The Impact of Social Media on Disaster Volunteerism: Evidence from Hurricane Harvey

2.1 Online Emergent Groups and Digital Volunteerism

The spread of information communication technologies (ICT) and social media platforms enabled individuals to volunteer in disaster response and relief in new ways. Using these technologies, people share information, ask for help, or coordinate actions through web pages, social media groups, hashtags, and crowdsourcing tools. These new ICT-based activities create a rich tapestry of digital volunteerism, which requires a careful examination and analytical perspective to understand. Palen and Hughes (2018) offer a useful classification of these activities. These are (i) citizen reporting, (ii) community-oriented computing, (iii) collective intelligence and distributed problem solving, and (iv) volunteerism. Citizen reporting drives significant attention from the public during and after disasters, especially with the help of smartphones and faster Internet connection. With advanced communication devices, individuals become sensors that feed information to the public and first responders (Avvenuti et al. 2016, 2018). Inevitably, such information flow shapes the public attitude towards disasters and how first responders take action. For example, during the 2007 wildfires in California, the Internet was instrumental in information seeking, community formation, and situational awareness Shklovski, Palen, and Sutton (2008). After the devastating earthquake in Haiti in 2010, people posted images of impacted areas and their personal experiences on social media platforms immediately. Within 48 h, the Red Cross received $8 million in donations (Gao, Barbier, and Goolsby 2011).

Community-oriented computing refers to the use of social media for alleviating the impacts of disasters on communities and helping communities become more resilient. Palen and Hughes (2018) argue that social media has been instrumental for citizens to express their grief, share their community activities, and create a sense of solidarity after disasters, which helped them return to normalcy.

Collective intelligence and distributed problem-solving refer to self-organizing human activity over social media that solves problems through collective action (Abdulhamid et al. 2021; Palen and Hughes 2018). For example, during the Tassie Fires in Australia in 2013, a woman started asking questions to her friends and family about the fire on her Facebook account as the disaster unfolded. As the details emerged, she decided to organize the flow of information on a Facebook page called “Tassie Fires – We Can Help” (TFWCH), which eventually turned into a communication channel between the citizens (victims and volunteers alike) and public officials (Irons et al. 2014). Through this newly emerged communication channel, citizens were able to track the whereabouts of the fires and as well as the wellness of people and animals almost in real time.

Digital volunteerism refers to the grassroots volunteer activities that take place remotely using online applications such as social media platforms and other tools. A notable example of digital volunteerism happened after the earthquake in Haiti in 2010. An open-source crisis map platform called Ushahidi (www.ushahidi.com) was established to connect data from multiple sources and users (Twitter, Facebook) and provide a real-time map to relief organizations (Starbird 2011; Starbird and Palen 2011). Similarly, after a Malaysia Airlines airplane had disappeared from the radar in 2014, thousands of volunteers across the world searched millions of images for signs of the potential wreck of the airplane (CNN 2014).

The categories of activities identified by Palen and Hughes (2018) can take place simultaneously in Facebook groups. Facebook groups, for example, can be used as forums for citizen reporting as well as expressing grief and finding solace at the same time (Irons et al. 2014). Likewise, grassroots volunteer movements turn Facebook into an organizing platform quickly. For example, the Cajun Navy uses Facebook and Twitter quite effectively for sharing information, organizing their activities, and reaching out to people (Smith et al. 2018). These examples emphasize that Facebook, like other social media platforms, creates more opportunities for volunteers to engage in disaster response in innovative ways (Abdulhamid et al. 2021). For example, some volunteer activities can solely be formed and practiced online, like in the case of the Malaysian Airlines accident, or, as in the case in the case of the Cajun Navy, social media platforms can facilitate the activities of volunteers on the ground. Such a wide range of social media use in disaster volunteerism indicates that social media has been transforming how and when volunteers organize and what functions they carry out. To better understand how social media shapes disaster volunteerism, it is important that we first understand the fundamental characteristics of disaster volunteerism and what they imply in the age of social media. The following sections will provide discussions on the fundamentals and characteristics of volunteer groups in disaster management and connect them to digital volunteerism.

2.2 Volunteer Groups in Disasters: Definition and Typologies

Disaster scholars have long studied collective human behavior following disasters. Researchers analyze how people act during and after disasters and document the emergence of collective behavior following disasters (Drabek and McEntire 2003). For example, contrary to common belief, only a little panic, looting, and other anti-social behaviors happen during disasters (Auf der Haide 1989; Fischer 1998; Quarantelli 1995). People usually demonstrate cohesive, unified, and altruistic behavior, which gives rise to volunteer activities (Helsloot and Ruitenberg 2004; Kaniasty and Norris 1995). Different terminologies (e.g., spontaneous volunteers, volunteerism, citizen response, emergent groups, emergent structures) are used to describe and analyze volunteer actions in the literature. Although there are nuances between these terms, they broadly point out to citizens’ activities after a disaster who willfully help victims without expecting compensation or any benefit in return (Fernandez, Barbera, and van Dorp 2006; Helsloot and Ruitenberg 2004; McLennan, Whittaker, and Handmer 2016; Stallings and Quarantelli 1985; Whittaker, McLennan, and Handmer 2015).

