Sage Journals: Discover world-class research

Abstract

Wildfires continue to threaten multiple regions in Canada, and evacuations are often the primary means of ensuring life safety. Understanding how people make decisions before and during wildfire evacuations is thus important in informing preparedness and planning. This research collected survey data between May and July 2023, from residents living in high to moderate fire-risk areas of Alberta and British Columbia (N = 2,868) to understand their intended evacuation behavior and choice-making during a future wildfire event. Our analysis focuses on underserved groups (people with disabilities, older adults, lower-income households, visible minorities, and carless residents), often neglected in evacuation planning processes. We contribute to the literature by uniquely focusing on decision-making within distinct underserved groups, rather than simply using these identities as variables within a broader model. Estimated logit models offer insight into factors affecting evacuation departure timing, destination and route choices, mode choices, and preferred shelter types. Results suggested that factors such as perceived risk, previous evacuation experiences, and intersecting vulnerabilities have a significant influence on group choices. For example, whereas risk perception significantly influenced evacuation timing among people with disabilities, sociodemographic characteristics were significant in determining shelter choices among older adults. These findings have important implications for enhancing equitable wildfire evacuations, pointing to the need for tailored strategies that consider the needs, barriers, and decision-making patterns of underserved groups. We provide several policy recommendations for local agencies, including ensuring multimodal evacuation plans with transit and shared mobility considerations and providing targeted support for those with intersecting vulnerabilities.

Keywords

evacuation planning wildfires underserved and vulnerable groups travel behavior transportation equity

Recent years have seen intense and long-lasting wildfires in North America, often requiring mass evacuations to ensure safety among residents. Specifically in Canada, there were 1,393 wildfire evacuations between 1980 and 2021, with more than 500,000 people evacuating ( 1 ). This challenge necessitates evacuation preparedness and effective logistical planning (i.e., determining evacuation rates, preparing transportation resources, identifying evacuation destinations, and managing evacuation shelters). However, previous disasters show that evacuation planning is still inadequate for underserved populations ( 2 ). For example, during the Alberta Fort McMurray wildfire in 2016, people with disabilities and older adults faced challenges evacuating. School buses used to transport these groups were not equipped with lifts or sufficient spaces to accommodate those with wheelchairs and walkers ( 3 ). Moreover, despite a growing body of research on modeling and understanding behavior and decision-making during evacuations, literature on equity considerations for underserved populations is sparse.

Consequently, this paper seeks to work toward an understanding of wildfire evacuation decision-making among underserved groups. We asked the following questions to guide our research:

How do underserved groups make evacuation/logistical decisions during wildfires?

What factors influence decision-making among underserved groups during wildfire evacuations?

How can local agencies equitably improve evacuation preparedness and planning for these groups?

To answer these questions, we collected survey data from 2,868 respondents in Alberta and British Columbia. Using a stated-preference format, we asked questions to understand choice-making and behavior during a wildfire evacuation. We focus our analysis on five groups (n = 1,622) that have been historically underrepresented in evacuation planning: people with disabilities, older adults, lower-income households, carless residents, and visible minorities. Findings from this study provide group-specific insight into evacuation decisions, preparedness, and logistical considerations during wildfire events.

Literature Review

In this section, we examine existing literature related to underserved groups in evacuation planning and evacuation behavior modeling. We conclude the section with key gaps our research aims to address.

Underserved Groups in Evacuation Planning

Underserved populations are often disproportionately affected by disasters owing to limited capacity to prepare for, respond to, and recover from the events ( 4 ). Particularly during evacuations, vulnerabilities can be heightened by physiological barriers (such as those involving physical and/or cognitive limitations), social barriers (including one’s culture, language, or race), or economic barriers (caused by the direct and indirect costs associated with evacuations) ( 5 ).

People with disabilities and older adults are particularly vulnerable as they are likely to experience health-related complications in addition to mobility challenges when evacuating ( 6 ). Consequently, research recommends creating partnerships with paratransit providers who are often familiar with the needs of those with limited mobility and can coordinate vehicles capable of transporting medical equipment during an evacuation ( 7 ). Studies show that carless residents, lower-income households, and immigrants/visible minorities primarily use public transit for daily mobility and are thus more likely to rely on transit during evacuations ( 8 , 9 ). Most cities, however, still lack public-facing evacuation plans with multimodal considerations ( 2 ). To meet these disadvantaged groups’ mobility needs, work has recommended that evacuation plans integrate transit use and include up-to-date information on pickup and dropoff locations as well as bus routes ( 10 ).

In addition to transportation needs, underserved groups often face unique challenges when seeking/staying at evacuation shelters. Many shelters, however, remain underprepared to meet these needs. For example, during Hurricane Irma, a designated special needs shelter in Monroe County, NY, was reported to have malfunctioning building elevators, limiting the movement of those using wheelchairs ( 11 ). The shelter further lacked licensed healthcare staff, which negatively affected evacuees with acute medical, physical, and/or mental health needs ( 11 ). Moreover, a study by Cutter and Smith on Hurricane Gustav found that lower-income and carless residents were transported by bus/train to shelters with insufficient restroom facilities, limited sleeping spaces, and long lines for food stamps ( 12 ). Unsurprisingly, research has shown that physical and social barriers in shelters may hinder evacuation compliance among those with access and functional needs ( 13 ). Inequities during emergency evacuations point to the need for tailored plans that consider the specific needs, barriers, and decision-making patterns of underserved groups.

Evacuation Behavior Modeling

Studies have increasingly employed discrete choice modeling as a tool to understand evacuee behavior and decision-making during an emergency. In discrete choice modeling, utility maximization is used as the primary decision rule—that is, given their characteristics/variables, a decision maker will choose the most attractive choice/alternative that maximizes their utility ( 14 ). During a wildfire evacuation, individuals make several key decisions including whether to evacuate or stay, when to depart, what transportation mode to use, what route to take, and where to seek shelter. Several researchers have used discrete choice modeling to understand the factors affecting an individual’s binary decision to evacuate or stay during a wildfire event. Factors such as gender (female), receiving a mandatory evacuation order, and higher risk perceptions were found to increase the likelihood of evacuation, whereas previous evacuation experiences, higher income, and age (younger), were found to increase the likelihood to stay ( 15 – 18 ). Moreover, studies on wildfire evacuations have shown that the presence of children in households, gender (men), and residence time (greater than 15 years) were found to lengthen departure time whereas experience with fire damage, residence ownership, and higher risk perceptions were found to decrease evacuation delays ( 19 , 20 ).

With regard to mode choices, a study by Wong et al. found that although households with children were more likely to take multiple vehicles during a wildfire evacuation, low-income households were less likely to use two or more vehicles ( 15 ). Whereas route choice modeling has been limited owing to its complexity, work by Wong et al. found evacuation orders to be influential in determining evacuees’ route choices ( 15 ). Specifically, those who received a mandatory order were likely to use highways whereas those who received a voluntary order were less likely to use highways ( 15 ). Finally, shelter choice models found that work requirements and age (older) significantly influenced the choice to stay with friends/family during a wildfire, whereas worry about finding housing was significantly correlated with staying at a hotel or private shelter ( 15 ). In all, wildfire evacuation behavior is generally understudied, and equity-centered analyses on this topic are sparse. Notable exceptions include work by Grajdura and Niemeier, who explored inequities in California wildfires using mixed methods (e.g., decision trees), and found significant inequalities in evacuation rates, sheltering, and displacement ( 21 ). Similarly, work by Sun et al. using ordinary least squares and weighted regression models, investigated vulnerabilities during the 2019 Kincade Fire, finding that disabilities and older age significantly delayed departure times ( 4 ).

Key Research Gaps

Despite considerable work in evacuation behavior modeling, several gaps remain. First, we observed a limited incorporation of equity considerations in wildfire evacuation research. Equity needs during hurricane evacuations have been examined, mostly as independent demographic variables (see Sadri et al. [ 22 ] for distinct models composed of disadvantaged groups). However, equity-centered research on wildfire evacuations remains sparse, beyond a few examples ( 4 , 21 , 23 ). Differences in evacuation strategies between hurricane and wildfire evacuations—such as shorter warning times and different hazard risks for wildfires—are likely to affect behavior and equity outcomes, particularly among underserved groups with limited resources ( 24 ). Second, we found that discrete choice models often incorporate underserved groups as variables within broader models of the general population. Models may thus fail to capture unique decision-making patterns within underserved groups. Finally, the impact of intersecting vulnerabilities on evacuation behavior is not well understood. Identifying these impacts is important as overlapping vulnerabilities can compound evacuation challenges and exacerbate inequities. In this research, we contribute to the literature by investigating wildfire evacuation choice-making among underserved groups to inform more equitable approaches to evacuation planning during wildfires.

Methodology

In this section, we outline the methodologies used for survey distribution and discrete choice modeling and present an overview of respondent demographics.

Survey Distribution

We distributed an online survey to Alberta (AB) and British Columbia (BC) residents living in areas with high to very high wildfire risk. An online format was selected to reach a broader population of respondents quickly and cost-effectively. Questions were designed to gather data on evacuation decisions, preparedness, and logistical considerations during wildfire events. The survey was conducted from May to July 2023 through both convenience and market research panel samples. Convenience sampling involves recruiting participants based on ease of access (including geographic locations) whereas a market research panel is often managed by third-party firms and provides access to prerecruited participants ( 25 , 26 ). Combining these two sampling methods enabled us to widen our reach and increase sample diversity. Similar mixed-method sampling strategies have been used in prior behavioral research (see, for example, Ciriaco and Wong [ 27 ]). To obtain participants for the convenience sample, assistance was received from local agencies in five communities—Canmore (AB), Whitecourt (AB), Nelson (BC), Quesnell (BC), and Salmon Arm (BC)—who distributed the survey through social media pages, official newsletters, and community events. These communities were selected to create a diverse set of high-risk fire locations with limited egress points and moderate population sizes (between 5,000 and 20,000). Participants were incentivized with the opportunity to win one of ten $75 gift cards to reduce selection and sampling bias. A market research panel was conducted by Qualtrics to increase the sample size. Data cleaning was carried out for both convenience and panel samples to remove incomplete responses, fast responses (<3 min), highly patterned responses, and responses with postal codes outside of AB/BC. The final sample consisted of 2,868 responses (1,497 from the convenience survey and 1,371 from the panel survey).

