The Dynamics of Information-Seeking Repertoires: A Cross-Sectional Latent Class Analysis of Information-Seeking During the COVID-19 Pandemic

ABSTRACT Understanding audiences’ information-seeking behaviors during a societal crisis it vital for effective crisis communication. Prior research has identified how individuals combine information sources during a specific crisis phase. However, there is a lack of studies analyzing the stability of such behavior across phases. Therefore, this study utilizes a four-wave panel study conducted in Sweden (N = 13,718) to examine information-seeking repertoires and potential drivers across phases with different threat severity during the COVID-19 pandemic. The results from a cross-sectional latent class analysis revealed four main types of information-seeking repertoires: pluralists, traditionalists, minimalists, and news junkies. Specifically, the findings show that individuals with different socio-demographic profiles broaden their information-seeking repertoire when threat severity is high, making socio-demographic factors a poor predictor of repertoire breadth. Instead, mainstream media trust seems to play a more important role as a potential predictor of broad information-seeking repertoires including non-mainstream sources. The dynamic nature of the repertoires cautions scholars not to make generalization about information-seekers and their characteristics across different phases of a crisis and underline the importance of future research to focus on factors beyond socio-demographics.

Effective crisis communication requires an understanding of citizens' information-seeking behaviors.Traditionally, scholars have focused on one source or compared diverse sources in isolation.Recent studies have sought to overcome this media-centerd perspective by examining how individuals combine multiple sources of information.Emerging evidence indicates that citizens tend to combine information sources in similar ways-referred to as information-seeking repertoires or information-seeking clusters-to cope with high threat situations (Houston et al., 2021;Kuttschreuter et al., 2014;Lee & Jin, 2019;Sommerfeldt, 2015;Wang & Ahern, 2015).This body of research has primarily focused on how individuals combine information sources during a specific time point.However, how individuals mix information sources may change during a prolonged societal crisis with distinctive phases.For example, the COVID-19 pandemic witnessed phases marked by high infection rates, casualties, and restrictions, followed by phases with lower numbers in both areas along with fewer restrictions (Johansson et al., 2023).Such variations are likely to have an impact on information-seeking behavior, affecting not only frequency of informationseeking but source preferences.
Due to the narrow focus on one specific phase, most studies do not problematize the stability of the identified types of information-seekers or their potential drivers across phases.Crisis communication research generally departs from the implicit assumption that audience segments remain relatively static.Scholars are therefore advised to consider how individuals with various socio-demographic traits perceive and search for information differently throughout a crisis (e.g., Lachlan et al., 2014;Sellnow et al., 2017Sellnow et al., , 2012)).However, transformations in source combinations across distinct phases of a crisis would necessitate more dynamic crisis communication strategies, including continuous adaptation of message designs and distribution strategies.Therefore, a better understanding of how individuals combine information sources and potential drivers of such behavior across phases is needed to develop crisis communication models applicable to prolonged societal crisis.
Against this background, the aim of this study is to investigate the stability of information-seeking repertoires and their potential drivers during an extended societal crisis.To this end, the prevalence of different information-seeking repertoires and their individual characteristics across phases with different levels of threat severity during the COVID-19 pandemic will be investigated, using a four-wave panel study conducted in Sweden (N = 13,718).Sweden is an intriguing case due to its high institutional trust (Ihlen et al., 2022), making it different from the United States (Uslander, 2018), where most studies on information-seeking repertoires or clusters have been conducted thus far.The stability of predictors that previous studies on information-seeking and news repertoires point out as important will be examined, including both socio-demographic factors, ideology, and mainstream media trust (Andersen et al., 2022;Spence et al., 2011;Strömbäck et al., 2018;Zhao et al., 2022).Thus, two different research fields will be combined with the aim to provide a basis for development of theoretical models on how audiences combine information sources applicable to prolonged crisis.

