#PuertoRicoSeLevanta: A Closer Look at the Language Used on the First-Year Anniversary of Hurricane Maria

Abstract

In September of 2017, Puerto Rico was hit by Hurricane María. Reactions to the hurricane were widely discussed on the social media site Twitter. The principal aim of the study was to examine the psychological processes of tweets one-year after Hurricane Maria and compare patterns of psychological processes within tweets originating from Puerto Rico relative to tweets originating from the continental United States and other countries. Also, researchers aim to geo-map the origin of tweets, as well as psychological processes exhibit in tweets world-wide. Researchers collected tweets (N = 1191) using #María, #PRSeLevanta, and #PuertoRico between September 20, 2018 to September 25, 2018. Linguistic Inquiry and Word Count software application was used to conduct a quantitative linguistic analysis of the sample of tweets, which classified the language utilized in the tweets across affective, social, psychological, and cognitive dimensions. A one-way between-groups multivariate analysis of variance investigated whether the affective, social, psychological and cognitive dimensions of the language utilized in the tweet varied as a function of where the tweet originated. Tweets varied in psychosocial dimensions as a function of where they originated, such that tweets originating from Puerto Rico used more language classified as positive emotion, cognitive processes, and referencing money in comparison to tweets originating elsewhere. These findings demonstrate how the particular use of words after a traumatic event can provide rich information about psychological processes and health outcomes experienced by individuals in the aftermath of disaster.

In September of 2017, Hurricane Maria – a category 4 storm – hit Puerto Rico. The hurricane devastated the island and was responsible for between 2,975 and 4,645 deaths (Kishore et al., 2018; Santos-Burgoa et al., 2018). Traumatic stress research frames natural disasters as experiences of “collective stress situations” (Santos-Burgoa et al., 2018). Disaster experiences are associated with a host of negative health outcomes among survivors, including posttraumatic stress, depression, substance abuse, and anxiety (Mesmar et al., 2016; Morganstein & Ursano, 2020; North, Kawasaki, Spitznagel, & Hong, 2004; Scaramutti, Salas-Wright, Vos, & Schwartz, 2019). Social media platforms such as blogs, Facebook, and Twitter have been described as a significant method of communication and coping for disaster survivors (Mesmar et al., 2016). Platforms such as Twitter provide a space for communication and management of collective trauma after a high-impact event, such as a natural disaster or mass shooting (Eriksson, 2016) and enable survivors to engage in community narrative formation (Harp, Grimm, & Loke, 2018), sometimes through the use of shared user-generated phrases called “hashtags”. Social media platforms provide a unique resource for post-trauma community support and collective narrative construction.

Potentially due to its ability to connect individuals with others experiencing similar hardships, Twitter is often a critical site of information-sharing during natural disasters. Social media platforms such as Twitter can use a mix of words, pictures, and videos to provide on-the-ground context about everything from damage to nature/property, physical and psychological injury, and area resources (Alam, Ofli, & Imran, 2018). Twitter facilitates the identification of at-risk communities and provides more targeted assistance in response to disasters (Chatfield & Brajawidagda, 2013; Shklovski, Palen, & Sutton, 2008; Takahashi, Tandoc Jr, & Carmichael, 2015). In the aftermath of Hurricane Maria, Twitter became a medium for Puerto Ricans living in Puerto Rico and elsewhere to share information about the specific challenges for recovery, to express gratitude to people providing assistance, to thank the media for keeping others informed, and to facilitate coping (Chatfield & Brajawidagda, 2013; García‐Ramírez, Bogen, Rodríguez‐Guzmán, Nugent, & Orchowski, 2020; Shklovski et al., 2008; Takahashi et al., 2015).

