Understanding homeowner proactive actions for managing wildfire risks

Wildfires have become increasing threats to residents, built environments, and ecosystems in the USA. Individual responsibility plays an important role in reducing structure-level ignitability, and in turn, overall community vulnerability. When wildfire cannot be completely prevented through risk reduction efforts, homeowners insurance can serve as the second line of defense by allowing homeowners to transfer risk. To understand homeowner decisions on wildfire-related proactive actions and the effects of such decisions on the housing recovery, this study conducted an online survey of homeowners living at high to extreme risk of wildfire in the Western United States and collected data related to two types of proactive actions: individual-level risk reduction actions and homeowners insurance. First, a regression model for each proactive action was estimated to identify key characteristics of homeowners and house/property that had the greatest impact. The results indicated that homeowner age and household income were the two common factors affecting their decisions about home hardening and insurance policies, while the only statistically significant factor in homeowner decisions about defensible space was satisfaction with the surrounding environment (e.g., scenic beauty, privacy). Moreover, the effects of each proactive action on the housing recovery process were evaluated. The results showed that home hardening was a more effective action in reducing wildfire damage to a house than defensible space was, which was consistent with homeowner perception. The survey results also indicated that homeowners with insurance were less likely to experience post-wildfire financial hardship, and subsequently were more likely to repair their damaged houses.


Introduction
Wildfire risk has increased significantly in recent years and is expected to grow across many parts of the USA. Eight out of the top ten costliest wildfires in the USA occurred between 2017 and 2020 (Insurance Information Institute 2020), with notable examples in California (e.g., the Carr Fire, the Woolsey Fire, and the Camp Fire, all of which occurred in 2018 and destroyed over 20,000 structures in California). Climate change may play an important role in increasing the frequency and severity (in terms of fire size and duration) of wildfire. The average global surface temperature has risen by 0.14 ˚F per decade since 1880, resulting in reduced winter precipitation, drier soil and vegetation, and longer summer dry seasons (Pausas and Keeley 2021). Under changing climate, more wildland-urban interface (WUI) areas have become fire-prone environments, and thus more houses in the WUI are exposed to wildfire danger. Moreover, growing population and human activities in the WUI have altered the environment and increased the exposure of human and highvalue assets to wildfire, making the problem even more complicated and severe (Westerling et al. 2011;Moritz et al. 2014;Smith et al. 2016). One-third of residential buildings in the USA are located in the WUI (Wisch and Yin 2019), and 4.5 million U.S. homes are at high to extreme risk of wildfire (Verisk 2019). Such population growth in the WUI, where flammable vegetation is dominant, has increased the likelihood of wildfire ignition (considering that 90% of wildfires are human-caused). Moreover, due to increased exposure in wildfire-prone areas, the small historical fires (e.g., Awbrey Hall Fire Oregon 1990) that did not cause significant economic losses could pose much greater risks if they were to occur today, as demonstrated by Wisch and Yin (2019). With the increasing occurrence of large wildfires and their higher consequences, in recent years, the federal/local governments, insurance companies, and the general public have paid greater attention to risk mitigation actions to protect communities from potential wildfires.
As part of wildfire risk reduction efforts, since the early 1900s, the U.S. federal and local governments have developed and implemented a variety of community-level wildfire protection plans. In the early stages, these policies were intended to reduce the consequences of wildfire by suppressing fires in an efficient and timely manner (Yin 2018), which included fire department support, fire crew allocation, adequate water supply and flow for firefighting, and road improvement for facilitated access. Wildfire policies have since evolved, and community-level efforts have placed more emphasis on reducing the likelihood of wildfire by removing fuel accumulation in high-risk areas (Yin 2018). For example, the Butte County Fire Prevention Program recently added multiple risk management activities such as pre-fire strategies and tactics, vegetation management (i.e., clearing fuels around communities and along roadways and evacuation routes), and fire break zones (Butte County 2021). The federal and local governments also enforce compliance with defensible space laws and regulations and home hardening to reduce property-level risk. To increase wildfire awareness and educate homeowners about wildfire risk reduction actions and evacuation preparedness, educational workshops and brochures are also provided as part of community-level efforts (Haines et al. 2004). Such additional components have been added because there has been an increasing awareness that, unlike other types of natural hazards, the probability of ignition, the rate of fire propagation, and the consequences of wildfire (all of which constitute wildfire risk) can be reduced by human efforts. Community Wildfire Protection Plans (CWPPs) are tailored to specifically address a community's unique conditions and risks related to wildfire (USFA, 2020) and have helped nearly 4,800 communities in the U.S. WUI in identifying their priorities for the protection 1 3 of life, property, and critical infrastructure (Communities Committee 2004;Jakes and Sturtevant 2013).
In addition to community-level wildfire risk management actions, individual homeowner proactive actions are also vitally important for enhancing community resilience. As described in Fig. 1, there are two types of proactive actions that homeowners can take prior to a wildfire event: individual-level risk reduction actions and homeowners insurance.