Information seeking during a crisis
In broad terms, information seeking refers to "planned scanning of the environment for messages about a specified topic" (Clarke & Kline, 1974, p. 233).Today's high-choice media environment provides individuals the opportunity to combine various types of information sources when scanning the environment for information about a societal crisis, including a plethora of channels, sources, and platforms (Van Aelst et al., 2017).Yet, research on information-seeking tends to focus on one source in isolation or compare different types of sources.Despite being channelcentric, such studies have provided important insights into people's information-seeking behavior in threatening situations.Specifically, this body of research points to an increased need for information during the acute phase of a societal crisis.People extend their information seeking by increasing their consumption of both channels they already use and channels they normally do not use (Ghersetti & Westlund, 2018;Westlund & Ghersetti, 2015).Traditional channels (e.g., television, newspapers) often get a larger audience during this phase.For example, research on media use in Western Europe at the beginning of the COVID-19 pandemic shows a steep rise in both the usage of and trust in public-service media and online newspapers (Van Aelst et al., 2021).
In addition to research on traditional news channels, the last decade has seen an uptick in the number of studies analyzing the usage and significance of social media platforms during a societal crisis (Rasmussen & Ihlen, 2017).Social media platforms enable individuals to seek out and share crisis-related information, both in public and with private social networks.Moreover, individuals are exposed to crisisrelated information on social media platforms without actively searching for it (Liu et al., 2019).Models like the Social Mediated Crisis Communication Model (SMCC) have provided insights about how the interplay between traditional news media channels, social media platforms, and interpersonal communication takes place during a crisis.In this interrelationship, social media plays an important role as a way for individuals to find new information as well as verifying uncertain information (Liu et al., 2019;Zhao & Tsang, 2021).Like consumption of traditional channels, recent research indicates that social media usage also increases during the acute phase of a crisis (Van Aelst et al., 2021).Generally, younger generations are more likely to seek information on social media than are older generations (Ghersetti & Westlund, 2018).Turning to non-mediated sources, interpersonal communication (e.g., face-to-face conversations, phone calls, texts) has also been identified as an important information source when people aim to make sense of a threatening situation (Vigsø & Odén, 2016).Returning to the SMCC model, findings show that individuals prefer offline interpersonal communication instead of mediated communication to cope with a crisis (Liu et al., 2019).
Like research on information seeking, audience research tends to analyze different news sources in isolation.However, recent studies have sought to overcome this media-centric perspective (e.g., Andersen et al., 2022;Edgerly, 2015;Mangold & Bachl, 2018).For instance, the news repertoire approach employs a cross-media view of people's news habits, in which a mix of different news sources constructs an individual news repertoire (Schrøder, 2011).The concept is a development of Hasebrink and Popp's (2006) framework of media repertoires, referring to the "entirety of media that a person regularly uses" (Hasebrink & Popp, 2006, cited in;Hasebrink & Domeyer, 2012, p. 759).Similar news repertoires have been identified across various settings, though their presence varies by country (Castro et al., 2021).Likewise, there have been attempts to overcome the mediacentric approach in crisis communication research by looking at how individuals combine sources during a societal crisis.For example, Houston et al. (2021) developed the communication ecology network (CEN) as a means of explaining the relationship between different crisiscommunication sources.The framework posits that related communication sources and their users cluster together within overarching CENs.Through network analysis, the authors identified five different information-seeking clusters in the United States during COVID-19: the liberal cluster, the conservative cluster, the television-oriented cluster, the public health resources-oriented cluster, and a broader cluster that includes a variety of mediated, organizational, and interpersonal communication resources.Similarly, Zhao et al. (2022) employed cluster analysis to analyze information-seeking convergence in the United States during the COVID-19 pandemic.The authors identified four types of information seekers one year into the pandemic: the traditionalists, the digital utilitarians, the deliberative actives, and the inactives.The deliberative actives frequently searched for information across a wide range of outlets and media types.The traditionalists and digital utilitarians, however, exhibited only moderate levels of information seeking.While the traditionalists mainly searched for information from official sources through traditional means, the digital utilitarians searched for information about COVID-19 through digital means from a wide range of sources, including friends, family, news sites, and professionals.Finally, the inactives rarely searched for information about COVID-19 from any source.
With its roots in the news repertoire approach, the information-seeking repertoire framework is yet another approach used to analyze how individuals combine information sources during a crisis.Generally, the informationseeking repertoire approach focuses on sources of information rather than news outlets.Key sources of information during a societal crisis include experts, politicians, and family and friends; the information from these sources can be found through various channels, including social media platforms, government websites, and newspapers.One example of a study employing an information-seeking repertoire approach is Sommerfeldt's (2015) study on how citizens combined information sources after the 2010 earthquake in Haiti, in which both traditional and elite information seekers were identified.The results are in line with previous research on how individuals combine information sources during a crisis, showing that some individuals tend to use a wider variety of sources serving complementary functions, while others use a narrower variety of sources serving convergent functions (Anthony et al., 2013;Yuan, 2011;Zhao et al., 2022).
While there have been attempts to assess how individuals combine sources during a societal crisis, such studies remain scarce.Additionally, previous studies have mainly focused on the United States, meaning that knowledge about the structure of information-seeking repertoires or clusters in other settings is either limited or non-existent.Yet, repertoires or clusters identified in societies characterized by high trust in institutions, such as Sweden (Ihlen et al., 2022), might differ from those identified in other societies with low levels of institutional trust, such as the United States (Uslander, 2018).Against this background, we pose the first research question as follows: RQ1: What information-seeking repertoires can be identified during the COVID-19 pandemic in Sweden?

