Positive and negative health events and trust

Abstract

Trust plays a vital role in shaping numerous socio-economic behaviors. However, little is known about how significant life events, such as adverse and beneficial health events, shape up trust. Therefore, we examine the impact of positive and negative health events on general trust using the LISS panel (3911 respondents) from the Netherlands. The primary health event variable is constructed by comparing self-reported health in 2017 with 2018, while trust measured in 2019 (after the health survey) is used as the outcome variable. We find a significant and negative impact of adverse health events on trust, while positive health events have an insignificant effect. We perform several robustness checks that support our main findings. Further investigation shows that negative health events impact trust by reducing psychological wellbeing and social interactions. The results provide fresh insights into the health events and trust nexus.

1 Introduction

Trust plays an important role in shaping up numerous behaviors. For example, in the domain of economics, trust impacts investment decisions and their efficiency (Dearmon and Grier 2009), influences schooling choices (Bjørnskov 2012), and human capital accumulation (Knack and Keefer 1997). Similarly, trust positively influences attitude towards democratic institutions, participation in civic organizations, and politics (Rothstein and Uslaner 2005). All these aspects determine micro development that provides foundations for macroeconomic growth. In the social domain, trust contributes to the societal wellbeing by making citizens tolerant towards minorities, promoting charity giving and volunteering (Uslaner 2002; Rothstein and Uslaner 2005). In the domain of health economics, trust promotes beneficial connections and partnerships among communities to collectively respond to epidemics, and to follow strategies designed to counter future emergencies (Nuriddin et al. 2018; Ryan et al. 2019; Yuan et al. 2022; Umer 2023a, b, c). In the psychological domain, trust influences personal satisfaction, optimism about one’s own life and happiness (Delhey and Newton 2003; Rothstein and Uslaner 2005).

The aforementioned avenues through which trust impacts socio-economic and psychological decisions necessitate the examination of how trust itself shapes up as individuals pass through different circumstances. The existing studies have already examined the impact of socio-economic, political (Delhey and Newton 2005), and psychological (Bogliacino et al. 2021) factors on trust. However, we have a limited amount of information about how positive and negative health events¹ influence trust. It is important to examine the impact of health on trust because both health and trust play an important role in daily activities and decisions; health events can be more frequent compared to the other shocks, such as financial shocks and natural disasters (Bai and Li 2021), and can have a heterogeneous impact across individuals. Also, understanding the impact of health on trust is important because health can be modified by individual decisions (unlike exogenous factors such as natural disasters), thus, indirectly, trust could also be modified if there is evidence for the impact of health on trust. Therefore, the current study examines the impact of both positive and negative health events on trust to further extend our understanding about the health-trust nexus.

Health shocks can influence trust through several channels. First, negative health events can limit social interaction by forcing sick individuals to take precautionary measures, and can hamper their physical strength required for healthy social interactions. A decrease in social interactions can eventually lead to a trust deficit. We can expect contrary outcomes in case of positive health events. Second, the negative (positive) health events can arouse negative (positive) emotions, which can translate into pessimism (optimism) and have a detrimental (beneficial) impact on trust. Third, trust entails the risk of betrayal. A negative health shock can make people more risk averse, (Decker and Schmitz 2016), and encourage them to minimize the risk of further betrayal by trusting less often. These potential mechanisms are suggestive links through which health shocks can influence trust. A conventional economic perspective however would suggest that trust, like other preferences, is stable² (Rabin 1998), and therefore, would not fluctuate due to health shocks. Therefore, a contrast between suggestive mechanisms and the conventional views sets up a perfect background to empirically test whether health shocks impact trust in the short run.