Proactive action 1: individual-level risk reduction actions
In response to wildfire incidents, individual responsibility plays an important role in reducing the likelihood of ignition and combustion of an individual house and removing potential ignition sources around the house. It has been demonstrated that structure-level wildfire risk can be significantly decreased by reducing its ignition vulnerability to firebrands and flames (i.e., home hardening) and creating defensible spaces surrounding a house (Communities Committee 2004;Syphard et al. 2014). Home hardening involves the use of non-combustible or ignition-resistant siding and trim, installing Class A fire-resistant roof assembly, and installing dual pane windows and/or fire sprinklers, all of which decrease the likelihood of ignition and combustion of an individual house. Designing and maintaining defensible space is also important given that the presence of flame and firebrand ignitions within 131' (or 40 m) of a structure would highly increase the chance of home ignition (Alexandre et al. 2016). Defensible space creates the buffer between a house and surrounding vegetation to slow the spread of wildfire and protect the house from fire. For example, the U.S. National Fire Protection Association (NFPA) defines three home ignition zones based on the distance from the exterior point of a house (NFPA 2021): immediate zone, intermediate zone, and extended zone, which are the areas of 0-5′, 5-30′, and 30′-100′ from the furthest attached exterior of the house, respectively. The NFPA suggests various risk reduction actions that should be taken for each zone ranging from removing flammable material to vegetation treatment (e.g., thinning tree canopies, spacing trees). The abovementioned individual-level risk reduction actions not only protect a house from wildfire but also affect the vulnerability of neighboring houses. Evans et al. (2015) found that such individual-level efforts can reduce the average home hazard by 20%.
The extent of structural damage induced by wildfire depends on various factors that include property-level parameters and landscape factors such as vegetation management, fuel characteristics, and topography. To gain knowledge about the relationship between wildfire Fig. 1 Role of homeowner proactive actions in the post-wildfire housing recovery process risk mitigation actions and the associated structural damages, several studies (Moore 1981;Radtke 1983;Syphard et al. 2012Syphard et al. , 2017Penman et al. 2014;Alexandre et al. 2016) have recently investigated the effects of individual-level risk reduction actions on building damage and its survival rate. As most of them have utilized computer-based simulation or laboratory experiments, until recently, existing wildfire risk mitigation actions have been driven by limited empirical studies that are based on restrictive assumptions on fire behavior, thus resulting in highly theoretical "best practices" (Syphard and Keeley 2019). Without having the actual damage and response dataset, it is nearly impossible to predict the best proactive plan. Hence, post-wildfire surveys and field studies are necessary to collect observations and validate simulation and/or laboratory results. While several researchers have attempted to conduct post-wildfire field surveys on schools and hospitals (Schulze et al. 2020), channel environment (Benda et al. 2003), etc., few studies have collected post-wildfire structural damage sustained by houses.
Extensive studies have been conducted to examine factors that influence wildfire mitigation actions taken by homeowners (e.g., McGee and Russell 2003;Schulte and Miller 2010;McFarlane et al. 2011;Ghasemi et al. 2020;Faulkner et al. 2009). Common factors identified as the key variables affecting homeowner mitigation decisions include property type and value, financial availability, mortgage situation, risk perception and attitude, previous experience with wildfire and other natural hazards, etc. For example, there is considerable evidence that people who have experienced natural disasters are more inclined toward taking risk mitigation actions (Peacock 2003;Ge et al. 2011;Lindell and Perry 2012). McGee and Russell (2003) suggested that several attributes of homeowners, including previous experiences with wildfires, involvement in agriculture and with the local fire brigade, and their social network, contributed to mitigation behavior, while wildfire preparedness within the community was affected by "a culture of self-reliance, experience with fires as part of farming, and social cohesion." The attributes of mitigation measures (such as costs and perceived efficacy) also play a key role in homeowner decisions (Winter and Fried 2000;McCaffrey 2008;Collins 2009). In addition to the characteristics of homeowners, communities, and mitigation measures, the spillover effect (normally termed as neighbor decisions) is also identified as one of the main factors affecting mitigation decisions given that wildfire risk is interdependent (Shafran 2008(Shafran , 2010Butry and Donovan 2008;Taylor 2019;Dickinson et al. 2020;Warziniack et al. 2019). If a structure ignites and burns due to flame or firebrand, the burning structure could become a new source of firebrand generation and threaten adjacent structures (Suzuki et al. 2014). Thus, one's mitigation behavior affects the risk levels of the neighboring environment and vice versa. For example, unmitigated properties or public lands adjacent to the property would reduce the willingness of owners to take mitigation actions, as they believe the efficacy of their own mitigation actions would be low (Brenkert-Smith et al. 2006, 2012. Inversely, mitigation actions taken by neighbors can cause the free-rider problem (NFPA 2020) also highlighted the importance of collaborative actions between neighbors to reduce their shared risk. As such, homeowner mitigation decisions have been well studied in the past two decades. However, many of them have focused on a single type of mitigation action and have not related these actions to housing damage and recovery, which is necessary for assessing their roles in community resilience.