Dynamics of information-seeking repertoires
Thus far, news and audience research has focused on which news repertoires that exist at a specific point in time.To overcome this shortcoming, recent research has sought to investigate the stability of news repertoires, concluding they-like general media consumption (e.g., LaRose, 2010; Taneja et al., 2012)-tend to be highly stable.For example, Andersen et al. (2022) studied news repertoires over a three-year period and found overall stability in how individuals compose their daily news consumption.
If transformations occur, they are likely to be linked to a change in supply or demand.
On the supply side, technological innovation may result in new media platforms, which may influence news repertoires (Ha et al., 2018).On the demand side, contextual changes, such as a societal crisis, might generate an increased need for information and cause individuals to transform their repertoires.Generally, the acute phase has gained most scholarly traction, resulting in a lack of knowledge about how individuals mix information sources across phases of a societal crisis.Nevertheless, these studies show that individuals tend to increase their information-seeking from multiple sources during the initial, acute phase (Van Aelst et al., 2021;Westlund & Ghersetti, 2015).The increased frequency of information-seeking may destabilize and broaden news habits, leading to homogenization of consumption patterns (Westlund & Ghersetti, 2015).In line with this expectation, Andersen et al. (2022) found clearer differentiation between news repertories during the first phase of the COVID-19 pandemic.
The few studies which have analyzed how individuals combine information sources departing from a longitudinal perspective suggest a decrease in citizens' information-seeking frequency across channels and sources as the crisis evolves (Vandenplas et al., 2021).This is in line with previous research on single source use, showing that individuals tend to experience communication fatigue in later stages of a prolonged societal crisis, resulting in a loss of interest in crisis messages and in the ability to process information (Shulman et al., 2021).Individuals also decrease their level of information-seeking from several sources to cope with negative emotions (Vandenplas et al., 2021).Specifically, Vandenplas and colleagues' study of news avoiders during the COVID-19 pandemic suggests that avoiding news constitutes a reconfiguration of consumers' media repertoires.Even news avoiders turned to the news at the beginning of the pandemic; however, as the influx of news began to be perceived as harmful, consumers engaged in situational news avoidance to shield themselves from harm.In line with the reasoning above, it is likely that information-seeking repertoires are dynamic-at least during a prolonged societal crisis with reoccurring acute phases.High threat severity is likely associated with increased information-seeking frequency from multiple sources, while low threat severity may lead to decreased information-seeking frequency from multiple sources, although this trend may dissipate over time as the crisis unfolds.However, research examining how individuals combine multiple sources of information during a prolonged crisis is limited or nonexistent, resulting in a lack of knowledge about the stability of such behavior.Thus, we pose the second research question as follows: RQ2: How stable are the identified information-seeking repertoires across phases of the COVID-19 pandemic?
Starting with the socio-demographic factors, results are inconclusive.Previous research indicates that individuals with higher levels of education are more likely to be classified as omnivorous or public service oriented (Sommerfeldt, 2015).Some scholars have found that individuals with lower levels of education are more likely to have a minimalist news repertoire (Bos et al., 2016), while others have found that those with higher levels of education are more likely to have a minimalist news repertoire (Strömbäck et al., 2018).In some studies, young people are linked to omnivorous repertoires (Van Eijck & Van Rees, 2000), while others conclude they tend to have a minimalist or social media-based repertoire (Andersen et al., 2022;Bos et al., 2016;Edgerly, 2015;Strömbäck et al., 2018;Trilling & Schönbach, 2013).Men and women also tend to have different repertoires, with women being more active information seekers during a crisis (Spence et al., 2011).
Turning to social factors, attitudes play important roles in media consumption patterns.For example, trust in mainstream news outlets is a significant driver of different news repertoires, mainly due to the rise of alternative news media (Andersen et al., 2022;Strömbäck et al., 2020).Not surprisingly, previous studies show that individuals with low trust in mainstream news media often seek out alternative sources of information (Fletcher & Park, 2017;Jackob, 2010;Tsfati, 2010;Tsfati & Cappella, 2003) and, in turn, are less likely to have a news repertoire comprising of mainstream news outlets (Mourao et al., 2018;Yuan, 2011).In addition, using social media as one's main source of news is related to lower media trust (Kalogeropoulos et al., 2019).Finally, ideology also plays a key role.A study by Dahlgren et al. (2019) on the mutual influence between selective exposure and ideology shows that ideology predicts which news sources individuals use, which in turn reinforces their ideological beliefs.In support of this argument, research shows individuals situated far to the right tend to use right-wing alternative media sources more frequently (Müller & Schulz, 2021).
While the above factors predict general media consumption and news repertoires, it is unclear whether they also predict information-seeking repertoires.Thus far, prior studies have analyzed potential drivers of information-seeking in one specific phase of a crisis, resulting in a lack of knowledge regarding the predictive power of individual characteristics on willingness to seek information across phases.Likewise, studies on how individuals combine information sources have mainly analyzed predictors at one specific phase of a crisis.For instance, Zhao et al.'s (2022) study on information-seeking convergence one year into the COVID-19 pandemic shows that age, gender, and education were significant drivers of diverse types of information-seeking clusters.Those who searched for information across a high number of channels as well as those with moderate levels of information seeking were younger and more educated.The traditionalists, who searched for information from traditional sources, and the inactives, who had low levels of information seeking, were older and less educated.The findings also show that the traditionalists were more likely to be male.
Thus far, studies on the potential drivers of information seeking or information-seeking repertoires have not employed a longitudinal perspective.If individuals transform their information-seeking repertoires, the predictive power of potential drivers are likely to be less stable or decrease in effect strength across phases.The question thus remains whether drivers of information-seeking repertoires are stable throughout a prolonged societal crisis, leading us to our third research question: RQ3: Are there any stable drivers connected to the identified information-seeking repertoires?