Volunteers’ activities take place in many different settings. For example, some volunteers simply converge on a disaster scene (e.g., a flood area) and help victims on their own. Others participate in disaster response and relief activities by working under the banner of formal organizations such as the American Red Cross or through volunteer groups whose tasks are pre-determined such as the Community Emergency Response Teams (CERT) as part of a FEMA program (Strandh and Eklund 2018). These structural differences between the settings within which volunteers may engage in disaster response and relief develop unique characteristics for volunteer groups. These structural differences are analyzed by scholars at the Disaster Research Center (DRC) extensively (Quarantelli 1995; Shaskolsky 1967; Stallings and Quarantelli 1985), and different typologies are developed to distinguish them.

The DRC typology classifies response organizations into four categories depending on whether their structure and tasks are old or new (see Table 1). The first category is labeled established groups that function with an old structure and have routine tasks. Examples of the established group include police, fire, and EMS services. The second category is called expanding groups (new structure, routine tasks) that include relief organizations such as the Red Cross that routinely engage in disaster relief activities but need to expand the structure to meet the demands of a new disaster. The third category is called extending organizations (old structure, new tasks) that refers to organizations that respond to disasters using their old structure. The fourth category is called emergent groups (new structure, new tasks) that refer to organizations that do not exist before a disaster and emerge with a new structure and new tasks (McLennan, Whittaker, and Handmer 2016; Quarantelli 1995; Strandh and Eklund 2018). Stallings and Quarantelli define emergent groups as “private citizens who work together in pursuit of collective goals relevant to actual or potential disasters but whose organization has not yet become institutionalized” (Stallings and Quarantelli 1985, 94). Most digital volunteer groups fall under this category, with some notable exceptions (Fathi et al. 2020), since the role of social media in disaster response is a relatively new phenomenon (King 2018; Demiroz 2020), and they rarely function as part of a formal organization (Hughes and Tapia 2015). Emergent groups manifest structural and functional characteristics that distinguish them from other types of organizations in the DRC typology.

2.3 Structural Characteristics of Emergent Volunteer Groups

Characteristics of emergent groups are documented by disaster scholars extensively. From a structural perspective, such groups lie in between established bureaucracies and isolated citizens who converge on the same problem. They do not function based on a pre-established structure when they respond to a disaster, and the tasks that they carry out are new to their members (Stallings and Quarantelli 1985; Twigg and Mosel 2017). In other words, these groups carry out new tasks within a new structure that emerges out of interactions among members of the group. Emergent volunteer groups have a flat hierarchy, meaning that the distance between the top and the bottom is relatively small. They rarely possess symbols or a specific physical location (e.g., an office), and one or more individuals can assume leadership roles, although the distinction between the leaders and the subordinates is usually not clear (Stallings and Quarantelli 1985). It is possible to observe a division of labor; however, tasks usually do not require expertise or specialization. Their structure is closely tied to the tasks they carry out. For example, tasks that require some level of supervision, such as handling injured survivors, may lead to a relatively more hierarchical structure in which a knowledgeable person such as a nurse oversees others (Drabek and McEntire 2003; Quarantelli 1984; 1995; Stallings and Quarantelli 1985; Twigg and Mosel 2017).

Online volunteer groups share similar characteristics with traditional groups, though there are some notable exceptions. For example, during the 2013 European Floods in Germany, people used Facebook and Twitter quite extensively. Twitter users were mostly using the platform for broadcasting information (i.e., information sharing) (Kaufhold and Reuter 2016). On Facebook, groups, and pages carried out more diverse functions than Twitter did, such as sharing tasks and documents. In both platforms, users carried out different tasks (i.e., division of labor) such as moderating, amplifying the message. These activities were distributed and lacked a formal hierarchical structure similar to that of a formal organization (Kaufhold and Reuter 2016). Within this distributed structure, group moderators occupy one of the most visible positions. Moderators’ core activities entailed the “mediation of volunteer activity and matched citizens’ demands with offers of help by utilizing and integrating social media functions and ICT which contributed to the process of structuring the rich information supply” (Kaufhold and Reuter 2016, 157). The dense communication network that the moderators are embedded in within a Facebook group inevitably grants them in a powerful position (Hanneman and Riddle 2005).

A notable exception regarding the similarities between online volunteer groups and traditional emergent groups is a product of the efforts for connecting online volunteerism and formal disaster response structures. In recent years, scholars and practitioners have sought ways to utilize online volunteers during disasters (Gambo et al. 2021). One of the products of these efforts is the Virtual Operations Support Teams (VOST) (Fathi et al. 2020; Hughes and Tapia 2015). VOSTs are not emergent structures; they are pre-established groups that build trust relationships with government officials. They are akin to Voluntary Organizations Active in Disaster (VOAD) structure with a presence only in the online realm and functions limited to certain activities (e.g., combing the social media and analyzing data) only as opposed to VOAD’s more diverse functions in the “real world.”

