Community recovery from a major natural hazard-related disaster can be a long process, and rebuilding likely does not occur uniformly across space and time. Spatial and temporal clustering may be evident in certain data types that can be used to frame the progress of recovery following a disaster. Publically available building permit data from the city of Joplin, Missouri, were gathered for four permit types, including residential, commercial, roof repair, and demolition. The data were used to (1) compare the observed versus expected frequency (chi-square) of permit issuance before and after the EF5 2011 tornado; (2), determine if significant space-time clusters of permits existed using the SaTScan™ cluster analysis program (version 9.7); and (3) fit any emergent cluster data to the widely-cited Kates 10-year recovery model. All permit types showed significant increases in issuance for at least 5 years following the event, and one (residential) showed significance for nine of the 10 years. The cluster analysis revealed a total of 16 significant clusters across the 2011 damage area. The results of fitting the significant cluster data to the Kates model revealed that those data closely followed the model, with some variation in the residential permit data path.
Although ample warning occurred (Paul and Stimers 2011), the 22 May 2011 Joplin, Missouri, tornado claimed 162 lives, over 2.3 times the 1990–2019 United States average of 68.4. This number is inflated due to the especially deadly 2011 season in which 553 people died (Simmons and Sutter 2012). Remove the 2011 values, and the 29-year average of deaths by a tornado in the United States stands at 51.7; under this scenario, the singular Joplin event killed 3.13 times as many people as the country’s annual average (SPC 2021). Compared to the 1950–2007 annual rate of fatalities per million residents for Missouri, the value of 0.7311 (Simmons and Sutter 2012) is eclipsed by the 2011 tornado, which caused 26.94 deaths per million (based on a 2010 population of 6.012 million) (USCB 2021), or 36.8 times the 1950–2007 average for the state and 74.1 times the U.S. average (0.4 deaths per million) (Simmons and Sutter 2012) for the same period. As of 2021, the tragedy’s 10th anniversary, it still stood as the deadliest and costliest tornado event since the 1950–1951 birth of the modern U.S. tornado record (2.8 billion 2011 USD actual, 3.382 billion 2021 USD adjusted) (SPC 2021), and the fifth deadliest tornado in the colonial United States and U.S. history (1680 to 2021) (Grazulis 1993; SPC 2021).
After entering the community from the west as a fairly weak EF1, the funnel grew into an EF5 wedge (estimated 200 mph, 320 kph), spanning approximately 20 city blocks north-to-south at its widest points and tracking nearly the entire west-east extent of the city (Fig. 1). It curved slightly southeast as it exited the community and continued through rural areas at EF0/EF2 strength. The centerline of the funnel caused catastrophic damage over roughly two-thirds of the path, damaging beyond repair or sweeping away thousands of structures altogether (~ 7500 total). Post-event damage assessments placed the total track length at 22.1 mi. (35.6 km) and the width a substantial 0.75 mi. (1.2 km), placing it in the top 1% of U.S. tornadoes since 1951 based on that measurement (SPC 2021).
The Kates (1977) reconstruction model provides the context of elapsed time from the event in four phases, including (1) emergency (approximately 1–2 weeks post-event); (2) restoration (approximately 1 week–5 months post-event); (3), reconstruction phase I (approximately 2 months–5 years post-event); and (4) reconstruction phase II (approximately 2–10 years post-event) (Fig. 2). Alexander (2002) re-examined Kates’ model regarding unit cost and extended the reconstruction phase II (post-disaster development) past 10 years. While conceived and published as hypothetical, Kates’ model has been used to frame the recovery and reconstruction periods of some of the most well-known disasters in U.S. history, including the 1906 San Francisco, CA, and 1964 Anchorage, AK, earthquakes (Kates 1977), and more recently Hurricane Katrina, which occurred in 2005 (Kates et al. 2006). To the authors’ knowledge, the Kates model has not previously been applied to a tornado event, likely because very few tornadoes (88 of 66,388, 0.13%, 1950–2019) (SPC 2021) in the United States have reached F5 or EF5 strength and thus a massive event such as what occurred in Joplin, MO, is exceedingly rare. Here the Kates model was used to frame statistically significant space-time-based clusters of permits, altering the model’s phase shapes to conform to the specific pattern uncovered by all significant space-time-based permit data.
2 Literature Review
Spatial analysis of post-disaster areas is widely used to understand better many aspects of the recovery process, including rebuilding. Comerio (2006) and Comerio and Blecher (2010) utilized building permit data to examine rebuilding in the wake of the 1989 Northridge and 1994 Loma Prieta earthquakes. The authors focused on estimating elapsed time from the disaster until a building was repaired or demolished and then rebuilt, and at what point the building could be occupied. McCarthy and Hanson (2008) incorporated building permit data and census and damage assessment data into their analysis of building type, damage state, and repair activity after Hurricane Katrina. In a similar study, Stevenson et al. (2010) examined the state of rebuilding from a spatiotemporal perspective in the wake of Hurricane Katrina using a spatial scan statistic developed by Kulldorff (1997) and refined by Kulldorff et al. (2005). Damage amounts and pre-event housing types were strong indicators of post-event space-time clusters of rebuilding based on the Monte Carlo statistic from the spatiotemporal analysis. Stevenson et al. (2010) concluded that uniformity in disaster rebuilding was not present in the study area. Similar to the work by Stevenson et al. (2010), Rathfon et al. (2013) used building permits and other data (census, ground-truthing) to examine the process of rebuilding; however, their goal was to devise a standard method for measuring and monitoring recovery. One of the declarations from their article emphasized that recovery is not an endpoint but a process.