After minimal news coverage of Hurricane Maria in early 2017, news coverage on the mainland US spiked in proximity to the one-year anniversary (Center-for-Puerto-Rican-Studies, 2018) which may have inspired online discourse. Importantly, many people were still suffering the consequences of the aftermath at the year anniversary. One year after the hurricane, survey research indicated that impacts of the storm were ubiquitous, with 83% of Puerto Ricans reporting that they were affected by the storm, and a quarter of Puerto Ricans indicating that they still experienced day-to-day life disruption as a result of the hurricane (DiJulio, Muñana, & Brodie, 2018). Negative consequences included financial loss, housing problems, physical and mental health challenges (Zorrilla, 2017). This survey focused exclusively on Puerto Ricans living on the island. We are aware of one investigation that examined mental health outcomes of Puerto Ricans residing on the island relative to those who had been displaced after the hurricane to Florida (US). This study assessed posttraumatic stress symptoms six months following the hurricane and suggested that displaced participants were significantly more likely to endorse diagnostic levels of posttraumatic stress than participants who remained in Puerto Rico (Scaramutti et al., 2019). Analysis of Twitter data permits further examination of post-hurricane functioning as linguistic indicators of affect, cognitive framing, and community.

Given the utility of social media to serve as an outlet for coping with trauma, and the importance of narrative formation in understanding a traumatic experience, it is critical to examine whether individuals who experience a collective trauma utilize social media to cope over time, or whether social media provides an immediate outlet that is later supplemented or replaced with more formal resources. On the first-year anniversary of Hurricane Maria in Puerto Rico Twitter hashtags including #PuertoRico #PRSeLevanta (‘Puerto Rico Rise Up’), and #Maria emerged online, indicating that web users were discussing the impact of the hurricane after a full year of recovery efforts. The present study provides a linguistic analysis of tweet content collected at the one-year anniversary of Hurricane Maria. The primary aims of this study were: 1) Describe the sentiment and psychological processes of online communications that unfolded on Twitter one-year after Hurricane Maria in Puerto Rico, 2) Compare the patterns of sentiment and psychological processes that emerged in tweets from Puerto Rico relative to tweets from the mainland United States and international locations, and 3) Explore geo-map differences between emotions and psychological processes tweeted from Puerto Rico, US and international countries.

METHODS

Procedures and Data Collection

The current study employed a cross-sectional design. Researchers collected Tweets including the hashtags #PuertoRico #PRSeLevanta (‘Puerto Rico Rise Up’), and #Maria utilizing NCapture, a Google Chrome extension of the NVivo. Tweets were captured and downloaded for five consecutive weekdays approximately at 11am during the week of the one-year anniversary of Hurricane María in Puerto Rico. Only publicly-accessible tweets were captured and identifying information associated with tweet content was removed prior to analysis. The research was Institutional Review Board approved. The inclusion criteria of tweets for the present study were tweets that: tagged relevant hashtags, referenced Hurricane Maria in Puerto Rico, included latitude and longitude data, and were in English or Spanish.

The initial dataset contained a total of 2,401 tweets that mentioned with at least one of the selected hashtags (#PuertoRico n = 991, #Maria n = 833, and #PRSeLevanta n = 577). Researchers deleted tweets using following requirements: posts with information about countries for which Hurricane Maria caused destruction that were not Puerto Rico, tweets including business promotions, tweets incorporating only videos, and links and/or pictures. Tweets including business propaganda were excluded from the analysis to focus on the sentiment and psychological process directly related to Hurricane Maria. Tweets with videos, links and pictures were excluded since they did not included text, needed for the linguistic analysis. Consistent with Twitter analysis literature, retweets were removed from the analysis to avoid prioritizing the sentiments of one single high-status user over others (Bogen, Bleiweiss, & Orchowski, 2019). Tweets presented here have been slightly re-worded to prevent reverse-identification of Twitter users, methodology that is considered as “best practice” (Ayers, Caputi, Nebeker, & Dredze, 2018). The final dataset contained tweets with one or more of the following hashtags: #PuertoRico #Maria #PRSeLevanta. Of the entire dataset (N=1191), 796 tweets were posted in English and 395 tweets were in Spanish. In terms of location, 400 tweets originated in the mainland US, 702 originated in Puerto Rico, and 89 tweets originated in international countries.