The pre-pandemic literature on the interplay of health shocks and trust is scarce.³ Chuang and Schechter (2015) examine the impact of sickness duration on trust, and find no significant impact. Their work however does not provide a causal impact of health shocks on trust. After the start of the pandemic, numerous studies have examined the impact of Covid-19 infections on trust. The findings do not converge. Most studies including Brück et al. (2020), Bellani et al. (2022) and Bogliacino et al. (2021) and Umer (2023c) find that the Covid-19 infections have insignificant impact on trust, while Gambetta and Morisi (2020) find that the Covid-19 infections have a positive impact on trust. These findings do not provide a unidirectional evidence for the impact of health shocks on trust. The divergence in findings can be due to the different nature of data used in these studies⁴ as well as different types of trust being used as outcome variable.⁵ Also, most studies conducted after the start of the pandemic used Covid-19 infections as a health shock. Therefore, findings cannot be applied to other common and frequent diseases that are fundamentally different from the Covid-19. Last but importantly, the existing studies do not examine the possible impact of positive health events on trust. Therefore, it is unknown whether and how improvements in health can influence trust.

In the current study, we try to bridge the gaps in the literature and examine the impact of positive and negative health events on general trust using the LISS panel data from the Netherlands (n = 3911). Our main health events variable compares self-reported health in 2017 with 2018 (no health event if health in 2017 and 2018 is the same; positive health event if health in 2018 improved and negative health event if health in 2018 depreciated compared to 2017), and subsequently, examines its impact on general trust elicited in 2019. The general trust in 2018 was elicited about four months prior to the health module. Therefore, to correctly identify the causal impact of health events on trust, we use trust data from 2019. We find robust evidence that the negative health events decrease trust, while positive health events have an insignificant impact. Our work also examines the two possible mechanisms (psychological wellbeing and social interactions) through which health events can affect trust. We find that the negative health event has a significant and detrimental impact on the general feelings, satisfaction with life and on social interactions, which in turn can have a detrimental impact on trust. The positive health event has an insignificant impact on these aspects.

As a robustness check, we perform a difference-in-differences (DID) analysis that controls for the individual fixed effects and provides causal inference. DID estimates also support our main findings. We also perform additional checks by using another measure of health events based on the self-reported assessment of changes in health during the last one year. This measure also shows a detrimental effect of adverse health events on trust, no significant impact of positive events, and supports the main findings.

Our study contributes to the existing literature through several avenues. First, we use data that is representative of the Dutch population and therefore, the findings have higher external validity at least in the Dutch context. Second, we examine the strength of the findings through different statistical checks and operationalization of health events and hence, the findings are robust. Third, apart from the negative health events frequently examined in the literature, we study the influence of positive health events as well. Therefore, we provide health event’s two-dimensional analysis (positive versus negative) that is important to understand the comprehensive impact of health changes on preferences.

The paper is organized as follows. The second section has data details, attrition, and endogeneity assessment. The third section reports the main findings, while the fourth section lists robustness checks. Fifth section discusses possible mechanisms behind the observed results, while the last section concludes the paper with relevant implications.

2 Data description

We use data from the Longitudinal Internet studies for the Social Sciences (LISS) panel that is administered to about 5000 households in the Netherlands, comprising 7500 individuals (at least 16 years old, Dutch speaking and permanently residing in the Netherlands). The respondents are selected based on a true probability sampling, and is representative of the Dutch population. The panel has 10 different modules available till 2022 that are part of the LISS core study.⁶

We do not use data from 2020 onwards because the pandemic can influence both health and trust at the same time. Therefore, the impact of health events on trust can be confounded by the pandemic’s effect and might lead to an incorrect assessment of health events on trust. To avoid this issue, we rely on data from 2017 to 2019 waves. We focus on a relatively narrow time window because of several reasons. First, a single baseline and follow up surveys are very suitable for studying diseases and health-related aspects because these are highly autocorrelated (McKenzie 2012). Second, as a robustness check, we perform a difference-in-differences (DID) analysis. The parallel trend assumption critical for the DID analysis is more likely to hold when the time between the survey waves is narrow.