Proactive action 2: homeowners insurance
Homeowners insurance, which covers damages to structures and personal belongings induced by wildfire and smoke, is another important proactive action that homeowners can take in case their properties are damaged by residual wildfire risks (Lee and Li 2021;Zhao et al. 2020). It helps policyholders in alleviating excessive financial burdens from repair/ reconstruction costs following a high-consequence wildfire event by transferring risks to a third party and over time (Pelling 2003). With sufficient insurance coverage, homeowners may receive reliable and timely claim payments that enable the expedited recovery processes of houses, and in turn, the community as a whole. The role of catastrophe risk insurance (e.g., earthquake insurance or flood insurance which is not covered by homeowners insurance and requires a separate endorsement) has been well investigated in several studies (Hazell 2001;Kunreuther and Pauly 2006;Paudel 2012;Charpentier 2014), but most of these studies have taken a qualitative approach to assessing its effect on community resilience. Due in part to the lack of quantitative assessment of catastrophe insurance, the effect of insurance has often been underestimated or neglected in many community resilience planning studies. While a limited number of studies have developed either theoretical or statistical models for the insurance purchase behavior of forest owners and/or homeowners through experimental economics and survey (McKee et al. 2004;Gan et al. 2014), most of these studies have considered only binary variables in their model (i.e., purchase insurance or not). However, given that homeowners with a mortgage (i.e., over 70% of the U.S. population) are required to purchase homeowners insurance, insurance coverage limits play a more significant role in housing recovery than insured status does. In many cases, uninsured and underinsured homeowners are low-(and middle-) income residents who are identified as economic and socially vulnerable groups in society (Eriksen and de Vet 2020;Priest et al. 2005;Mockrin et al. 2015), and therefore their post-disaster financial availabilities may greatly affect community resilience. Moreover, few studies have attempted to relate homeowner purchasing behavior to their financial availability following a hazard event (Zhao et al. 2020).
To address the significant research gaps identified above, this paper examines potential factors that may affect homeowner proactive actions for managing wildfire risks and the role of such actions in the housing recovery. To achieve the goal, a post-wildfire online survey of homeowners was conducted in multiple counties at high to extreme risk of wildfire in California and Washington. Based on the online survey results, we identified the key factors that contributed most to homeowner decisions about each proactive action and estimated a set of logistic regression models in order to relate the characteristics of houses and homeowners to their willingness to invest in the proactive action. Then, the effect of proactive actions on the housing recovery process was explicitly modeled by assessing (a) the effect of homeowners insurance policy on delay time (T delay ), and (b) the effect of individual-level risk reduction actions on housing repair time (T repair ); see Fig. 1. The rest of this paper is organized as follows. Section 2 describes the investigation methods including the online survey data collection method and statistical analyses. In Sect. 3, the results obtained from the data analyses are discussed. Finally, Sect. 4 presents summary and conclusions.

Research questions
This study is motivated by the following research questions: 1. What factors affect homeowner proactive actions for managing wildfire risks? 2. How do the proactive actions taken prior to a wildfire event affect the housing recovery process following the event?
To answer the research questions, this study uses a quantitative approach to identifying independent variables that are likely to influence homeowner proactive actions for managing wildfire risks and assessing their impacts on the housing recovery processes through an online survey. The quantitative models developed based on the survey data will provide federal/local government with preliminary insights into how to motivate homeowners to take proactive actions and which proactive action could be more effective in expediting the housing recovery process.