2.4 Life Span of Emergent Volunteer Groups

Emergent volunteer organizations also differ from other categories of organizations concerning their lifespan and relations with their environment. The lifespan of emergent volunteer groups is considerably short (e.g., hours or a few days) (Stallings and Quarantelli 1985; Strandh and Eklund 2018; Twigg and Mosel 2017). Their life and size change depending on the ebb and flow of members and the need for volunteers to carry out their tasks (Twigg and Mosel 2017). Their relationship with their environment is limited to their lifespan unless they manage to crystallize their tasks, function, and structure and maintain them in the long term. In many cases, it is unlikely to see a formally assigned liaison person to manage relations with other organizations and other external actors (Stallings and Quarantelli 1985). If organizations manage to maintain their structure and function, which rarely happens, they may designate individuals to manage relationships with their environment.

2.5 Functions of Emergent Volunteer Groups

Volunteers engage in various activities and try to fill the gaps that formal organizations do not or cannot fill. Their activities and functions are categorized in different ways by different scholars. For example, Stallings and Quarantelli (1985, p. 94) categorize emergent volunteer groups into three broad groups based on their functions. The first type of group is labeled damage assessment groups. These groups inform public officials regarding the damage that occurred in their community. The second type of group is labeled operation groups. These groups carry out tasks such as collecting and distributing food and donations, helping clean up the impacted areas from debris and other things. The third group is labeled as the coordinating groups. These groups set the direction for their community moving forward after a disaster and try to settle down disputes and problems in their community Stallings and Quarantelli (1985).

Twigg and Mosel (2017) follow a different approach and divide volunteer activities into ten different categories based on more specific functions they carry out (see Table 2). These categories are medical (e.g., search and rescue, first aid, donating blood), information/communications (e.g., registration of victims, translating, sharing information and messages), psychological and bereavement (e.g., psychological counseling), shelter, supplies, and provisions (e.g., collecting, transporting, and distributing relief supplies), building and services (e.g., removing debris and clearing streets, cleaning up after disasters), coordination and security (e.g., informal coordination of other groups and activities), preparedness (e.g., issuing warnings, helping with evacuations), advocacy (e.g., challenging actions and practices of official response agencies, maintaining the security of property, lobbying public officials), and others (e.g., raising funds for victims, taking care of animals).

Similar volunteer functions are identified in the literature, though not organized. For example, search and rescue, medical assistance, donations, sheltering, cleaning up, provision of food, emotional and social support, and animal care are widely mentioned in disaster research (Drabek and McEntire 2003; Fernandez, Barbera, and van Dorp 2006; Helsloot and Ruitenberg 2004; Majchrzak, Jarvenpaa, and Hollingshead 2007; Majchrzak and More 2011; McLennan, Whittaker, and Handmer 2016; Stallings and Quarantelli 1985; Twigg and Mosel 2017; Whittaker, McLennan, and Handmer 2015; Wolensky 1979). The functions and capabilities of emergent volunteer groups expand into new directions with the proliferation of sophisticated yet affordable and user-friendly technology such as smartphones, the Internet, and social media platforms.

4.1 Data Source

This research analyzes data obtained from Facebook groups that were created during Hurricane Harvey. Facebook allows its users to communicate on pages and groups, which are more structured communication environments than #hashtags used in other platforms such as Twitter. The structured environment on Facebook makes analyzing groups and users’ activity easier for a few reasons. First, the boundary of a Facebook group is clear, and it is possible to observe the size of a group and its activities. Second, Facebook users are more likely to use their real identity and location and communicate with people they are connected with in real life (i.e., community). Third, the lack of anonymity requires users to communicate with others within acceptable social norms. Facebook groups often provide a description of the page, explain its purpose, and group creators set rules for communication in the group (e.g., no advertisement, political, or offensive posts). Such characteristics of groups define their purpose (i.e., function), determine what kind of information could be shared in that group, and may suggest a group structure. For example, moderators of a group facilitate group communication and may delete some posts they deem unsuitable to the group’s purpose. The existence of a clear boundary of groups, interactions between real users, and predictable rule structure (i.e., social norms) in communication among group members make Facebook a more suitable platform for volunteer groups to emerge.

4.2 Data Collection

The data for this research were collected from Facebook. Two types of data were collected. The first data type is the post of group members as text data. The second type of data that was collected was group metadata. The following steps were followed in data collection. In the first step, the keywords “Hurricane Harvey” and “Harvey” were searched on Facebook to identify all the groups that were created regarding Hurricane Harvey. The search identified 195 groups that were about Hurricane Harvey. Twenty-one of these groups were closed groups whose data were not available to the public, and five groups were inaccessible for unknown reasons. A total of 169 groups’ data were accessible.