Demographic Overview

Our survey revealed a relatively varied age distribution among participants with a median age of 40 and an age range of 18 to 90 years. More than half of the respondents in our survey identified as female (54.5%), 44.2% identified as male, and other genders (nonbinary, two-spirit, transgender) comprised 1.7% of the respondents. Survey respondents were predominantly white (78.3%), with visible minorities making up 11.0% of the sample. Additionally, 28.6% of the respondents reported having a seen or unseen disability. With regard to household incomes, 29.4% earned $100,000 or more, 37.2% made between $50,000 and $99,999, and 18.5% reported incomes less than $50,000 (in Canadian dollars). Many respondents (71.0%) had higher education degrees (i.e., diploma/four-year/graduate/doctorate degrees). This aligned with the employment distribution in our survey (70.0% were employed full- or part-time). More than half of the respondents lived in a single-family home (52%) and most owned their residences (74.6%). Many respondents reported owning at least one vehicle (95.5%), with more than half owning two or more personal vehicles (56.0%). Compared with the 2021 Canadian census data (Table 1), our survey oversampled employed individuals and those with higher education levels and undersampled visible minorities. Nonetheless, other demographic variables were relatively consistent with census data and enabled us to obtain valuable insights from the sample. For the full list of survey questions asked for this research, see Appendix Table A1.

Table 1.

Demographic Overview and Comparison with 2021 Census Data

Category	Sample	Canada census 2021 (AB and BC combined) ( 28 )
Median age	40.0	40.8
Visible minorities, %	11.0	30.8
Average household size, n	2.9	2.5
Household income (CAD), %	(in 2022)	(in 2020)
Under $10,000	1.2	1.9
$10,000 to 19,999	3.3	2.5
$20,000 to 29,999	4.1	6.7
$30,000 to 39,999	5.4	6.2
$40,000 to 49,999	4.6	6.8
$50,000 to 59,999	5.9	6.8
$60,000 to 69,999	6.5	6.7
$70,000 to 79,999	7.6	6.4
$80,000 to 89,999	8.0	6.0
$90,000 to 99,999	9.2	5.6
$100,000 and over	29.4	44.3
Gender identity, %
Woman	54.5	50.5
Man	44.2	49.5
Diploma/four-year/graduate/doctorate degrees, %	71.0	48.2
Employed (full-time/part-time), %	70.0	48.2
Home ownership, %	74.6	68.6
People with disabilities, %	28.6	28.1

Note: AB = Alberta; BC = British Columbia; CAD = Canadian dollars.

Underserved Groups and Intersectionality

Our research focused on five groups that have been historically underrepresented in evacuation planning: people with disabilities (has one or more seen or unseen disabilities), older adults (65 years or older), lower-income households (household income below $50,000 in Canadian Dollars [we defined lower-income households as those with an annual income less than $50,000. This threshold aligned with the Canadian Low-Income Cut-Offs (LICOs) before tax, which for a family of three (for geographies with more than 500,000 people) is $44,966 in Canadian dollars ( 29 ). Given that the average household size in our survey was approximately 2.9 people and our income categories were in $10,000 increments, we used <$50,000 as an approximate cutoff to identify lower-income respondents.]), carless residents (without a personal vehicle), and visible minorities (not in a dominant ethnic group and may suffer discrimination based on physical/cultural traits). Table 2 summarizes key evacuation statistics for each group. It should be noted that, although household-level information (e.g., presence of older adults or individuals with disabilities in a household) was also collected, underserved groups in this study were defined based on respondents’ self-reported individual characteristics. This approach was taken to ensure that the analysis reflected the respondents’ identities and evacuation behavior and to avoid assumptions about the decision-making of other household members.

Table 2.

Key Descriptive Statistics—Stated Evacuation Logistics and Choices by Group

Scenario prompt: In this part of the survey, you will answer questions regarding an evacuation. Consider a situation where a wildfire is burning a few kilometers from your residence. The current temperature is high, with strong winds, and low humidity.
Evacuation statistics	People with disabilities (n = 821)	Older adults (n = 524)	Lower-income households (n = 532)	Carless residents (n = 117)	Visible minorities (n = 315)
Percentage of sample (N = 2,868)	28.6	18.3	18.5	4.1	11.0
Evacuation choice	(n = 821)	(n = 524)	(n = 532)	(n = 117)	(n = 315)
Will evacuate, %	99.1	98.3	98.9	99.1	99.4
Will evacuate promptly, %	88.3	86.6	89.1	85.5	88.9
Preparation time	(n = 821)	(n = 524)	(n = 532)	(n = 117)	(n = 315)
<30 min, %	29.2	21.4	38.5	34.2	37.8
30–59 min, %	28.3	26.9	22.9	19.7	22.5
1–1.99 h, %	23.5	31.3	20.5	26.5	18.4
2 or more hours, %	19.0	20.4	18.0	19.7	21.3
Mode choice	(n = 805)	(n = 514)	(n = 521)	(n = 108)	(n = 315)
One personal vehicle (or RV), %	47.0	60.1	55.7	12.0	48.9
Two or more personal vehicles, %	39.3	35.6	27.4	8.3	38.7
Shared mobility (ride-source, carpool), %	4.1	2.5	7.7	34.3	4.1
Public transit (bus, train), %	6.2	0.6	3.6	18.5	5.4
Active modes (walk, bike), %	3.5	1.2	5.6	26.9	2.9
Within community evacuation	(n = 610)	(n = 221)	(n = 400)	(n = 113)	(n = 277)
Yes, %	13.1	6.3	11.8	14.2	18.8
Route choice	(n = 783)	(n = 468)	(n = 492)	(n = 107)	(n = 305)
Highways, %	41.0	59.4	40.7	42.1	32.1
Major roads (including arterial roads), %	15.3	17.1	13.6	10.3	19.0
Local roads, %	24.5	13.0	23.8	22.4	24.9
Rural roads, %	19.2	10.5	22.0	25.2	23.9
Shelter type	(n = 788)	(n = 497)	(n = 504)	(n = 108)	(n = 311)
Friend’s residence, %	17.4	12.1	22.4	21.3	21.5
Family member’s residence, %	42.9	48.9	37.3	35.2	35.0
Second residence, %	5.3	8.0	4.8	0.9	6.1
Public shelter, %	8.9	5.0	9.9	16.7	12.9
Community center, %	6.6	4.0	6.7	11.1	6.4
Short-term rental, %	1.4	0.8	2.0	1.9	2.9
Hotel/motel, %	11.7	11.7	12.3	10.2	13.2
RV, %	5.8	9.5	4.6	2.8	1.9

Note: RV = Recreational Vehicle.

We conducted cross tabulations to explore intersectionality between the underserved groups in our survey. The chi-square test analysis (using p-values) helped identify significant associations between the intersecting groups. These are summarized in Table 3.

Table 3.

Group Associations Using the Chi-Square Test

	Group intersections (percentage of total sample, N = 2,868)
Underserved groups	People with disabilities	Lower-income households	Older adults	Carless residents	Visible minorities
People with disabilities	na	8.3% (↑) (0.000) **	5.3% (0.873)	2.0% (↑) (0.000) **	2.8% (0.201)
Lower-income households	8.3% (↑) (0.000) **	na	3.1% (0.405)	2.6% (↑) (0.000) **	2.9% (↑) (0.000) **
Older adults	5.3% (0.873)	3.1% (0.405)	na	0.4% (↓) (0.016) *	0.6% (↓) (0.000) **
Carless residents	2.0% (↑) (0.000) **	2.6% (↑) (0.000) **	0.4% (↓) (0.016) *	na	0.7% (0.088)
Visible minorities	2.8% (0.201)	2.9% (↑) (0.000) **	0.6% (↓) (0.000) **	0.7% (0.088)	na

Note: (↑) denotes a positive likelihood that those in group A also belong to group B based on cross tabulations.; (↓) denotes a negative likelihood that those in group A also belong to group B based on cross tabulations.

p < .05; **p < .01.

na = not applicable.

Notably, we observed that carless residents were significantly and positively associated with multiple underserved groups (people with disabilities and lower-income households), highlighting potential compounding challenges for those without personal vehicles during wildfire events.

Discrete Choice Modeling

To understand the factors affecting wildfire evacuation behavior among underserved groups, we leveraged discrete choice modeling techniques as outlined by Train ( 30 ) and Ben-Akiva and Lerman ( 31 ). Under this framework, an individual n is assumed to evaluate a set of discrete alternatives and choose the one that provides the highest utility. The utility associated with alternative j is denoted by $U_{nj}$ , which is composed of an observable (systematic) component and a random unobservable component,

U_{nj} = β_{j} X_{nj} + ε_{nj}

(1)

where $X_{nj}$ is vector of observed attributes and $β_{j}$ is vector of parameters capturing their influence on utility. Assuming the unobserved terms $ε_{nj}$ are independently and identically distributed with a Gumbel distribution, the probability that individual n selects alternative j from the set J can be obtained by the multinomial logit model,

P_{nj} = \frac{e^{β_{j} X_{nj}}}{\sum_{k ϵ J} e^{β_{k} X_{nk}}}

(2)

The above formulation can be used for two or more alternative scenarios with only two alternatives, this simplifies to the binary logit model, where the probability of choosing an alternative i becomes

P_{n} = \frac{1}{1 + e^{- β X_{n}}}

(3)

We developed binary logit models for two choices: 1) evacuate promptly or not and 2) evacuate within the community of origin or to a location outside the community. We additionally developed multinomial logit models for decisions involving multiple alternatives. These included departure time, mode choice, route choice, and shelter type. All models were developed using the Python package Biogeme 3.2.13 ( 32 ). The models incorporated variables related to sociodemographic characteristics, risk perception, travel behavior, and previous hazard/evacuation experiences. Only variables relatively independent of each other (correlation coefficients under 0.3) were included in the models. We retained variables that were statistically significant (at the 95% level), were behaviorally important, and had intuitive coefficient signs. In some models, we retained insignificant variables if they were behaviorally relevant, met a priori expectations, and decreased model bias, consistent with previous research (see, for example, Ben-Akiva and Lerman [ 31 ] or McFadden [ 33 ]). We opted for simpler models (binary and multinomial) as these were behaviorally clear and easy to interpret. Future research could conduct analyses using other discrete choice model types and further consider other historically underserved groups during evacuations (e.g., those with limited English proficiency).