Modeling and assessing the effects of disasters and recovery has been undertaken in several other forms. Specific to the Joplin, MO, tornado, Smith and Sutter (2013) reviewed the pace and success of recovery in the community following the 2011 tornado through the lens of comparing Joplin, MO, to New Orleans, LA, recovery efforts after Hurricane Katrina. The authors claimed that community involvement and effective and efficient local government assistance in Joplin, MO, worked in tandem to begin the recovery process, thus setting the community up for a successful recovery (unlike the results seen in New Orleans, LA). Santos et al. (2014) created an inoperability input-output model (IIM) focused on economic disruptions following a disaster. The IIM furthered the understanding of how particular events can alter the path of recovery post-disaster by applying them to the Nashville, TN, region. The results showed that variation across sectors of the economy requires striking a balance between prioritizing recovery policies and interventions.
Cui et al. (2015) studied hurricane landfall from three significant storms (Frances in 2004, Katrina in 2005 and Ike in 2008) and the relationship to building permit issuance. The authors claimed issuance varied by county and that losses and recovery might be pushed into the future if the impact is severe and permanent. Hashemi-Parast et al. (2017) examined the reconstruction process in Bam, Iran, following the 26 December 2003 earthquake. Government data and satellite imagery facilitated the authors’ findings that while progress occurred in the years following the event (through 2014), lesser damaged areas were further along than more heavily damaged locations, highlighting the unevenness with which an area may recover post-event. Tax assessment, census, land assessment, and modeling efforts have all proven helpful in approaching the problem of describing post-disaster recovery. This research may add to the body of literature by analyzing permit data on one of the most destructive tornadoes in U.S. history.
In this study we examined restoration and reconstruction I and II phases within the damage zone in the community of Joplin, MO, using a chi-square analysis and an analysis of the spatiotemporal aspects via Monte Carlo simulations and time series clustering with the SaTScan™ platform (version 9.7) focused on residential, commercial, roof repair, and demolition permits. In the restoration phase, structures are demolished, with longer-term rebuilding in the reconstruction I and II phases. Those three phases built the conceptual framework for this analysis. The emergency phase of the Kates model was not considered here, as emergency operations such as initial response and rescue do not typically require building permits.
3.1 Building Permit Data Cleaning
Two data types described the recovery pace in the restoration phase, which is phase 2 in the Kates model. First, demolition permit data indicated the progression through the restoration phase. Second, permits that identified roof repair as the singular scope of work for that permit determined if damage to roofs would space-time cluster in (primarily) the least damaged areas and be repaired within the restoration period. Two additional data types were used entering the reconstruction phases. First, residential housing rebuilding is a critical metric in recovery process assessment, as shelter is a basic human need (Maslow 1943). Quarantelli (1982) identified four stages of housing recovery, including emergency shelter, temporary shelter, temporary housing, and permanent housing—the latter was examined here. Economic development is a crucial concept in urban planning and post-disaster recovery (Chamlee-Wright and Storr 2008; Simmons and Sutter 2011, Smith and Sutter 2013); thus, commercial rebuilding permit data were considered descriptors of recovery. Residential and commercial were assumed to represent the reconstruction phases (I and II) in the Kates model, although either may appear beginning in the restoration phase.
Building permit data, spanning the period 23 May 2011–31 December 2020 (data beginning point of 1 day after the 22 May 2011 event), were requested and delivered from the city of Joplin, MO, City Clerk’s Office in spreadsheet format and imported into an 18,750-row single-table (unindexed heap) database for cleaning, and then into a GIS for geocoding (row removal values noted below in parentheses). Rows with no useable address or vague location descriptions (for example, Rangeline Road; 20th Avenue, 39 rows), missing or a vague description of the permit’s scope of work (45 rows), errant rows included and labeled as “business license” (192), mismatched dates (stated calendar year did not align when compared to the year in the dd/mm/yyyy format of the permit data, 12 rows), dates outside of the requested range (4 rows), or missing dates (25 rows) were eliminated (317 rows removed, 18,433 remained). The geocoding process resulted in a 99.3% match rate, leaving 18,377 rows of usable geodata. Using spatial data on damaged structure location (Paul and Stimers 2011, 2014) and 2010 census block files, all permit data located outside of a census block not recorded as having experienced any damage on 22 May 2011 were removed (9694 rows), leaving 8683 rows of useable permit data. Demolition rows that contained a calendar year value, but no specific date value were set to 15 August of the stated calendar year (309 rows).
Next, based on the permit description column, queries were written against the GIS data to extract only those permits identifiable as residential housing, commercial building, roof repair, or demolition through text string fuzzy matching and manual review of the results. The results were compared to the original data through several rounds of sorting, reviewing, and re-querying, producing 4925 rows of data, including (1) 1721 residential building permits; (2) 170 commercial building permits; (3) 1668 demolition permits; and (4) 1366 roof repair-only permits. Data were then displayed in a GIS to isolate and separate the four classes of permit data into new shapefiles for the spatial analyses.