In the second step, a Facebook application programming interface (API) that researchers created and the Rfacebook package (Barbera et al. 2017) available in the R programming environment were used for downloading group data. The data collection process started in March 2018; however, during this period, Facebook changed its protocols for third-party users’ access to Facebook data because of the Cambridge Analytica scandal (Wylie 2019). Changes in Facebook protocols for data collection created significant technical and time barriers for researchers. Consequently, the data collection process ended before data from all groups were collected. Data from 74 out of 169 accessible groups were collected (43.78% of all groups). The group data provide information about group members such as individual users’ unique numerical ID, users’ name, posts in the group, timestamp of posts, type of content they posted in the group (e.g., photo, status, or link), link to the content posted (if any), unique numerical id number of the group, like count, comment count, and shares count (see Table 3).

4.3 Text Data

For the purpose of this research, the unique numerical group id, the timestamp, and the messages as the text data were used. All the other data (e.g., user’s name, likes count, shares count) were deleted from the dataset.

The text data collected from groups were analyzed to identify the topics that groups discussed, which was instrumental in identifying their functions. Packages in the Python programming language are used for the text data analysis with the following steps. First, the data were preprocessed and cleaned. Preprocessing includes removing punctuations, HTML, and XML tags from the data. Stop words (e.g., “a”, “am”, “it”) and common/generic words (“houston”, “harvey”, “texas”) were removed using the nltk e2 library. Then, stemming and lemmatization procedures were conducted using the spacy e3 library. The purpose of this step is to convert a word to a common base form (going – > go). Finally, Latent Dirichlet Allocation (LDA) topic modeling was conducted via the Gensim framework. LDA is a probabilistic topic model, and each document is viewed as a mixture of topics. LDA identified topics discussed in each group category in week one through week six and provided evidence regarding how discussions in group categories evolved in this period.

4.4 Group Metadata and Categorization

The metadata includes group description (if provided by the group founders), group location (if available), size, creator, and creation date. Each group on Facebook was visited, and the group metadata was manually entered into an Excel spreadsheet. Next, each group’s function was extracted from its description, and then groups were categorized based on their function. To ensure the accuracy of group descriptions, they were compared to groups’ content. If a group’s description and content were consistent, the group description was used for extracting group functions and categorizing them. If there were any inconsistencies between a group’s description and its content, or no group description was provided, researchers used the best judgment to identify that group’s function and categorize it.

Groups were categorized based on the volunteer functions identified by Twig and Mosel (2017) in Table 2. New categories were added, or existing categories were modified as needed. For example, some groups contained posts without a specific purpose, and some others contained posts that were not relevant to Harvey relief (e.g., commercials). Such groups were labeled as general and irrelevant^[2], respectively. Also, some functions that were mentioned under a different category by Twigg and Mosel (2017) were made a separate category due to their importance. For example, Twigg and Mosel (2017) classified animal care under the other category; however, due to its importance in Harvey, animal care was made a category of its own in this article. Final categories with their description are presented in Table 4.

The purpose of clustering groups in categories based on functional similarities is to make data handling, analysis, and presentation more manageable. Presenting results from 74 individual groups would be prohibitively long and visually impossible. Consequently, they were categorized according to their function and analyzed in categories rather than individual groups. Also, this will provide which functions are crucial and get more attention from volunteers during the Hurricane Harvey.

5.1 General Overview

Findings from the analysis of group descriptions and functions show that the most frequently established group type is information/communication (25 groups). The supplies and provisions category (14 groups) ranks second, and the animal care category (8 groups) ranks third (see Table 5 for details). Two groups were labeled irrelevant because their content was not related to Hurricane Harvey.

The metadata (i.e., group names and descriptions) show that a group’s function is closely related to a broader goal/purpose that group was organized around. The data shows two broad goals/purposes that groups were organized around, hence forming two clusters. The first broad goal is reaching out and helping a specific community. Here a community can be a Church group, a city or county, or a geographic region. In groups that are organized around a community, usually, multiple functions are carried out (information sharing, cleaning up, donations, etc.). For example, people living in a specific city or county use their Facebook group for sharing information, asking for donations, or sharing resources. The second broad goal is accomplishing a specific function. In such groups, group members try to accomplish a specific function (e.g., search and rescue, animal care, donations) and usually limit themselves to that function. The functions of these groups are not limited to a specific community. For example, if a group was created for the specific function of information sharing, Hurricane information from all impacted areas can be shared in that group.

The groups that were not organized only around a community or specific function constitute the third cluster. Groups in the third cluster can be hybrid or multi-purpose groups. A hybrid group is one that carries out a specific function for a specific community. Here, both the community and function are equally important. A multi-purpose group is one with no community affiliation or a specific purpose.