Results

In this section, we present results from binary and multinomial logit models that were developed to understand factors related to: 1) evacuating promptly, 2) destination choice, 3) departure timing, 4) mode choice, 5) route choice, and 6) shelter type. Models of the same underserved group may have different sample sizes depending on response rates. It should be noted that for categorical variables included in the models (e.g., age, income, gender), one category was selected as the reference (base) category. For example, where “young age” and “low income” were included in a model, the omitted categories (representing all other age or income groups, respectively) served as reference categories. In all tables, “NS” represents non-significant variables.

Evacuate Promptly

Although most previous research has modeled the choice to evacuate/stay during an evacuation ( 14 , 15 ), our research opted to examine the binary decision to evacuate promptly or not, owing to the high number of respondents who would evacuate (>96%). In our survey, the question on prompt evacuation did not provide a specific time frame, allowing participants to interpret promptness based on their own understanding. A subsequent question asked respondents to indicate their actual intended time of departure (see the section covering departure [preparation] time). In this model, the decision not to evacuate promptly is used as the base choice (reference category). As such, a positive coefficient indicates that the variable increases the likelihood of evacuating promptly, whereas a negative coefficient denotes a decrease in the likelihood of evacuating promptly. Results from the binary logit model are presented in Table 4.

Table 4.

Binary Logit Model—Evacuate Promptly

Choice 1: Will evacuate promptlyChoice 2: Will not evacuate promptly: base
Variables	People with disabilities		Older adults		Lower-income households		Carless residents		Visible minorities
Variables	Est coeff	p-value	Est coeff	p-value	Est coeff	p-value	Est coeff	p-value	Est coeff	p-value
Constant	2.154	0.000 **	1.271	0.000 **	2.734	0.000 **	2.043	0.002 **	2.333	0.002 **
Household characteristics
Household with at least one child	NS	NS	NS	NS	NS	NS	−1.992	0.002 **	NS	NS
Individual characteristics
Young adult (18–35 years)	0.682	0.006 **	NS	NS	NS	NS	1.007	0.080	NS	NS
Woman	NS	NS	0.964	0.001 **	0.562	0.050 *	NS	NS	NS	NS
Visible minority	NS	NS	NS	NS	NS	NS	1.787	0.025 *	NS	NS
Risk perception
Willing to take risks¹	NS	NS	NS	NS	NS	NS	−1.587	0.016 *	NS	NS
Tries to reduce risk to family^a	NS	NS	NS	NS	−1.157	0.036 *	NS	NS	−1.470	0.044 *
Residence is likely to be threatened by wildfires next season^b	0.529	0.034 *	NS	NS	NS	NS	NS	NS	0.972	0.010 **
Staying increases odds of harm^c	NS	NS	NS	NS	0.605	0.047 *	NS	NS	0.744	0.045 *
Evacuating increases odds of losing home^c	NS	NS	NS	NS	NS	NS	1.556	0.037 *	NS	NS
Evacuation experiences/decisions
Experienced discomfort from wildfire smoke	−0.691	0.009 **	NS	NS	−0.891	0.009 **	NS	NS	NS	NS
Previously learned about wildfires	NS	NS	0.568	0.041 *	0.643	0.036 *	NS	NS	NS	NS
Job was affected by wildfires	−0.645	0.009 **	NS	NS	NS	NS	NS	NS	NS	NS
Number of observations	821		524		532		117		315
ρ2 (fit)	0.51		0.46		0.53		0.53		0.53
ρ2 (adjusted fit)	0.50		0.45		0.52		0.45		0.51
Initial log-likelihood	−569.07		−363.21		−368.75		−81.10		−218.34
Final log-likelihood	−278.34		−196.76		−171.75		−38.37		−103.33

Note: Est coeff = estimated coefficient; NS = not significant.

Respondents rated the statement as very or somewhat true.

The possibility of residence being threatened by a wildfire next season was rated as very or somewhat likely.

Respondents indicated that the action (evacuate/stay) would greatly or somewhat increase the odds of the stated outcome.

p < .05; **p < .01.

Overall, we observed that those who had previously experienced discomfort from wildfire smoke, from both the individuals with disabilities and lower-income groups, were less likely to evacuate promptly. Moreover, those with disabilities whose jobs were previously affected by wildfires were less likely to evacuate promptly. Previous experiences with wildfires could lead to a reduced perception of risk and a delayed evacuation response. Similar tendencies have been observed in hurricane evacuations. Research by Wong et al. on hurricane evacuee behavior found that experience with at least one prior hurricane and being a previous evacuee decreased the likelihood of evacuation ( 14 ). Conversely, we found that older adults and low-income households who had previously learned about wildfires were more likely to evacuate promptly. Learning about wildfire risk and appropriate evacuation actions may enhance risk awareness, contributing to faster evacuations. We additionally observed that lower-income earners and visible minorities who perceived that staying would increase their odds of being harmed were more likely to evacuate promptly. The results suggest that heightened risk perception influences quicker evacuations. Interestingly, low-income households and visible minorities who would try to reduce the risk to their families would take longer to evacuate. This could reflect the additional time needed to organize, care for, or secure dependents and family members before evacuating. The demographic variables revealed interesting insights into intersecting vulnerabilities. For example, we found that older women and lower-income women were more likely to evacuate promptly. We also observed that carless residents with at least one child were less likely to evacuate promptly. Evacuating with children without access to a personal vehicle could present logistical challenges, particularly if public transportation is not reliable. This may in turn lead to delays in responding to evacuation orders.

Destination Choice (Within Community)

Previous literature has shown that wildfire evacuees tend to travel shorter distances to reach their destinations ( 15 , 34 ). As such, we sought to examine the factors affecting different groups’ decisions to evacuate to a location within their community as a binary logit model. The decision to evacuate to a location outside an individual’s home community was used as the base choice. The results are presented in Table 5.

Table 5.

Binary Logit Model—Destination Choice

Choice 1: Destination location is within home communityChoice 2: Destination location is outside of home community: base
Variables	People with disabilities		Lower-income households		Carless residents		Visible minorities
Variables	Est coeff	p-value	Est coeff	p-value	Est coeff	p-value	Est coeff	p-value
Constant	−2.400	0.000 **	−1.148	0.000 **	−2.758	0.000 **	−1.656	0.000 **
Individual characteristics
Young adult (18–35 years)	0.729	0.004 **	NS	NS	1.287	0.048 *	NS	NS
Has a disability	NS	NS	NS	NS	1.321	0.025 *	−1.349	0.008 **
Higher income (≥ $100,000 CAD)^a	NS	NS	NS	NS	NS	NS	0.737	0.042 *
Is part of a community group	NS	NS	−0.905	0.019 *	NS	NS	NS	NS
Risk perception
Tries to reduce risk to family	1.479	0.007 **	NS	NS	NS	NS	NS	NS
Staying increases odds of harm	−0.571	0.044 *	−0.666	0.034 *	−1.456	0.018 *	NS	NS
Evacuating increases odds of losing home	NS	NS	NS	NS	1.223	0.052	NS	NS
Evacuation experiences/decisions
Has previous hazard experience	−0.916	0.001 **	NS	NS	NS	NS	NS	NS
Experienced wildfire property damage	0.968	0.013 *	NS	NS	NS	NS	NS	NS
Previously learned about wildfires	−0.652	0.018 *	NS	NS	NS	NS	NS	NS
Job was affected by wildfires	0.783	0.007 **	NS	NS	NS	NS	NS	NS
Will evacuate after a voluntary order	NS	NS	−0.763	0.036 *	NS	NS	−1.009	0.005 **
Will evacuate after seeing neighbors evacuate	NS	NS	NS	NS	NS	NS	1.099	0.002 **
Primary evacuation mode: public transit	−2.593	0.021 *	NS	NS	−1.763	0.045 *	NS	NS
Primary evacuation mode: shared mobility	NS	NS	NS	NS	NS	NS	1.730	0.007 **
Will tow items behind their vehicle	−0.797	0.030 *	NS	NS	NS	NS	NS	NS
Number of observations	619		408		93		263
ρ2 (fit)	0.51		0.51		0.45		0.39
ρ2 (adjusted fit)	0.49		0.50		0.35		0.35
Initial log-likelihood	−429.06		−282.80		−64.46		−182.30
Final log-likelihood	−209.29		−135.50		−35.68		−110.59

Note: Est coeff = estimated coefficient; CAD = Canadian dollars; NS = not significant.

Some respondents preferred not to answer the income question (>5.3%); this group is included in the base category along with respondents from other income levels.

p < .05; **p < .01.

Among the risk variables, we found that people with disabilities, lower-income households, and carless residents who perceived that staying would increase their odds of being harmed were more likely to evacuate to locations outside their communities. This may show that a perception of risk associated with staying leads to longer-distance evacuations. Moreover, we observed that people with disabilities who would try to reduce the risk to their families were more likely to evacuate to locations within their communities of origin. This may be because of the proximity to more familiar and accessible facilities or established support networks within the community. Among evacuation experiences, individuals with disabilities who experienced wildfire property damage were more likely to evacuate to locations within their communities, possibly because of a desire to stay near their homes and protect their properties from future damage. In contrast, those with disabilities who previously learned about wildfires were more likely to evacuate to locations outside their communities, suggesting that prior wildfire education may increase risk awareness leading to long-distance evacuations. We further observed that people with disabilities and carless residents who would evacuate using public transit were more likely to evacuate to locations outside their communities, possibly reflecting a reliance on intercity routes during a wildfire. Among the demographic variables, the results showed that young adults from both the people with disabilities and carless residents groups were more likely to evacuate to locations within their communities. Younger adults from these groups may face mobility or resource constraints that would influence a preference for shorter evacuations. Interestingly, we observed that although carless residents with a disability would evacuate within their communities, visible minorities with a disability were more likely to evacuate outside the community. This may point to mobility challenges among those with disabilities who lack access to a personal vehicle. Destination choices are likely to go hand in hand with shelter preferences and availability and could be modeled through a joint framework in future research. No variables were significant in determining destination choice for the older adults group.