Measures

Emotional Content

Linguistic Inquiry Word Count (LIWC) a computerized text-based analysis transforms words to psychologically-relevant percentages, was used to conduct a linguistic analysis of tweets (Pennebaker, Boyd, Jordan, & Blackburn, 2015). LIWC analysis was conducted at the tweet level; accordingly, LIWC categories represent the percentage of a tweet that reflects a particular sentiment. In other words, sentiment analysis of the hypothetical tweet “I’m furious” would show 50% negative affect at the tweet level. Original generated categories by LIWC developers include: linguistic dimensions (e.g., pronouns), other grammar words (e.g., verbs), psychological processes (e.g., positive emotion), cognitive processes (e.g., insight, cause, discrepancies tentativeness, certainty, and differentiation), perceptual process (e.g., see), biological process (e.g., health), time orientations (e.g. past focus), personal concerns (e.g., work), and informal language (e.g., swear words). LIWC analysis also produces a reliable and valid measure of narrative expression of emotion (Pennebaker et al., 2015). The tweets in Spanish were analyzed using the Spanish dictionary which relied on the 2001 LIWC version and the English tweets relied on the 2015 LIWC version. The Spanish dictionary includes more words than the English dictionary due to the higher rate of verb synonyms and gender specific conjugations (Ramirez-Esparza, Chung, Kacewicz, & Pennebaker, 2008). For the analysis, some output categories were omitted to permit focus on categories associated with post-trauma narratives and post-trauma functioning and physical health (Alvarez‐Conrad, Zoellner, & Foa, 2001). LIWC permits analysis of linguistic content and has shown associations with symptoms of posttraumatic stress disorder, seasonal affective disorder and attention-deficit/hyperactivity disorder (Coppersmith, Dredze, & Harman, 2014; Coppersmith, Dredze, Harman, & Hollingshead, 2015; Pennebaker et al., 2015). Prior research documents the utility of applying text analysis to examine tweets addressing mental health concerns, tweets relating to hurricanes and typhoons (Andrei, Elson, & Zarrella, 2015).

Location

Tweets were categorized based on their Twitter user’s IP address location to originate from: a) Puerto Rico - including the main island and municipalities of Vieques and Culebra, two smaller islands off the eastern coast of the main island; b) Mainland-US – including the mainland, Alaska, and Hawaii; c) International – comprised of other countries besides the mainland US and Puerto Rico. Geographical Information System (GIS) was used to visually map Twitter user’s IP addresses. Geographic analysis was conducted in ArcGIS, and output maps were constructed utilizing ArcMap. Map data was retrieved from The Natural Earth and US Census Cartographic Boundary Files, a public map dataset domain.

RESULTS

Emotional Content

To characterize psychological processes of online communications on Twitter one-year after Hurricane Maria made landfall in Puerto Rico, researchers used LIWC English and Spanish dictionaries to analyze percentage of the total words represented in theoretically meaningful categories. Analyses focused on the ‘tweet’ level, automatically providing a 280-character limit. Seventeen linguistic dimensions were included in the analysis: they (3^rd personal pronoun), affective process (e.g. happy), positive emotion (e.g. love), negative emotion (e.g. hurt), sadness (e.g. crying), social processes (e.g. talk), family (e.g. daughter), cognitive processes (e.g. cause, know, ought), time orientations (past, present, and future), money (e.g. audit), death (e.g. kill), time (e.g. end) (e.g. end), health (e.g. clinic), home (e.g. kitchen), and religion (e.g. church).

Table 1 shows the average percentages for LIWC dimensions. Tweets coming from Puerto Rico presented a higher content of cognitive processes, followed by social process, and present-focus. Tweets posted from the mainland US focused on social processes, followed by present-focus, and cognitive processes. On average, international tweets contained dimensions of cognitive processes, present-focus, and social processes. Online conversation discussed social process dimensions (e.g., referencing family, friend). Also, users shared present time orientation words (e.g., present tense verbs, present oriented words). Most users shared posts containing cognitive processes dimension (e.g., words about cause/effect).

Table 1.