Case selection
The study is based on a four-wave panel survey conducted in Sweden during the COVID-19 pandemic.Sweden is interesting for two main reasons.First, it represents a high-trust society, in which trust toward mainstream media remains relatively high (Andersson, 2021;Ihlen et al., 2022).Consequently, it constitutes a very different case from the United States (Uslander, 2018), where most studies have been conducted thus far.Second, like many other countries, Sweden is currently experiencing changes in its media consumption patterns.Mainstream news readership is declining alongside an increasing use of social media for news consumption (Newman et al., 2021).A prolonged societal crisis might further intensify such trends.The COVID-19 pandemic in Sweden therefore offers an interesting context in which to analyze how individuals combine sources in an evolving media environment characterized by high-trust, as well as the potentially dynamic nature of such behavior.
Sweden's first COVID-19-related casualty occurred in March 2020 and the country soon experienced high death tolls in April and May (Folkhälsomyndigheten, 2023a).The Swedish strategy relied on citizens voluntarily following recommendations rather than coercive measures, and the society therefore remained relatively open throughout the pandemic (Johansson & Vigsø, 2021).The first wave of infections ended during the summer of 2020, and wave two did not hit until the end of October 2020, resulting in high infection rates and casualties around Christmas and, in turn, expanded restrictions.The Delta variant of SARS-CoV-2 prolonged the second phase, leading to several restrictions remaining in place throughout spring 2021.Around the same time, the Swedish vaccine program was enrolled, and the death tolls decreased.In the summer of 2021, a high number of citizens were vaccinated, causing infection rates to decrease and the government to lift several restrictions (Folkhälsomyndigheten, 2021(Folkhälsomyndigheten, , 2023b)).

Data
The Citizen Panel at the SOM-institute, University of Gothenburg, was used for the analyses. 1The first survey wave was sent out in April 2020 (during the first infection phase, high threat severity), and the other survey waves were sent out in September 2020 (before the second infection phase, low threat severity), December 2020 (during the second infection phase, high threat severity), and April 2021 (during the second infection phase but at the same time as the vaccination program was rolled out, middle threat severity).The data was thus gathered at different levels of threat severity (two phases with high threat severity, one phase with low threat severity, and one phase with middle threat severity), enabling comparison of informationseeking repertoires and potential drivers across such phases.The panel survey had a net sample size of 13,718, including both opt-in individuals and a randomized sample.It is important to note that the inclusion of opt-in participants resulted in the final sample being skewed toward middle-aged, older individuals with high levels of education.The sample also included more men than women.The analyses are based on individuals who participated in wave one and responded to each indicator in each survey wave.Therefore, the final number of analyzed respondents varies by wave (9,240, 7,897, 4,476, and 8,790).The skewness of the sample as described above remained in each wave.