3.2 Tabular Statistical and Simple Spatial Analyses
Simple descriptive statistical analysis (DSA) was performed on the four permit data types, including a count of permit type by year, percent of the total in that year for the permit type, and percent of permits for that type compared to the entire permit total. Data for the 10 years before the date of the tornado event (22 May 2001–31 December 2010, aligned with the post-event dataset) were requested from the Joplin, MO, City Clerk’s Office, received, cleaned, and geocoded following the procedures described above (inside and outside the damage census blocks and separated by permit type) to create a baseline by permit type for chi-square analysis. The chi-square analysis focused on determining if building permits by type and year in and out of the damage zone for each of the 10 years post-event (2011–2020) showed a statistically significant difference over the baseline average by permit type in and out of the damage zone over the 10-year pre-event period. Although it is reasonable to assume that permit issuance will increase following a significant disaster, empirical evidence based on statistical significance bolsters the case for clustered data to fit an existing model, such as Kates’, if such observed data are outside of an existing model’s expected range. Here, it was assumed that a statistically significant difference of an increase was related to the rebuilding efforts following the 2011 disaster, which allowed for the determination of in what (if any) year permit issuance ceased to show a statistically significant difference from the pre-event 10-year mean, thus revealing a time-based cluster.
3.2.1 Chi-square Hypothesis
H01: No statistically significant relationship exists between pre- and post-disaster permit issuance by 10-year average (pre, 2001–2010) and year (post, 2011–2020) in the damage zone compared to outside the damage zone.
3.3 Spatiotemporal Analysis Using SaTScan™
A second examination of the data entailed utilizing the epidemiological modeling software platform SaTScan™ to search for significant clusters of permit point data across the four permit types using a Monte Carlo test (Hammersley and Handscomb 1964) and space-time clustering (Kulldorff 1997), and the space-time permutation model developed by Kulldorff et al. (2005). The SaTScan™ application accepts simple tabular data as input, as long as either latitude/longitude or Cartesian coordinates are included, which can be exported from an attribute table in a GIS (here, QGIS™ was used), and the results are easily imported back into a GIS environment. The export file contained the number of cases (1 row = 1 case, or a 1:1 relationship), the issue date of the permit, and a unique identifier (primary key, or PK) composed of the calendar year of the permit and the permit number assigned by the city (for example, the 250th permit issued in 2013 was coded 20130250, creating a unique PK). The SaTScan™ program requires the user to make several selections to prepare the data for analysis. No assumptions concerning the number of clusters are required, however. The study period was set from 23 May 2011 to 31 December 2020. The type of analysis chosen was “space-time,” with a “space-time permutation probability model” and a scan area of “high to low” rates. The time aggregation unit was set to “days.” Basic output settings included “cluster information” and “location information.” The criteria for reporting clusters were set to “most likely clusters, hierarchical,” with “no geographical overlap” in the clusters. Finally, significant clusters were defined as having p \(\le\) 0.05.
The SaTScan™ space-time model identifies clusters and then employs the chosen spatial scan statistic (for example, the space-time probability model, retrospective, used herein) to determine significance using a Monte Carlo test. The maximum scanning window size (MSWS) was set to use a time aggregation of 147 days for commercial permits, 1406 days for demolition permits, 1555 days for residential permits, and 1248 days for roof repair permits, and scanned for both high and low rates. As noted by Han et al. (2016) and expanded upon by Lee et al. (2021), the default maximum spatial cluster size (MSCS) may lead to overly large clusters and obscure smaller significant ones. However, the correction for this, the Gini index, can be selected only when employing a Bernoulli or Poisson model. Further, arbitrarily adjusting the scan window exposes the analysis to the modifiable areal unit problem (MAUP), a concern in any spatial dataset in which the scale can be adjusted (Openshaw 1978; Openshaw and Taylor 1979; O’Sullivan and Unwin 2003; Duque et al. 2018). Multiple MSCS values were tested, and only minimal variations in cluster radius, case counts, Monte Carlo ranks, X2 test statistics, and p-values were found. Given the pre-analysis runs resulting in apparent stability, the MSCS default value of 50% was used, as suggested by Kulldorff (2021).
Once the scan is complete, the maximum likelihood is examined to determine if a cluster occurred by random chance. An expected value of points is then derived based on the initial number of actual observations and using what Kulldorff et al. (2005) defined as spatial cylinders, in which the height of the cylinder accounts for the temporal component while the base of the cylinder accounts for the spatial component, thousands of passes are made to compare expected versus observed for identification of cluster significance. The results feature the p-value to identify significant clusters, the cluster radius, Monte Carlo rank, and X2 test statistic, with the results output to a shapefile containing all needed tabular and spatial information to import the results into a spreadsheet or GIS for further observation and analysis.
3.3.1 SaTScan™ Space-Time Permutation Model Hypotheses
H02: No statistically significant space-time clusters exist for post-event residential building permits inside the damage zone.
H03: No statistically significant space-time clusters exist for post-event commercial building permits inside the damage zone.
H04: No statistically significant space-time clusters exist for post-event roof repair (roof repair as the only scope of work) building permits inside the damage zone.
H05: No statistically significant space-time clusters exist for post-event demolition permits inside the damage zone.
3.4 Space-Time Cluster Data Fit to the Kates Recovery Model
Cluster data were examined to create histograms of permit frequency issuance by year within that cluster; those histograms formed the basis for fitting the data to the Kates model. Since the total permit counts by cluster and type showed an extensive range—min = 0 (value appeared in all permit counts), max = 477 (demolition permits, 2011)—to avoid singularly tall bars on any histogram, the data were transformed using min-max scaling. Scaled values were then displayed by a histogram, one per year per permit type, with a log10 x-axis (following Kates’ horizontal axis) and used to create custom groups and peaks unique to the Joplin, MO, event.
The results are presented below in the form of descriptive statistics, chi-square analyses, and space-time cluster analyses. The chi-square analyses describe the differences in permit issuance between the pre- (2001–2010) and post-disaster (2011–2020) phases (10-year periods), and the space-time cluster analyses describe the significance of clusters based on permit type.