5.2 Emergence and Lifespan

The emergence of groups happens in four time periods (see Table 5). These are (i) before landfall (August 24), (ii) during the impact (August 25–29), (iii) immediate aftermath (August 30 – September 1), and (iv) post-disaster (September 2 – November 7). On August 24, one day before Harvey made the first landfall near Rockport, three Facebook groups were created. During the impact (August 25–29), forty-five new groups were created. In the immediate aftermath of the disaster (August 30 – September 1), fifteen new groups were created. During the post-disaster period (September 2 – November 7), eleven new groups were created. The trend in the creation of new groups had an upward trajectory from August 24 through August 27. The trend is stabilized on August 27 and 28, and it followed a downward trajectory on August 29 and onwards. Detailed information about the creation times of groups is presented in Table 5.

The results show a close association between the time period that groups were created in and their functions. For example, if a group is created during the immediate impact period, it is very likely to be categorized under a response-related category (e.g., medical). The only exception to this observation is information/communication groups, which emerged in all time periods. Groups that were created before the landfall and during the early days of the impact are either for information communication or general-purpose. Starting from August 27, which is the impact period, different categories of groups started to emerge. The earliest groups categorized as supplies and provisions, animal care, buildings and services, missing persons, medical, fundraising, and special needs were created between August 27–29, which is the impact period. In the immediate aftermath period (August 30 – September 1), fifteen new groups were created. Five of these new groups were in supplies and provisions, four were in animal care, and two were in information/communication categories. One new group was created under each of the medical, building and services, shelter and housing, and other categories. During the post-disaster period, eleven new groups were established. Four of these groups were in information and communication, four of them were in supplies and provisions, one was in animal care, one was in shelter and housing, and in other categories.

Groups of all categories follow similar patterns in their activity timeline. The analysis of group content shows that almost all categories of groups generate most of their content in the few days following their inception, and their activities drop dramatically in subsequent days (see Figure 1). For example, groups in information/communication, supplies and provisions, animal care, general, medical, building and services, other categories show an initial spike in the amount of content generated in the early days of the Hurricane and decline significantly in subsequent days. Despite a decline in content generation over time, groups rarely dissolve or stop generating content. Most group categories continue to generate content even several months after the Hurricane, albeit in low volumes.

Figure 1:

The number of posts created per day for each category. The y-axis in graphs shows the amount of content generated and the x-axis shows days (day 0 = August 24).

Groups classified under missing persons, shelter and housing, fundraising, special needs, and irrelevant deviate from this pattern. These groups have an initial spike in the content generated in the early days; however, they either stop generating content (e.g., missing persons, fundraising, special needs, irrelevant) after a certain time or show irregular patterns (shelter and housing) in content.

5.3 Structure

For examining group structure, two different measures were used. These are the group size and the number of active members on a given day. The group size shows the number of members in each group. Group sizes vary significantly across group categories. Groups in the general category have the biggest number of members cumulatively (16,677 members) even though there are only six groups in this category. Medical groups have the second largest (14,193 members), and information/communication groups (11,536 members) have the third largest membership in cumulative terms. However, the distribution of group size varies from one category to another. Categories that contain a relatively large number of groups, such as information/communication (25 groups), supplies and provisions (14 groups), and animal care (8 groups), follow the power law in the distribution of group size. That means there are a few groups in a particular category with a large number of members, and many groups in the same category have significantly smaller membership sizes. The median group size for each category also points out the power law in group structures (see Table 6). For example, the information/communication category consists of 25 groups. The largest three groups in this category have 3807, 1233, and 1161 members, respectively, while the smallest three groups have 1, 1, and 4 members. The power law is evident in the distribution of group size in the medical category as well. The group that was created by the Cajun navy has over 12,000 members, while other groups under this category have 1424, 611, 121, and 28 members. Categories with three groups or fewer do not follow the power law. In other words, as the number of groups under a category exceeds three, the power law in the distribution of group size emerges.

The second important measure for understanding group structure is the number of active members in a group on a given day. Active group members are users who generate content in the group. The number of active members in a group can provide evidence regarding how discussions in the group are shaped and dominated. Groups with a relatively large size but a small number of active members are likely to be dominated by that small group of active members. Depending on the group category, such dominant users may generate content for informing the group members, organizing search and rescue activities, helping pet and livestock owners find shelter for their animals, etc.

In contrast, a large number of active users in a group means group activities and functions are shared by a wide variety of users and not dominated by a person or a small group of users. Figure 2 visualizes the number of active group members for each category of groups during Hurricane Harvey. As shown in Figure 2, almost all group categories show a spike during the periods before landfall (August 24), the impact (August 25–29), and the immediate aftermath (August 30 – September 1). This period (August 25 – September 1) can be labeled as the peak period since it is when groups are most active. Groups remain active even weeks after the peak period, albeit at significantly lower levels.