Measurements
We measured information seeking with items identified as important in crisis communication research, including Swedish news media, foreign news media, social media, government web pages, family/friends, and alternative media.More specifically, the following question was listed alongside each of those items: "How often have you searched for information about COVID-19 in/from the following sources?"The question was followed by a five-point scale, ranging from 1 = several times a day to 5 = never.For further details about the indicator variables, see Table S1 in the Appendix in the online supplementary materials).
The covariates are as follows: age, sex, educational level, mainstream media trust, and ideology.Sex was measured through a categorical variable with three levels: female, male, and other.It was re-coded into a factor variable with two levels: female and male with other made into NA.We measured age by asking participants which year they were born in and then re-coded their responses into three groups: 16-39 years, 40-69 years, and 70+ years.Respondents' educational level was measured on a scale ranging from 1 = did not complete elementary school) to 9 = PhD degree and was re-coded into a factor variable with three levels: low education (elementary school or less), middle education (high school and post-secondary education), and high education (undergraduate and graduate studies).To measure trust in mainstream media, respondents were asked about the degree of trust that they have in various mainstream media outlets, ranging from 5 = very little trust to 1 = a lot of trust (reverse-coded).The outlets include: Sveriges Television, Swedish public-service television, SVT; Sveriges Radio, Swedish public-service radio, SR; and TV4 large commercial TV channel, Channel 4. Principal component analysis revealed that the items loaded on one factor for each wave explaining 75-76% of the variance.Thus, the items were rescaled and made into additive indexes for each wave of mainstream media trust, ranging from 0 to 1 (M = .70,.69,.69,.69;SD = .22,.23,.23,.23;Cronbach's α = .91,.92,.90,.92).Finally, to measure ideology, participants were asked to place themselves on a scale ranging from 0 = far-left to 10 = far-right.The responses were re-coded into a factor variable with three levels-left, right, and center-with an unlabeled 5 representing the center of the scale.

Analytical approach
Latent class analysis (LTA) and its counterpart for continuous variables, latent profile analysis (LPA), are person-centered mixture-modeling techniques which identify subgroups sharing similar characteristics in a larger population.These subgroups, known as latent classes or profiles, are determined by analyzing response-pattern probabilities to observed variables.First, respondents are assigned to latent classes or profiles based on their item-response probabilities.Second, class conditional probabilities are calculated, enabling analysis of likely response patterns across the identified groups (McCutcheon, 1987).The probabilistic nature of LCA and LPA results in less biased classification estimates than other person-centered methods, such as cluster analysis (Karnowski, 2017), making them the most suitable for identifying subgroups in a larger population (McCutcheon, 1987).
Since a 5-point scale was used for measuring the indicator variables, LCA (treating the indicator variables as categorical) was preferred over LPA.Further, cross-sectional LCA was chosen instead of latent transition analysis (LTA) due to the potentially dynamic nature information-seeking repertoires, making interpretation of LTA results more difficult due to noninvariance (Collins & Lanza, 2009).Nonetheless, it is worth noting that LCA can be extended to longitudinal data, wherein transition probabilities between classes are calculated based on response-probabilities (Collins & Lanaza, 2010).
The first step entailed conducting an LCA without covariates.Due to restrictions related to the number of indicator variables, a maximum of five latent classes were tested for each wave.To identify the best-fitting model, multiple estimation approaches were applied, including listwise deletion of cases and Expectation Maximization Likelihood Estimation (EM) of data missing at random.The best solution was chosen by examining different information criteria, including Bayesian information criteria (BIC), sample size-adjusted Bayesian information criteria (aBIC), consistent Akaike criterion (CAIC), and the bootstrap likelihood ratio test (BLRT).To be precise, the information criteria were used to determine whether an increase in classes improved the model's fit (Nylund et al., 2007;Spurk et al., 2020).Since the CAIC performs worse on skewed data (Morgan et al., 2016), more weight was placed on the BIC, aBIC, and BLRT.
Based on the modal class assignments, weighted multinomial regressions were conducted on each wave to explore potential drivers of the information-seeking repertoires.Due to the high number of classes and waves, the reference group was not systematically rotated for each regression. 2 To mitigate the distortion of estimates by treating probabilities as exact measurements, the class assignments based on posterior probability were extracted and used as weights.While this is currently considered one of the most appropriate methods of including covariates in LCA, it should be noted that the errors still will be distorted to a certain degree (Clark & Muthén, 2009).The predicted probabilities for each covariate were calculated based on the regression models, providing the probability of a covariate response given the assigned latent class.