4.1 Descriptive Statistical Analysis and Chi-square
Descriptive statistics for the four permit types over the 23 May 2011–31 December 2020 period revealed that for residential permits, 75.2% of the 1721 permits issued over the period were issued by 2015. The 2 years with the highest percentage of permit type totals were 2011 (21.5%) and 2012 (24.8%). Commercial permits reached the near-three-quarter mark in 2014, with 73.5% of the cumulative type total. Most permits were issued in the first 2 years following the disaster for roof repair and demolition, with roof repair accumulating 76.5% of the type total in 2011 and demolition reaching 78.5% in the same year (Table 1).
A comparison of each permit type in and out of the damage zone by year post-disaster (2011–2020) against the 10-year average by permit also separated by damage zone was performed using a chi-square analysis (chi-square values are reported with no continuity correction and significance at p < 0.05). Residential permits showed statistically significant results (p < 0.001) for the years 2011–2019, with only 2020 allowing for the acceptance of the null hypothesis. Permits inside the damage zone exceeded the 10-year pre-event mean for all years, while permits outside the damage zone were below the mean in 2012–2019 and above the mean for 2011 and 2020. These results indicated that while the raw numbers slowly moved closer to the pre-event 10-year mean, the building pace in the post-disaster decade was significantly higher than expected in a no-disaster scenario. Commercial permits showed a similar result for the first half of 2011–2020. The increase in permit issuance in the damage zone for 2011–2015 was statistically significant (Table 1), but this was not the case for 2016–2017. Significant results reappeared in 2018, although below the mean both in and out of the damage zones.
Roof repair permits, issued largely in the lesser-damaged areas labeled limited or moderate, showed a massive increase in 2011 (1045 permits inside the damage zone, compared to the pre-event mean of 56). This increase was statistically significant (p < 0.001), as was the following year (2012, p = 0.008). In 2013, no significant results appeared, but significance reappeared in all remaining years (2014–2020). At first, this would seem to contradict the assumption that roof repair (and roof repair only) would be completed quickly in the months following the disaster. However, the years 2014–2015 were below the 10-year pre-disaster mean, not above, which supports the assumption that roof repairs were completed in a short period. In 2016, roof permit issuance was below the mean in the disaster zone but above in the non-affected areas. Examining the raw data coupled with its significance in the years 2017–2020 became interesting. In 2017, 2018, and 2020, the number of roof repair permits was below the 10-year mean inside the damage zone and above the mean inside the damage zone in 2019. Throughout 2017–2020, permits issued each year exceeded the 10-year pre-disaster mean (by multipliers ranging from 2.5 to 3.1). However, those statistically significant increases were likely not owed to continued roof repairs based on 2011 tornado damage; Joplin witnessed one hailstorm in each year, 2017, 2018, and 2020, and three in 2019, all with 1-inch (25.4mm) stones recorded except for the 10 October 2019 hailstorm, which brought 1.75-inch (44.45mm) hailstones. It was hypothesized that those hail events drove the roof repair values higher for those years, although that hypothesis was not tested here.
An examination of demolition permit data showed results aligned with what one would reasonably expect to be the case in the years following a major disaster. Over the period 2011–2015, the number of demolition permits issued inside the damage zone was greater than the 10-year pre-disaster mean and statistically significant. Outside the damage zone, permits were higher than normal for 2011 but below the mean for 2012–2020. It was speculated that the demolition activity outside the damaged areas slowed down due to the increased demand for those resources inside the affected areas. From 2016 to 2020, demolition permits inside the damaged areas were below the 10-year pre-disaster mean, indicating that most demolition activities had ceased by 2016. It was further surmised that most demolition was completed in 2011, with 78.5% of all permits for demolition inside the damage zone issued in that year. By 2015, the last year that showed statistical significance, 97.0% of all 2011–2020 demolition permits had been issued.
4.2 Space-Time Cluster Analysis
Using the space-time permutation model, which covered all four permit types—residential, commercial, roof repair only, and demolition used in this analysis—4925 permits were examined. Significant clusters were identified as those showing p < 0.05. All permit types returned at least one significant cluster, with the majority returning four or more (Table 2).
4.2.1 Residential and Commercial Clusters
Seven residential clusters were significant (Fig. 3). Thus, H02 was rejected, and Ha2 was accepted; at least one statistically significant cluster exists for post-event residential building permits inside the damage zone. Residential cluster 1 (n = 76, p < 0.001, radius [r] = 0.50 mi2.) lies from 22nd Avenue north to E 16th Street and extending west-east from S Pennsylvania Avenue to S Michigan Avenue. The second residential cluster (n = 48, p < 0.001, r = 0.22 mi2.) was situated north-south between W 24th and 20th Streets and west-east between Byers and S Joplin Avenues. The third significant residential cluster, the largest in both permit total and area (n = 175, p < 0.001, r = 0.55 mi2.), was found on a heavily damaged north-south stretch between E 24th and E 16th Streets and spanning west-east across S Patterson Avenue and S Range Line Road/U.S. Highway 77 Business. Residential cluster 4 (n = 103, p = 0.002, r = 0.42 mi2.) occurred at the western edge of the tornado’s path, spanning from W 32nd Street/Newton Road north to Gabby Street/W 26th Street and west-east from S Winfield Avenue to S Tyler Avenue (southeast extent) and S Adele Avenue (northeast extent). The fifth residential cluster was the smallest in permit count and size (n = 11, p = 0.003, r = 0.03 mi2.), straddling Maryland Avenue between E 22nd and E 20th Street. Residential cluster 6 was the second-largest cluster in permit count, behind only residential cluster 3, but the fourth largest in radius behind clusters 1, 3, and 4 (n = 149, p = 0.003, r = 0.42 mi2.), showing permits extending from E 25th Street to E 21st Street on a north-south axis and running west-east from State Highway 43/Main Street to Missouri Avenue. The seventh and final residential cluster was the second smallest in both permit count and radius (n = 16, p = 0.004, r = 0.10 mi2.). It runs mostly north-south along S Pearl Street between W 28th Street (southern extent) to W 26th Street (northern extent), with a smaller collection of the cluster on S Wall Street between W 28th Street and W 27th Street. Four commercial clusters emerged, but only one was significant (n = 8, p = 0.011, r = 0.12 mi2.), containing just eight permit data points and aligning along a north-south path on Rex Avenue just north of E 20th Street. Thus, H03 was rejected, and Ha3 was accepted; at least one statistically significant cluster exists for post-event commercial building permits inside the damage zone.
