Social media facilitate the creation and sharing of information. They are recognized as the key communication channels during crisis, such as Twitter, Facebook, Sina Weibo, and WeChat. They can provide a large amount of volunteered information, such as observation, opinion, feeling, and psychological demand,Reference Takahashi, Tandoc and Carmichael1-Reference Hadi and Fleshler3 which helps decision makers to take action more quickly through providing collective intelligence.Reference Finch, Snook and Duke4, Reference Hossain, Kam, Kong and Wigand5 A growing number of studies have explored the use of social media by emergency managersReference Jamali, Nejat and Ghosh6 and government officials,Reference Wukich7 as well as the general public.Reference Xu, Yin and Chen8 It is unclear what drives and affects the use of social media, especially during emergency.Reference Ma9 Understanding the reasoning behind social media usage will contribute to the optimal use of social media.Reference Ngamassi10
There are no time and space constraints in sharing and collaborating on social media.Reference Díaz, Carroll and Aedo11 But there is a significant difference in the social media usage pattern in terms of distance to the disaster site.Reference Simon, Goldberg and Adini12, Reference Kumar and Singh13 Citizens from affected areas turn to social media to share their experiences, seek help, and coordinate their response. Remote audiences participate in online interactions as spectators or volunteers who provide social support.Reference Starbird, Dailey and Walker14 Not all users participate in online discussion during disasters. It is more likely that affected people will share information on social media compared with users from the relative, which indicates that location of citizens and distance from the disaster place play an important role on social media usage. Different from traditional distance, the distance during disasters is a subjective distance, which is more complex and may change given the situation. For example, when a disaster happens, 1 minute for affected people feels much longer than that for those who are farther away from the disaster.Reference Chandran and Menon15 In this paper, the psychological distance is introduced to measure this kind of subjective distance because it changes people’s mental representation of an event.Reference Huang, Burtch, Hong and Polman16
The concept of psychological distance, proposed by Beckerman in 1956,Reference Trope and Liberman17 measures the distance between distal object and our location based on an individual’s direct experience and subjective judgment.Reference Liberman, Trope and Stephan18 Relating to psychological distal objects helps manage humans’ life effectively and has the ability to plan and coordinate with other individuals.Reference Liberman and Trope19 Psychological distance as a composite indicator is quantified by the integrated distance of geographic distance, time delay, relationship between events and people, and the probability of similar events occurring.
This paper focuses on exploring the impact of location and distance on social media usage patterns. Online engagement means that Internet users go online to participate in discussion by posting messages, comments, and pictures on a political or social issue on a website or using their blogs to explore information, which is adopted to quantify the social media usage behavior of the public.Reference Smith, Schlozman and Verba20 It is usually expressed as publishing, liking, commenting, or sharing messages.Reference Kortelainen and Katvala21 It is measured as the total number of articles,Reference Knight22 the number of tweets,Reference Van Lent, Sungur and Kunneman23 or the number of comments per user in response to events.Reference Strekalova24 In this study, online engagement as the macro characteristics of users’ social media usage behavior is measured by the number of original messages related to the event on social media. Further, we explore the relationship between online engagement and event location to improve our understanding on citizens’ social media usage behavior.
RELATED WORK
Online Engagement and Psychological Distance
Social media usage behavior includes posting messages and forwarding them,Reference Xhafa, Patnaik and Yu25 as shown in Figure 1. When disaster happens, social media users become aware of the situation, and some of them will post various messages online. Important information can then be propagated by their followers to a considerable extent.Reference Liu, Wang and Wu26 Most current studies focus on the forwarding behavior. However, as the source of information, the behavior of information posting should be investigated. This study focuses on social media usage behavior from the information posting behavior.
The Factors in Information Posting and Propagation Process.
A relationship exists between online behavior of the public and their psychological distance of an event.Reference Trope and Liberman17 Psychological distance has been used to test its effects on consumer online reviewsReference Huang, Burtch, Hong and Polman16 and on the linguistic concreteness of natural language use.Reference Snefjella and Kuperman27 Psychological distance also helps enhance the understanding of users’ online behavior during disasters. People may evaluate whether the disaster is physically close or distant, just occurring, or may happen for a while, pertains to themselves or others, and may happen again, to affect their behavior on how to respond to the disaster.Reference Carmi and Kimhi28 Lent et al.Reference Van Lent, Sungur and Kunneman23 observe that people might not fear being infected with the epidemic unless it became psychologically close. They examine that psychological distance increases when the Ebola outbreak becomes physically close. Similar to Lent et al.’s research, our study aims to explore the relationship between online engagement of the public and psychological distance during disaster.