Results
The BIC, aBIC, and BLRT values consistently suggested four or five classes as the best model fit across the survey waves (see Figures 1-4 in the Appendix in the online supplementary materials).While the BIC and aBIC values decreased sharply between one and three classes, the fiveclass solution had the lowest BIC and aBIC values.It should be noted that the BIC and aBIC values may have continued to decrease if we had added more classes to the model.By doing a separate BLRT using the mclust package in R, the results showed a significant value up to five clusters in each wave.Adding interpretability to the equation, five classes were chosen as the best-fitting model for each wave.Due to similarities in terms of composition of sources and frequency of use (authors own interpretation), the classes were categorized into four main classes and variations of these across waves.This was done to illustrate that certain classes appeared less distinctly differentiated from one another, which is likely to explain the relatively low Entropy-value (approximately 0.6) (Muthén, 2008;Weller et al., 2020). 3nswering RQ1, the four main types are pluralists, traditionalists, minimalists, and news junkies.The identified repertoires somewhat align with findings in previous studies, which provided inspiration and ideas for the labels (e.g., Andersson, 2021;Castro et al., 2021;Zhao et al., 2022).The pluralists are characterized by their relatively high use of all channels in pursuit of information about COVID-19.While they did not exclude any of the channels, the frequency with which they used specific channels differed between waves.Three variations of pluralists were identified, including the mainstream pluralists (frequent users = FU), the mainstream pluralists (less frequent users = LFU), and the alternative pluralists.The alternative pluralists are characterized not only by their high usage of all channels but also by their comparatively high consumption of alternative media daily for information-seeking purposes.The item-response probabilities for the alternative pluralists are visualized in Figure 1 (probabilities from wave one, representing a general pattern).This stands in stark contrast to the mainstream pluralists, who were more likely to use Swedish news media for information-seeking purposes.Like the alternative pluralists, however, they did not exclude any channels.The two groups of mainstream users mainly differed in terms of information-seeking frequency: less frequent use (LFUs) and frequent use (FUs).Compared to the FUs, the LFUs were more likely to use alternative media more often (once or more times a week).However, both groups were less likely than the alternative pluralists to use alternative media.The itemresponse probabilities for the mainstream pluralists are visualized in Figures 2 and 3 (probabilities from Wave 1, representing a general pattern).
The tendency to not exclude any channels is shared with the news junkies.However, the news junkies differed from the pluralists in significant ways.They were far more likely to frequently use Swedish news media (several times a day).In addition, they were more likely than the mainstream pluralists to use alternative, social, and foreign media several times per day for information-seeking purposes.In general, the news junkies sought information more frequently than the pluralists.The item- response probabilities for the news junkies are visualized in Figure 4 (probabilities from Wave 1, representing a general pattern).
Both the news junkies and the pluralists stand in stark contrast to minimalists, who were less likely to use any of the considered channels for informationseeking purposes.Two variations of the minimalists were captured: minimalists and selective minimalists.The selective minimalists were slightly more likely  than the minimalists to use Swedish news media daily to seek information about COVID-19.The minimalists, however, had a slightly higher likelihood to use alternative, foreign, and social media one or more times a week.Compared to the selective minimalists, the minimalists thus had a somewhat broader information-seeking repertoire, although both classes seldomly sought information about COVID-19.The item-response probabilities for both types of minimalists are visualized in Figures 5 and 6 (probabilities from Wave 2, representing a general pattern).Finally, the traditionalists share features with both the minimalists and the pluralists in that they use some of the channels for information-seeking purposes but exclude others.The likelihood that the traditionalists used Swedish news media, government websites, and interpersonal communication was relatively high.However, they were more likely to exclude social, foreign, and alternative media.Two variations of the traditionalists were identified: the news-oriented traditionalists were more likely to use Swedish news media, while the socially-oriented traditionalists were slightly more likely to use family and friends as sources one or more times per week.Moreover, the socially-oriented traditionalists had a higher probability of using social, foreign, and alternative media one or more times per week.The item-response probabilities for both variations of traditionalists are visualized in Figures 7 and 8 (probabilities from waves 1 and 4, representing a general pattern).
Answering RQ2, the variations of the main types were identified across different phases of the COVID-19 pandemic, indicating that informationseeking repertoires are relatively dynamic as shown in the timeline in Figure 9.At the onset of the pandemic (phase one, high threat severity), four classes with rather broad repertoires and with relatively high to high information-seeking frequency were captured.More specifically, we identified the news junkies alongside three variations of pluralists: the mainstream (FU), the mainstream (LFU), and the alternative (FU).Finally, the news-oriented traditionalists were also captured-the only class with a comparatively narrow repertoire.In September 2020 (phase two, low threat severity), three classes with rather narrow repertoires and low information-seeking frequency were identified: the selective minimalists, the news-oriented traditionalists, and the minimalists.Alongside these, we also captured two variations of pluralists (the mainstream FUs and the mainstream LFUs).In December 2020 (phase three, high threat severity), the news junkies that were identified at the onset of the pandemic appeared again.The newsoriented traditionalists and mainstream pluralists (both FU and LFU) remained alongside the selective minimalists.The minimalists were, however, no longer identified by the model.
In April 2021 (phase four, middle threat severity), a mix of both narrow and broad repertoires appeared.The minimalists were identified again, who were not prominent in Wave 3. In addition, we captured the alternative pluralists, who had not appeared since Wave 1, alongside the mainstream pluralists (FU), the news-oriented traditionalists, and the socially oriented traditionalists.
Another indicator that individuals transform their information-seeking repertoires relates to changes in the sizes of the repertoires.The mainstream pluralists (FU) reached their largest points when threat severity was high (phase one and three), while the mainstream pluralists (LFU) peaked in phase two (low levels of threat severity), indicating that some mainstream pluralists (FU) in Wave 1 might have moved to the mainstream pluralists (LFU) in Wave 2. Both the news junkies and the alternative pluralists were largest in phase one (high threat severity) and decreased in size as the crisis evolved.While the selective minimalists remained fairly equal in size, the estimated population share of the minimalists was at its largest during middle threat severity (phase four), indicating that some individuals transformed their repertoires by decreasing their information-seeking frequency from multiple sources.The news-oriented traditionalists also remained fairly equal in size throughout all phases.Notably, the frequency of information seeking decreased for all classes as the crisis evolved.