Departure (Preparation) Time

We analyzed evacuation preparation time as a multinomial logit model. Table 6 is sectioned by underserved group, enabling a comparison across time intervals within each group. Less than 30 minutes was used as the reference category.

Table 6.

Multinomial Logit Model—Preparation Time

Choice 1: <30 min: baseChoice 2: ≥ 30 min, < 1 hChoice 3: ≥ 1 h, < 2 hChoice 4: ≥2 h
Variables	≥ 30 min < 1 h		≥ 1 h < 2 h		≥2 h
Variables	Est Coeff	p-value	Est Coeff	p-value	Est Coeff	p-value
People with disabilities (n = 821)
Constant	−0.179	0.192	−1.115	0.008 **	−1.049	0.000 **
Individual characteristics
Visible minority	−0.936	0.004 **	−0.914	0.014 *	NS	NS
Carless resident	NS	NS	0.756	0.020 *	NS	NS
Lower income (<$50,000 CAD)	−0.496	0.012 *	−0.662	0.002 **	NS	NS
Risk perception
Willing to take risks	NS	NS	−0.463	0.008 **	NS	NS
Tries to reduce risk to family	NS	NS	1.254	0.001 **	NS	NS
Residence will be threatened by wildfires next season	NS	NS	NS	NS	−0.547	0.009 **
Evacuation experiences/decisions
Has previous hazard experience	NS	NS	−0.456	0.022 *	NS	NS
Experienced wildfire property damage	NS	NS	−1.151	0.004 *	−1.172	0.003 **
Experienced discomfort from wildfire smoke	NS	NS	0.624	0.002 *	1.235	0.000 **
Experienced wildfire injury	0.930	0.001 **	NS	NS	NS	NS
Will evacuate after can no longer defend home	0.604	0.003 **	0.698	0.001 **	0.795	0.000 **
Will evacuate after household is ready	NS	NS	NS	NS	0.452	0.021 *
Will evacuate after seeing neighbors evacuate	NS	NS	NS	NS	−0.729	0.000 **
Primary evacuation mode: public transit	1.205	0.000 **	NS	NS	1.185	0.022 *
ρ2 (fit)	0.10
ρ2 (adjusted fit)	0.08
Initial log-likelihood	−1,138.15
Final log-likelihood	−1,022.08
Older adults (n = 524)
Constant	0.671	0.019 *	0.609	0.081	0.122	0.710
Individual characteristics
Lower income (<$50,000 CAD)	−0.710	0.028 *	−0.671	0.035 *	−0.795	0.026 *
Has lived in residence more than 10 years	NS	NS	NS	NS	0.510	0.027 *
Risk perception
Residence will be threatened by wildfires next season	−0.626	0.040 *	−0.878	0.003 **	−0.723	0.026 *
Staying increases odds of harm	NS	NS	0.500	0.047 *	0.589	0.050 *
Evacuation experiences/decisions
Attended a public meeting on wildfires	NS	NS	−0.795	0.004 **	NS	NS
Will evacuate after can no longer defend home	NS	NS	0.619	0.003 **	0.535	0.027 *
Will evacuate after seeing neighbors evacuate	0.489	0.019 *	NS	NS	NS	NS
ρ2 (fit)	0.05
ρ2 (adjusted fit)	0.03
Initial log-likelihood	−726.42
Final log-likelihood	−692.95
Lower-income households (n = 532)
Constant	−0.799	0.000 **	−0.604	0.019 *	−2.140	0.000 **
Household characteristics
Household has at least one child	NS	NS	−1.094	0.000 **	NS	NS
Individual characteristics
Young adult (18–35 years)	NS	NS	−0.795	0.001 **	NS	NS
Woman	NS	NS	0.863	0.001 **	NS	NS
Has disability	NS	NS	NS	NS	0.695	0.004 **
Visible minority	NS	NS	−1.094	0.010 *	NS	NS
Risk perception
Staying increases odds of losing home	NS	NS	NS	NS	−0.632	0.022 *
Evacuation experiences/decisions
Experienced discomfort from wildfire smoke	0.589	0.007 **	NS	NS	1.032	0.000 **
Saw wildfire flames	NS	NS	NS	NS	0.647	0.011 *
Previously learned about wildfires	NS	NS	0.521	0.031 *	0.608	0.016 *
Primary evacuation mode: public transit	NS	NS	NS	NS	1.256	0.009 **
ρ2 (fit)	0.11
ρ2 (adjusted fit)	0.09
Initial log-likelihood	−737.51
Final log-likelihood	−655.49
Carless residents (n = 117)
Constant	−0.534	0.170	−1.126	0.038 *	−2.030	0.001 **
Individual characteristics
Has disability	0.754	0.166	0.976	0.066	1.220	0.032 *
Risk perception
Staying increases odds of harm	NS	NS	1.217	0.018 *	1.261	0.027 *
Evacuation experiences/decisions
Primary evacuation mode: personal vehicle	−1.765	0.025 *	−2.006	0.014 *	−0.854	0.230
Primary evacuation mode: public transit	NS	NS	NS	NS	1.320	0.023 *
ρ2 (fit)	0.12
ρ2 (adjusted fit)	0.04
Initial log-likelihood	−162.20
Final log-likelihood	−143.18
Visible minorities (n = 315)
Constant	−0.490	0.003 *	−0.875	0.000 **	−1.455	0.000 **
Individual characteristics
Older adults (≥65 years)	1.147	0.032 *	NS	NS	NS	NS
Lower income (<$50,000 CAD)	NS	NS	−1.119	0.009 **	NS	NS
Evacuation experiences/decisions
Experienced wildfire property damage	−0.871	0.045 *	NS	NS	NS	NS
Saw wildfire flames	NS	NS	NS	NS	0.599	0.037 *
Previously learned about wildfires	NS	NS	NS	NS	0.865	0.003 **
Will evacuate after can no longer defend home	NS	NS	0.971	0.002 **	0.656	0.029 *
ρ2 (fit)	0.08
ρ2 (adjusted fit)	0.05
Initial log-likelihood	−436.68
Final log-likelihood	−402.70

Note: Est coeff = estimated coefficient; CAD = Canadian dollars; NS = not significant; RV = Recreational Vehicle.

p < .05; **p < .01.

Among the people with disabilities’ group, we observed that those who had previously experienced wildfire smoke were likely to take more than an hour to evacuate. Conversely, older adults who previously attended public meetings on wildfires were more likely to take less than 30 min preparing to evacuate. This is possibly because of a better understanding of evacuation protocols and preparedness procedures facilitated by the public meetings. Furthermore, we observed that lower-income visible minorities were likely to take less than 30 min preparing to leave their residences during an evacuation. Moreover, lower-income individuals who believed that staying would increase the odds of losing their homes were more likely to prepare to evacuate in less than 30 min. This finding suggests that these individuals may prioritize personal safety over defending their property, especially if they have a low confidence in reducing the risk to their property. Interestingly, among the carless residents group, only those who selected a personal vehicle as their intended evacuation mode were likely to evacuate in less than 30 min. It is likely that some respondents interpreted the term “personal vehicle” more broadly to include non-owned vehicles used for personal purposes. This may suggest that some respondents anticipate borrowing/renting a vehicle during an evacuation or expect to have temporary access to a vehicle to improve mobility or reduce evacuation delays during a wildfire. Finally, among the visible minorities group, we found that those who had previously experienced wildfire property damage were more likely to evacuate in less than 30 minutes, possibly owing to a heightened risk perception and an awareness of potential wildfire dangers.

Mode Choice

We divided evacuation mode choices into four categories: 1) shared modes, 2) personal vehicles, 3) public transit, and 4) active modes. Evacuating with shared modes was used as the base choice for this model (Table 7).

Table 7.