Figure 2:

The number of users active in each category per day. The y-axis in graphs show the number of active users (users that generate content) in the group and the x-axis shows days (Day 0 = August 24).

To identify whether a group is dominated by a small group or not, the number of active users in peak time was compared to the total number of users in that group. For example, groups categorized as information/communication have more than 11,000 users, whereas the number of active users in its peak time varies between 20 users and 160 users (see Table 7). That means a small number of people (between 20 and 160 users depending on the day) generate and share content with other 11,000 users while the others in the group become passive consumers of content. The domination of a relatively small group of people over the entire group makes them gatekeepers of information in the group.

Table 7 presents a measure for group domination in groups by dividing the median value of the number of active users within the first twenty-five days by the median value for group size^[3] (see Eq. (1)).

Equation (1) Formula for calculating the value for relative group domination

For example, the median group size for groups in the information/communication category is 154, and the median value for active users during the peak time is 90. A value is calculated by using Equation 1. For the information/communication category, the result would be

A value close to one means more members in a group actively participate in group discussions; a value close to 0 means fewer members in a group actively participate in group discussions.

When the same equation is applied to all categories of groups (see Col seven in Table 7), the result would be a value for the relative group domination for each category. This value makes comparing structural domination in group categories possible. The results show a negative correlation between median group size and group domination value in Col 7 (r = −0.247). In other words, categories and groups with many members tend to be dominated by a relatively small group of users, while categories and groups with fewer members have a more balanced relationship among their members.

5.4 Function

The analysis of the content generated in each category of groups provides evidence regarding their functions and their evolution over time. First, to understand their function, word frequencies in each category of groups were extracted. The most commonly occurring words in almost all groups are help and need (see Table 8). However, animal care, building and service, and special needs are exceptions to this pattern. Some of the most common words under the animal care category are find, help, pet, animal, and horse. The most common words under the building and service category are school, teacher, help, need, and classroom. Finally, some of the most common words under the special needs category are need, help, deaf, interpret, want, and disaster.

Second, to understand the evolution of groups’ function, a topic analysis was conducted for each category of groups for each week between week 1 and week 6^[4]. A topic analysis shows what topics are being discussed in a text by clustering words into small groups. A breakdown of discussion topics for each group category for the given six weeks period is presented in Table 9. The results show that discussion topics in groups are consistent with their category in the earlier weeks. However, group discussions converge on recovery, relief, and reconstruction subjects as time passes. That does not mean that all groups end up talking about the same topics (e.g., insurance claims, contractors, etc.) by week 6. Groups mostly preserve their functional identity, and yet, there is a natural transition from immediate response to recovery and rebuilding in group discussions between week 1 and week 6. For example, in week 1, information/communication groups discuss information for help and rescue related topics, groups under the supplies and provisions category discuss volunteers and immediate needs such as baby formula, and groups under the building and services category discuss teachers’ needs and teachers’ requests for rebuilding classrooms in their schools. In week 3, groups in each of these categories discuss common topics such as damage repairs, inspections, construction, insurance claims, and adjusters with nuances that represent each group category’s characteristics.

There are notable exceptions to these general findings. Animal care groups almost exclusively discuss animal related issues. The evolution of discussion in these groups happens from mostly horse related topics in week 1 to only pet related issues in week 6. Also, special needs groups also consistently discuss similar topics, such as finding interpreters and the Americans with Disabilities Act (ADA), in six weeks.

6.1 Emergence and Lifespan

Findings show numerous distinctions between the evidence regarding traditional groups in the literature and Facebook groups examined in this article regarding their emergence and lifespan. First, the literature suggests that it is difficult to pinpoint when traditional volunteer groups form and dissolve. Also, changes in functions of traditional volunteer groups may go unnoticed by researchers since observing these groups in action is a difficult task. In contrast, data from online groups provide specific time information regarding their establishment. Such rich data enable researchers and first responders to observe at what stage (e.g., response, immediate relief, etc.) new groups emerge, the purpose of these groups (e.g., information sharing, search and rescue), and their structural characteristics (e.g., size). A notable finding of Facebook groups is that one of them was created before the hurricane made landfall. This shows that volunteers can start getting organized even before a disaster hits. Numerous groups emerged in the initial impact, immediate aftermath, and post-disaster stages, which are indicators for the needs of disaster victims. For example, a group that was created for the needs of the deaf people is a symptom of the needs of a specific vulnerable group in the population. Similarly, a group that was created by survivors in a specific city or neighborhood three weeks after a disaster can provide evidence regarding the recovery (or problems with it) in that area.

Second, findings show that, unlike traditional volunteer groups, members of Facebook groups continue to communicate long after the Hurricane even though their communication may not be disaster related. These posts usually cover a wide variety of topics, such as the announcement of a happy day or recovery stories; they are sporadic, and they do not always align with group descriptions. The long-term effects of these online groups on their members (e.g., psychological relief and recovery) have not been explored in-depth and can be a topic of future research.