Measurement of Psychological Distance
Psychologically distal objects are regarded as the things that do not happen to us right now, or may happen somewhere else or may happen to other people.Reference Liberman, Trope, McCrea and Sherman29 According to this definition, psychological distance is quantified by 4 dimensions: spatial distance, temporal distance, social distance, and likelihood (probability or hypotheticality), along with the dimensions of time, space, social network, and probability.Reference Trope and Liberman17, Reference Liberman and Trope19 This study focuses on the first 3 dimensions. The fourth dimension, probability, is not considered here.
Spatial distance can be measured by the geographic distance to the reference point, which is an important factor in users’ online behavior. People tend to make different decisions on the objects that are spatially near or far away.Reference Fujita, Henderson and Eng30 Temporal distance is the distance to the reference point in the temporal dimension, which is found to influence the perceived hierarchy of incoming threats.Reference Carmi and Bartal31 Social distance is defined as the degree of the personal closeness, personal involvement, or the willingness to engage in the relationship with the person.Reference Ayazi, Lien and Eide32 The measurement of each dimension is a challenging problem. Some examples are provided in Table 1.
Some Examples of the Measurement of Each Dimension of Psychological Distance
Based on the measurement of each dimension, the effects of psychological distance on humans’ behavior will be considered quantitatively. However, 2 controversial questions should be answered first. How many dimensions should be considered? Are they considered simultaneously? Some researches (in Table 2) illustrate these answers. As shown in Table 2, 2 methods (questionnaire survey and regression analysis) are applied to test the effects of the psychological distance from various dimensions.
Some Examples of the Quantitative Researches of Psychological Distance
METHODOLOGY
Regression analysis is conducted to explore the relationship between online engagement of the public and psychological distance during disaster. Three dimensions of psychological distance (spatial, temporal, and social distances) are selected to test the relationship. A linear regression model is applied to explore the relationship between online engagement and psychological distance. To remove the distributional skewness, online engagement is log-transformed. We assume that there is a linear relationship between the log-transformed online engagement and 3 dimensions, as shown in Formula 1.
$$\eqalign{ & \log (Online\,Engagement) = {a_0} + {a_1} \cdot Spatial\,Distance + {a_2} \cdot \cr & \,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,Temporal\,Distance + {a_3} \cdot \cr & \,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,Social\,Distance \cr} $$
where a 0 is the constant, a 1 is the coefficient of spatial distance, a 2 is the coefficient of temporal distance, and a 3 is the coefficient of social distance. We also propose the relationship between psychological distance and various dimensions can be measured by a linear model as Formula 2.
$$\eqalign{ & Psychological\;Distance = {b_1} \cdot Spatial\;Distance \cr & + {b_2}\; \cdot \;Temporal\;Distance + {b_3} \cdot Social{\mkern 1mu} Distance \cr} $$
where b 1 is the coefficient of spatial distance, b 2 is the coefficient of temporal distance, and b 3 is the coefficient of social distance. At last, the relationship between the log-transformed online engagement and psychological distance may be defined by a linear model as Formula 3.
$$\log (Online\ Engagement) = {c_0} + {c_1} \cdot Psychological\ Distance$$
where c 0 is the constant, and c 1 is the coefficient of psychological distance.
The regression analysis is used to determine these coefficients.
CASE STUDY
The Tianjin Port explosion is used as an example to analyze the impact of the location and distance on citizens’ online engagement from the perspective of psychological distance.
TIANJIN PORT EXPLOSION INCIDENT
The Tianjin Port explosion, a serious safety crisis, happened in China in 2015. A hazardous material warehouse, located in the Tianjin Binhai District, exploded on August 12, 2015. There were 2 large blasts, as shown in Figure 2. The first blast occurred at 11:34 PM on August 12, 2015. After 30 seconds, the second one occurred.Reference Huang and Zhang37, Reference Chung and Kim38 The explosions caused a great damage to the local community; 165 people were killed, 798 were injured, and a large number of buildings were damaged. Cumulative loss was up to US$9.923 billion by December 10, 2015. The explosions resulted in fierce discussion on social media. Many users expressed their opinion, experience, needs, and concerns on Sina Weibo and WeChat to provide a large amount of information.