4.2.2 Roof Repair Permits Only Clusters
Application of the space-time permutation model statistic to permits for roof repair (permit was issued explicitly for roof repair only or included in a permit that was not issued for a full rebuild) yielded 11 clusters, four of which were significant (Fig. 4). Thus, H04 was rejected, and Ha4 was accepted; at least one statistically significant cluster exists for post-event roof repair building permits inside the damage zone. Roof cluster 1 contained the most permit points and was the largest in the area (n = 109, p < 0.001, r = 0.56 mi2.), encompassing an area between E 30th Street at the southern extent to E 21st Street at the northern extent, and running west-east from S Joplin Avenue to Kansas Avenue. Cluster 2 in the roof repair category (n = 44, p = 0.020, r = 0.31 mi2.) sits between E 32nd Street to the south and E 24th Street to the north, with its westernmost point at Park Avenue and the easternmost near S Ozark Avenue. The third significant roof repair cluster (n = 65, p = 0.045, r = 0.09 mi2.) was located just to the northeast of cluster 2 and extended south of E 24th Street north past E 22nd Street and from S Florida Street at the western boundary to Texas Avenue at the east. The fourth and final significant roof cluster (n = 32, p = 0.046, r = 0.10 mi2.) identified was at the western section of the funnel’s entry into Joplin, MO, stretching from Gabby Street/W 26th Street north to W 23rd Street with a westernmost permit located on the western edge of W 24th Street and a collection of permits running north-south along S Adele Avenue at the eastern damage extent.
4.2.3 Demolition Permit Clusters
The fourth permit type included all demolition permits issued. The SaTScan™ analysis resulted in five clusters, four of which revealed statistical significance (Fig. 5). Thus, H05 was rejected, and Ha5 was accepted; at least one statistically significant cluster exists for post-event demolition permits inside the damage zone. Demolition cluster 1 (n = 50, p < 0.001, r = 0.64 mi2.) runs from W 23rd Street in the south to W 18th Street in the north (although the radius buffer extends another four blocks north) and spans a west-east range from Annie Baxter Avenue to S Wall Street. Demolition cluster 2 was the largest significant cluster in both permit count and area of the demolition permit type and of any type (n = 644, p < 0.001, r = 1.53 mi2.). This massive cluster covered the south-north area between E 32nd Street and E 15th Street and included a west-east range running from S Joplin Street to Texas Avenue. Cluster 3 (n = 12, p = 0.007, r = 0.28 mi2.) encompasses an area delineated by E 24th Street at the south of the buffer to W 20th Street and a west-east span including S Willard Street to S Tyler Avenue. The fourth and last significant demolition cluster (n = 194, p = 0.009, r = 0.57 mi2.) contained the second-largest count of permits for both the demolition type and for all types. It was situated in the southwestern portion of the damage path, stretching from Mahaska Avenue north to W 23rd Street and Annie Baxter Avenue east to S Pearl Street.
The discussion section details the ground-truthing efforts in September 2021 to verify the cluster analyses. Specifics concerning each cluster and progress are discussed. The data are then fit to the Kates recovery model to test the alignment of empirical data to a hypothetical model.
5.1 Cluster Verification, Recovery Progress, and the Landscape
Results of the space-time cluster analysis were ground-truthed in September 2021. Where possible, significant clusters were visited to visually examine the state of rebuilding or repair and verify the accuracy of the cluster size and homogeneity of repair type. Some demolition clusters, by the nature of the type (structures removed then replaced), could not be directly observed, except in areas where no rebuilding had yet occurred, as indicated by still-empty lots. Thus, it stands to reason that some demolition clusters overlapped residential and commercial points in clusters, matching an exact address. Damage clusters are not discussed in a separate section but with other clusters nearby. Broadly, the space-time cluster analysis algorithm performed well in that significant clusters of permits were easy to identify on-site, and the type of repair or rebuilding as indicated by the permit data and grouped by a given cluster stood out when visually compared to surrounding structures that were not a part of the cluster under examination. However, some variation in structure type and repair and rebuilding status was noted and is discussed by cluster and type.
Residential clusters (prefixed by RC) are presented in order as determined by the p-value and Monte Carlo rank and numbered RC1–RC7; the lone commercial cluster (CM1) is discussed alongside RC3, the closest residential cluster geographically. Demolition permit clusters (DC1–DC4) are discussed, where appropriate, in locations where geographic alignment was found to the residential clusters and summarized at the conclusion of this section. A feature common across all residential clusters was the mix of styles, regardless of the damage level inflicted on the area. Construction age variation indicates that not all homes in a cluster were wholly demolished, which is supported by the assertion that variation in damage can be found along the tornado’s path, even in areas where damage was, as a whole, catastrophic. Additionally, the pace of reconstruction, and thus, the appearance of the built landscape, can be thought of as anisotropic across the community space. Characteristics of the affected population, whether economic or cultural (although the economic standing is the more probable candidate), are likely to affect recovery progress (Peacock et al. 1997).