We construct health events variable based on the subjective information about health reported in the module “Health” administered in November–December 2017 (wave 10) and later in November–December 2018 (wave 11). We do not use health data from 2019 because the “Personality” module of the LISS panel that elicits trust is administered before the health module in May–June every year. Therefore, to correctly identify the impact of health events on trust, we rely on health data from 2017 and 2018 waves. To construct the outcome variable ‘Trust’, we rely on the personality module administered in May–June 2019. However, to perform baseline checks, we rely on “Personality” module administered in May–June 2017 prior to the “Health” modules used for the construction of health events. For demographic and income controls, we use data from the module “Background Variables.” This module is administered every month. We however use these variables reported in December 2017 (before treatment) for baseline checkups and for our main analysis. We also provide a timeline of the surveys in Table 1 for the convenience of readers. Merging the background information (2017 and 2019) with personality modules (2017 and 2019) and health information (2017 and 2018) provides us with 3994 matched observations.

Table 1 Survey timeline

2.1 Health events variable

Multiple methods to construct health shocks or events are used in the existing literature. Several studies relied on changes observed in the self-reported health of individuals (Clark and Etilé, 2002; Disney et al. 2006; García-Gómez 2011; Islam and Maitra 2012; Banks et al. 2020). Some studies relied on direct questions about health shocks (Gloede et al. 2015; Bünnings 2017). Others used severe health conditions or the strength of hand grip as proxies for health shocks (Decker and Schmitz 2016; Banks et al. 2020; Rice and Robone 2022; Smith et al. 2022). In the development economics literature, studies also used the number of days off from work due to illness as a proxy for health shock (Islam and Maitra 2012).⁷

To construct health events variable, we follow studies from the existing literature that use changes in self-reported health information. Specifically, we use the following self-evaluation of health by respondents.

How would you describe your health, generally speaking?⁸

$$1 \, = \, Poor \quad 2 \, = \, Moderate \quad 3 \, = \, Good \quad 4 \, = \, Very \, Good \quad 5 \, = \, Excellent$$

We compare the answers to the above question from 2017 with 2018 surveys to construct the following health event variable.

$$\begin{aligned} Health \, Event \, = & \,\, \, 0 \, if \, health \, reported \, in \, 2017 \, and \, 2018 \, is \, equal \, \left( {health \, unchanged \, or \, no \, event} \right); \\ & \,\,1 \, = \, if \, health \, improved \, in \, 2018 \, compared \, to \, 2017 \, \left( {positive \, event} \right); \\ & \,\,2 \, = \, if \, health \, deteriorated \, in \, 2018 \, as \, compared \, to \, 2017 \, \left( {negative \, event} \right) \\ \end{aligned}$$

For example, if a respondent reported a ‘Poor’ health in the 2017 wave, and reported anything other than ‘Poor’ in the 2018 wave, that respondent experienced an improvement in the health (positive health event). Similarly, if a respondent reported ‘Excellent’ health in the 2017 wave, and reported anything other in the 2018 wave, that respondent experienced a deterioration in health (negative health event).

The average change in the reported health for the positive and negative health events is 1 unit. This is because only a few respondents reported extreme positive changes in health from 2017 to 2018 (poor to excellent: n = 0; poor to very good: n = 2), or extreme negative changes from 2017 to 2018 (excellent to poor: n = 1; excellent to moderate: n = 1). Therefore, due to lack of reasonable variation in the data, we do not analyze the impact of extreme changes in health on trust. Rather we focus only on whether a change (positive or negative) is reported in the two survey ways.

The health events variable is operationalized to capture the impact of positive and negative health events as well as no event. This categorization is more helpful than the unilateral negative event because it provides wider insights into the interplay of health and trust. In literature, most studies have focused only on the impact of the negative health events, and therefore, relatively little is known about the impact of a positive health events on preferences. In this aspect, our work is different and enriched with broader analyses targeting both positive and negative health events.

The subjective nature of the health events variable has some apparent drawbacks. The respondents can either over- or under-state their health condition to conform themselves with the social desirability bias, or, to avoid social discrimination and stigma. Also, we do not know with surety whether any doctor has diagnosed the respondent with a health condition, and therefore, an incorrect self-assessment cannot be completely ruled out. Therefore, we perform additional robustness check by using another proxy for health events. We use the self-reported comparison of the current health compared to the last year. This additional measure also supports our main findings.

