The network structure of trust in the COVID-19 pandemic

Research on trust is very extensive in the social sciences and trust is often referred to as a cornerstone of democratic society that facilitates or enables some key processes and functions. Above all, trust allows easier cooperation and unification for common goals (Warren 2018) and is also positively linked to the participation and voluntary association of people (Putnam 1993, 2000; Uslaner 2018). Since there is a consensus in the literature that trust is important for the functioning of democratic societies under normal circumstances (Warren 2018), it is no surprise that in times of crisis or sudden events, which the coronavirus pandemic undoubtedly represents, trust becomes a subject of interest (Bavel et al. 2020; Devine et al. 2020).

When we look at some studies of trust from the previous 2009 A(H1N1) pandemic, in a broader sense they emphasised the role of cooperation in helping society overcome the pandemic. Those studies focused on the relationship between trust and desirable behaviour such as compliance with hygiene measures (Prati et al. 2011) or immunisation by vaccination (Rönnerstrand 2013). Prati et al. (2011) found that respondents who complied with hygiene measures were the ones who had higher trust in the media and the Ministry of Health. Similarly, Rönnerstrand (2013) notes that both social and institutional trust are positively associated with the acceptance of vaccination. According to Bavel et al. (2020), it is the connection between trust and required behaviour that is one of the key aspects of coping with the current pandemic.

At the same time, several studies on trust in the COVID-19 situation have been published. As can be seen from the review article by Devine et al. (2020: 4), the topics of these studies focus on adherence to measures, policy implementation, risk perception, the relationship between trust and mortality, and especially change in trust or its different types.

It is important to note that the scholars who have studied trust in crisis situations, in our case specifically in a pandemic, distinguish between social and political trust (see e.g. Rönnerstrand 2013; Esaiasson et al. 2020; Schraff 2020). Social and political trust are relatively broad concepts, and, as Newton et al. (2018) note, it is possible within the framework of social trust to distinguish between various subtypes – in particular, generalised trust, which relates to people in general, without knowing them personally, and particularised trust, directed at loved ones and friends. Likewise, political trust has multiple layers. Political trust is directed at political actors, namely institutions that are impartial, such as the courts or the police, and political institutions or representatives such as the government, parliament, or president (Newton et al. 2018; see also Newton and Zmerli 2011; Uslaner 2018). There is an ongoing discussion whether social and political trust are two separate concepts (Uslaner 2018) or are more interconnected than previous studies have suggested (Newton et al. 2018). This issue was recently explored by Zhang et al. (2020), who studied different types of trust and the relations among them. They discovered that different types of trust can be grouped under either social or political trust, thus these two kinds of trust are indeed distinct phenomena. They also demonstrated that a kind of connection exists between social trust and political trust. The method they used, network analysis, allowed them to examine the network structure of trust, and thus provide new insight into this issue. Because studies on trust during a pandemic distinguish between social and political trust, we decided to follow Zhang et al.'s (2020) study to examine whether the division into these two types of trust is valid during a coronavirus crisis, and to what extent the two trusts are interconnected. In our study, we use data from the Czech Republic collected in May 2020. The Czech Republic is a country that had very few cases of coronavirus infection in the spring of 2020 and at the same time introduced very strict measures from the very beginning, such as restrictions on movement and the mandatory wearing of face masks in public. These measures significantly affected people's lives.

Studies on trust in epidemics distinguish between social and political trust, and the effect of these two types of trust is the same in some cases – i.e. both types of trust contribute to desirable behaviour (Rönnerstrand 2013), or a higher level of both trusts is associated with the later introduction of restrictive policies (Toshkov et al. 2020). However, there are cases where the impact of each trust may be different (see Goldstein and Wiedemann, 2020), where a higher level of social trust may be associated with a lower level of willingness to follow action. Thus, the relationship between political and social trust may vary depending on the conditions in each country and may also differ over time during a pandemic, which may be due to its severity (see Schraff 2020).

The dependence of trust on context has also been noted by Newton et al. (2018), and they moreover explain that various theories look at these two types of trust differently: according to one approach, trust is universal, according to another one it is a matter of context. According to the latter approach, trust can change if a situation changes. The relationship between social and political trust may weaken or disappear altogether if society is in serious difficulties (Newton et al. 2018: 40), and a pandemic is certainly such a situation.