Third, there is always a possibility that online groups can change their name and description and repurpose themselves for a different goal. The data show that some groups changed their names multiple times as they transitioned from one stage (e.g., immediate aftermath) to another (e.g., long term recovery). Another interesting case was the transformation of a group from one disaster to another without changing its core function (i.e., response). The Facebook group named Cajun Relief Foundation (categorized as medical) was created on August 27, 2017, as a response to Hurricane Harvey. When Hurricane Maria made landfall in Puerto Rico in the following weeks, the group changed its name and description to reflect the activities they did in Hurricane Maria response. This suggests that online volunteer groups can emerge and function during a disaster. Then, they can go dormant when there are no disasters and become active when there is a need for them. This is an advantage for online volunteer groups because the technological structure of the group remains intact when the group goes dormant and can help the group become active faster in the future. Technology also helps these groups crystallize their structure and have a more formal appearance than other groups. This is helpful for local officials to identify these groups as potential partners and include them in the disaster management system as done in the case of VOSTs and volunteer and technical communities (van Gorp 2014). However, such groups should prove to be reliable and develop trust with emergency management agencies as barriers for establishing such partnerships are high (Fathi et al. 2020; van Gorp 2014; Reuter, Marx, and Pipek 2012; Reuter and Kaufhold 2018).

In summary, three types of groups were evident in the data with respect to their lifespan. The first type of groups emerge, function, and stop communicating after the disaster. The second type of groups emerge, function, and maintain communication at a low level and the topic of communication is different from the original group function. The third type of groups emerge, function, then go dormant when there is no disaster, and become active whenever there is a need. These groups change their group and description depending on the circumstances, and it is safe to assume their members have strong personal ties in the real world. Out of these three types of groups, the third type is best suited for partnering with the formal disaster management system, although such partnerships require time and trust to be established and flourish. Groups of the second type are expected to have some impact on their users’ psychological recovery and coping capability in the long term. The contributions of support groups for the psychological and physical wellbeing of disaster survivors is well documented in the literature (Bonanno et al. 2008; Bott, Ankel, and Braun 2019; Gargano et al. 2017; Palen and Hughes 2018); thus, the scope of the impact of Facebook groups on psychological resilience of survivors can be examined in future research. Groups of the first type are least likely to develop a strong relationship with their environment and maintain their function and structure in the long term. Their contribution to practitioners would be the unique information they generate in their posts. For example, groups whose names and functions are distinct from others (e.g., group named Deaf Hurricane Harvey Survivors) are likely to provide unique information for practitioners about what is happening in the field during and after a disaster.

6.2 Structure

Two types of measures were used for examining group structure: group size and number of active members in a given day. The results show a power-law distribution in group size within categories with four or more groups. That means one or two groups have a significantly high number of members while a large number of groups in the same category have a considerably small number of members.

There can be several reasons for the emergence of power-law distribution in group size. The first reason might be the purpose and location of the group. If a group was created for residents of a specific geographical region (e.g., Harris County, Houston, Corpus Christi), people outside of that region would not join in those groups. For example, groups created for the Houston area are more likely to be larger than those target smaller cities or counties.

The second reason might be the legitimacy of the founders of a group. Groups established by people affiliated with a specific organization (e.g., churches, Cajun Navy) might be more attractive to users with a particular interest in the functions of these groups. For example, a Facebook group that was created by a megachurch for collecting and distributing donations might attract people who know this Church and want to donate to victims.

The third reason might be the snowball effect on group growth. Network scientists suggest that when making a decision, people tend to choose things that are preferred by others Fowler and Christakis (2011). The same argument can be made for group growth as well. Groups that gained better traction in growth than others were more likely to grow faster than other groups because of the snowball effect created by users’ behavior.

An important implication of power law in group size distribution is that not all groups survive and continue to function. Groups that fail to attract members are likely to lose their ability to form a structure and carry out their functions. Groups with a relatively large member pool can form a structure depending on their functions. However, the size of a group can have important implications for its function as well. A large number of members can be facilitator or an inhibitor for a group. If a group’s function is to share information, more members might bring a higher volume of information flow, which can be beneficial for members. In most information-sharing groups, group creators set the rules for posting content (e.g., no political or offensive posts) and remove the content that does not comply with the rules. In other words, group creators play the role of facilitators. If a group’s function is action-oriented such as search and rescue and donations, however, coordinating a large member base might be challenging and may require a division of labor. In such a case, things such as specialized teams, a coordinating body, and alternative means of communication, and other elements of a hierarchical structure become more visible.