Means, standard deviations and test of between-subjects for linguistic dimensions between location

	Puerto Rico (n=400)	United States (n=702)	International (n=89)
Linguistic Variable	Mean (SD)	Mean (SD)	Mean (SD)	df	MS	F	Sig.	partial η2
Linguistic Dimension
They (3rd personal pronoun)	1.27 (2.94)	0.73(2.04)	0.82 (1.84)	2	37.33	6.67	0.001*	0.01
Psychological Process
Affective process	5.55 (10.16)	5.53 (6.31)	4.89 (4.85)	2	21.79	0.36	0.695	0.00
Positive Emotion	4.33 (9.93)	3.05 (5.09)	2.52 (3.81)	2	249.34	5.04	0.007*	0.00
Negative Emotion	1.35 (3.83)	2.58 (4.40)	2.55 (3.71)	2	199.89	11.51	0.000*	0.02
Sadness	0.34 (1.16)	0.80 (2.31)	0.71 (1.56)	2	27.58	7.31	0.001*	0.01
Social process	7.07 (7.57)	8.30 (8.08)	6.52 (6.50)	2	264.11	4.33	0.013*	0.00
Family	0.44 (1.53)	0.45 (1.66)	0.45 (1.67)	2	0.04	0.02	0.985	0.00
Cognitive process	11.54 (11.58)	7.60 (7.46)	7.70 (6.17)	2	2045.64	25.36	0.000*	0.41
Time Orientation
Focus in the past	1.86 (3.91)	2.95 (4.48)	2.39 (3.55)	2	151.6	8.72	0.000*	0.14
Focus in the present	6.34 (6.08)	7.83 (6.85)	6.88 (6.14)	2	291.43	6.89	0.001*	0.01
Focus in the future	0.34 (1.52)	0.86 (2.36)	1.05 (3.29)	2	40.8	8.37	0.000*	0.14
Personal Concerns
Money	1.58 (3.30)	0.89 (2.42)	1.35 (3.00)	2	50.51	6.49	0.002*	0.01
Death	0.20 (0.89)	0.59 (1.81)	1.06 (2.28)	2	34.9	13.52	0.000*	0.22
Time	4.43 (7.45)	7.31 (8.61)	5.73 (6.44)	2	807.88	12.34	0.000*	0.20
Health	0.42 (1.70)	0.47 (1.70)	0.49 (1.48)	2	0.45	0.16	0.855	0.00
Home	0.53 (1.82)	0.36 (1.25)	0.58 (1.78)	2	4.92	2.17	0.115	0.00
Religion	0.31 (1.36)	0.29 (2.18)	0.14 (0.80)	2	1.08	0.31	0.734	0.00

^*p value < .05

Analysis based on content and Tweet Location

A one-way between-groups multivariate analysis of variance (MANOVA) was performed to explore sentiment and psychological process based on tweet location. The dependent variables were the seventeen LIWC dimensions and the independent variable was location (Puerto Rico, mainland US, international). There was a statistically significant difference between location on the combined dependent variables, F (34, 2344) = 6.04, p =.00; Wilks’ L = .85; partial ² = .08.

Analysis of variances on the dependent variables was conducted as a follow-up test to the MANOVA. Findings supported significant location differences at the p < .05 level in the scores of the variables they, positive emotion, negative emotion, sadness, social process, cognitive processes, time focus (past/present/future), death, and time orientation. There were no significant location differences observed for affective process, family, health, home, and religion.

Inspection of means scores using a Bonferroni adjusted alpha level of p < 0.002 indicated that tweets originating from Puerto Rico included significantly more of the pronoun they compared to tweets originating elsewhere. Tweets coming from Puerto Rico showed significantly more words related to positive emotion, cognitive processes, and money than tweets from the US or international locations. Tweets originating in Puerto Rico shared less content related to negative emotion, sadness, social processes, time focus orientation (past/present/future), death, and time than posts coming elsewhere.

Geographic Information System Analysis

GIS results presented the expansion of the online conversation around the anniversary of Hurricane Maria in Puerto Rico. Tweets frequently included cognitive processes. For example, one user from the mainland US shared: “@DavidBegnaud is always a great reporter. I love his style; his reports are usually done as selfies; I don’t know if he even has a crew. But he gets the stuff that everyone else misses (he was brilliant after #Maria in Puerto Rico). If David’s here, you know #Florence is bad.” This user reflects about the impact Hurricane Florence made in North Carolina around the same time the data was collected. The user is also comparing the media coverage made in Puerto Rico during Hurricane Maria and the efforts North Carolina. A user from Puerto Rico shared: “After #Hurricane #Maria many people were stranded without #Medicine. #PuertoRico produces #urgent care #medicine need for many #patients in the #US. FEMA (Federal Emergency Management Agency) is supposed to make sure that urgent care need meaning #medicine still quickly reaches patients.” A user from an international location shared: “Now, the governor of Puerto Rico explains his support to the government’s funded report about Hurricane Maria’s death toll.”

