the factors of the economic costs of the event, the level of 696 development in the country, and the strength of logistical 697 infrastructure having proven critical. In short, developed 698 nations facing larger disasters with strong logistical 699 infrastructure seem less likely to encounter friction during 700 crisis. Unfortunately, less developed nations facing smaller 701 scale disasters and using weaker infrastructure are more the 702 norm, and more likely, based on these results, to have a 703 lack of coordination. 704 The problem of balancing top-down and bottom-up 705 approaches and ensuring smooth coordination will require 706 creative political and administrative frameworks, perhaps 707 ones not influenced strongly by political regime changes 708 (Wood and Waterman 1991). While this article has focused 709 partly on extreme weather-related events (including 710 typhoons and floods), the pressing challenge of climate 711 change will require simultaneous bottom-up adaptation and 712 top-down legislative and institutional efforts that need to be 713 coordinated well to avoid waste and achieve efficiency.
714 One experimental approach to climate change adaptation 715 has involved social learning, knowledge exchange, and 716 social networks across levels of agency (Butler et al. 2015).
717 Similarly large scale, cross-national private sector projects 718 require horizontal and vertical integration of teams across 719 firms and countries, and research has underscored the 720 importance of identification and empowerment of leaders