Location and Local Map of the Tianjin Port Explosion.
Measurement of Online Engagement
Online engagement is measured by the number of original messages related to the event on Sina Weibo (one of the most popular microblogs in China). The keyword-based method is used to search and collect the disaster-related social media data based on the Sina Weibo API (the open application programming interface). The disaster-related data were collected through 2 sets of key words in Chinese, including “Tianjin” and “explosion” or “Tianjin explosion.” The data were collected from August 12 to August 27, 2015 (16 days). Only Chinese microblogs were collected and there was no location constraint in the process of data searching. The messages posted in China or overseas matching the key words were added to the dataset. The volume of the dataset was 27 635. Searching results give the detailed information of the data, which are divided into 3 categories: (1) the user’s information, including the nickname, ID, region, birthday, short introduction, and labels; (2) the information on the message’s text, including content, hashtag, post time, visual information (eg, pictures or videos), and URL; and (3) posting media (eg, mobile phone, computer).
The noisy data should be filtered from the dataset, because they can easily prevent crisis managers from acquiring the meaningful information. Some data collected through the machine key word-matching may not be related to the disaster. The manual classification is performed to filter unrelated data.Reference Deng, Liu and Zhang39 Twenty graduate students participated in labeling the collected data into the disaster-related or unrelated categories. Each student read and labeled the message manually because each microblog message was a short text. The message label unrelated for the first time is double-checked by the second student in case of a wrong classification. If both students make the same conclusion, the message is removed. But when the second student label is related, the message is added to the dataset. The region of the author, message content, and post time were critical in this study. The data with incomplete information therefore were removed from the dataset. Finally, 5171 pieces of the messages were labeled unrelated in all 27 635 messages; 22 464 pieces of messages remained. All valid data were with the geographic location of the author (user’s region) and post time.
In order to measure the online engagement of different areas, we used the spatial aggregation analysis on disaster-related data based on the geographic location. The messages posted by the citizens in the same province were classified into 1 category. The data from overseas were removed from the dataset. Online engagement was non-dimensional.
Measurement of Psychological Distance
In this study, we used spatial distance, temporal distance, and social distance to analyze the relationship between online engagement and psychological distance. Psychological distance is considered at the provincial level. In the Tianjin Port explosion, the reference points of spatial distance, temporal distance, and social distance are Tianjin, the time of the explosion, and the citizens in Tianjin, respectively.
Spatial Distance
Spatial distance measures the geographic distance from the reference point (Tianjin city) to a study location. In order to test the effects of psychological distance on the restaurant review online, the geographic distance is defined as the distance from the reviewer’s place to the reviewed restaurant.Reference Huang, Burtch, Hong and Polman16 Similar to this study, it is assumed that the region in the user’s profile is the location of the posting messages. However, the region may be the place of residence, and some people may not be in this place when disaster happens. For example, the users may be on a business trip to Tianjin. The users will perceive greater risk and have a higher probability to post messages about the explosion. Because the real-time geographic data are difficult to obtain, it is assumed that all users are in the registration place on Sina Weibo when the explosion occurs.Reference Simon, Goldberg and Adini12
The circle distance, the shortest distance between 2 points on a circle, is widely used to calculate the spatial distance.Reference Huang, Burtch, Hong and Polman16 The reachability between 2 locations is not represented in this method. In order to describe the reachability between other locations and Tianjin, the distance is measured based on the transportation route between 2 locations. In this paper, the spatial distance is calculated by the shortest highway distance between 2 locations that are the province of the user’s region and Tianjin. Spatial distance between 2 locations is quantified based on the route data from Google Maps. The distance between different provinces ranges from 1 km to 3671 km (M = 1393.23, SD = 880.99). To remove the distributional skewness, spatial distance is defined as the log-transformed form of the distance. The dimension of the spatial distance is log(km). If the user’s region is in Tianjin, the spatial distance is assigned as zero.
