1 Introduction

Social media platforms, such as Reddit, X (formerly Twitter) and Facebook, have become valuable sources of real-time human interaction data, attracting researchers seeking insights into various socioeconomic and psychological issues [1, 2]. These platforms offer spaces for users to discuss their attitudes, behaviors, opinions, and beliefs on topics such as health communication, vaccine misinformation, the spread of diseases like influenza, and self-remedy. Reddit, known for its anonymous user environment and diverse subreddits based on interests, played a significant role in facilitating discussions about the emergence and impacts of COVID-19.

As the pandemic, which has claimed approximately 6.9 million lives worldwide [3], continues to unfold, there are growing concerns about its long-term socioeconomic consequences, including post-pandemic health issues, job loss, and economic disruptions. While extensive research has focused on the immediate impacts of COVID-19, there is a relative lack of knowledge about its longer-term effects. Research into the first year of the pandemic determined topic modeling as a helpful method in exploring leading topics in COVID-19 discourse on Reddit [4]. This study aims to extend research into the following year, 2021, examining user discussions on the darker consequences of the pandemic to shed light on the post-pandemic “new normal.”

Understanding these discussion patterns is essential for tailoring support and interventions for individuals facing specific post-COVID-19 challenges. As the virus mutates and becomes endemic, this research provides deeper insights into the lived experiences of people, contributing to the broader understanding of the ongoing impact of COVID-19 on society.

2 Previous works

COVID-19 strained an already burdened healthcare system, endangered immunocompromised individuals, and led to delayed treatment of other health concerns, with potential consequences for future fatalities [5]. Beyond the virus’s immediate direct health consequences such as infections and death, COVID-19 resulted in indirect detrimental effects globally, including the psychological repercussions of future pandemic and the need to adapt to a pandemic-affected lifestyle [6]. Extensive research showed the considerable effects of COVID-19 on mental health [7,8,9,10,11,12] Some studies focused on specific populations at potentially greater risk, like students, older adults, women, nurses, and people of color [9, 11]. For example, university students grappled with pervasive anxiety (97%), largely linked with COVID-19-related stressors such as economic concerns, disruptions in daily life, delays in educational outcomes, and social support [9]. The pandemic demonstrated heightened depression and anxiety rates among older adults, with women having a higher rate than men [11]. Frontline healthcare workers, particularly nurses, experienced unique emotional and psychological stressors and turned to social media to express their anger, anxiety, sadness, and exhaustion [10, 13].

Research also uncovered the association between economic factors like unemployment and mental health issues. A projection analysis demonstrated how increased unemployment due to COVID-19 could lead to increased death by suicide and recommended preventative economic strategies like wage subsidies, small business governmental support, and community-based programs [14]. The panic buying phenomenon that marked the early months of COVID-19 was explored using a crowd psychology theoretical perspective to determine its consequences [15]. Other research has explored gendered trends in COVID-19 consequences, finding economic concern for women [16].

Studies harnessed the power of social media, with Facebook and X (formerly Twitter) emerged as the primary platforms for data collection [8, 10, 12, 13]. In the U.S., surveys distributed via a social media campaign during the pandemic unveiled alarming statistics, including a 29% increase in alcohol use among adult respondents, with 47% experiencing COVID-19 associated anxiety symptoms and 30% experiencing symptoms of depression [8]. A Twitter-based study tracking the first 5 weeks of infection outbreak observed trends in real-time emotional fluctuations, including anxiety, sadness, anger, and positive emotions, mirroring the evolving impacts of pandemic [12]. This study not only documented sustained emotional expressions of anxiety, sadness, anger, and positivity but also explored collective emotions, including “changes in the quality, magnitude, and duration of the emotional experience,” within the Twitter community [12]. Another study utilizing Twitter data detected increased signals of depression due to COVID-19 [17]. Additionally, an examination into gendered trends in COVID-19-related discourse on Twitter highlighted predominant discussions about women focused on medical bills and domestic violence concerns [16].