Multinomial Logit Model—Mode Choice

Choice 1: Shared mobility (carpooling, ride-sourcing): baseChoice 2: Personal vehicle (one or more vehicles, RV)Choice 3: Public transit (bus, rail)Choice 4: Active modes (walking, biking)
Variables	Personal vehicle		Public transit		Active modes
Variables	Est Coeff	p-value	Est Coeff	p-value	Est Coeff	p-value
People with disabilities (n = 805)
Constant	1.958	0.000 **	−2.422	0.001 **	−0.042	0.919
Individual characteristics
Woman	NS	NS	−1.206	0.002 **	NS	NS
Young adult (18–35 years)	−0.821	0.003 **	NS	NS	−1.323	0.005 **
Visible minority	NS	NS	1.062	0.024 *	NS	NS
Higher income (≥$100,000 CAD)	1.808	0.000 **	NS	NS	NS	NS
Has lived in residence more than 10 years	1.091	0.002 **	NS	NS	NS	NS
Risk perception
Willing to take risks	−0.844	0.001 **	NS	NS	NS	NS
Tries to reduce risk to family	0.964	0.001 **	NS	NS	NS	NS
Residence will be threatened by wildfires next season	1.014	0.001 **	1.467	0.012 *	1.637	0.005
Staying increases likelihood of losing home	2.178	0.000 **	2.166	0.000 **	1.331	0.034 *
Evacuation experiences/decisions
Previously learned about wildfires	NS	NS	−1.076	0.008 *	NS	NS
Prepared for an evacuation	0.687	0.012 *	NS	NS	NS	NS
Will evacuate to a destination within origin community	NS	NS	NS	NS	1.635	0.000 **
Will evacuate after a mandatory order	NS	NS	1.274	0.005 **	NS	NS
Will evacuate after a voluntary order	NS	NS	0.811	0.012 *	NS	NS
Will evacuate after seeing neighbors evacuate	−0.596	0.008 **	NS	NS	NS	NS
ρ2 (fit)	0.70
ρ2 (adjusted fit)	0.68
Initial log-likelihood	−1,115.97
Final log-likelihood	−334.66
Older adults (n = 514)
Constant	1.644	0.016 *	−1.466	0.022 *	−0.956	0.069
Household characteristics
Household has at least one child	−1.192	0.012 *	NS	NS	NS	NS
Individual characteristics
Visible minority	−1.753	0.008 **	NS	NS	NS	NS
Has disability	−1.007	0.029 *	NS	NS	NS	NS
Owns residence	1.532	0.002 **	NS	NS	NS	NS
Risk perception
Tries to reduce risk to family	1.890	0.000 **	NS	NS	NS	NS
Evacuation experiences/decisions
Will evacuate to a destination within origin community	11.380	0.000 **	NS	NS	13.961	0.000 **
ρ2 (fit)	0.87
ρ2 (adjusted fit)	0.86
Initial log-likelihood	−712.5
Final log-likelihood	−92.16
Lower-income households (n = 521)
Constant	1.113	0.001 **	−1.031	0.139	−1.499	0.001 **
Individual characteristics
Has disability	NS	NS	1.111	0.032 *	1.490	0.001 **
Visible minority	NS	NS	1.778	0.001 **	NS	NS
Has lived in residence more than 10 years	1.282	0.002 **	NS	NS	NS	NS
Risk perception
Willing take risks	−0.567	0.028 *	NS	NS	NS	NS
Tries to reduce risk to family	0.900	0.006 **	−1.208	0.039 *	NS	NS
Staying increases likelihood of losing home	0.757	0.038 *	NS	NS	0.939	0.061
Evacuation experiences/decisions
Saw wildfire flames	0.539	0.043 *	NS	NS	NS	NS
Will take <30 min to prepare for an evacuation	NS	NS	−1.556	0.020 *	NS	NS
Will evacuate after can no longer defend home	NS	NS	0.993	0.039 *	NS	NS
Will tow items behind their vehicle	1.164	0.001 **	NS	NS	NS	NS
ρ2 (fit)	0.60
ρ2 (adjusted fit)	0.58
Initial log-likelihood	−722.26
Final log-likelihood	−288.35
Carless residents (n = 108)
Constant	−1.920	0.000 **	−14.107	0.000 **	0.425	0.209
Individual characteristics
Young adult (18–35 years)	NS	NS	NS	NS	−2.309	0.001 *
Owns their residence	1.615	0.011 *
Evacuation experiences/decisions
Experienced wildfire property damage	2.273	0.032	NS	NS	NS	NS
Will evacuate promptly	NS	NS	13.65	0.000 **	NS	NS
Will take <30 min to prepare for an evacuation	1.348	0.017 *	NS	NS	NS	NS
Will evacuate to a destination within origin community	NS	NS	NS	NS	2.289	0.001 **
ρ2 (fit)	0.21
ρ2 (adjusted fit)	0.15
Initial log-likelihood	−149.72
Final log-likelihood	−118.17
Visible minorities (n = 315)
Constant	2.707	0.000 **	0.268	0.467	−1.515	0.049 *
Individual characteristics
Has disability	−0.893	0.017 *	NS	NS	NS	NS
Evacuation experiences/decisions
Saw wildfire flames	1.141	0.013 *	NS	NS	NS	NS
Previously learned about wildfires	−0.939	0.018 *	NS	NS	NS	NS
Job is related to wildfires	−1.039	0.048 *	NS	NS	NS	NS
Will evacuate after a mandatory order	0.993	0.012 *	NS	NS	NS	NS
Will evacuate after seeing neighbors evacuate	0.694	0.120	NS	NS	2.100	0.019 *
ρ2 (fit)	0.68
ρ2 (adjusted fit)	0.65
Initial log-likelihood	−436.68
Final log-likelihood	−141.92

Note: Est coeff = estimated coefficient; CAD = Canadian dollars; NS = not significant.

p < .05; **p < .01.

Among the individuals with disabilities, we observed that those who try to reduce the risk to their families were more likely to use a personal vehicle to evacuate a wildfire. On the other hand, we found that people with disabilities who were also young adults were more likely to use shared modes or public transit during a wildfire evacuation. Furthermore, we observed that individuals with disabilities who would evacuate to locations within their home communities were more likely to use active modes. Those without access to personal vehicles or those unable to drive because of mobility limitations may rely on walking or mobility devices to complete shorter evacuation trips. Within the older adults group, those who had disabilities and those who were visible minorities were both less likely to use personal vehicles during an evacuation. Among lower-income households, we observed that visible minorities were more likely to evacuate with transit over other modes, perhaps owing to financial constraints. Interestingly, carless residents who would take less than 30 minutes preparing for an evacuation were more likely to use a personal vehicle over other modes. Finally, among the visible minorities group, those whose jobs were related to wildfires would use shared mobility, public transit, or active modes compared with personal vehicles.

Route Choice

Similar to Wong et al., we classified route choices into four: major roads, local roads, highways, and rural roads ( 14 ). We then asked respondents to rank road types based on how much time they would spend on each during an evacuation. Discrete choices were formed by considering road types with the highest rank (most amount of time). Table 8 presents the multinomial logit model results, with major roads as the base choice.

Table 8.

Multinomial Logit Model—Route Choice

Choice 1: Major (arterial) roads: baseChoice 2: Local roadsChoice 3: HighwaysChoice 4: Rural roads
Variables	Local roads		Highways		Rural roads
Variables	Est Coeff	p-value	Est Coeff	p-value	Est Coeff	p-value
People with disabilities (n = 783)
Constant	0.355	0.056	0.625	0.000 **	0.081	0.690
Household characteristics
5+ people in the household	0.790	0.044 *	NS	NS	NS	NS
Individual characteristics
Young adult (18–35 years)	0.672	0.000 **	NS	NS	0.684	0.001 **
Higher level degree (graduate/professional/doctorate)	NS	NS	0.570	0.014 *	NS	NS
High income (≥$100,000 CAD)	−0.649	0.011 *	NS	NS	NS	NS
Has lived in residence more than 10 years	NS	NS	NS	NS	−0.472	0.047 *
Risk perception
Staying increases likelihood of losing home	0.374	0.033 *	NS	NS	NS	NS
Evacuation experiences/decisions
Experienced wildfire evacuation	0.465	0.037 *	NS	NS	0.725	0.001 **
Experienced wildfire injury	NS	NS	−1.189	0.001 **	NS	NS
Prepared for an evacuation	−0.394	0.052	NS	NS	NS	NS
Will evacuate after a mandatory order	NS	NS	0.474	0.006 **	NS	NS
Will evacuate after a voluntary order	NS	NS	NS	NS	−0.568	0.003 **
Will evacuate after seeing lower traffic congestion	−0.720	0.006 **	NS	NS	NS	NS
Will evacuate after seeing household is ready	−0.366	0.044 *	NS	NS	NS	NS
ρ2 (fit)	0.11
ρ2 (adjusted fit)	0.10
Initial log-likelihood	−1,085.47
Final log-likelihood	−963.84
Older adults (n = 468)
Constant	0.193	0.548	1.227	0.000 **	0.645	0.107
Individual characteristics
Has disability	0.657	0.035 *	NS	NS	NS	NS
Employed full-time	NS	NS	−0.651	0.028 *	NS	NS
High income (≥$100,000 CAD)	NS	NS	0.536	0.024 *	NS	NS
Risk perception
Tries to reduce risk to family	NS	NS	NS	NS	−1.565	0.000 **
Staying increases odds of losing home	NS	NS	NS	NS	0.734	0.026 *
Evacuation experiences/decisions
Experienced wildfire evacuation	0.850	0.039 *	NS	NS	NS	NS
Experienced discomfort from wildfire smoke	−0.875	0.006 **	NS	NS	NS	NS
Will take <30 min to prepare for an evacuation	0.729	0.029 *	NS	NS	NS	NS
Will evacuate after a voluntary order	−0.837	0.005 **	NS	NS	NS	NS
ρ2 (fit)	0.25
ρ2 (adjusted fit)	0.23
Initial log-likelihood	−648.79
Final log-likelihood	−485.05
Lower-income households (n = 492)
Constant	−0.729	0.091	0.432	0.261	0.401	0.152
Individual characteristics
Has disability	0.641	0.004 **	NS	NS	NS	NS
Young adult (18–35 years)	0.590	0.016 *	NS	NS	0.862	0.001 **
Older adults (65+ years)	NS	NS	0.591	0.031 *	NS	NS
Has lived in residence 10+ years	−0.640	0.048 *	NS	NS	NS	NS
Risk perception
Tries to reduce risk to family	0.903	0.009 **	NS	NS	NS	NS
Evacuation experiences/decisions
Previously learned about wildfires	NS	NS	0.456	0.020 *	NS	NS
Job is related to wildfires	NS	NS	−1.478	0.005 **	NS	NS
Will evacuate after a voluntary order	−0.794	0.012 *	−0.690	0.018 *	−0.594	0.064
Will evacuate promptly	NS	NS	0.831	0.022 *	NS	NS
Will take <30 min to prepare for an evacuation	0.569	0.011 *	NS	NS	NS	NS
ρ2 (fit)	0.11
ρ2 (adjusted fit)	0.09
Initial log-likelihood	−682.06
Final log-likelihood	−606.36
Carless residents (n = 107)
Constant	0.429	0.301	0.183	0.739	0.905	0.033 *
Risk perception
Staying increases odds of losing home	1.243	0.030 *	NS	NS	1.248	0.030 *
Evacuation experiences/decisions
Will evacuate after a mandatory order	NS	NS	1.361	0.009 **	NS	NS
Will evacuate after cannot defend home	NS	NS	NS	NS	−1.084	0.032 *
Primary evacuation mode: public transit	NS	NS	1.214	0.040 *	NS	NS
ρ2 (fit)	0.15
ρ2 (adjusted fit)	0.10
Initial log-likelihood	−148.33
Final log-likelihood	−125.83
Visible minorities (n = 305)
Constant	−0.122	0.734	−0.999	0.072	−0.446	0.197
Household characteristics
5+ people in the household	NS	NS	NS	NS	1.017	0.012 *
Individual characteristics
Has disability	NS	NS	NS	NS	−0.693	0.058
Young adult (18–35 years)	0.573	0.057	NS	NS	NS	NS
Risk perception
Willing to take risks	−0.777	0.006 **	NS	NS	NS	NS
Residence will be threatened by wildfires next season	NS	NS	NS	NS	0.681	0.044 *
Staying increases odds of harm	0.715	0.017 *	NS	NS	NS	NS
Evacuation experiences/decisions
Experienced wildfire evacuation	−0.992	0.013 *	NS	NS	NS	NS
Experienced discomfort from wildfire smoke	NS	NS	0.742	0.005 **	NS	NS
Experienced wildfire injury	NS	NS	−1.424	0.012 *	NS	NS
Job is related to wildfires	1.245	0.006 **	NS	NS	NS	NS
Will evacuate promptly	NS	NS	1.446	0.009 **	NS	NS
Will evacuate after packing	NS	NS	NS	NS	0.550	0.053
ρ2 (fit)	0.08
ρ2 (adjusted fit)	0.05
Initial log-likelihood	−422.82
Final log-likelihood	−387.47

Note: Est coeff = estimated coefficient; CAD = Canadian dollars; NS = not significant.

p < .05; **p < .01.