As Newton et al. (2018) state, social and political trust are intertwined, and recent studies agree that there are mostly positive associations among different types of trust. However, the relationships between these types of trust can change over time. Therefore, it depends on the specific conditions in the given historical period, what the levels of the different types of trust are, and how strong the associations between them are (Newton et al. 2018).

Examining trust during a pandemic should not just focus on the level of trust, whether social or political, or the role that trust plays in people adhering to hygiene measures or implementing policies. As Devine et al. (2020) note, studying trust in this kind of crisis also sheds light on the relationship between social and political trust, its dynamics and the mechanism behind it. We want to contribute to this knowledge with our study and further explore the relationship between these two types of trust.

In our text, we follow the study by Zhang et al. (2020) who examined the relations between different types of trust in 11 countries. In their analysis they use a 21-item measure that was reduced to 7 factors: trust in representative government, trust in governing bodies, trust in security institutions, trust in financial institutions, trust in knowledge producers, trust in community, and trust in close relations (Zhang et al. 2020: 3). As mentioned above, two clusters appeared in the network that clearly demonstrated that social and political trust are two separate concepts, but also that there is a connection between them. Trust in governing bodies (judiciary, tax system, etc., see Liu et al. (2018) for more details) has a central position in the resulting network. This type of trust links strongly to other types of political trust, namely trust in representative government and trust in security institutions, but also to one type of social trust – namely, trust in community (trust in neighbours, citizens of the country, etc.). Trust in community is also connected to trust in security institutions (Liu et al. 2018).

Using network analysis to understand the structure and interconnection of attitudes is a relatively new approach (see, for example, Boutyline and Vaisey 2017; Brandt et al. 2019; Zhang et al. 2020). In general, the basic idea behind this method is that individual attitudes are nodes in a given network and the correlations between them are edges. This method allows us to examine individual attitudes in terms of their position in the network and to assess which attitudes are central and which are peripheral, thereby determining how important and interconnected they are.

An important finding from Zhang et al.'s (2020) research for our analysis is that the described network has a similar structure in all countries studied, and therefore we can consider the results of this analysis to be relatively robust. Although the Czech Republic was not among the countries examined, the diversity of these countries and the consistency of the results suggest that this network would have a similar structure here, too.

Since we can assume that a pandemic is a situation that can affect the relationships between different types of trust (see Newton et al. 2018), it is important to explore what is the relationship of different types of trust during the coronavirus crisis, whether their division into two types of clusters applies, and how strong are the relationships between various types of trust and their clusters.

In this connection we test the following hypotheses. The first is that governing bodies play an important role and based on the literature (Zhang et al. 2020) we could expect that trust in these institutions occupies a central position in the trust network (H1). On the other hand, we could expect that their importance will decrease and that executive institutions (i.e. representative government) will be at the centre of the network, as these institutions play a major role during the crisis (H2), or we could expect security institutions to be at the centre, as they are in some countries, according to Zhang et al. (2020), and they can also play an important role in crisis management (H3).

Studies on trust in a pandemic situation usually also examine trust in institutions that are directly involved in resolving the crisis, such as the Ministry of Health (Prati et al. 2011) or health professionals (Dryhurst et al. 2020). For this reason, we have included this specific form of trust in our study. Trust in the media (Prati et al. 2011; Kye and Hwang 2020), which can be both the carrier of the message but also the actor itself, is also explored. While medical staff (H4a) and the media (H4b) are institutions that can play a crucial role during the coronavirus crisis, we assume that they could occupy a more central position in the trust network.

An important element in the study of trust in a pandemic is the emotional response, i.e. how much people are concerned about the epidemic (Prati et al. 2011; Bavel et al. 2020). People who are worried about the disease are more likely to follow proposed hygiene measures (Prati et al. 2011) and express a willingness to vaccinate themselves (Rönnerstrand 2013). How concerned people are is also closely related to different types of trust. People who are worried (measured as a part of risk perception) have more trust in science and medical professionals, and concern is negatively associated with trust in government (Dryhurst et al. 2020). Therefore, we can also assume that different levels of concern about the coronavirus pandemic may induce different network structures of trust (H5). In our analysis, we distinguish the structure of trust for those who show high and low levels of concern. Although we can expect that the relationships between various types of trust differ for people who are more and those who are less concerned (see Dryhurst et al. 2020), we cannot state exactly how these associations will differ in strength or in the positivity or negativity of ties. However, we suppose that for people who are more concerned, the context, i.e. the pandemic situation, is important.