2.2 Outcome variable: trust

Our outcome variable ‘trust’ is measured through a standard trust question reported as follows.

Generally speaking, would you say that most people can be trusted, or that you can’t be too careful in dealing with people? Please indicate a score of 0 to 10.⁹

$$0 \, = \, You \, can^{\prime}t \, be \, too \, careful \;\;\;\;\;\;\;\;10 \, = \, Most \, people \, can \, be \, trusted$$

Two respondents in 2017 and five respondents in 2019 did not provide their answer to this question, and hence, they all are dropped from the analysis. Furthermore, a fraction of respondents (n = 83) do not answer the trust question in both waves, and thus, are also eliminated from the analysis. We also use several controls observed in 2017 (before the treatment) for our main analysis. Further details about the outcome variable, health events, and controls are reported in Table 2. We have a comparable number of respondents who reported a positive or negative health events, while most of the respondents (about 70%) reported identical health conditions in 2017 and 2018.

Table 2 Data summary

2.3 Attrition analysis

Merging the background module (December 2017) with trust and health information leads to 5,131 observations in 2017. A subset of respondents (n = 1005) did not participate in 2019 data (obtained by merging background information obtained in December 2019 with the personality module), leading to an attrition rate of about 20%. If attrition is non-random, it can lead to bias in estimates. Moreover, it is also possible that respondents with relatively stable preferences, specifically trust, remain in the sample while those with volatile preferences leave the panel. Therefore, as a first step of the analysis, we examine the randomness of attrition in data. The literature suggests several methods to perform attrition analysis.¹⁰ We examine whether any systematic differences exist in the outcome variable between attrite and non-attrite samples, as this is the most relevant assessment in the current context. A simple means comparisons tests show no significant difference in trust between attrite (average value = 5.93, n = 988) and non-attrite samples (average value = 6.04, n = 4045; t-stat = 1.34, p value = 0.18). We further analyze the randomness in attrition by using regression analysis and report the outcomes in Table 3 while providing the complete output in Appendix A.

Table 3 Summary of attrition analysis

The first regression is without controls, the second has all controls (measured in 2017) except income, while the third regression has income as an additional control. The coefficient for attrition variable is insignificant in all three regressions. In the fourth regression, we added an additional variable (interaction of attrition and health status in 2017). The coefficient for attrition is still insignificant. In fact, all coefficients for the interaction term (Attrite # Health) are also insignificant. Therefore, we do not find any evidence to suggest that the attrition is non-random.

The trust variable is elicited in 2017. The variable Attrite equals to one if the individual surveyed before did not participate in 2019 data. Demographic controls include age, family size, number of children, household head's living arrangement with partner, marital status, housing status, education, and income. Robust standard errors are in parentheses. *** p < 0.01, ** p < 0.05, *p < 0.10.

2.4 Examination of baseline differences in trust

It is important to examine whether trust at the baseline elicited prior to the health information is identical across people who reported a positive or negative health event. If trust at the baseline is different, it can cause endogeneity bias. The evidence specifically after the outbreak of the Covid-19 shows that trust is positively associated with preventive health measures (Bargain and Aminjonov 2020; Umer 2022). Therefore, differential trust at the baseline can influence health behaviors and hence cause health events, leading to an endogeneity problem. We perform several assessments to examine this possible endogeneity before we proceed to the main analysis. First, we perform a simple means comparison test to check if trust elicited at baseline in 2017 differs across three groups forming our health events variable. We do not find any evidence that trust differs across the three groups (output in Table 4).

Table 4 Comparison of trust at baseline: means comparisons

Second, we perform a regression analysis to further assess whether trust elicited in 2017 differs across different groups constituting the health events variable and summarize the outcomes in Table 5 while provide complete output in Appendix B. In the first regression we do not use any controls. In the second and third regressions, we use demographic and income controls measured at the baseline in December 2017. In all three regressions we do not find any significant coefficients to suggest that trust at the baseline is different among respondents who reported positive or negative health events in comparison to those who reported unchanged health. The results in Tables 4 and 5 together provide no evidence to suggest that trust at the baseline is different among the three groups.