Data collection
To collect data used to support quantitative models, we conducted an online survey of homeowners in California and Washington where at least one wildfire event occurred in the past five years. The affected areas were identified by being overlaid with the layer containing wildfire perimeters in the past five years in ArcMap. Participants were recruited in January 2021 by Qualtrics research service, a professional organization that uses prequalified respondents to achieve significant response rates for the purpose of validity. While online convenience sampling was used to recruit participants, these samples collected by Qualtrics panels could be demographically and politically representative, as demonstrated in Boas et al. (2020). The inclusion criteria for participating in this survey were to be homeowners in the study area ( Fig. 2) who were at least 18 years of age and whose houses were damaged by at least one wildfire event (i.e., house experienced at least a minor damage state due to wildfire) in the past five years. Since the survey was designed to assess the effect of pre-wildfire proactive actions on post-wildfire house damage and delay time, homeowners whose houses were not damaged by wildfires were excluded. Figure 2 shows the study areas and the number of valid survey responses received from each area.
In the study area, about 30,000 structures (i.e., the estimated total population size) were damaged or destroyed by major wildfires. Based on the approach to determining sample size presented in Krejcie and Morgan (1970), 64 responses were a sufficient sample size to generate a 90% confidence interval and 10% margin of error (MoE), while 95 sample size was required for 95% confidence interval and 10% MoE. We collected 85 responses from the study area and excluded five invalid responses if the answers to the survey questions did not meet the following requirements: (a) one's mortgage balance should be less than or equal to his/her property value; (b) the dwelling coverage limit should be less than or equal to his/her property value; and (c) given the dwelling and content coverage limits, the insurance premium and deductible should be within a reasonable range. After the exclusion, total 80 valid responses (71 responses from California and 9 responses from Washington) generated a 93% confidence interval and 10% MoE. Although 10% MoE may be considered large, there was a barrier to increasing the sample size because (a) only homeowners who experienced at least minor wildfire damage to their houses in the past five years could participate in this survey, and (b) convenience sampling was used.
To enable participants to clearly understand the four possible damage states (i.e., none, minor damage, major damage, and destroyed) sustained by a house due to wildfire, a detailed information page describing the damage states was provided at the beginning of the survey. The page presented the images of a house experiencing each damage state along with a detailed description as illustrated in Table 1. The online survey consisted of a set of closed-ended quantitative and qualitative questionnaires, including demographic information; property type and value; risk perception/attitude; previous experience with wildfire and other types of natural hazards; homeowners insurance policy; homeowner financial availability, mortgage situation; proactive actions taken at the time of the most recent wildfire event; and time to initiate and complete the recovery of their houses. Moreover, participants were encouraged to have their homeowners insurance policy documents (the policies they held at the time when their properties were damaged by the most recent wildfire) at hand so that they could answer the questions related to their insurance policies, such as dwelling/content coverage limit and deductible, additional living expenses (ALE) coverage limit, and annual insurance premium. All study protocols were approved by the Washington State University Institutional Review Board before the study commenced (IRB #18084).

Statistical analyses
The data were analyzed in two steps. First, to investigate the first research question, we assessed the effects of independent variables on homeowner decisions about two different types of proactive actions and identified the key variables that should be included in the logistic regression models. Then, to address the second research question, we constructed the quantitative relationship between homeowner proactive actions and the delay and repair times of a damaged house and examined their impact on the housing recovery process.

Regression models for proactive actions
To identify key variables affecting each type of proactive action (i.e., individual-level risk reduction actions and homeowners insurance), we performed a logistic regression analysis for each action based on the survey data. The independent variables considered at the initial step of estimating the regression models were the variables describing the characteristics of house/property and homeowners and are summarized in Table 2 and Appendix A. We first coded all these independent variables as dummy variables to analyze qualitative data such as categorical representation (e.g., house damage state, demographic information). As a homeowner had decided whether or not to take a certain type of proactive action before his/her property was damaged by the most recent wildfire, a decision variable can take only two values, 1 (take action) or 0 (do not take action), which are mutually exclusive and collectively exhaustive. Thus, the probability ( p ) that an individual homeowner had taken a certain type of proactive action is expressed by the following logistic regression equation: in which = the vector of coefficients for independent variables; and X = the vector of independent variables. Higher p indicates a higher probability that a homeowner had taken the proactive action. The regression model for each proactive action was estimated using a backward stepwise regression approach. It began with all the independent variables and at each step gradually eliminated the least significant variables from the regression model until only statistically significant variables were left in the model. At each step, the Wald Chi-Square Test was used to select the variable that should be eliminated: the variable with a P value greater than a 5% level of significance was eliminated. This process was repeated until all the remaining variables did not meet the specified level for elimination. The final model obtained from this approach was compared with the models with more independent variables (obtained from the previous steps) and was found to be the optimal one based on the Akaike Information Criterion (AIC) and/or Bayesian Information Criterion (BIC). First, logistic regression models were estimated for two types of individual-level risk reduction actions (i.e., home hardening and defensible space), respectively. The survey asked participants to indicate the types of individual-level risk reduction actions they took at the time when their properties were damaged by the most recent wildfire and then classified them into two categories: home hardening (e.g., use of non-combustible or ignitionresistant siding and trim, installation of Class A fire-resistant roof assembly, installation of multi-pane windows or ideally tempered glass, installation of fire sprinklers) and defensible space (e.g., design and maintenance of the immediate, intermediate and/or extended zones). Hence, the dependent variables for the regression models were whether or not home hardening or defensible space was adopted at the time of the most recent wildfire. A logistic regression model for homeowners insurance was also estimated, where the dependent variable was binary: a homeowner purchased or did not purchase homeowners insurance at the time when his/her property was affected by the most recent wildfire. However, as explained in Sect. 1, a binary logistic regression model for insurance purchase could be meaningful only if homeowners can fully decide whether to buy insurance based on their preference. However, about 70% of U.S. households have a mortgage, and lenders require homeowners with a mortgage to insure houses to protect their investment. Consequently, these homeowners do not have much room to decide whether or not to purchase homeowners insurance. However, they may have more freedom to choose insurance coverage (although some lenders still have minimum requirements for dwelling coverage). Moreover, insurance coverage plays a critical role in estimating the financial availability of homeowners following a wildfire event. If a house is not fully insured, homeowners still experience financial hardship and tend to delay house recovery until they obtain financing. In light of this, in addition to the binary logistic regression model for insurance purchase, homeowner decisions about insurance policy were also investigated. Homeowners insurance policy includes four types of coverage: dwelling, personal property, additional living expenses (ALE), and liability protection. The survey did not include any questions pertinent to liability protection because the study focused on structural damage to the house, attached structure, and personal property, and the associated economic losses. Dwelling coverage is the main component of homeowners insurance policy, and the other two coverages are often determined by the percentages of the selected dwelling coverage, such as 50% to 70% of dwelling coverage for personal property and 20% of dwelling coverage for ALE. Thus, a linear regression model for dwelling coverage limit was estimated. The ordinary least squares approach was used to estimate unknown coefficients, and then a stepwise backward approach with t-tests was used to eliminate insignificant independent variables.