Potential drivers of information-seeking repertoires
In the next step, we examined the stability of potential drivers of the information-seeking repertoires.A predictor is considered stable if it shows the same predictive trend across phases (see Table S8 in the Appendix in the online supplementary materials).Starting with the broad repertoires, the mainstream pluralists (FU) had a higher probability of being middle-aged, age 40 or older.The mainstream pluralists (LFU) were slightly more likely to be centrist or rightleaning, more likely to have low or middle levels of education (elementary school or high school and post-secondary education), and low levels of mainstream media trust.Finally, the alternative pluralists were more likely to have low levels of education (elementary school or less), slightly more likely to be centrist or right-leaning, and more likely to have low levels of mainstream media trust.Finally, the news junkies were slightly more likely to be female and more likely to have middle to high levels of education (high school and post-secondary education or higher).Among the factors mentioned, mainstream media trust showed the strongest predictive trend.Turning to the narrow information-seeking repertoires, the selective minimalists had a higher probability of being middle aged or older (40+), low levels of education (elementary school or less), and low or middle levels of mainstream media trust.Finally, they were also slightly more likely to be centrist.The news-oriented traditionalists were more likely to have high levels of education (university and PhD studies) and slightly more likely to be left-leaning.Compared to the other classes, they were also more likely to have high levels of trust toward mainstream media.Of the factors mentioned, education showed the strongest predictive trend for the selective minimalists, while education as well as mainstream media trust showed the strongest predictive trends for the news-oriented traditionalists.Finally, the minimalists were more likely to be younger (ages 16-39), have low levels of education (elementary school or less), and were slightly more likely to be centrist or right-leaning.Among these, education and age showed the strongest predictive trends.4