A mix of old and new homes was evident in RC1, aligned with DC1, especially near Parr Hill Park between Kansas and New Hampshire Avenues on 18th Street. A stand-out visual in this area was on 18th and New Hampshire; new homes with multiple roof hips sitting among older homes damaged but not destroyed. Newer construction became more prominent east of 18th Street and St. Louis Avenue and near 16th Street and Indiana Avenue. Residential cluster 2 (RC2), also the southeastern portion of DC1, showed a similar layout, with a mix of new and old homes with multiple development styles.
The heavily damaged north-south extent between East 24th and East 16th Streets and west-east across South Patterson Avenue and South Range Line Road/U.S. Highway 77 Business contained RC3, which provided striking visual evidence of the juxtaposition of styles and rebuilding progress. Along East 24th and Delaware Avenue, a scattered mix of old and new homes were present, much like what appeared in RCs1–2. However, more empty lots still stand, old concrete (as evidenced by the color) abuts new homes, and large lots (for example, corner of East 20th and Delaware Avenue) remain empty. Further, some lot sizes in this cluster area were noticeably larger—a phenomenon for a potential future study. Residential cluster 4 (RC4), situated near the location of the now-removed St. Joseph’s Hospital, featured similar characteristics as the first three locations in that the mix of old and new homes was abundantly clear, as was the construction style.
Residential cluster 5 (RC5) was the smallest in both permit count and size but presented perhaps the most interesting pocket of development within the centerline of the catastrophic damage path. This space-time cluster was evident when running just two blocks, from East 22nd to East 20th Streets along Maryland Avenue. Most structures along this stretch were rebuilt as multi-family dwellings that easily emerged visually from their pre-WWII neighbors, as evidenced by one home at the southern extent of the cluster that can be accessed only by an alley. A few homes in the cluster’s northern extent resembled those found in RC3. Several open lots remain in the northern portion of the cluster (although a bank and some realtor signage indicate those lots have been rezoned as commercial).
Residential cluster 6 (RC6), situated near the western edge of the massive DC2 cluster, follows the pattern established in four of the five clusters discussed thus far, with a mix of old and new homes. Residential cluster 7 (RC7), the last of the significant clusters, was a small grouping south of RC6 and situated on the southern periphery of the catastrophic damage area, and within DC4 also shows mixed designs. However, here we noted some homes sitting atop clearly older foundations (again, as evidenced by the concrete wall’s faded coloring). This cluster also featured new and ongoing construction and the development of walking trails as amenities for the neighborhood residents.
Four demolition clusters emerged from the space-time analysis across the damaged area, as noted within the RC1–7 discussion. Demolition cluster 1 (DC1), situated in the northwest portion of the damage path, in a mix of catastrophic to limited damage, contained only a small portion of RC2 and featured empty lots through the cluster zone. The largest cluster in radius by far, DC2 logged 644 demolition permits (508, or 78.9% of which were issued in 2011–2020). It covered parts of all but two of the 12 significant non-demolition clusters (excluded only were RC4 and roof repair cluster 4). Demolition cluster 3 (DC3), the westernmost of the clusters, was the smallest in the area and was fairly well-built back, but some empty lots were still present. Demolition cluster 4 (DC4), the second-largest demolition cluster behind DC2, was observable like the others in that the area was built back to some degree, but not entirely, and several empty lots were present.
Following the residential permit naming convention, roof repair permits (prefixed by FC) are presented in order as determined by the p-value and Monte Carlo rank and numbered FC1–FC4. For roof repair permits significant clusters were hypothesized to appear along the periphery of the path’s center (a line defining the center of the most heavily damaged area). Roof repair cluster 1 (FC1), mainly located on the central southern portion of the path, sits roughly halfway along the more lightly damaged periphery and the heavier damaged center. Throughout the southern portion of FC1, lighter damage was evident in that nearly every home that featured construction styles from the 1950s to 1970s was topped with a new roof. Moving north into the more heavily damaged areas, a mix of old homes with new roofs and completely new homes was found. Roof repair clusters 2 and 3 (FC2 and FC3) are situated close together, with the northern edge of FC2 congruent with the southern edge of FC3. Characteristics of FC2 and FC3 through each cluster are similar to one another and FC1; a clear sign of lighter damage was evident through new roofs capping old homes. The final cluster that showed significance, FC4, was located at the damage path’s western portion near the old hospital in an area where the tornado had not yet gained full EF5 strength but was likely already close to or at EF4 strength. However, wind speeds would have been lower peripheral to the centerline (Paul and Stimers 2012). Scattered new construction was visible, and new roofs on old houses, as was the case in the other roof clusters, were also clearly visible.
5.2 Data Examined Using the Kates Recovery Model
Immediately following the event, significant demolition activity commenced and continued through most of 2011, as evidenced by the chi-square results and significant spatial clusters of demolition permits (Fig. 6). Demolition can be considered a restoration, return, and patching (phase 2) activity, as permits are issued to demolish buildings that may pose a hazard to residents and workers. Further, once the emergency phase passes, demolition permit issuance should continue, signaling the period in which any unsalvageable structure is removed to create the needed space for reconstruction I to commence. The demolition peak aligned closely to the generalized model conceived by Kates (1977). The largest demolition cluster (DC2) and the smaller DC1, 3, and 4 clusters all showed peaks in issuance around weeks 16–17. Roof repair followed a similar pattern and can likewise be considered a restoration, return, and patching (phase 2) function described by Kates. Roof repair can pose severe problems for homeowners (primarily exposure to the elements, inviting leaks, and further structural damage). However, it is a relatively simpler repair process than complete demolition, site preparation, and rebuilding. Also, like demolition, the four roof repair clusters showed a peak in issuance early in the process, around weeks 10–11.