Table 5 Comparison of trust at baseline: OLS Regressions

2.5 Empirical equation

We examine the impact of health events on trust by estimating the following equation with the help of OLS regressions.

$$Trust_{i } = \beta_{0} + \beta_{1} {\text{Positive Event}}_{ i} + \beta_{2} {\text{Negative Event}}_{ i} + \beta_{3k} X_{i} + \in_{i}$$

(1)

where $Trust_{i }$ is the outcome variable for individual i measured in 2019. ${\beta }_{1}$ and ${\beta }_{2}$ are coefficients of interest that measure the impact of positive and negative health events on trust with respect to individuals who reported no change in health. A significant ${\beta }_{1}$ and ${\beta }_{2}$ will indicate a significant impact of health events on trust with respect to individuals who did not experience any change in health. The vector ${X}_{i}$ contains control variables measured in 2017 while ${\epsilon }_{i}$ is the error term.

We estimate Eq. 1 by using three different regressions. In the first regression we do not include any controls. In the second regression we include all controls except income while in the last regression we include income as well. There are the least number of observations for income (n = 3717; Table 2) and therefore, to perform analysis with the enhanced number of observations, we exclude income in two of the three regressions. We consider coefficients to be non-zero if at least two of the three regressions provide significant results at the traditional significance levels.

3 Results

For brevity reasons, in Table 6 we report the main findings focusing on health events while report complete results in Appendix C. In comparison to respondents with unchanged health, we do not find any significant evidence to suggest that positive health events influence trust, irrespective of whether we do not add any controls (regression 1) or add controls (regressions 2 and 3), ceteris paribus. However, we do find a negative and significant coefficient for the negative health events in all three regressions, ceteris paribus. Specifically, the existence of negative health events decreases the level of trust by 3.2% (= 0.193/6.119) over the mean of the outcome variable, as indicated in column (3). It means a negative health event deteriorates trust in the short-run.

Table 6 Impact of health events on trust

As the three regressions in Table 6 use different numbers of observations, this could be a possible reason that can influence results. Therefore, we use the minimum number of observations used in regression 3, and re-run regressions 1 and 2. The coefficient for negative health events remains negative and significant in both regressions (p < 0.10). The output is reported in Appendix D.

We also added the health and trust information elicited in 2017 to the regressions to control for the possible mean reversion. The coefficient for the detrimental health events remains negative and significant (p < 0.01) in all three regressions. Furthermore, the coefficient for the positive health events also becomes significant in two regressions (p < 0.10), and provides some support that positive health events can improve trust (output in Appendix E). However, in Sect. 5, no robustness check provides any support to suggest that the positive health events improve trust.

Health events can be non-random and might cause bias in estimates. Therefore, we used propensity score matching (PSM). PSM selects individuals from the large control group who have similar pre-treatment characteristics to those in the treatment group. The differences in the outcome variable between the selected individuals (from the control group) and the treatment group are therefore attributable to the treatment (Caliendo and Kopeinig 2008). We implement PSM with all controls including income measured in 2017 (pre-treatment controls) that might be related to the treatment variable health events, and probit model to approximate a randomized experiment and to estimate a causal impact. PSM analysis shows insignificant impact of the positive health events. However, PSM gives a negative and significant (p < 0.10) impact of negative health events on trust (output in Appendix F). PSM findings also support the main findings in Table 6.