Quantitative relationship between proactive actions and housing recovery
To quantitatively assess the impacts of individual-level wildfire risk reduction actions and homeowners insurance on the housing recovery processes, we examined the relationship between proactive actions and variables related to house damage and recovery. First, we investigated how homeowner risk reduction actions taken prior to a wildfire event affected post-wildfire damage states. Four different groups of homeowners were considered as independent variables: homeowners without any risk reduction actions, with home hardening only, with defensible space only, and with both actions. The frequency distributions of house damage state (minor damage, major damage, and destroyed) for each group were constructed. In this study, the houses that had been already hardened before homeowners bought them were not considered because many homeowners were not likely to be aware of structural hardening adopted before they moved in.
We also assessed the effect of insurance on housing recovery by considering homeowner repairing decisions, post-wildfire financial availability, and delay time as dependent variables. The participants were divided into three groups, including Group A (homeowners with full dwelling coverage), Group B (underinsured homeowners), and Group C (uninsured homeowners). Then, for these three groups, homeowner repairing decisions given a housing damage state were compared. Moreover, several questions about homeowner post-wildfire financial situation were asked during the survey, including homeowner out-of-pocket expenses and financial hardship they experienced due to a wildfire event, and the most helpful financial source for them to recover from wildfire-induced damage. The relationships between independent variables (three groups of homeowners) and dependent variables were constructed. Lastly, in this study, delay time was defined as the time between wildfire containment and the initiation of the house repair/reconstruction process. It can be induced by many different impeding factors, such as post-disaster inspection; engineering mobilization and review/redesign; financing; contractor mobilization and bid process; and permitting (Lee et al., 2019;Zhao et al., 2020). Specifically, financing delay was calculated as the sum of the time required for homeowners to secure funding sources and the time due to delayed payments associated with insurance, loan, or government assistance. To estimate the total delay time, the survey question also asked participants about the time taken to initiate their structural (house) repair/reconstruction process since the wildfire in the community had been contained. Then, we examined how delay time was impacted by the insurance coverage limit.

Results and discussions
The demographic characteristics of the sample are presented in Table 2. Sixty percent of the participants were male, and 40% of them were female. The age group between 30 and 39 years old comprised the highest proportion in the sample, and 68.75% of the respondents were Caucasian. The majority of the respondents (60%) attended college. At the time when their properties were damaged by the most recent wildfire, 72.5% of the respondents were employed, and 43.75% of the respondents reported an annual household income before taxes of $25,000 to $99,999, followed by $100,000 to $149,999 (22.5%) and less than $24,999 (17.5%). Census demographic data (in percentage) are also summarized in Table 2. The sample mean of each demographic characteristic was compared with its population mean through a hypothesis testing approach to demonstrate if the collected sample could represent the population well. The P values are summarized in Table 2. For most of the demographic characteristics, the null hypothesis that there was no significant difference between the population mean and sample mean was accepted, which implies the representativeness of the collected sample. However, at the 5% significance level, the null hypothesis was rejected for age and educational background.