Residential permit issuance, shown to be significantly increased via the chi-square analysis in 9 of the 10 years, revealed a multi-modal pattern based on the cluster analysis, with peaks occurring around weeks 30, 150, and 400 (the week 400 peak comprised three individual peaks, which were consolidated here into one mode to conform to the Kates structure of a broad view as opposed to granular details). Peak 1, sitting at roughly week 30 on the graph and continuing (though trailing off) through week 100, represents the first phase of reconstruction; reconstruction I, rebuild/replace, was constructed mainly of clusters RC3, RC4, and RC6. Residential clusters 3 and 6 (RC3 and RC6) are located along the centerline of catastrophic damage, while RC4 sits at the western edge of the path. The former two clusters are also fully contained in DC2 (the largest) and represent rebuilding efforts at the start of the effort to restore neighborhoods to pre-tornado conditions. A continued rebuilding effort in RC3 was responsible for the second peak on the graph’s residential line. Residential clusters 2, 5, and 7 (RC2, RC5, and RC7) created the third peak at approximately week 400, and well into reconstruction II, major construction and improved and developed. While also situated along the catastrophic centerline, these three neighborhoods did not see significant rebuilding efforts for 6–8 years post-event. The single commercial cluster along Rex Avenue and north of 20th Avenue created a single peak around week 200. Unlike the spatial analysis, however, it was noted that the chi-square results showed significant increases in commercial permit issuance through the first 5 years following the event.
Plotted against Kates’ (1977) peaks by each of the three phases (initial emergency phase excluded), it can be observed in these data that the functions one would think of as restorative, the drive to fix what can be fixed quickly and set the community on a path to recovery, occurred following the Joplin, MO, event, and aligned well with Kates’ model. Residential rebuilding, undoubtedly an activity that should be categorized as reconstruction I, was shown in these data to have occurred earlier than in the Kates configuration models for a disaster event. Further, while the first peak occurred near the same temporal location as Kates’ reconstruction I high point, the second peak created a multi-modal data distribution. Returning to a closer alignment with Kates, the third of three peaks in the residential building type aligned closely with the reconstruction II phase. Overall, it was concluded that the 2011 Joplin, MO, event fits well into the Kates model when considering significant space-time clusters; this supports previous work in using the Kates recovery model and further encourages additional research of significant events and associated recovery time using the same.
The 2011 Joplin, MO, tornado was a historic, massive, and devastating event that ushered in no less than a decade of rebuilding, and recovery continued as of the 10th anniversary. Disaster recovery is a complex and multi-faceted process, and following a significant event such as this one can span many years. This research has shown that cluster analysis can effectively determine the pace and progress of recovery following a significant natural hazard-related disaster. Future research directions born of this undertaking may include an analysis of the change in lot size and use type and where new clusters of residential and (likely more evident) commercial properties have emerged following the 2011 Joplin, MO, tornado. Further, this research sets the stage for continued analysis at the 20th anniversary in 2031, at which point, and potentially coupled with lot size and use type analysis, could extend the Kates model past the 10-year point and into Alexander’s (2002) post-disaster development stages. It is hoped that this research might add to the knowledge base by furthering the understanding of how patterns can emerge from readily available data, such as building permits.
Alexander, D.E. 2002. Principles of emergency planning and management. Oxford, UK: Oxford University Press.
Chamlee-Wright, E.L., and V.H. Storr. 2008. The entrepreneur’s role in post-disaster community recovery: Implications for post-disaster recovery policy. Mercatus Policy Series, Policy Primer No. 6. George Mason University, Fairfax, Virginia, USA. https://www.mercatus.org/system/files/MPS_PP6_community_and_entrepreneurs-web.pdf. Accessed 18 May 2022.
Comerio, M.C. 2006. Estimating downtime in loss modeling. Earthquake Spectra 22(2): 349–365.
Comerio, M., and H. Blecher. 2010. Downtime data on residential buildings after the Northridge and Loma Prieta earthquakes. In Proceedings of the 9th U.S. National and 10th Canadian Conference on Earthquake Engineering, 25–29 July 2010, Toronto, Canada.
Cui, Y., D. Liang, and B.T. Ewing. 2015. Empirical analysis of building permits in response to hurricane landfalls. Natural Hazards Review 16(4): 1–10.
Duque, J.C., H. Laniado, and A. Polo. 2018. S-maup: Statistical test to measure the sensitivity to the modifiable areal unit problem. PLoS One 13(11): e0207377.
Grazulis, T.P. 1993. Significant tornadoes, 1680–1991. St. Johnsbury, VT: Environmental Films.
Hammersley, J.M., and D. Handscomb. 1964. Monte Carlo methods. London: Methuen & Co.
Han, J., L. Zhu, M. Kulldorff, S. Hostovich, D.G. Stinchcomb, Z. Tatalovich, D.R. Lewis, and E.J. Feuer. 2016. Using Gini coefficient to determining optimal cluster reporting sizes for spatial scan statistics. International Journal of Health Geographics 15(1): Article 27.
Hashemi-Parast, S.O., F. Yamazaki, and W. Liu. 2017. Monitoring and evaluation of the urban reconstruction process in Bam, Iran, after the 2003 Mw 6.6 earthquake. Natural Hazards 85(1): 197–213.