The data come from a special round of data collection conducted within the ‘Our Society’ project. Respondents (N = 1043) were selected using quota sampling and the quota variables were sex, age, education, economic status, frequency of internet use, region (NUTS 3), and size of place of residence. CAWI and CATI were the modes of data collection used and the data are representative of the population of the Czech Republic aged 18 and over. Fieldwork was conducted from 7 May to 23 May 2020. To add context to the data collection information, it should be noted that the state of emergency declared in the Czech Republic in response to the coronavirus lasted from 12 March to 17 May 2020, and at the time of data collection the population had had approximately two months of experience with anti-coronavirus measures and government intervention, but at that point a slight relaxation of some of the measures was being introduced.

We include in our analysis the types of trust that were presented in Zhang et al.'s (2020) study, but we do not measure them using the same number of items, and the items in our study are the ones given here in italics: trust in government (national government, local government, prime minister or president), trust in governing bodies/non-partisan government bodies (judiciary/courts, government surveillance agencies, trust in election outcomes, the tax system), trust in security institutions (police, military), trust in financial institutions and corporations (banks, stock market, multinational corporations, oil companies), trust in knowledge producers (scientists, universities), trust in community (neighbours, one's own ethnic group, other citizens of one's country), trust in close relations (friends/close ones, trust in extended family, trust in immediate family). The items we omitted were either irrelevant for the Czech context (oil companies) or were not included in the research due to the length of the questionnaire. Conversely, the items that have been included are ones that are regularly examined, and their factor structure is known. In the type of analysis that we conduct, omitting some items from a given type of trust may not necessarily be an issue.

Because we wanted to explore the specific situation of the coronavirus crisis, we added to the original items also trust in two other institutions – medical professionals and media – so that we could examine their position in the network.

We measured the trust items on a four-point scale (I definitely trust – I definitely distrust). The wording of the questions and descriptive statistics can be found in the online appendix.

To distinguish between people more concerned about the coronavirus pandemic and people who were less concerned, we used a scale of 8 items that measured people's concern about health, the economy, supplies, and the capacity of hospitals given the spread of the disease (α = 0.73). We then divided the respondents into two groups based on the median. The wording of the questions and descriptive statistics can be found in the online appendix.

When we were preparing the data for our analysis, we recoded the ‘don't know’ responses as missing values. We also excluded from the analysis respondents with more than 7 missing items out of 22, resulting in a sample size of N = 1005. In other cases, we performed data imputation using the method of multivariate imputation by chained equations (Buuren and Groothuis-Oudshoorn 2011) for all items. We imputed data based on sociodemographic variables (gender, age, education) and the other items of trust examined. We also performed an analysis on data with listwise deletion (N = 696), and the results are very similar to the results using imputed data but are statistically less significant. The results based on data with listwise deletion are available in the online appendix.

We created our network by replicating Zhang et al.'s (2020) procedure, but instead of factor scores we used the average item scores if a particular type of trust was measured by more than one item. The original robustness analysis performed by Zhang et al. (2020) shows that similar results are obtained using the factor scores and the average values of the scale. All the values are thus calculated in a similar way. The resulting network is based on regularised partial correlations (Epskamp and Fried 2018). For regularisation we used graphical LASSO (least absolute shrinkage and selection operator) with a tuning parameter chosen using the Extended Bayesian Information criterium (EBIC). The R-package ‘bootnet’ was used for the calculations and the R-package ‘qgraph’ for network visualisation. Confidence intervals and significance tests are based on bootstrapping with 1000 resamples. Like Zhang et al. (2020), we used strength centrality to determine the importance of an item (particular trust) within the network. We also calculated closeness centrality and betweenness centrality with results that are similar to those of strength centrality.

As we can see in the network of trust for the whole sample, when medical staff is not added to the network, trust in security institutions is the most central node that interconnects the other parts of the network. Security institutions are to a statistically significant degree the most central of all the nodes in the network. The second most central position is held by trust in scientists, but their position does not differ to a statistically significant degree from the centrality of the remaining nodes. Conversely, the position of trust in courts (as a representative of governing bodies) is rather average and does not differ to a statistically significant degree from most other nodes. The role of trust in representative government is quite marginal and is to a statistically significant degree the least central of all the other nodes except trust in close relations.