Although the pandemic appears to be waning, a growing awareness has emerged regarding the extensive and long-lasting ramifications of the pandemic. What is especially salient in the current literature is how these COVID-19 consequences intersect, often resulting in mental health consequences. It is also evident that these ramifications persist and encompass a wide spectrum of societal dimensions, including healthcare, economics, mental health, and social dynamics. It is therefore crucial to understand the enduring effects of COVID-19, not just as a historical record but as a source of insights to inform evidence-based decision-making and policy formulation, enabling governments and institutions to better prepare for similar crises in the future. As we transition into a post-pandemic world, the insights gained from this research will play a pivotal role in devising strategies to facilitate preparedness for future crises and building a more resilient and adaptable society.

3 Material and method

AskReddit is the second largest subcommunity, or “subreddit”, on the social media website Reddit.com, with about 40 million members. A question—“What are some of the darker effects COVID-19 has had that we don’t talk about?”—was posed on AskReddit. This particular post garnered a substantial response of around nearly 24,000 comments and is presently archived on Reddit, preventing any further comments. For the purposes of this study, responses posted between the original posting date and October 30, 2021, amounting to a total of 23,957 posts, were used.

Initially, we (the research team) employed a qualitative approach, wherein two human coders engaged in the qualitative coding of the posts. However, it soon became evident that qualitative coding of these posts was an exceedingly human-intensive and time-consuming process, requiring coders to engage in close and thorough examination of this large dataset with over 2500 pages and 1,061,824 words. This raised the question if it was feasible to manually read and code the entire dataset by human coders. To overcome this challenge posed by the sheer volume of data, we turned to a machine learning approach: probabilistic topic modeling. Topic modeling is a popular natural language processing tool that analyze large volumes of text data to extract the underlying topics or themes within the datasets [18]. The choice of topic modeling was driven by the fact that it shares commonalities with qualitative approaches, as both approaches focus on identifying patterns or themes within data [19]. Human judgement is essential to both qualitative approaches and topic modeling [18, 19]. It is worth noting, however, that both approaches have their strengths and limitations. While qualitative approaches are time- and resource-intensive when it comes to analyses of a large corpora of data, topic modeling, although less time consuming, lacks the nuanced insights that a human can provide. In the following paragraphs, we explain both qualitative approach and topic modeling. It is important to clarify that qualitative coding was not integrated with topic modeling. Since topic modeling is an unsupervised machine learning approach, we did not use qualitative codes in training the topic model.

For the qualitative analysis, we employed a stepwise systematic approach, following the Gioia methodology [20], a widely used grounded theory approach in management literature [21]. The Gioia methodology has been described in detail in a paper by Gioia, Corley, and Hamilton [20]. In brief, this methodology consists of several steps. First, researchers generate first-order categories based on their initial understanding of respondents’ terminology. Then, they identify similarities and differences in these categories and stipulate phrasal descriptions to these categories. This is followed by the creation of second-order themes, where researchers act as knowledge agents and seek theoretical dimensions that inform their understanding of the phenomena being studied. Finally, researchers seek if it is possible to distill the second-order themes further into aggregate dimensions [20].

In this study, we followed the Gioia methodology. Initially, two team members of the research team completed initial coding (first-order categories) of 500 posts each based on their understanding and terminologies used in the Reddit posts. Based on these first-order categories, the research team members developed a codebook independently. Subsequently, the team members collaborated to identify similarities and differences in their codes, grouping them accordingly. At this step, both the team members observed if the codes revealed or explained the phenomena that was being observed and assigned phrasal descriptions to the grouped codes. They revisited their respective posts and assign themes based on the final codebook. An independent research team member, not involved in the initial coding process, reviewed the initial coding schema, codebook, and theme assignment for accuracy and validation. Finally, two team members evaluated the themes by analyzing the codes, phrasal descriptions, and existing literature to determine whether the results had precedents and if any new concepts emerged in the process. Table 1 provides information about codebook, number of codes for each theme, and examples.

Table 1 Qualitative data analyses codebook
Full size table

Topic modeling was performed on all responses, utilizing Latent Dirichlet Allocation (LDA) applied over n-grams [22, 23]. LDA is a popular technique in natural language processing and employs a probabilistic modeling approach to uncover themes or topics within a large collection of documents or texts [18]. It assumes that each document in the collection is a mixture of various topics, and each topic is associated with a distribution of words. Through statistical modeling, LDA helps to identify these underlying topics, hidden semantic structures, and their associated word distributions without needing predefined categories or labels [18].