Temporal Distance
The temporal distance is quantified by the time when users post their event-related messages in reference to the explosion. For users, the temporal distance is quantified by the time difference between disaster time and their first messages. For a province, the data are the messages posted by all citizens in the province, and temporal distance of the province is the average time for all data.
There is a question on how to choose the number of messages used in the calculation for each province. In this study, we have compared 3 widely used methods, which are the average of all data, the value of the first data, and the average of a fixed number of the data. The comparative analyses are shown in Figure 3. When all data are used, the values range between 120.7 and 148.6. There is no significant difference for the provinces. The reason is that there are different numbers of microblogs on the explosion in different provinces, and the different number of microblogs reduces their difference of temporal distance. For the messages in the same province, temporal distance of each user fits the normal distribution. Though the users in different provinces have a quick response on social media, the average value becomes very large because there is a large number of the messages produced, and some users post messages after a long delay. For the Qinghai province, the users have a relatively slow response, and the average value becomes small because the total number of messages is small. The range of temporal distance of different provinces is from 0 to 6.48 hours if the first message is used. Uncertainty is obvious by using this method. Therefore, the same number of the data in different provinces is selected to quantify temporal distance. Becaues there are only 14 messages in the Qinghai province, temporal distance is quantified as the average of the first 14 data in all provinces, which gives a better explanation for temporal distance than the previous methods. The dimension of the temporal distance is hours.
The Scatter Plots of Different Measurements of Temporal Distance.
Social Distance
The social distance between a province and Tianjin is measured based on the interaction activity between 2 places. A simple kind of interaction is transportation. Planes, trains, and cars are the 3 most popular ways of transportation between 2 cities in China. When the distance is long, people tend to choose planes or trains, but driving a car or taking a train may be their preference in the short distance. The threshold in this study is the distance that requires 4 hours to drive a car on the highway, and planes and trains will be used beyond 4 hours. We first discuss the measurement of social distance in long-distance. The flight information is obtained from the Ctrip website (http://www.ctrip.com/), the official website for air ticket booking. The train information is obtained from the website of 12306 (http://www.12306.cn/mormhweb/), the official website for train ticket booking. For each province, we only count the number of flights and trains between the provincial capital and Tianjin. The number of trains is easy to measure from the website of 12306. However, the flight information needs further processing. Searching flight information on the Ctrip, we can obtain the number of direct flights and transit flights. However, not all transit flights are taken into consideration. The transit flights that need to transfer 2 or more times are ignored. The transit flights that stop over 5 hours are also ignored. To measure the weights of flights and trains, we normalize the number of flights and trains separately. Social distance is defined as the reciprocal of the result. The equation is shown in Formula 4 in long-distance.
$$s{d_i} = {1 \over {{{n{f_i}} \over {sf}} + {{n{t_i}} \over {st}}}}$$
where sd i is the social distance from the province i to Tianjin; nf i is the number of flights from the province i to Tianjin; sf is the sum of flights from all provinces to Tianjin; nt i is the number of trains from the province i to Tianjin; st is the sum of trains from all provinces to Tianjin. The social distance of Tianjin is set to be 1. The social distance is non-dimensional.
Trains and cars are used in the short distance. The social distance is measured similarly to the situation in long-distance. We can obtain the cars’ information from the website of China Highway (http://www.china-highway.com/).
Regression and Analysis Results
The relationship between social media usage behavior of the public and several kinds of distance is shown in Table 3.
Values of All Variables and Their Descriptive Statistics
Online engagement ranges from 14 to 3272 (M = 618.74, SD = 696.37), which is the log-transformed for further analysis.
Correlation Analysis
For the regression analysis, we analyze the correlation between the dependent variables and independent variables using SPSS 20.0 (IBM Corp, Armonk, NY). The results are presented in Table 4. All correlations between the log-transformed online engagement and psychological distance are significant at the 0.01 level (2-tailed). It proves that there is a significant relationship between social media usage and psychological distance.
Results of Correlation Analysis
* Correlation is significant at the 0.05 level (2-tailed).
** Correlation is significant at the 0.01 level (2-tailed).
Regression Analysis
The relationship between online engagement and psychological distance is regressed by SPSS 20.0 (IBM Corp, Armonk, NY). The regression model fits the relationship relatively well (R 2 = 0.718, adjusted R 2 = 0.687), and the coefficients of all variables are significant at the 0.05 level, as shown in Table 5.