We observed that people with disabilities who had previously experienced a wildfire evacuation were more likely to use local or rural roads compared with major roads or highways. Past experiences with evacuation congestion may lead them to opt for less popular routes. Interestingly, within the older adults group, those who tried to reduce the risk to their families were more likely to use major roads during a wildfire evacuation. Those who are risk-averse may choose major roads owing to their higher speed limits and easier connections to destinations outside their community. Lower-income households under the age of 35 were more likely to use local or rural roads during an evacuation. Among the carless residents, we noted that those who would evacuate using transit were more likely to opt for highways. The selection of highways among transit users may reflect an assumption that transit services would use major highway corridors during an evacuation. Finally, among the visible minorities group, we found that prompt evacuees were likely to use highways.

Shelter Choice

We sought to understand the factors affecting shelter choice through a multinomial logit model, using a family/friend’s residence as the base choice (Table 9).

Table 9.

Multinomial Logit Model—Shelter Type

Choice 1: Family/friend’s residence: baseChoice 2: Government-provided (public shelters, community centers)Choice 3: Hotel/motel/AirbnbChoice 4: Second residence/mobile home
Variables	Government shelter		Hotel/motel/Airbnb		Second residence/ mobile home
Variables	Est Coeff	p-value	Est Coeff	p-value	Est Coeff	p-value
People with disabilities (n = 788)
Constant	−0.586	0.019 *	−1.099	0.000 **	−1.997	0.000 **
Individual characteristics
Older adults (65+ years)	NS	NS	NS	NS	0.549	0.040 *
Higher level degree (graduate/professional/doctorate)	0.753	0.006 **	NS	NS	NS	NS
High income (≥$100,000 CAD)	NS	NS	NS	NS	0.577	0.022 *
Owns residence	NS	NS	NS	NS	0.661	0.023 *
Risk perception
Staying increases likelihood of losing home	0.477	0.030 *	NS	NS	NS	NS
Staying increases odds of harm	−0.553	0.011 *	NS	NS	NS	NS
Residence will be threatened by wildfires next season	NS	NS	NS	NS	−0.601	0.018 *
Evacuation experiences/decisions
Experienced wildfire smoke	−0.685	0.002 **	NS	NS	NS	NS
Previously learned about wildfires	−0.696	0.003 **	NS	NS	NS	NS
Will evacuate after a voluntary order	NS	NS	−0.534	0.014 *	NS	NS
Will tow items behind their vehicle	−0.476	0.040 *	−0.724	0.006 **	NS	NS
Primary evacuation mode: shared mobility	1.055	0.013 *	NS	NS	NS	NS
ρ2 (fit)	0.24
ρ2 (adjusted fit)	0.23
Initial log-likelihood	−1,092.40
Final log-likelihood	−826.84
Older adults (n = 497)
Constant	−1.905	0.000 **	−0.890	0.000 **	−1.193	0.000 **
Individual characteristics
Woman	NS	NS	NS	NS	−0.714	0.004 **
Has disability	0.877	0.013 *	NS	NS	NS	NS
Higher income (≥$100,000 CAD)	−1.508	0.018 *	NS	NS	NS	NS
Higher level degree (graduate/professional/doctorate)	NS	NS	−0.889	0.015 *	NS	NS
Employed full-time	1.287	0.001 **	NS	NS	NS	NS
Lived in residence 10+ years	NS	NS	NS	NS	0.476	0.050 *
Risk perception
Evacuating increases odds of losing home	0.795	0.023 *	NS	NS	NS	NS
Evacuation experiences/decisions
Will evacuate after a voluntary order	−1.070	0.003 **	−0.584	0.035 *	NS	NS
Will evacuate when can no longer defend home	−0.896	0.016 *	NS	NS	NS	NS
Will tow items behind their vehicle	NS	NS	−1.159	0.017 *	NS	NS
Primary evacuation mode: shared mobility	2.054	0.001 **	NS	NS	NS	NS
ρ2 (fit)	0.28
ρ2 (adjusted fit)	0.26
Initial log-likelihood	−689.00
Final log-likelihood	−493.51
Lower-income households (n = 492)
Constant	−0.790	0.001 **	−1.464	0.000 **	−0.326	0.486
Individual characteristics
Employed full-time	NS	NS	NS	NS	1.059	0.001 **
Evacuation experiences/decisions
Will evacuate after a mandatory order	NS	NS	NS	NS	−0.932	0.005 **
Will evacuate after a voluntary order	NS	NS	−0.674	0.016 *	NS	NS
Will evacuate promptly	NS	NS	NS	NS	−1.426	0.001 **
Will take <30 min to prepare for an evacuation	0.511	0.037 *	NS	NS	NS	NS
Will evacuate after can no longer defend home	NS	NS	0.643	0.016 *	−1.002	0.015 *
Primary evacuation mode: personal vehicle	−0.917	0.001 **	NS	NS	NS	NS
ρ2 (fit)	0.25
ρ2 (adjusted fit)	0.23
Initial log-likelihood	−698.70
Final log-likelihood	−524.88
Carless residents (n = 107)
Constant	0.538	0.358	−0.214	0.716	−1.307	0.094
Household characteristics
Household has at least one child	1.212	0.047 *	NS	NS	NS	NS
Individual characteristics
Woman	−1.160	0.021 *	NS	NS	NS	NS
Young adults (18–35 years)	−0.947	0.062	NS	NS	NS	NS
Risk perception
Staying increases odds of losing home	NS	NS	−2.640	0.006 **	NS	NS
Residence will be threatened by wildfires next season	1.078	0.039 *	NS	NS	NS	NS
Evacuation experiences/decisions
Experienced wildfire evacuation	−1.471	0.033 *	NS	NS	NS	NS
Saw wildfire flames	−1.380	0.010 *	NS	NS	NS	NS
Previously learned about wildfires	NS	NS	−1.459	0.054	NS	NS
Will evacuate after a mandatory order	NS	NS	−1.445	0.027 *	NS	NS
Will evacuate after a voluntary order	−1.311	0.034 *	NS	NS	NS	NS
Will evacuate promptly	NS	NS	NS	NS	−1.982	0.060
Will evacuate after household is ready	NS	NS	1.565	0.024 *	NS	NS
ρ2 (fit)	0.38
ρ2 (adjusted fit)	0.28
Initial log-likelihood	−149.72
Final log-likelihood	−93.53
Visible minorities (n = 305)
Constant	−0.546	0.166	−2.228	0.000 **	−2.413	0.000 **
Household characteristics
Household has at least one child	NS	NS	NS	NS	1.143	0.020 *
Individual characteristics
Older adult (65+ years)	1.432	0.005 **	NS	NS	NS	NS
Higher income (≥$100,000 CAD)	−0.869	0.035 *	NS	NS	−1.307	0.035 *
Owns residence	NS	NS	0.701	0.049 *	NS	NS
Risk perception
Staying increases odds of harm	NS	NS	NS	NS	−1.088	0.020 *
Evacuation experiences/decisions
Experienced wildfire property damage	0.804	0.038 *	NS	NS	NS	NS
Job is related to wildfires	NS	NS	NS	NS	1.369	0.028 *
Will take <30 min to prepare for an evacuation	0.819	0.011 *	NS	NS	NS	NS
Will evacuate after can no longer defend home	NS	NS	1.224	0.000 *	NS	NS
Will tow items behind their vehicle	NS	NS	NS	NS	1.174	0.010 *
Primary evacuation mode: personal vehicle	−1.155	0.003 **	NS	NS	NS	NS
ρ2 (fit)	0.28
ρ2 (adjusted fit)	0.24
Initial log-likelihood	−431.14
Final log-likelihood	−312.63

Note: Est coeff = estimated coefficient; CAD = Canadian dollars; NS = not significant.

p < .05; **p < .01.

Among the individuals with disabilities, we noted that those from high-income households and those who owned their residence were both more likely to stay at a second residence/mobile home. These individuals may prefer more private sheltering accommodations and may have personalized accessibility options at their second residences. Notably, older adults who would evacuate using shared mobility were more likely to stay at public/government shelters. Among lower-income residents, those taking less than 30 minutes to prepare for an evacuation were more likely to stay at government-provided shelters. We further observed that carless residents who previously experienced an evacuation were more likely to shelter with friends/family. This pattern may show an inclination within the group to stay with loved ones for comfort in a future emergency. Finally, among the visible minorities’ group, those who would evacuate in less than 30 minutes were more likely to stay at a government-provided shelter, similar to lower-income residents.

Discussion and Policy Recommendations

Based on our research findings, we offer several recommendations for wildfire evacuation planning and policy (Table 10). Each recommendation is accompanied by a discussion of relevant findings. Recommendations are aimed at local emergency management offices as well as transportation agencies to make wildfire evacuations more equitable for underserved groups.

Table 10.