Many tweets reflected commentary referencing social processes. A tweet originating in Puerto Rico shared: “On this night many Puerto Rican brothers and sisters will remember #Maria with fear but even more the 60,000 families that still live under blue tarps, and Ricardo Rossello’s government and FEMA hasn’t done anything.” One tweet from mainland US said: “Katrina victims received 27 months of extended help. #Maria victims are only getting 11 months. #PuertoRico Hurricane Maria victims scramble for housing as FEMA vouchers expire.” An international tweet reflected: “#Hurricane #Maria hit women the hardest. A disaster reverses the advances of development & reverses progress for women as well. Therefore, women in more developed regions may end up even more disproportionately impacted by Hurricane Maria.”

Many tweets included words discussing time orientation, specially focusing in the present. A tweet coming from Puerto Rico stated: “A dark night today without electricity in #PuertoRico new reality after one year after #MARIA.” For instance, a tweet from mainland US shared: “Today marks one year since Hurricane #Maria killed nearly 3,000 people in #PuertoRico. More than ever, we remember the island and our efforts to lend a helping hand to our brothers and sister in law enforcement. A year later, Puerto Rico is still recovering.” An international tweet shared: “One year after #Irma and #Maria, hurricane affected government & people working hard to #BuildBackBetter, changing from #EarlyRecovery to long-term #resilience.”

Tweets mentioned words about positive emotion. A tweet from the mainland US said: “If I was to be born again and given the chance to pick the perfect place to grow up in and call home I would ask to do it over again in #puertorico my beautiful island, I love you Puerto Rico! You’ll always be home sweet home #maria.” A tweet from Puerto Rico reflected on positive emotions: “In #PR we go above and beyond when helping our fellows in need. I thank God and all the employees, physicians, artists, and peers that got together to bless many people.” A tweet from an international location posted: “Can’t wait for the day that I go back to Puerto Rico, that small island really made me feel at home in such a short space of time, with gods will I will be back very soon #PRSELEVANTA.”

DISCUSSION

On the year anniversary of Hurricane Maria, tweets with hashtags #PuertoRico, #Maria, and #PRSeLevanta, were observed to evidence differences in linguistic framing across Puerto Rico, mainland US, and other international locations. Relative to tweets originating elsewhere, tweets originating from Puerto Rico included more use of they, positive emotion, language related to cognitive processes, and content related to money. Additionally, relative to tweets originating elsewhere, tweets originating in Puerto Rico involved less negative emotion, sadness, social processes, death, and time orientation (focus on past/present/future).

It is interesting that tweets originating on the island evidenced more positive emotion and less negative emotion relative to those originating elsewhere. Given the scale of the Puerto Rican diaspora living in the mainland even prior to Hurricane Maria, it is likely that some of tweets originating outside of Puerto Rico were from Puerto Ricans (Hinojosa & Meléndez, 2018). It is possible that individuals who had chosen not to leave the island or who had returned by the time of the year anniversary were generally more positive and optimistic about Puerto Rico; indeed, it is reasonable to assume that people who did not feel safe to remain and/or returned would have less positive affect on the anniversary of the hurricane. This finding is also consistent with a study which reported greater posttraumatic stress disorder symptoms among Puerto Ricans displaced in Florida relative to those who remained on the island (Scaramutti et al., 2019).

It is important to consider the socio-political context of Puerto Rico and specific background on the anniversary of Hurricane Maria. Puerto Rico’s recovery from the hurricane had been slow, with delays in aid, blackouts, and the visible reminders of homes with FEMA blue tarps as a new hurricane season loomed (DiJulio et al., 2018; Zorrilla, 2017). Tweets originating in Puerto Rico were more likely to include use of they, possibly reflecting a sense that recovery and rebuilding progress was controlled by others outside of the island. Consistent with other research exploring wellbeing of Puerto Ricans after hurricane Maria (Capielo Rosario, Abreu, Gonzalez, & Cardenas Bautista, 2020). it is possible that this language may reflect dissatisfaction about how they (federal/state government, FEMA, the mainland)(Center-for-Puerto-Rican-Studies, 2018) have handled the recovery after the hurricane. This context may also be important to understanding the higher prevalence of tweets referencing money in Puerto Rico relative to other regions. Although congress had approved the allocation of millions of dollars to the recovery of Puerto Rico, delays in actual provision of the budget had multiple negative downstream effects as well as the assignment of a US-appointed Fiscal Board responsible for managing emergency aid (Center-for-Puerto-Rican-Studies, 2018; DiJulio et al., 2018).