The estimation of treatment effects with matching estimators is based on the conditional independence assumption (CIA), that is selection based on the observable characteristics. However, if there are unobserved variables which affect treatment assignment and the outcome variable simultaneously, a 'hidden bias' might arise. We follow the strategy of Altonji et al. (2005), which utilizes selection on observables to evaluate potential bias from unobservables. Let ${\beta }^{R}$ denote the estimated coefficient for negative health events from the regression with no control variables (i.e., utilizing the coefficients in column (1) of Table 6; ${\beta }^{R}$ = − 0.211), and let ${\beta }^{F}$ denote the estimated coefficient from the PSM specification (i.e., column (2) in Appendix F; ${\beta }^{F}$= v0.242). Subsequently, the ratio can be calculated as $\left|{\beta }^{F}/{(\beta }^{R}-{\beta }^{F})\right|$. A larger ratio is more favorable due to two key intuitions: a smaller difference between ${\beta }^{R}$ and ${\beta }^{F}$ implies less impact on the estimate from selection on observables, and a stronger selection on unobservables is required (relative to observables) to explain away the entire effect.

Our results suggest that to completely negate the negative relationship between the negative health event and trust, the influence of remaining unobservable factors would need to be 8 times greater than that of observable factors. This scenario is reasonably unlikely, and therefore, we expect the potential bias that can be caused by unobservable variables is trivial.

4 Robustness checks

We assess the robustness of our findings by conducting several robustness checks. In the first part, we apply a difference-in-differences approach to obtain a causal impact of health events while controlling for the individual fixed effects. In the second part, we use an alternate measure of health events and examine its impact on trust.

4.1 Difference-in-differences (DID) approach

The DID method serves as our primary approach to conduct a robustness check, aiming to control for individual fixed effects, such as risk and time preferences, as well as extroversion. The objective is to establish a precise link between health events and their impact on trust. To implement the DID method, we utilize responses from baseline surveys (2017–2018) and follow-up surveys (2018–2019), constructing a balanced panel at the individual level.

The identifying assumption of the DID method is often referred to as the parallel trend assumption or the common trends assumption. This assumption states that, in the absence of the treatment (in this case, health events between 2017 and 2018), the average outcomes for the treatment and control groups would follow a parallel trend over time. In other words, any observed differences in the outcomes between the treatment and control groups before the treatment is applied can be attributed to pre-existing differences, but the trends themselves would be similar.

In ensuring the parallel trends assumption, we focus our analysis on individuals who did not report any change in their health status at the baseline in 2017.¹¹ This deliberate restriction ensures a consistent health profile for both the treatment and control groups, signifying the absence of health events between 2016 and 2017. By doing so, we aim to establish a baseline environment where any subsequent observed differences in outcomes can be more confidently attributed to the introduced health events in and after 2017. This reinforcement enhances the validity of the parallel trends assumption within our study and strengthens the causal relationship between health variations and changes in trust.

We incorporate the interaction term between our primary health events variable from the earlier section and the wave variable T in our DID estimation. Here, T takes the value of 0 for baseline survey and 1 for the follow-up survey. The model can be represented mathematically as follows:

$$Trust_{it } = \beta_{0} + \beta_{1} ({\text{Positive Event}} \times T)_{it} + \beta_{2} ({\text{Negative Event}} \times T)_{it} + \beta_{3k} X_{it} + \gamma_{i} + \delta_{t} + \in_{it}$$

(2)

where ${Trust}_{it}$ is the outcome variable for individual i at time t, and two health events variables denote the presence of positive and negative health changes between 2017 and 2018, as previously discussed. Each of these variables interacts with the binary variable for the wave indicator $T$. The resulting interaction terms capture the differential effects of health events post-intervention, indicating whether there is a significant change in the outcome due to positive or negative health events after 2017. To account for time-invariant individual characteristics and eliminate common trends across time, we include individual fixed effects and year fixed effects. Additionally, we account for time-variant characteristics, including age, family size, number of children, household head's living arrangement with partner, marital status, housing status, education, and income. Recognizing that the variations in some covariates, such as education, may be limited between 2017 and 2019, we gradually introduce these covariates to ensure the robustness of our results. ${\epsilon }_{it}$ is the error term.