The size of a group also has important implications for its structure. Larger groups tend to be dominated by a small number of members, while smaller groups tend to have a flatter structure. These findings can imply multiple things. First, depending on a group’s function, users may join a group not to contribute or volunteer for anything but simply to increase their situational awareness. This is understandable for information sharing groups. Second, the relationship between group size and group capacity is more complex than one may expect. For example, more people in a Facebook group may not translate into more volunteers willing to contribute to the group’s functions. Third, the ratio of media group size and peak time active users can be interpreted as evidence for collective action problems in Facebook groups during disasters. Future research may focus on the conversation rate for groups’ sizes and the actual number of volunteers from those groups, and underlying structural factors contributing to the conversation rate. Should group domination be taken as a proxy (admittedly an imperfect one) for group hierarchy, one can argue that these structural characteristics of larger Facebook groups contradict those of traditional volunteer groups. The literature suggests a flat hierarchy for traditional volunteer groups, which may be challenging to observe and analyze in real-time.

6.3 Functions

The findings from the analysis of word frequencies show a significant level of functional overlap among groups. Regardless of the original group functions (i.e., the functions/goals declared in the group description), users seek information, ask for help, and discuss storm-related issues across almost all group categories. Additionally, the most common category group is information/communication. These pieces of evidence show that people use social media primarily for receiving and sharing information. This is not surprising since that is the primary purpose of social media platforms in daily use. Researchers who study social media’s role in disasters can focus on two things. First, they can focus on the information-sharing aspect of social media and analyze the messages being shared. For this line of research, Shannon’s entropy, as well as text analysis, sentiment analysis, and communication network tools can be useful. Second, researchers can separate other functions from information sharing and focus on those functions. For this second line of research, a more in-depth analysis of group function, members, and structure can be explored through qualitative methods such as interviews.

An important distinction point between online groups and traditional groups is the divergence in the scope of group functions even when they are in the same category. For example, groups that were labeled as information/communication in the data diverge from traditional information/communication groups regarding the specific functions they carry out. The literature suggests that groups that are classified as information/communication carry out functions such as registration of victims, displaced persons, and evacuees, translating, issuing, and sharing information and messages. Groups classified as information/communication in this article, however, predominantly function as information sharing channels. The type of information shared in these groups includes, but is not limited to, flooded areas, road closures, storm impact and damage, and requests for specific information. This divergence between online groups and traditional groups points out a more significant issue regarding group classifications.

The more significant issue with group classifications is that classifications mentioned in the literature do not adequately cover the functions of all online volunteer groups, and there is a need for better classification functions. For example, animal care was not mentioned among Twigg and Mossel’s categories. Classifications of online volunteerism offered by other scholars also use broad categories that do not capture animal care specifically (Abdulhamid et al. 2021; Palen and Hughes 2018; Reuter and Kaufhold 2018; Reuter, Marx, and Pipek 2012). It is important to emphasize that we do not suggest that animals in disasters is a neglected subject in the literature. Instead, we argue for more granular analyses and classifications that would capture a wider variety of volunteerism types. Another example is the Facebook group that was created for deaf people. The group served as a communication channel and provided resources and guidance to deaf individuals. Online groups like these show a significant diversification in online volunteerism, and there is a need for better classification of online groups for better addressing people’s needs.

Finally, discussions in groups evolve along with transitions from one phase of disaster to the other. In the first week of the disaster, most discussions were about immediate response related issues. As time passed, immediate response related needs were replaced by medium term relief (e.g., finding a rental apartment or hotel) and, later, long term recovery (e.g., contractors, construction) related discussions. The evolution of social media users’ response to disasters was documented in the literature (Wang, Ye, and Tsou 2016; Wang and Ye 2018), and our findings enrich the evidence regarding the spatial characteristics of online volunteering.

7.1 Limitations and Future Research

This article has some limitations. First, less than 50% of all groups identified were analyzed because Facebook changed data access protocols for third parties, and access to the data for other groups was not possible. Second, the data is limited to Facebook groups only. Online volunteer activities on Twitter, Nextdoor, Facebook Pages, etc., could enrich the findings. However, the nature of each social media platform is different, and data from each platform require different analytical tools. Combining data from multiple social media platforms might make analysis and interpretation very difficult. Third, the study mostly relies on quantitative tools for text analysis. Although quantitative tools are very useful in analyzing large amounts of data, a greater insight could be obtained if the quantitative analysis were complemented by qualitative analyses such as interviews or content analysis of posts. A study with such a broad scope would better fit a book project; hence, it can be a topic of future research.

This article opens doors for several other potential future research as well. First, future research can examine the long-term effects of Facebook groups on disaster recovery, and psychological relief can shed light on other aspects of online volunteerism. Second, future research can examine the evolution of volunteer groups on Facebook over time and identify to what extent these groups become active again in another disaster. Third, researchers can focus on specific group categories (e.g., special needs, animal care) and conduct a detailed analysis of specific topics being discussed in these groups. Such information can help detect the needs of the people in these groups and identify some lessons learned for disaster preparedness and response. Finally, there is a need for a more comprehensive and up-to-date classification of online volunteer groups. This can be accomplished by analyzing data from multiple disasters and triangulation.