Before performing topic modeling, the text data was preprocessed to enhance the quality of the analysis. Preprocessing involved cleaning the data to remove irrelevant content, such as special characters, URLs, stop words and formatting. The text data was then tokenized into words and phrases. N-grams (n = 1, 2, 3) were obtained to capture multi-word expressions and phrases, and subsequently, LDA, a topic modeling technique, was employed over several combinations of unigrams (n = 1), bi-grams (n = 2), and tri-grams (n = 3). To ensure the quality and relevance of the topics generated, quantitative (coherence score) and qualitative evaluations were performed for these combinations. Topic coherence scores suggested the model trained over the combinations of bigrams and trigrams provide quantitatively better results, with 30 topics (see Fig. 1).

Fig. 1
figure 1

Coherence values for topic models. Models were trained with different number of topics, for combinations of unigram+bigram_trigram (UBT) and bigram_trigram (BT). The BT topic model with 30 topics provides higher coherence values for best representation

Full size image

An independent coder was assigned to evaluate these topics. They first reviewed each topic and its related keywords. Next, they examined the corresponding samples clustered within each topic. The coder reviewed and categorized the topics based on their content and meaning, ensuring the accuracy of the algorithm-generated topics. They chose phrases representative of the samples in relation to each topic, independent of the qualitative codes derived earlier. An investigator, who was not involved in the coding process, performed a final review of the topics and evaluations provided by the coder to maintain consistency and reliability.

4 Results

While we employed both qualitative analysis and topic modeling in our study, it is important to clarify that these methods were applied separately. As mentioned earlier, we used topic modeling because qualitative analysis of such a large dataset would have been too time- and labor-intensive. Specifically, topic modeling was performed on the entire dataset, whereas qualitative analysis was conducted only on the first 1000 posts. Consequently, the codes generated from topic modeling (see Table 2) differed from those produced through qualitative analysis (see Table 1). Since these two analyses involved different scopes and methodologies—qualitative analysis for in-depth and nuanced insights, and topic modeling for broader and overall coverage—we present the results of both analyses separately. The results of the qualitative analysis are presented first, followed by the results of the topic modeling.

Table 2 Results of topic modeling
Full size table

4.1 Qualitative analyses

Qualitative analyses of posts revealed twenty second-order themes (Table 1). Approximately 16% of the posts were either satirical, deleted, unrelated, duplicated, or unintelligible comments and, as a result, were excluded from further examination. For parsimony, we discuss the top five themes that emerged in our qualitative analyses.

4.1.1 Mental health

Users primarily discussed some of the serious consequences of COVID-19 on their mental health (13%). Some users reported experiencing exacerbated mental health issues, including depression, anxiety (both social and generalized), phobia, thoughts of suicide and self-harm, and heightened fear of death or another pandemic. For example, a user indicated how COVID-19 aggravated their agoraphobia—“I have agoraphobia and the lockdowns knocked me out of my routine of going out and made me afraid to leave the house again.” Others discussed their friends’ experiences or made observations and predictions on how COVID-19 have or might impact people. For example, a young professional stated that “many of the people in my circle (friends, acquaintances etc.) are also professionals—in the last ~ 18 months, I have personally known 7 people who have committed suicide likely because they are (in no particular order) [1] overworked, [2] isolated, [3] feeling trapped. Make sure to check in on your friends—and if YOU need help, there are lots of resources.”

4.1.2 Death and dying

A substantial number of users directed their attention to discussions regarding direct and indirect deaths resulting from COVID-19 (10%). Discussions on direct deaths focused on vulnerable groups such as children, people from low socioeconomic status, nursing home residents, and older adults. For example, a user who stated that they were pediatrician highlighted that they “have had multiple babies and toddlers brought to the hospital by police for “found alone in the home with caregiver deceased.” Discussions on indirect deaths implied collateral damage such as delayed treatment resulting in death or decline from isolation resulting in death. A user responded, “Dementia and Alzheimer’s patients have died at higher rates due to their brains not being exercised by human contact. They’re calling them “collateral deaths.”