Subsequently, the results are presented in Table 7 (complete results in Appendix G). In column (1), we conduct a regression without any individual-level covariates, while column (2) includes demographic controls, and column (3) incorporates an additional income control. Across all specifications, no significant coefficient is observed for the interaction term between positive health events and the year. However, in two out of the three regressions, a significant and negative coefficient is identified for the interaction term between a negative health events and the year. These DID estimates align with previous findings, indicating that negative health events have an adverse impact on trust, with an effect size equal to 2.6% in columns (1) and (2).

Table 7 Health events and trust: DID Estimation

4.2 Using alternate measures of health events

We consider an additional health event variable based on the subjective health reports to examine the robustness of findings reported earlier. This variable is constructed from the following question administered in 2018:

Can you indicate whether your health is poorer or better, compared to last year?¹²

1) Considerably poorer	(n = 70; 1.79%)
2) Somewhat poorer	(n = 674; 17.24%)
3) The same	(n = 2645; 67.65%)
4) Somewhat better	(n = 448; 11.46%)
5) Considerably better	(n = 73; 1.87%)

As the question already elicits health reports for the last year, it automatically provides comparative evaluation. We merge the first two categories to construct negative, and the last two categories to construct positive health events. We term this new variable as Health Event 2:

$$\begin{aligned} Health \, Events \, 2 \, = & \;\; \, 0 \, if \, health \, reported \, is \, same; \\ & \;\;1 \, = \, if \, health \, improved \, somewhat \, or \, considerably \, \left( {positive \, event} \right); \\ & \;\;2 \, = \, if \, health \, deteriorated \, somewhat \, or \, considerably \, \left( {negative \, event} \right) \\ \end{aligned}$$

The above variable is slightly different from the main health event variable reported in Sect. 2. The main health events variable is constructed by comparing the self-reported health in 2017 with that reported in 2018. On the other hand, the Health Events 2 variable is constructed based on the self-assessment of respondents of their health in the last year. While the operationalization of the two health events variables is different, they both essentially capture the subjective evaluation of one’s own health. Moreover, both variables also capture positive and negative health events, as well as unchanged health. Therefore, both variables are similar. We add a cross table that illustrates the alignment between the baseline and alternative health event measures (Appendix H).

The regression output with the Health Events 2 variable is reported in Table 8, while the complete output is in Appendix I. There is no significant impact of positive health events on trust in all three regressions. The negative health events have a detrimental impact on trust, and the coefficient is highly significant in all three regressions. Specifically, negative health events decrease the level of trust by almost 10% (= 0.604/6.120) over the mean of the outcome variable, as indicated in column (3). These findings support the main results reported in Table 6. The magnitude of impact of negative events however is quite large in case of Health Events 2 variable compared to the main variable (Health Events). This difference is likely associated with the differences in the construction of the two health events variables. Nevertheless, both variables provide strong support for the detrimental impact of negative health events on trust.

Table 8 Health events and trust: using alternate health events variable

5 Mechanisms

While there can be several reasons behind the detrimental impact of the negative health events on trust, here are the ones that appear to be the most relevant. First, negative health events can restrain social interactions of sick people by limiting them to home or requiring them to spend time in the hospital. The limited social interaction can be severe and prolonged especially for people suffering from fatal diseases. The diminished social interaction due to the negative health events can be a possible reason for the decrease in trust. Moreover, sick people might also feel uncomfortable around people due to the symptoms of their diseases, which in some cases can be quite visible and distinct. This discomfort can in turn have a detrimental impact on trust. Second, occasions like negative health events can induce negative feelings and emotions and cause psychological dissatisfaction with a greater intensity¹³ (Terpstra 2011), and these can ultimately have a detrimental impact on trust.¹⁴ We do not find a significant impact of positive health events perhaps because an improvement in health is a natural and expected condition of human beings. Moreover, such a condition might not influence social interactions, emotions and probably risk taking. In this section we explore these mechanisms with the help of available data from the LISS panel.

5.1 Health events and psychological wellbeing

We examine the impact of health events on feelings and satisfaction (proxies of psychological wellbeing) with the help of the following two survey items from the Personality module administered in May–June 2019.