Key Recommendations and Discussion of Findings

Key recommendations	Discussion
Provide targeted support for those with previous wildfire experiences	Those who experienced discomfort from wildfire smoke and those whose jobs were previously affected by wildfires from the people with disabilities group were less likely to evacuate promptly. In contrast, older adults who previously learned about wildfires and those who attended a public meeting on wildfires were more likely to evacuate promptly and take less time preparing for an evacuation.These results highlight the need for targeted education and support among residents, particularly those who have been previously affected by wildfires. Emergency officials may consider specialized outreach programs for these individuals, accurately depicting wildfire risk and appropriate evacuation actions. Public meetings and personal preparedness programs may prove effective in reducing the time between a mandatory order and an evacuation, helping to increase safety during wildfires ( 35 ).
Implement preparedness programs to enhance timely evacuations
Provide accessible and affordable long-distance transit vehicles during wildfire evacuations	Visible minorities and those with disabilities from lower-income households were more likely to use public transit as an evacuation mode during a wildfire. Carless residents who would evacuate promptly were also more likely to use public transit. In addition, transit users from both the carless residents and individuals with disabilities groups were more likely to evacuate to destinations outside their home community.Emergency management agencies should collaborate with transit agencies to ensure the presence of reliable transit services during evacuations. Ongoing partnerships with paratransit providers could be formed to offer specialized transportation for those with disabilities. Transit agencies may further consider implementing fare-free policies during emergencies for evacuees from lower-income households ( 10 ). Given that some transit users would use highways to evacuate outside their home communities, agencies may consider providing intercity bus/shuttle services for long-distance evacuations. These may further assist in mitigating congestion from single-occupant vehicles during evacuations.
Strengthen partnerships with ride-sharing companies	Individuals with disabilities and carless residents under the age of 35 were more likely to use shared mobility (e.g., ride-sourcing, carpooling) during a wildfire evacuation. Visible minorities and older adults with a disability were also more likely to use shared mobility. In addition, those who had previously learned about wildfires from the carless residents group were likely to use shared modes during a wildfire evacuation.As noted by Sadri et al., shared mobility could produce equitable outcomes during emergencies, including for areas with limited transit accessibility ( 23 ). As such, emergency management officials may consider creating partnerships with ride-sourcing companies (e.g., Uber, Lyft) to ensure vehicle availability and cost subsidization during wildfire evacuations.
Consider traffic management strategies to address capacity challenges for major roads and highways during wildfire evacuations	Our survey revealed that most respondents would spend most of their evacuation time on highways. Those who would evacuate after a mandatory order were generally more likely to use highways whereas those who would evacuate after a voluntary order were likely to use major roads during a wildfire evacuation. Moreover, more than 80% of survey respondents from each underserved group indicated that they would evacuate using one or more personal vehicles (excluding carless residents).Major routes including highways and arterial roads should be prepared to handle high volumes of traffic during a wildfire evacuation. As Zehra and Wong noted, effective rerouting can help mitigate congestion during evacuations ( 36 ). Agencies may consider implementing rerouting strategies outlined in recent research (such as by Shahparvari and Abbasi [ 37 ]) to optimize traffic flow and ensure smooth evacuations during wildfires. Additionally, strategies such as contraflow lanes and emergency shoulder usage may be further considered to address capacity challenges. Creating evacuation zones based on wildfire risk and implementing staggered evacuation times could further help distribute traffic and minimize congestion on highways and major roads.
Equip private and public shelters to accommodate diverse needs during an emergency	Older adults with disabilities and carless residents with children were more likely to stay at a government shelter compared with other shelter types. Lower-income households and visible minorities who would evacuate in less than 30 min were also more likely to stay at a government shelter. Interestingly, among the people with disabilities and older adults groups, no variables were positive for sheltering at a hotel/motel/Airbnb. This result may point to accessibility issues at private establishments.Government shelters (including public shelters and community centers) should be equipped with accessibility features (e.g., ramps, accessible restrooms, wide doorways/hallways) as well as medical accommodations for older adults and those with disabilities. Moreover, amenities should be provided to offer comfort and sufficient privacy for those evacuating with children. In addition, government shelters should be ready for use after a mandatory order to accommodate fast evacuees during a wildfire. Private accommodations (i.e., hotels, motels, and Airbnb) should ensure adequate accessibility features to promote usage among older adults and evacuees with disabilities.
Ensure highly localized sheltering resources during a wildfire	Our survey showed that many respondents would evacuate to locations within their respective communities. Moreover, discrete choice modeling results revealed that young adults from both carless residents and individuals with disabilities groups were likely to evacuate to destinations within the community. Among the visible minorities, high-income earners and those willing to take risks were also likely to evacuate to locations within their communities.As Cova et al. noted, most evacuees tend to travel shorter distances during wildfire evacuations ( 34 ). As such, agencies should be prepared to provide localized sheltering resources to ensure safe and convenient accommodations. In cases where public shelters are limited, agencies could create partnerships with recreation centers, local hotel chains, as well as Airbnb to provide accessible and affordable sheltering options. Moreover, shelters should be easily accessible by public transit and active modes to accommodate carless residents evacuating to destinations within their respective communities.
Identify intersecting vulnerabilities for more equitable outcomes	Together, findings from our research ultimately point to the necessity of identifying and addressing the needs of those with intersecting vulnerabilities during wildfire evacuation planning. Identifying overlapping barriers to evacuation can help prevent vague planning and support equitable outcomes.Establishing and maintaining evacuation registries can be helpful in pre-identifying medical, transportation, and sheltering needs among underserved groups ( 10 ). Evacuation registries should offer accessible registration methods and incorporate transparency in how information will be used. Agencies may further consider working with nongovernmental organizations (NGOs) and community-based organizations (CBOs) representing underserved groups to ensure effective outreach and communication with these groups during the planning process ( 10 ).

Study Limitations

Although this study provides important insights into wildfire evacuation decision-making among underserved groups, several limitations should be considered. First, because we opted for an online survey format, those with limited internet access were restricted from participation. However, we note that an online format enabled us to widen our reach and obtain respondents from diverse cities and municipalities in AB and BC. Moreover, we enlisted the help of fire departments and community officials, who distributed the survey more broadly through social media, newsletters, and community events. Second, we acknowledge that our survey may have had self-selection bias as respondents opted into the study for both the panel and the convenience sample, especially since respondents had to live in a high fire-risk area. Incentives for both cases and quotas for the panel may have reduced some of this bias. Third, we note that our survey oversampled residents who owned at least one personal vehicle, therefore undersampling the number of respondents who were carless. Our survey also oversampled high-income and highly educated populations. Additionally, our survey asked respondents to mark their expected origin and destination locations during a wildfire evacuation. This question received few responses, most likely owing to technical difficulties associated with the embedded map interface. Consequently, the sample size for our destination choice model was limited, which may have affected variable significance and conclusions. Future work may incorporate alternative methods for collecting location data. Although we collected data on evacuation origin and destination points, we did not gather information on respondents’ household locations. Future research could incorporate household location data to better understand their role in evacuation decisions such as mode and route choices. Moreover, future iterations of this study could identify spatial and network characteristics of each origin and destination community as additional modeling variables.

From a modeling standpoint, in this paper, decisions such as mode choice, shelter choice, and destination choice were modeled separately for each underserved group. Although valuable, we acknowledge that this approach may not fully capture the correlated nature of these decisions. Future work may consider employing a joint modeling framework to better capture the joint nature of evacuation choice-making among underserved groups. Future work may also consider comparing the evacuation choice-making patterns of underserved groups with those of non-underserved populations. This comparative analysis (see for example Ciriaco et al. [ 38 ] for resilience hub usage) could help uncover key differences in decision-making processes and highlight unique needs and challenges faced by underserved groups during wildfire evacuations. Other modeling techniques, such as those in machine learning (ML), could also be considered in future iterations of the research for improved predictive capabilities, though careful consideration must be made, given that ML models do not have any underlying behavioral theory. Explainable artificial intelligence or ML-informed discrete choice modeling (see for example Van Cranenburgh et al. [ 39 ]) would offer a roadmap for alternative analyses.

Conclusion

Although there is a growing body of research on wildfire evacuation decision-making, the specific behavior of underserved groups remains underexplored. To address this gap, our study utilized survey data from 2,868 respondents in AB and BC. We employed discrete choice modeling to examine the effects of evacuation experience, risk perception, and sociodemographic variables on key wildfire evacuation choices among underserved groups. In summary, we found that,

There is notable heterogeneity among underserved groups, affecting evacuation timing, destination, mode, and shelter choices.

Overlapping vulnerabilities and intersectionality can significantly influence wildfire evacuation choices among underserved groups.

Previous wildfire experiences influenced longer preparation and departure times, whereas prior learning about wildfires significantly influenced prompt evacuations.

Higher risk perceptions were associated with shorter preparation times, a higher likelihood of evacuating promptly, and a greater tendency to select destinations outside communities of origin.

Individual and household characteristics were significant in determining evacuation shelter choices among underserved groups.

Ultimately, our findings highlight the role of tailored strategies in meeting the evacuation needs of underserved groups. These insights have important implications for equitable emergency planning, emphasizing the need to move beyond broad strategies and instead develop context-specific strategies to address the needs of underserved groups. Local agencies can enhance preparedness and equity by 1) providing targeted support for those with previous evacuation experiences, 2) ensuring multimodal evacuation plans with shared modes and transit considerations, 3) equipping public shelters to accommodate diverse needs, and 4) addressing intersecting vulnerabilities. Future research could target rural areas that often have fewer transportation options and evacuation resources. Research focused on evacuation choice-making among underserved residents in rural areas may uncover challenges and decision-making processes that differ from those in more urbanized areas. Future work can further build on this research by expanding our methodology to other regions in North America and globally, and considering additional underserved groups such as immigrants, unhoused individuals, or those with limited English proficiency.

Supplemental Material

sj-docx-1-trr-10.1177_03611981251380279 – Supplemental material for Wildfire Evacuation Choice-Making among Underserved Groups in Alberta and British Columbia

Supplemental material, sj-docx-1-trr-10.1177_03611981251380279 for Wildfire Evacuation Choice-Making among Underserved Groups in Alberta and British Columbia by Veronica Wambura and Stephen D. Wong in Transportation Research Record

Footnotes

Acknowledgements

We thank Jen Beverly from the University of Alberta and Amy Kim from the University of British Columbia for their assistance with community engagement. Thank you to Syeda Narmeen Zehra and Mohammad Babaei for assisting in data collection and cleaning. We finally thank the local agencies that assisted in finding survey participants.