4.1.3 Impact on vulnerable populations (children and elderly)

To some users, it was impact on vulnerable population groups. As one user stated—“The majority of childcare and eldercare during the pandemic was done by women, and economists are estimating that this (along with them not returning to the same roles at the same capacity due to ongoing instability in other family needs, etc.) will set back women’s lifelong earnings and take a long time to climb back up to where we were.” Beyond health, there were discussions on educators who played a crucial role in adapting teaching and learning during the pandemic—“The hateful rhetoric I’ve witnessed accusing teachers of laziness, of not want to return to school, of not caring for students, is appalling. People attack teachers for quitting, too, claiming their hearts must be ice to leave so many children in the lurch. But did any one of the 4917 comments here spare a thought for teachers’ physical and mental well-being? The endless 12-h workdays? Being forced back into schools without mask or vaccination mandates for slashed pay and benefits?”.

4.1.4 Social and political division

Some users also voiced concerns about growing social and political divisions among couples, families, relationships, friends, and work colleagues (6.8%). For instance, a user mentioned that “the darkest effect [imo] is the levels of dishonesty about everything and the divisions it has caused.” Another user stated that “the rifts/divides it’s causing families and friends; people being cut out of others’ lives due to believes and support.” Users of online communities often respond with satirical remarks. A user, for example quipped that it is a “good way to thin out your friends list.” Vaccination debate was also a prominent part of the discussion on how Covid19 was creating a rupturing society and how people celebrate death of people not agreeing with their opinions. For example, a user stressed—“Division…. People dehumanising populations for varying stances on covid. I see redditors casually celebrate every death of an unvaccinated person. Really morbid, especially considering how mainstream it is, depressing.”

4.1.5 Work-related issues

Six percent of responses focused on work-related issues such as burnout, thoughts of changing professions, insufficient staffing layoffs, furloughs, unemployment, and attrition among a variety of sectors such as health care, education, service industry. Writing about their bartending experiences, a user shared how they feel disassociated with their job—“People who were furloughed had time to realize they hate their job. I can only speak in service industry, but it feels like a soul crushing job more than ever. Guests are worse and more entitled than ever, and staff shortages have made shifts harder and longer.

Another user stated how burnout led experienced professionals to leave their job, creating a gap which is being filled by inexperienced people in healthcare—“the over work and over stress of health care providers making a lot of them leave the line of work. We are talking about experienced people leaving and never looking back. Means a lot of the people replacing them aren't being trained properly.” Discussing their personal experience, a user supported the burnout experienced by healthcare providers—“From housekeepers to physicians, every functional employee of the hospital system is burned out. They’ve been treated like garbage- like this was the job they signed up for, and how dare they expect a shred of respect or help. I left my job as an RN in December and would sooner work at Wendy's than put on a pair of scrubs ever again.”

4.2 Results of topic modeling

Table 2 presents an overview of the results of our topic modeling analysis, offering insights into key topical things and examples from users’ posts. Although the key topics that emerged prominently among users’ online conversations during this period were physical health, mental health, vaccinations, job losses, the impact on the young population, the loss of relatives, and suicidal thoughts, the primary focus of these conversations revolved around physical and mental health. Physical health-related discussions frequently centered on COVID-19 symptoms, such as the loss of the sense of smell, cognitive decline, and brain fog, among other related issues. Mental health was a recurring theme, including topics focusing on depression, anxiety, and suicidal thoughts.

Conversations also centered around the emotional and personal impact of losing loved ones due to the pandemic, highlighting the significant human cost of the crisis. Alongside these deeply personal narratives, user posts also discussed the sociopolitical consequences of the pandemic, particularly the measures taken to combat it. This included conversations around financial implications, vaccination efforts, insufficient healthcare resources, and the prevalence of misinformation and distrust toward systems and institutions. The polarization of opinions and beliefs became evident through these conversations, with the spread of low-credibility information in online discussions, reflecting the challenges in disseminating critical information and building trust among people during the pandemic. Furthermore, as a result of stay-at-home measures that kept families confined to their homes, users engaged into conversations that reflect signs of family stress and tension, particularly among the younger population.