Key factors affecting homeowner proactive actions
This subsection identifies independent variables that are likely to influence homeowner proactive actions for managing wildfire risks to answer the first question specified in Sect. 2.1.

Individual-level risk reduction actions
Based on the survey results, 65% of the respondents adopted home hardening, while 58.75% of the respondents adopted defensible space. As shown in Table 3, the key independent variables that affected decisions to adopt home hardening were homeowner age, household income, total wealth, and willingness to invest in individual-level wildfire risk mitigation actions. It should be noted that some of these actions (e.g., siding, asphalt roof shingles, dual pane windows) could have been taken for aesthetic purposes or simply to replace worn items. However, given that homeowner willingness to invest in wildfire risk mitigation actions was identified as one of the most statistically significant factors, the results imply that many homeowners took such actions to harden homes. There was no significant evidence to claim that age was correlated with other key independent variables (i.e., their willingness to invest in risk mitigation actions, household income, and total wealth). Homeowner decisions about home hardening obtained from the survey were plotted against the simulated ones from the regression model in Fig. 3. It showed 80% accuracy in estimating decision variables. The logistic regression results for defensible space are summarized in Table 4. Satisfaction with the surrounding environment (e.g., scenic beauty, proximity to recreation, privacy) was identified as the only variable that significantly affected homeowner decisions to design and maintain defensible space surrounding their homes. Seventy percent of the simulated results from the regression model for defensible space matched the survey results. As presented in Table 4, the variables describing the characteristics of homeowners and houses did not have any statistically significant impacts on homeowner decisions about defensible space. It can be interpreted that, in this survey, homeowners perceived defensible space as a proactive action to mitigate wildfire risks to the surrounding environment rather than a house itself, whereas home hardening was adopted to protect their houses from wildfires. We also performed a regression analysis for homeowners who adopted both types of wildfire risk mitigation actions and summarize the results in Table 5. While the level of their confidence in the structural resistance of the houses against wildfire had negative impact on their decisions, prior experience with natural hazards encouraged them to take both actions. Moreover, people who purchased homeowner insurance showed higher probability of adopting both risk mitigation actions. This result is consistent with other studies (e.g., Meldrum et al. 2019) revealing that risk reduction decisions and insurance decisions could be jointly determined.
Additionally, we performed a hypothesis test to see if a wildfire-related policy could be one of the key independent variables affecting homeowner proactive action. Considering that only California requires homeowners living in the State Responsibility Area (SRA) to create a 100 ft defensible space, we set up the null hypothesis that there was no significant difference in the ratios of homeowners creating defensible space between California and Washington (which does not have such policy). The null hypothesis was accepted, which indicated that the requirement for defensible space in SRA did not affect the ratio of Fig. 3 Comparison of homeowner decisions about home hardening between simulated and survey results homeowners creating defensible space significantly. Given that only 9 responses were collected from Washington, however, this result cannot be generalized to the entire population or other types of wildfire-related policies.

Homeowners insurance
The survey data indicated that 80% of the respondents had homeowners insurance in place at the time of the wildfire. As presented in Table 6, the homeowner mortgage balance and neighbor proactive actions were identified to have statistically significant impacts on homeowner decisions about insurance purchases. Considering that homeowners with a mortgage were required to purchase homeowners insurance, the finding that a positive mortgage balance was the most significant factor in this regression model was well-supported. In addition, we also examined the independent variables that might affect homeowner decisions about insurance policy (more specifically dwelling coverage limit) through linear regression analyses. The results are summarized in Table 7. As expected, property value had the greatest impact on homeowner decisions about dwelling coverage limits. Interestingly,  the other two independent variables (i.e., age and household income) in the regression model were also the key variables that were likely to affect homeowner decisions about home hardening (see Table 3). Given that the binary regression model for insurance purchase does not reflect homeowner preference appropriately, homeowner age and household income can be considered the two most statistically significant factors affecting homeowner decisions to take proactive actions.

Effect of proactive actions on postfire housing recovery
This subsection quantitatively assesses the impacts of pre-wildfire proactive actions on the housing recovery processes to answer the second question specified in Sect. 2.1.