In general, how do you feel? 1 = Very bad; 7 = very good. [n = 3911; Mean = 5.71]¹⁵

I am satisfied with my life. 1 = Strongly disagree; 7 = Strongly agree [n = 3911; Mean = 5.41]

The main findings are in Table 9, and complete output in Appendix J. Negative health events have a significant and negative impact on feelings (Panel A) and life satisfaction (Panel B) in all three regressions. Specifically, negative health events decrease general feelings by 1.6% (column 3) and satisfaction with life by 2% (column 6). However, positive health events have an insignificant impact on these feelings. We also conducted a DID analysis by using information on general feelings and life satisfaction from the 2017 and 2019 waves. The DID results also show a negative and significant causal impact of only negative health events on the psychological wellbeing (results in Appendix K). These findings provide support to the earlier argument that the negative health events can have a detrimental impact on feelings and satisfaction with life, which in turn can reduce trust.

Table 9 Impact of health events on feelings

5.2 Health events and social interactions

We examine the impact of health events on social interactions with the help of the following two survey items that were a part of the Personality module administered in May–June 2019.¹⁶

Feel comfortable around people. 1 = Very Inaccurate; 5 = Very Accurate. [n = 3906; Mean = 3.81]

True friendship. 1 = Extremely unimportant; 7 = Extremely important. [n = 3896, Mean = 5.95]

The main findings are reported in Table 10 while complete output is in Appendix L. The negative health events have a significant and negative impact on feeling comfortable around people (Panel A, Regressions 1 and 3) and on the importance of true friendship (Panel B, in all regressions). Specifically, negative health events decrease comfort around people and the value of true friendship by about 2% (columns 3 and 6). However, positive health events have an insignificant impact on these behaviors. We also conducted a DID analysis by using the information on social interactions from 2017 and 2019 waves. The DID results also show a negative and significant causal impact of only negative health events on social interactions (results in Appendix M). These findings also provide support to the earlier argument that negative health events can reduce social interactions by having a negative impact on friendship, and can make sick people uncomfortable around others probably due to their symptomatic diseases.

Table 10 Impact of health events on social interactions

6 Conclusions and implications

In this study we examined the impact of positive and negative health events on the general trust using the LISS panel data from the Netherlands. We find robust evidence that negative health events have a detrimental effect on trust, while positive health events have no significant impact on trust. We explore psychological wellbeing and social interactions as the possible mechanisms through which health events can influence trust. We find evidence that only negative health events have a significant and detrimental impact on both psychological wellbeing and social interactions, which in turn can have a detrimental impact on trust. We also conduct several robustness tests to scrutinize our main findings, and all tests show that only negative health events have a significant impact on trust.

Our work provides systematic evidence on the impact of both negative as well as positive health events on trust. We also provide an interesting perspective on the possible mechanisms through which health events can affect trust. We hope that these findings will provide relevant information to future studies that intend to examine the influence of health events on preferences. It is worth exploring the possible heterogeneous impact of different diseases on other social and economic preferences. Also, our work relied on only subjective measures of health events, and therefore, the findings might change if an objective measure based on the actual disease information is used for analysis. We leave these avenues for the future research.

Our findings have several practical implications. First, negative health events not only have a direct impact on the body, but can have spillover negative effects on a range of economic outcomes via a decline in trust. The negative impact on economic outcomes can ultimately influence the trajectory of micro wellbeing at the individual level, and at the macro level as well. Second, a decline in trust due to negative health events can also cause psychological unrest by diminishing happiness and personal satisfaction. Both of the aforementioned aspects suggest that we need to look beyond the direct impacts of negative health events and consider the indirect impacts as well that can occur through a decline in trust. Accounting for these indirect impacts can help in formulating a comprehensive policy to combat the impact of negative health events on socio-economic outcomes. The absence of a significant impact of positive health events on trust is an unexpected outcome, and shows that health improvements do not necessarily repair the damage done to trust due to the negative health events. Therefore, special efforts are required to repair trust after negative health events.