It is somewhat surprising that tweets outside of Puerto Rico were more likely to reference death. It is possible that, for individuals on the island, death was experienced and discussed in real time during Maria and in the weeks to months that followed and so there was less focus on death at the time of the anniversary. In contrast, for individuals living elsewhere, commentary on lives lost may have been a way to describe the overall impact of the hurricane. Additionally, a revised estimate of the death toll of Hurricane Maria had been released by researchers on the mainland US, attributing a substantially higher mortality than those officially reported by the government (Kishore et al., 2018; Santos-Burgoa et al., 2018). It is possible that individuals outside of Puerto Rico, who had not experienced the scale of the morbidity, were especially surprised and distressed by these revised numbers.

Another difference observed in this research between individuals tweeting from Puerto Rico relative to elsewhere was that individuals tweeting from Puerto Rico appeared to be less focused on time (past or future) and negative emotions. It is possible Puerto Ricans who are still in the midst of a stressful event could show limited temporal focus because they are still “in” the event (Tausczik & Pennebaker, 2010); a focus on practical matters such as money, may be a better match during this ongoing stress for individuals who remain in Puerto Rico.

LIMITATIONS AND FUTURE RESEARCH

Although a strength of this research is the inclusion of tweets in both English and Spanish, entire academic disciplines have been dedicated to the study of differences in representations of the world through different languages. To our knowledge, this is the first study of linguistic features with Twitter to use two different languages. A limitation of this study was the absence of individual-level contextual data; we did not know which individuals tweeting from outside of Puerto Rico may have been displaced by the hurricane. Additionally, we don’t know if tweets came from Puerto Ricans who had relocated to the mainland before the hurricane and still have family members living in PR, and are using Twitter to express guilt, empathy, or helplessness. Given our focus on geospatial aspects of the anniversary, we were limited to examining tweets that permit geolocation.

Future research would be strengthened by efforts to complement publicly available content with individual level data. Other areas of future research include resilience differences between groups, in this case Puerto Ricans in the island and those leaving in the US. Future research cold also includes to focus on the importance of the education to create a presence of mental and behavioral health professionals in social media.

IMPLICATIONS

Implications of the present research span global public health and clinical domains. As natural disasters occur with increasing frequency and result in repeated displacement of individuals throughout the globe, scenarios such as the one described here are likely to be repeated. Given the current findings, there may be differences in the feelings and focus of individuals who remain in an environment after a natural disaster relative to those who live outside of that setting. In situations like Puerto Rico, whereas outsiders and the diaspora may want to focus on time (either hoping to predict a brighter future or reflect on the past) or on indicators of the severity of an event (such as the number dead), survivors who remain in disaster affected areas may be more focused on basic and imminent needs. Survivors who remain in a disaster-affected region may feel a sense of being a small community of we with a looming sense of they if survivors are asking for things (such as financial aid) that are out of sync with what outsiders appear to want to provide. Using geocoded data may help responders to better identify and more fully respond to the needs of communities. Clinicians may consider incorporating social media communication into therapy - inviting clients to share digital interactions that provide a window daily life and social interactions (Nugent, Pendse, Schatten, & Armey, 2019).

CONCLUSION

The present analysis reflects the utility of examining the way in which individuals discuss their experiences on social media in order to understand the ways in which individuals are coping with natural disasters. As a naturalistic methodology, LIWC allows researchers to understand sentiments and emotions that individuals express when communicating about natural disasters. Individuals directly affected by Hurricane Maria may turn to social media sites such as Twitter to express their feelings about the impact of the disaster. These findings highlight the enduring trauma associated with Hurricane Maria, as well as the need for continued support for those directly impacted by the storm.