Author Contributions

The authors confirm contribution to the paper as follows: study conception and design: V. Wambura, S. Wong; data collection: V. Wambura, S. Wong; analysis and interpretation of results: V. Wambura, S. Wong; draft manuscript preparation: V. Wambura, S. Wong. All authors reviewed the results and approved the final version of the manuscript.

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This study was funded by Infrastructure Canada (now Housing, Infrastructure and Communities Canada) through its Research and Knowledge Initiative (RKI). We thank the sponsors for their support.

ORCID iDs

Veronica Wambura

Stephen D. Wong

Supplemental Material

Supplemental material for this article is available online.

References

Christianson

A. C.

Johnston

L. M.

Macnab

J. A.

Oliver

Watson

Young

MacDonald

Little

Bautista

N. G.

Wildland Fire Evacuations in Canada from 1980 to 2021. International Journal of Wildland Fire, Vol. 33, No. 7, 2024, p. WF23097.

Renne

J. L.

Mayorga

What has America Learned Since Hurricane Katrina? Evaluating Evacuation Plans for Carless and Vulnerable Populations in 50 Large Cities across the United States. International Journal of Disaster Risk Reduction, Vol. 80, 2022, p. 103226.

Haznet. Advice for Emergency Managers from the First Nations Wildfire Evacuation Partnership. HazNet. https://haznet.ca/advice-emergency-managers-first-nations-wildfire-evacuation-partnership/ Accessed May 25, 2018.

Sun

Forrister

Kuligowski

Lovreglio

Cova

Zhao

Social Vulnerabilities and Wildfire Evacuations: A Case Study of the 2019 Kincade Fire. Safety Science, Vol. 176, 2024, p. 106557.

Shaheen

Bell

Cohen

Yelchuru

Travel Behavior: Shared Mobility and Transportation Equity. Report Number: PL-18-007. U.S. Department of Transportation, Federal Highway Administration, Washington, D.C., 2017.

Gibson

M. J.

We Can Do Better: Lessons Learned for Protecting Older Persons in Disasters. AARP Public Policy Institute, Washington, D.C., 2006.

Feng

Vodopivec

Miller-Hooks

Supporting Mobility-Impaired Populations in Emergency Evacuations. Transportation Research Record: Journal of the Transportation Research Board, Vol. 2532, No. 1, 2015, pp. 118–128.

Allen

Farber

Greaves

Clifton

Sarkar

Levinson

D. M.

Immigrant Settlement Patterns, Transit Accessibility, and Transit Use. Journal of Transport Geography, Vol. 96, 2021, p. 103187.

Pulcinella

J. A.

Winguth

A. M. E.

Allen

D. J.

Gangadhar

N. D.

Analysis of Flood Vulnerability and Transit Availability with a Changing Climate in Harris County, Texas. Transportation Research Record: Journal of the Transportation Research Board, Vol. 2673, No. 6, 2019, pp. 258–266.

10.

Wambura

Wong

S. D.

Incorporating a Public Transit Equity Lens in Evacuation Planning. Transportation Research Record: Journal of the Transportation Research Board, Vol. 2678, No. 11, 2024, pp. 1679–1690.

11.

Holmes

T. J.

Williams

P. C.

Wong

Smith

Bandzuh

J. T.

Uejio

C. K.

Assessment of an Evacuation Shelter Program for People with Access and Functional Needs in Monroe County, Florida during Hurricane Irma. Social Science & Medicine, Vol. 306, 2022, p. 115108.

12.

Cutter

S. L.

Smith

M. M.

Fleeing from the Hurricane’s Wrath: Evacuation and the Two Americas. Environment: Science and Policy for Sustainable Development, Vol. 51, No. 2, 2009, pp. 26–36.

13.

Karaye

I. M.

Thompson

Perez-Patron

Taylor

Horney

J. A.

Estimating Evacuation Shelter Deficits in the Houston–Galveston Metropolitan Area. Risk Analysis, Vol. 40, No. 5, 2020, pp. 1079–1091.

14.

Wong

Shaheen

Walker

Understanding Evacuee Behavior: A Case Study of Hurricane Irma. University Of California, Berkley, CA, 2018.

15.

Wong

S. D.

Broader

J. C.

Walker

J. L.

Shaheen

S. A.

Understanding California Wildfire Evacuee Behavior and Joint Choice Making. Transportation, Vol. 50, 2022, pp. 1165–1211.

16.

Kuligowski

E. D.

Walpole

E. H.

Lovreglio

Mc Caffrey

Modelling Evacuation Decision-Making in the 2016 Chimney Tops 2 Fire in Gatlinburg, TN. International Journal of Wildland Fire, Vol. 29, No. 12, 2020, pp. 1120–1132.

17.

Toledo

Marom

Grimberg

Bekhor

Analysis of Evacuation Behavior in a Wildfire Event. International Journal of Disaster Risk Reduction, Vol. 31, 2018, pp. 1366–1373.

18.

Lovreglio

Kuligowski

Gwynne

Strahan

A Modelling Framework for Householder Decision-Making for Wildfire Emergencies. International Journal of Disaster Risk Reduction, Vol. 41, 2019, p. 101274.

19.

Forrister

Kuligowski

E. D.

Sun

Yan

Lovreglio

Cova

T. J.

Zhao

Analyzing Risk Perception, Evacuation Decision and Delay Time: A Case Study of the 2021 Marshall Fire in Colorado. Travel Behaviour and Society, Vol. 35, 2024, p. 100729.

20.

Grajdura

Qian

Niemeier

Awareness, Departure, and Preparation Time in No-Notice Wildfire Evacuations. Safety Science, Vol. 139, 2021, p. 105258.

21.

Grajdura

Niemeier

Inequities & Injustices in Large-Scale California Wildfire Evacuations & Post-Evacuations. https://papers.ssrn.com/abstract=4022006. Accessed July 1, 2024.

22.

Sadri

A. M.

Ukkusuri

S. V.

Murray-Tuite

Gladwin

Analysis of Hurricane Evacuee Mode Choice Behavior. Transportation Research Part C: Emerging Technologies, Vol. 48, 2014, pp. 37–46.

23.

Wong

S. D.

Broader

J. C.

Shaheen

S. A.

Can Sharing Economy Platforms Increase Social Equity for Vulnerable Populations in Disaster Response and Relief? A Case Study of the 2017 and 2018 California Wildfires. Transportation Research Interdisciplinary Perspectives, Vol. 5, 2020, p. 100131.

24.

Zhao

Wong

S. D.

Developing Transportation Response Strategies for Wildfire Evacuations via an Empirically Supported Traffic Simulation of Berkeley, California. Transportation Research Record: Journal of the Transportation Research Board, Vol. 2675, No. 12, 2021, pp. 557–582.

25.

Ahmed

S. K.

How to Choose a Sampling Technique and Determine Sample Size for Research: A Simplified Guide for Researchers. Oral Oncology Reports, Vol. 12, 2024, p. 100662.

26.

Moss

A. J.

Hauser

D. J.

Rosenzweig

Jaffe

Robinson

Litman

Using Market-Research Panels for Behavioral Science: An Overview and Tutorial. Advances in Methods and Practices in Psychological Science, Vol. 6, No. 2, 2023, pp. 1-25.

27.

Ciriaco

T. G. M.

Wong

S. D.

Review of Resilience Hubs and Associated Transportation Needs. Transportation Research Interdisciplinary Perspectives, Vol. 16, 2022, p. 100697.

28.

Government of Canada, S. C. Census Profile, 2021Census of Population. https://www12.statcan.gc.ca/census-recensement/2021/dp-pd/prof/index.cfm?Lang=E. Accessed July 5, 2023.

29.

Government of Canada, S. C. Low Income Cut-Offs (LICOs) before and after Tax by Community Size and Family Size, in Current Dollars. https://www150.statcan.gc.ca/t1/tbl1/en/tv.action?pid=1110024101. Accessed April 30, 2025.

30.

Train

K. E.

Discrete Choice Methods with Simulation. Cambridge University Press, Cambridge, 2009.

31.

Ben-Akiva

M. E.

Lerman

S. R.

Discrete Choice Analysis: Theory and Application to Travel Demand. MIT Press, Cambridge, MA, 1985.

32.

Ortelli

Hillel

Pereira

F. C.

De Lapparent

Bierlaire

Assisted Specification of Discrete Choice Models. Journal of Choice Modelling, Vol. 39, 2021, p. 100285.

33.

McFadden

The Measurement of Urban Travel Demand. Journal of Public Economics, Vol. 3, No. 4, 1974, pp. 303–328.

34.

Cova

T. J.

Sun

Zhao

Liu

Kuligowski

E. D.

Janfeshanaraghi

Lovreglio

Destination Unknown: Examining Wildfire Evacuee Trips Using GPS Data. Journal of Transport Geography, Vol. 117, 2024, p. 103863.

35.

McGee

T. K.

Public Engagement in Neighbourhood Level Wildfire Mitigation and Preparedness: Case Studies from Canada, the US and Australia. Journal of Environmental Management, Vol. 92, No. 10, 2011, pp. 2524–2532.

36.

Zehra

S. N.

Wong

S. D.

Systematic Review and Research Gaps on Wildfire Evacuations: Infrastructure, Transportation Modes, Networks, and Planning. Transportation Planning and Technology, Vol. 47, No. 8, 2024, pp. 1–35.

37.

Shahparvari

Abbasi

Robust Stochastic Vehicle Routing and Scheduling for Bushfire Emergency Evacuation: An Australian Case Study. Transportation Research Part A: Policy and Practice, Vol. 104, 2017, pp. 32–49.

38.

Ciriaco

T. G. M.

Zehra

S. N.

Wambura

Wong

S. D.

Equitable Transportation and Resilience Hubs: Analysis of Underserved Population Needs, Usage, and Travel Behaviour. Transportation Research Interdisciplinary Perspectives, Vol. 29, 2025, p. 101314.

39.

van Cranenburgh

Wang

Vij

Pereira

Walker

Choice Modelling in the Age of Machine Learning - Discussion Paper. Journal of Choice Modelling, Vol. 42, 2022, p. 100340.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.02 MB