Individual-level risk reduction actions
We assessed the effect of individual-level wildfire risk reduction actions on house damage state. Only 12.5% of the respondents did not take any pre-wildfire risk reduction actions (i.e., either home hardening or defensible space) at the time when their houses were affected by the most recent wildfire. As shown in Table 8, 60% of the homeowners without any risk reduction actions experienced minor structural damage to their houses, while 40% of their houses were destroyed by a fire. The vast majority (87.5%) of the respondents adopted at least one individual-level wildfire risk reduction action. To determine whether there was a statistical relationship between mitigation action and structure damage state, the null and alternative hypotheses were defined as follows: H 0 : There is no relationship between mitigation action and structural damage state.
H a : There is a relationship between mitigation action and structural damage state. The chi-square statistic was 6.3317, and the P value was not significant (0.3897), indicating that the null hypothesis was not rejected. Given that the chi-square test is very sensitive to sample size, however, it is hard to conclude that there is no relationship between these two variables.
More specifically, homeowners who adopted only fuel treatment in defensible space had similar house damage state distribution as those who did not adopt any mitigation actions. Minor structural damage comprised the highest proportion (56%), followed by destroyed (33%) and major damage (11%). Based on the results, it was not clear if designing and maintaining defensible space was effective in reducing wildfire damage to a house. It might be because defensible space reduces the chance of firebrand ignitions in the surrounding environment rather than mitigating wildfire consequences if a home has already ignited. Since homeowners who did not experience home ignition were screened out at the beginning of the survey, the reduced chance of ignition (i.e., the efficacy of defensible space) was not captured in the survey results. The result was also consistent with our interpretation that defensible space was perceived by homeowners as a proactive action to mitigate wildfire risks to the surrounding environment rather than reducing post-wildfire house damage (see Sect. 3.1.1). On the other hand, homeowners who adopted only home hardening or both types of mitigation actions were less likely to experience a destroyed damage state (i.e., 13% and 21%, respectively) compared to those who adopted only defensible space or none of these actions. Hence, it can be inferred from the results that home hardening could be a more effective action to reduce wildfire damage to a house.

Homeowners insurance
First, we examined the effect of insurance on homeowner repairing decisions. Among the three groups, Group B (underinsured homeowners) comprised the highest proportion of the participants, followed by Group A (homeowners with full dwelling coverage) and Group C (uninsured homeowners). Table 9 summarizes the repairing decisions of these three groups. While the damage state distributions of the three groups were similar, Group C homeowners were much less likely to repair their houses, especially when the houses were completely destroyed. It should be noted that it would be possible that Group C homeowners were more likely to be seasonal or second homeowners and were not willing to rebuild their homes that were not their primary residences. However, the result generally suggested that insurance could ensure homeowners to be financially secure following a wildfire event and help them repair their houses. Hence, it can be expected that a community with a higher insurance take-up rate (and more homeowners having full coverage) would have a higher rate of housing repair after a wildfire event.
We also assessed the effect of homeowners insurance on the financial availability of homeowners following a wildfire event because their post-wildfire financial situation determines the financing delay time of the housing recovery process as well as repairing decisions. First, we estimated homeowner out-of-pocket expenses by measuring the difference between their estimated total wealth (including housing equity, vehicles, retirement, life insurance, fixed-income investment, managed assets, common stock and mutual fund shares, liquid assets, farms, business equity in other real estates, and net worth) prior to and following a wildfire event. The out-of-pocket expenses of the three groups were compared in Fig. 4a. As expected, Group A experienced the least out-of-pocket expenses ($23,913), which indicated that full dwelling coverage reduced the financial burden of Group A homeowners following a wildfire event. However, contrary to our expectation, the mean value of the out-of-pocket expenses of Group B ($125,000) was much higher than the mean value of Group C ($35,938). It may be because most of the homeowners in Group C decided not to repair their damaged houses due to lack of financial availability as shown in Table 9, and thus their out-of-pocket expenses could not necessarily reflect repair/reconstruction costs. On the other hand, higher portion of the homeowners in Group B decided to repair their houses because they received payment from insurance companies. They still had to pay deductibles and the remaining repair/reconstruction costs that were not covered by homeowners insurance, which induced higher out-of-pocket expenses. Therefore, it would not be wise to conclude that homeowners insurance was not effective in reducing the financial burden of underinsured homeowners. The effectiveness of homeowners insurance in post-wildfire financial situation was further supported by the survey results. Fifty percent of Groups A and B homeowners (who  . 4 Comparison of financial situations between three groups having different insurance dwelling coverage had either full or partial insurance coverage) indicated that insurance was the most helpful financial source for them to recover from the wildfire event, followed by loans (23.43%), government assistance (18.75%), and other sources (7.82%), while 62.5% of Group C homeowners reported government assistance as the most helpful financial source. Participants were also asked to answer a question about the type of financial hardship they experienced due to wildfire damage (e.g., mortgage default, mortgage forbearance, selling their properties, and huge loan or debt). As shown in Fig. 4b, only 25% of Group C homeowners responded that they did not experience any financial hardship after experiencing wildfire, while homeowners who did not report any financial hardship in Groups A and B were 43.5% and 31.7%, respectively. Moreover, Groups A and B homeowners answered that, on average, 75.38% of repair/reconstruction costs were covered by insurance. All these results supported the effectiveness of homeowners insurance in reducing post-wildfire homeowner financial burden. Moreover, the effect of homeowners insurance on delay time (T delay ) was examined. The mean values of the delay time of Group A (8.1 months) and Group B (8.9 months) were longer than the mean value of Group C (6.8 months), which was contrary to our common belief. Since delay time was not only induced by lack of financing but also induced by other impeding factors and only a limited number of Group C homeowners repaired their damaged properties, this result might not be able to capture the effect of homeowners insurance on the delay time. In this regard, we further quantified the effect of different types of financial sources on financing delay time (which was defined as the time between claim application and payment). However, government financial aid showed a shorter delay time (3.05 months), followed by homeowners insurance (3.21 months) and loans (3.23 months). In conclusion, the findings of this study indicated that homeowners insurance did not reduce financing delay, while it encouraged homeowners to repair their damaged houses by relieving the financial burden from repair/reconstruction costs.