Regression Results
* P ≤ 0.1.
** P ≤ 0.05.
*** P ≤ 0.001.
The model is shown in Formula 5.
$$\eqalign{ & \log (Online\,Engagement) = 3.515 \cr & \,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\, - (0.233 \times Spatial\,Distance \cr & \,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\, + 0.007 \times Temporal\,Distance \cr & \,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\, + 0.005 \times Social\,Distance) \cr} $$
Based on the regression analysis, there is a multivariate linear relationship between log-transformed online engagement and psychological distance. Social media usage is negatively correlated with spatial distance. It means that people will be aware of low risk if they are far from the place. People who live close to the event location tend to pay more attention to the disaster. Social media usage is negatively correlated with temporal distance, which indicates people may pay less attention when the disaster has taken place for a long time. Social media usage is also negatively correlated to social distance, which indicates that citizens in an area that interacts rarely with Tianjin may pay less attention to the disaster.
We are aware that the value of R 2 is not very high in the regression analysis. The reason may be that some provinces have similar psychological distance, but their online engagement is quite different. For example, the spatial distances from Shanghai and Shaanxi to Tianjin are close, but online engagement data of 2 provinces are 1003 and 467, respectively.
Based on the characteristics of Formula 5, psychological distance is measured by a linear model, as shown in Formula 6. Psychological distance is positively correlated with 3 dimensions. Then the relationship between online engagement and psychological distance is described by Formula 7. The model indicates that there is a linear relationship between log-transformed online engagement and psychological distance. Social media usage of the public is negatively correlated with psychological distance, which is consistent with that in Liberman.Reference Liberman, Trope, McCrea and Sherman29
$$\eqalign{ & Psychological\,Distance = 0.233 \times Spatial\,Distance + 0.007 \cr & \,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\, \times Temporal\,Distance + 0.005 \cr & \,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\, \times Social\,Distance \cr} $$
$$\log (Online{\kern 1pt} {\kern 1pt} Engagement) = 3\hbox{.}515 - Psychological{\kern 1pt} {\kern 1pt} Distance $$
We proved that there is a linear relationship between social media usage of the public and psychological distance during a major disaster like Tianjin Port explosion, which happened suddenly and caused severe damage to the city areas. The relationship model in this study can be applied to the similar disasters. Physical and social distances between 2 locations can be calculated in advance. Once a similar disaster happens, the location of the disaster can be determined, and physical and social distances can be obtained directly. Online engagement is then the function of temporal distance. The social media usage of the public can be assessed quickly after a disaster. In other events, such as a hurricane, we should consider the probability of the event to measure psychological distance. This study helps increase our understanding of social media usage to support decision-making.
CONCLUSION
This study helps improve the understanding of the impact of location and distance on social media usage. We uses the Tianjin Port explosion as an example to collect multi-source data. Citizens’ online engagement is estimated by the disaster-related social media data collected from Sina Weibo. The impact of location and distance is measured using 3 dimensions of psychological distance. Based on the correlation analysis and regression analysis, the following conclusions can be made.
First, location and distance have a great impact on online engagement of the public during disasters. Three dimensions (spatial, temporal, and social distances) of psychological distance are used to evaluate the location and distance. Citizens who live close or have more interaction with the location of the disaster tend to pay much more attention to the disaster. Second, there is a linear relationship between social media usage behavior and psychological distance. Online engagement is negatively correlated with psychological distance. Third, psychological distance can be measured by a linear model using 3 dimensions. Psychological distance is positively correlated with spatial, temporal, and social distances. Fourth, a linear relationship exists between social media usage behavior and psychological distance, and the intensity of social media usage decreases as psychological distance increases.
This study gives a new insight into emergency management. On one hand, it helps for social media users to participate in the response process. On the other hand, it helps emergency managers improve the understanding of online engagement influenced by location and distance.
Acknowledgments
This work is supported by National Key R&D Program of China (No. 2017YFC0803300), National Science Foundation of China (Grant Nos. 91646201, 71704091, U1633203), the China Postdoctoral Science Foundation (Grant No.2017M610917), and the China Scholarship Council (CSC) (No. 201606210380).