Therefore, the results support Hypothesis 3, and not Hypotheses 1 and 2. This means that the most central role in the time of the coronavirus crisis is occupied by those institutions that have been playing an important role during the crisis, such as security institutions or scientists, who are important for helping to resolve the crisis.

To further elaborate this hypothesis, we added to the networks the item of trust in medical staff, who have played a crucial role in the fight against the coronavirus. Healthcare professionals occupy the central position in the whole sample network and this centrality is statistically significant. It is healthcare professionals and security institutions that connect the different parts of the network, and their central role is statistically significant. These results support Hypothesis 4a and again confirm that the role of trust in the institutions that are directly dealing with the crisis is important.

In addition to healthcare professionals, we also tried to add the media to the networks, as we consider them to have been a crucial actor during the crisis, but they were found to occupy a more peripheral position in the networks, which does not support Hypothesis 4b. More detailed results can be found in the online appendix.

The aim of our study was to examine the relationship between different types of trust during the coronavirus crisis. We found two clusters: one cluster of social trust and one cluster of political trust. These clusters are interconnected, but in a different way than in Zhang et al.'s (2020) study. Based on a network analysis, we showed that the institutions directly dealing with the crisis are the ones that are at the centre of the network. This applies in particular to security institutions and medical staff.

We also found that the structure of trust differs for people with different levels of concern about the coronavirus crisis. People with a higher level of concern have a more structured network of trust, while for people with a lower level of concern both networks (with and without the medical staff) are less structured. The differences in the network structure of trust are all the more important given that the overall level of trust is quite similar for people with different levels of concern. At the same time, these different networks suggest that different structures but not necessarily different levels of trust arise in relation to how much people are worried about the coronavirus crisis. We can conclude that different levels of concern create a different context for linking social and political trust.

Boutyline and Vaisey (2017) consider the central nodes of the network of attitudes to be the points from which other elements of the network derive. In this respect, it is important to recognise that it is trust in the institutions dealing with a crisis that connects and can affect trust in other institutions. Based on their analyses, where governing bodies were the central points of the network, Zhang et al. (2020) concluded that trust is mediated by institutions that effectively carry out non-political activities in the functioning of the state, and thus, through trust, they connect the ‘bottom-up and top-down structures of society‘ (Zhang et al. 2020: 8). By analogy, we can say that at the time of the coronavirus epidemic, key functions performed by the state are being shifted to other institutions, which are becoming these interconnectors.

As Zhang et al. noted (2020: 9), in research on trust, great attention is paid to the level of trust, but the study of trust could be extended to other aspects that can help us understand the connection between different types of trust and the implications this has for the functioning of democracy. In our case, the study of trust in a time of crisis or during a pandemic should focus not only on the level of trust but also on the relationships between individual trusts, as well as the role these relationships play in the required behaviour and policy implementation (see the call for more research on trust in Devine et al. (2020)).

Our study has, however, several limitations. We cannot be sure whether this structure is universal or is typical for the Czech Republic. The stability of the network of trust across 11 very different countries, some of which are relatively culturally close to the Czech Republic (Estonia, Poland), suggests nevertheless that the structure described in the original article is relatively general. Another possible source of bias is that we do not use all the items from the original scale and we also do not use factor scores. Nevertheless, the original robustness analysis shows that results similar to the factor scores are obtained using the average values of the scale.

It is important to mention that our study also has the same limitations as Zhang et al.'s (2020) research. We performed the analysis on cross-sectional data, and we have the same limitation on the number of types of trust as they do, though the types of trust used are quite varied.

Although trust may vary between countries and contexts, and the results of a study conducted in only one country need to be taken with caution, some characteristics of the trust networks we examined could be valid in countries with a similar cultural tradition and where the pandemic has followed a similar course.

No potential conflict of interest was reported by the author(s).

Matous Pilnacek is a researcher in the Public Opinion Research Centre at the Institute of Sociology of the Czech Academy of Sciences. His areas of interest include survey research methodology, pre-election surveys, and likely voter models.

Paulina Tabery is a researcher in the Public Opinion Research Centre at the Institute of Sociology of the Czech Academy of Sciences. Her research interests include opinion formation, interpersonal and media communication, opinion polls, and survey research methodology.