5 Discussions and implications

Findings of this study demonstrate that a wide range of expressions, comprising of experiences, emotions, and insights are shared within online communities. Our study focused on examining the patterns in the users’ discussions on the after effects of COVID-19 in an online community. Impact on mental health was a common theme in the online discussions. Users shared their experiences of heightened anxiety, stress, and depression due to the pandemic, which was compounded by the narratives of those who lost their loved ones. Our findings confirm previous research into COVID-19 increases in symptomatic mental health expressions [6]. Emotional responses to the pandemic were multifaceted and included fear, anger, and grief, with their consequences evident in individuals’ personal lives. Findings also showed users’ engagement with health issues that arose indirectly due to the pandemic, indicating a long-term consequence on healthcare systems in the future. This finding is particularly useful in illustrating healthcare system changes in patient experience noted previously, such as personal safety concerns [24]. Users discussed lingering health issues faced by those who had recovered from COVID-19 (or “long COVID”).

Our analyses also indicated fragmentation in society, both directly through differing viewpoints on the pandemic and indirectly through its socioeconomic consequences. The widespread polarizing effect of social media discussion is well-established in the literature [4]. Economic consequences that emerged as a recurrent theme in our study, with users discussing the impacts of the pandemic on their employment and financial stability, reflect the far-reaching implications for vulnerable populations beyond challenges relating to managing personal finances.

Concerns about the potential long-term effects of the pandemic on future generations were also prominent which align with published findings from other studies [5]. Users’ posts also demonstrated a consistent pattern in raising concerns about development and well-being of children. Users expressed worry about the pandemic’s effect on education, development, and emotional well-being of children. The challenges of limited social interaction, particularly during lockdowns, also were shared experiences within the online communities.

Whereas some of the aftereffects of COVID-19 received less attention in the discussions among reddit users, they indicate critical areas to be studied in the future. For example, changes in relationships received lesser attention but were a significant aspect of these conversations, as individuals grappled with strained, broken, or changed relationships after the pandemic. Some users engaged in deep philosophical discussions, reflecting on the fragility of life and existential questions prompted by the pandemic. Most notably, some individuals turned to substances as a coping mechanism, leading to discussions on addiction and dependence.

Findings of this study have several implications for researchers and practitioners, particularly in the field of public health, mental health, and economic wellbeing. The post-pandemic impact on mental health and subjective well-being underscores the need for accessible mental health support and interventions [7]. Resources for coping with the emotional loss should be readily available, including trained practitioners on providing support for coping with multilayered grief. Online access to these resources can potentially improve connection to health information [25]. Preparing healthcare systems for indirect health consequences is essential as they cope with post-pandemic healthcare needs [26]. Long-term strategies must consider efforts to mitigate social fragmentation and countering low-credibility information and misinformation among general public. In the wake of growing inflation, addressing economic disparities should be a priority, including safety nets for vulnerable populations. Additionally, the changes in relationships brought about by the pandemic [6] necessitate resources for healing and adaptation, as is increased attention to substance dependence and addiction support [8].

Although our study offers insights into the challenges individuals faced during the COVID-19, it has some limitations which should be considered when interpreting the findings. First, the reliance on data from an online community introduces potential sampling bias, as participants engaging in discussions about the darker effects of COVID-19 may not represent the broader population. This limits the generalizability of the study beyond the specific online platform studied. Second, despite employing two unique analytical approaches, the subjective nature of qualitative coding and human interpretations of topics from the list of words introduce potential biases in the identification of themes and topics. Additionally, one significant limitation of the study is the lack of integration between qualitative coding and unsupervised topic modeling. While we employed both approaches separately, integrating these in future studies (e.g., using supervised LDA, which can effectively incorporate qualitative priors to guide the topic extraction process) might result in identifying topics that align well with the qualitative codes [27]. Finally, the focus of our study on a specific timeframe during the pandemic may not fully capture evolving public sentiment or issues over time. Utilizing longitudinal data or conducting periodic analyses could provide a more comprehensive understanding of how concerns and perceptions shifted throughout different phases of the pandemic.

6 Conclusion

In sum, the long-term darker effects of COVID-19, as expressed within online communities, have shown diverse human experiences that require a holistic approach for recovery and resilience. It is also evident from the current study that the online communities have become invaluable spaces for seeking support, providing empathy, and having dialogue. The organic nature of user interactions in online communities and their significance in providing a voice to the collective experiences of individuals can serve as a useful tool to collect naturalistic data which is not possible by traditional research methods such as surveys and interviews. Moreover, the lessons learned from online conversations can inform strategies for unforeseen challenges or crises develop a more resilient and compassionate society.