Kates, R.W. 1977. Major insights: A summary and recommendations. In Reconstruction following disaster, ed. J.E. Haas, R.W. Kates, and M.J. Bowden, 261–293. Cambridge, MA: The MIT Press.
Kates, R.W., C.E. Colten, S. Laska, and S.P. Leatherman. 2006. Reconstruction of New Orleans after Hurricane Katrina: A research perspective. Proceedings of the National Academy of Sciences 103(40): 14653–14660.
Kulldorff, M. 1997. A spatial scan statistic. Communications in Statistics, Theory, and Methods 26(6): 1481–1496.
Kulldorff, M. 2021. SaTScan™ user guide for version 9.7. https://www.satscan.org/cgi-bin/satscan/register.pl/SaTScan_Users_Guide.pdf?todo=process_userguide_download. Accessed 18 May 2022.
Kulldorff, M., R. Heffernan, J. Hartman, R.M. Assunção, and F. Mostashari. 2005. A space-time permutation scan statistic for the early detection of disease outbreaks. PLoS Medicine 2: 216–224.
Lee, S., J. Moon, and I. Jung. 2021. Optimizing the maximum reported cluster size in the spatial scan statistic for survival data. International Journal of Health Geographics 20(1): 1–14.
Maslow, A.H. 1943. A theory of human motivation. Psychological Review 50(4): 370–396.
McCarthy, K.F., and M.A. Hanson. 2008. Post-Katrina recovery of the housing market along the Mississippi Gulf Coast, No. 511. Santa Monica, CA: Rand Corporation. https://www.rand.org/pubs/technical_reports/TR511.html. Accessed 18 May 2022.
Openshaw, S. 1978. An empirical study of some zone-design criteria. Environment and Planning A 10(7): 781–794.
Openshaw, S., and P.J. Taylor. 1979. A million or so correlation coefficients, three experiments on the modifiable areal unit problem. Statistical Applications in the Spatial Science 21: 127–144.
O’Sullivan, D., and D. Unwin. 2003. Geographic information analysis. New York, NY: John Wiley & Sons.
Paul, B.K, and M.J. Stimers. 2011. Tornado warnings and circumstance of deaths: The case of 22 May 2011 tornado in Joplin, Missouri. Quick Response Report QR226. The Natural Hazards Research Center, Institute of Behavioral Science, Colorado University, Boulder, CO, USA. https://hazards.colorado.edu/uploads/quick_report/paul_2011.pdf. Accessed 18 May 2022.
Paul, B.K., and M.J. Stimers. 2012. Exploring probable reasons for record fatalities: The case of 2011 Joplin, Missouri, Tornado. Natural Hazards 64(2): 1511–1526.
Paul, B.K., and M.J. Stimers. 2014. Spatial analyses of the 2011 Joplin tornado mortality: Deaths by interpolated damage zones and location of victims. Weather, Climate, and Society 6(2): 161–174.
Peacock, W.G., H. Gladwin, and B.H. Morrow. 1997. Hurricane Andrew: Ethnicity, gender and the sociology of disasters. New York, NY: Routledge.
Quarantelli, E. 1982. Sheltering and housing after major community disasters: Case studies and general conclusions. Disaster Research Center, Ohio State University, Columbus, OH, USA. https://udspace.udel.edu/handle/19716/1132. Accessed 18 May 2022.
Rathfon, D., R. Davidson, J. Bevington, A. Vicini, and A. Hill. 2013. Quantitative assessment of post-disaster housing recovery: A case study of Punta Gorda, Florida, after Hurricane Charley. Disasters 37(2): 333–355.
Santos, J.R., K.D.S. Yu, S.A.T. Pagsuyoin, and R.R. Tan. 2014. Time-varying disaster recovery model for interdependent economic systems using hybrid input-output and event tree analysis. Economic Systems Research 26(1): 60–80.
Simmons, K., and D. Sutter. 2011. Economic and societal impact of tornadoes. Boston, MA: American Meteorological Society.
Simmons, K., and D. Sutter. 2012. Deadly season: Analysis of the 2011 tornado outbreaks. Boston, MA: American Meteorological Society.
SPC (Storm Prediction Center). 2021. Storm Prediction Center WCM Page. https://www.spc.noaa.gov/wcm/. Accessed 18 May 2022.
Smith, D.J., and D. Sutter. 2013. Response and recovery after the Joplin tornado: Lessons applied and lessons learned. The Independent Review 18(2): 165–188.
Stevenson, J.R., C.T. Emrich, J.T. Mitchell, and S.L. Cutter. 2010. Using building permits to monitor disaster recovery: A spatio-temporal case study of coastal Mississippi following Hurricane Katrina. Cartography and Geographic Information Science 37(1): 57–68.
USCB (United States Census Bureau). 2021. Decennial census of population and housing datasets. https://www.census.gov/programs-surveys/decennial-census/data/datasets.2010.html. Accessed 18 May 2022.
The authors acknowledge the City Clerk’s Office of Joplin, MO, for providing the permit data used in this research.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Stimers, M., Lenagala, S., Haddock, B. et al. Space-Time Clustering with the Space-Time Permutation Model in SaTScan™ Applied to Building Permit Data Following the 2011 Joplin, Missouri Tornado. Int J Disaster Risk Sci 13, 962–973 (2022). https://doi.org/10.1007/s13753-022-00456-9
- Joplin tornado
- Space-time clustering
- Space-time permutation model
- Building permit data
- Tornado recovery