Conclusion
Understanding how citizens combine information sources throughout a prolonged crisis is critical for effective crisis communication.Yet, there is a lack of research examining how people combine information sources across distinct phases of a crisis.Therefore, the stability of such behavior and its potential drivers remains poorly understood.This study aimed to shed light on this issue by investigating how individuals combine information sources across phases characterized by various levels of threat severity and potential drivers of such behavior.By applying a latent class analysis across four waves of a panel study conducted in Sweden during the COVID-19 pandemic, four main types of information-seeking repertoires were discovered: the pluralists, the traditionalists, the minimalists, and the news junkies.The types that were identified bear some resemblance to those found in previous studies (see Castro et al., 2021;Zhao et al., 2022).However, they were discovered in distinct phases of the crisis.A higher number of broad repertoires characterized by high information-seeking frequency were identified during high threat severity, while a higher number of narrow repertoires characterized by low information-seeking frequency were detected amid low threat severity.As expected, an increasing number of individuals became minimalists one year into the crisis (April 2021) although threat severity remained relatively high, indicating a communication fatigue or emotional overload resulting in decreased information-seeking from multiple sources (Shulman et al., 2021;Vandenplas et al., 2021).
Turning to the potential drivers, the study did not identify any stable socio-economic predictors of repertoire breadth across crises phases, although age and education showed relatively stable trends for individual (broad) repertoires.However, a stable pattern for those who were likely to have minimalistic or narrow repertoires was identified.The data suggests that low education levels predict minimalistic repertoires, while higher education levels drive greater information-seeking in mainstream sources and exclusion of non-mainstream alternatives.Shifting attention away from the socio-demographic factors, low levels of trust in mainstream news media were associated with two of the pluralistic repertoires characterized by more frequent use of alternative news sources.This is in line with previous research, showing that those who distrust mainstream media are more likely to rely on alternative sources (Fletcher & Park, 2017), potentially resulting in broader repertoires.Finally, the study found that the selective minimalists also tended to have less trust in mainstream news media, although these tendencies were comparatively weak.As previous research on news consumption suggests, low levels of institutional trust might cause individuals to decrease their overall news consumption (Goyanes et al., 2021).
The findings contribute to crisis communication research by enhancing our understanding of the dynamic nature of information-seeking in the modern media landscape, providing a basis for theoretical models applicable to prolonged crisis.While previous studies have shown an increased use of multiple sources in the initial phase of a crisis (Val Aelst et al., 2021), this study further reveals fluctuation across phases.Specifically, the results indicate that individuals with different socio-demographic profiles tend to broaden their information-seeking repertoires amid high threat severity, making socio-demographic factors a poor predictor of repertoire breadth.However, mainstream media trust seems to play an important role-at least in predicting a specific type of breadth including non-mainstream sources.In other words, although it may be easy to reach a great deal of the population amid times of high threat through mainstream channels, those who distrust mainstream media are likely to turn to additional sources serving complementary functions (Anthony et al., 2013).From a crisis communication perspective, this may be problematic, indicating that individuals with low levels of trust are exposed to a higher level of conflicting information when the crisis is most acute.
The dynamic nature of information-seeking repertoires challenges implicit assumptions about stable audience segments (Lachlan et al., 2014;Sellnow et al., 2017) and advises against generalizing characteristics of different information-seekers across crisis phases.Future crisis communication research needs to account for the broadening of repertoires amid high threat phases and develop models that mitigate the potential adverse effect of increased exposure to conflicting information.Related to this, the findings suggest that scholars should focus on drivers beyond sociodemographics when examining how individuals combine information sources across phases with different threat severity.Additional research is needed to determine how and under what conditions media trust affects repertoire breadth.For instance, distrust is likely to be a stronger predictor of broad information-seeking repertoires in societies with low levels of institutional trust due to the potentially increased tendency to search for information from non-mainstream sources (Fletcher & Park, 2017).Moreover, the directionality of the relationship needs to be properly understood.A reinforcing relationship is likely to exist in which repertoire breadth reinforces mainstream media distrust and vice versa during a prolonged crisis (Slater, 2007).Finally, factors which have shown to be of importance for individuals' general willingness to seek information, such as individual variations in negative emotions or perceived threat severity (Yang & Kahlor, 2012), may also play important roles.Hence, the study paves the way for future theory-building, highlighting the dynamic nature of how individuals combine information sources during a prolonged crisis and the need for further research aimed at understanding the underlying factors influencing changes in source combinations.
While this study provides important theoretical insights, it is important to acknowledge its limitations.First, considering the results and discussion about the role of socio-demographic factors it should be noted that an important factor that may influence individuals' information-seeking repertoires-ethnicity-was not included in this study (Spence et al., 2007).Second, the sample was skewed toward males, older, middle-aged, and well-educated individuals, meaning that the repertoires may not be generalizable to the broader population.It is noteworthy that the five-class solution was identified as the best-fitting model despite the homogeneity of the sample, indicating that five or more information-seeking repertoires may exist in the broader population.Second, restrictions on the maximum number of identifiable solutions may have prevented us from identifying classes, such as the minimalists in phase one and four.Future research should thus extend the information-seeking repertoire approach by measuring specific sources that individuals turn to when using different platforms.Finally, the results indicate repertoire transformations, but do not show which repertoires individuals are inclined to move toward or the individuals who are most inclined to move.Therefore, we hope that this study serves as inspiration for future scholars to explore movements between information-seeking repertoires and to include a higher number of indicator variables to capture information-seeking repertoires that this study may not have identified.

Disclosure statement
No potential conflict of interest was reported by the author(s).
of Gothenburg.He holds a master's degree in European studies, politics and public administration from the University of Gothenburg and the University of Konstanz, and a bachelor's degree in European studies from the University of Gothenburg.His research interests are media-effects and public-opinion via media influence on individual perceptions of society, which he primarily addresses using computational methods.

Figure 1 .
Figure 1.Item response probabilities: alternative pluralists (wave 1).Note: Range is from high to low: dark = high use; light = low use.Missing data modeled at listwise deletion.Total N = 9240.

Figure 4 .
Figure 4. Item response probabilities: news junkies (wave 1).Note: Range is from high to low: dark = high use; light = low use.Missing data modeled at listwise deletion.Total N = 9240.

Figure 5 .
Figure 5. Item response probabilities: minimalists (wave 2).Note: Range is from high to low: dark = high use; light = low use.Missing data modeled at listwise deletion.Total N = 8790.

Figure 6 .
Figure 6.Item response probabilities: selective minimalists (wave 2).Note: Range is from high to low: dark = high use; light = low use.Missing data modeled at listwise deletion.Total N = 8790.

Figure 7 .
Figure 7. response probabilities: news-oriented traditionalists (wave 1).Note: Range is from high to low: dark = high use; light = low use.Missing data modeled at listwise deletion.Total N = 9240.

Figure 8 .
Figure 8. Item response probabilities: socially-oriented traditionalists (wave 4).Note: Range is from high to low: dark = high use; light = low use.Missing data modeled at listwise deletion.Total N = 7898.

Figure 9 .
Figure 9. Timeline of identified information-seeking repertoires according to phases.