Summary, limitations, and conclusions
This study conducted a post-wildfire online survey and statistical analyses of homeowners living in multiple counties at high to extreme risk of wildfire in California and Washington to understand homeowner decisions on wildfire-related proactive actions and the effect of such actions on the housing recovery. The online survey was targeted at homeowners in these counties whose properties were damaged by wildfires in the past five years. The survey results revealed that homeowner age and household income were the common key independent variables affecting homeowner decisions about both home hardening and homeowners insurance, while the only key independent variable in the regression model for defensible space was satisfaction with the surrounding environment. Although the survey results indicated that home hardening was found to be more effective in reducing wildfire damage to a house than defensible space did, it would not be wise to generalize this conclusion due to the limited sample size. Moreover, the results clearly implied that the effect of homeowner insurance on post-wildfire financial availability was significant, and homeowners with higher coverage limits were more likely to repair their damaged properties. However, contrary to our initial expectation that homeowners with full coverage might receive expedited claim payments that could speed up the housing recovery processes, its effect on reduced financing delay was not supported by the findings. The results from this study can provide guidance to federal/local government on (a) how to motivate homeowners to adopt proactive actions by identifying the key factors affecting homeowner decisions (i.e., the answer to the first research question in Sect. 2.1), and (b) how to effectively enhance community resilience by understanding the effect of proactive actions on housing recovery process (i.e., the answer to the second research question).
There are some limitations and room for improvement in this study that could be addressed in future research. First, to generalize the results, the sample size needs to be increased, and the collected sample should be able to represent the general population. Considering that the conclusions were drawn from the homeowners whose houses sustained at least minor wildfire damage, the results and conclusions of this study cannot be applied to the houses that survived a wildfire without any damage. It should be noted that the comparison between the samples that reported wildfire-induced structural damage and the samples that did not report any wildfire-induced damage could provide further insight into the effect of pre-wildfire proactive actions on house survivability. Second, this study assumed that the survey respondent was representative of his or her household or primary decision-maker and used the individual-level variables (e.g., age, gender, education, employment) to predict household-level proactive mitigation actions-home hardening, defensible space, and homeowners insurance-in the logistic regression models. Given that this assumption may introduce bias due to representative issues (Hung and Wang 2022;Seebauer et al. 2017), future studies should address this issue by restricting the predictors to household-level variables or by conducting an in-depth interview with all members of a household. Third, post-wildfire reconnaissance surveys can be conducted to better quantify the effects of individual-level risk reduction actions on structural damage to houses, and in turn, the repair time of houses. These results will greatly complement the results obtained from the online survey because the impact of defensible space on wildfire risks to houses was not clearly quantified in this study. Moreover, based on our content validity, some respondents seemed to have difficulty quantifying their total wealth, property values, expenditures, economic losses, among others. Therefore, in-depth individual interviews with homeowners will also be helpful to provide sufficient information and guidance on such quantification processes to interviewees. 10.00 $750,000-$999,999 = 8 5.00 $1,000,000-$1,499,999 = 9 6.25 $1,500,000-$1,999,999 = 10 3.75 $2,000,000-$2,999,999 = 11 0.00 > $ 3,000,000 = 12 1.25 Homeowner perceived probability that wildfire will occur in his/her community in the next 5 years 01 2.50 0-1% = 2 1.25 1-4% = 3 2.50 5-9% = 4 2.50 10-19% = 5 12.50 20-39% = 6 13.75 40-59% = 7 22.50 60-79% = 8 17.50 80-100% = 9 22.50 Homeowner perceived probability that his/her property will be damaged by a wildfire event