In this section we present the estimation results, using the methodology outlined in the previous section. In addition, we supplement the results with an analysis of subpopulations of compliers and non-compliers with social distancing and face covering measures. We dropped the observations for individuals who were indifferent in trusting the public institutions. Thus, the number of observations is lower than the sample size, i.e. 934 observations, which we still deem sufficiently large for the robustness and representativeness of our analysis.
We predominantly focus on the impact of trust in different institutions on compliance. In the first column we present the results considering trust of respondents in health institutions. The second column considers trust in government, while the specification used in the third column considers trust in the pandemic commission. Finally, in column four we evaluate jointly the effect of trust variables on compliance with introduced measures. The significance of socio-economic characteristics is considered in each regression as well.
A. Effect on compliance with social distancing measures
Estimation results in column 1 of Table 2 suggest that trust in health institutions has a positive and statistically significant impact on compliance with social distancing measures. We find that those who trust health institutions are 59 percent more likely to follow social distancing. This estimate is significant at 1 percent level. Considering the trust estimate in the second column we find that those who trust the government are approximately 43 percent more likely to comply with social distancing measures, and this estimate is significant at 5 percent level. If we consider the individual effect of trust in the pandemic commission (column 3), the results suggest that those who trust this institution are highly likely to comply with the given measure. Particularly, those who trust the pandemic commission, are almost 86 percent more likely to practice social distancing while this estimate is statistically significant at 1 percent level of significance. Finally, if we consider the trust variables jointly, the estimates indicate that only trust in pandemic commission has a positive and statistically significant impact on compliance with social distancing measures. Individuals trusting the pandemic commission are 88 percent more likely to follow these measures and this estimate is significant at 1 percent level. Since the pandemic commission consists of health experts as well as health institution representatives and government officials, this might be the reason why other trust variables are not significant in this particular analysis. Thus, it may be the case that trust in pandemic commission partly captures trust in other institutions too.
In both analyses, where the trust in the permanent pandemic commission, which during the first wave of the COVID-19 pandemic, was the main expert advisory body of the government, this variable was significant at 1 percent level. Also, other trust variables were significant as explained above. Thus, this indicates that trying to enhance the compliance with social distancing measures, the government should build trust of their citizens in the government itself and in other relevant public institutions.
As for the other factors significantly affecting the compliance with social distancing measures, we evaluate the impact of other socio-economic characteristics. As can be seen in Table 2, the compliance of people, who are employed tends to be significantly larger than those, who have other employment status. The estimates suggest that employed individuals are 60 to 65 percent more likely to comply with social distancing measures than people with other employment status and this estimate is statistically significant at 1 percent level. This is an understandable finding, since in the work place the virus can spread easily and compliance with social distancing measures can contribute to decreasing its spread.
As can be seen in columns 3 and 4, the related regression analysis has shown that pensioners are also significantly more likely to comply with social distancing measures, but these estimates are statistically significant only at 5 percent level. This can be explained by older people being more vulnerable to the virus. Moreover, the estimated results suggest that those who feel in danger of COVID-19, are about 50 percent more likely to follow social distancing measures, while these estimates are significant at 5 percent level for each set of the analysis performed. This is a sign that people who feel in danger tend to comply with the rules more, which implies the importance of mass objective information about the risks of this particular disease (but also in other contexts) to make people fully aware of health consequences of non-compliance.
In our study, gender and the number of people living in a household do not statistically significantly affect the compliance of people with social distancing measures. However, it is interesting that individuals who have more mobile phones tend to statistically significantly comply less with social distancing measures. The estimates indicate that individuals who have more mobile phones are up to almost 45 percent less likely to comply with these measures. Perhaps, this variable could be used as a proxy for more extensive use of online media (e.g. social networks), including those relating to the diverging/fake news about the virus and related effects. And thus, more extensive use of online media could create confusion and consequently lower compliance with social distancing measures.
In addition to the regression analysis of the causal relationship between exogenous and dependent variables of our interest, we also looked into the proportions of compliers, who trust public institutions.
As can be seen from Table 3, those who comply with social distancing measures, are more likely to trust public institutions. The results suggest that almost 80 percent of compliers with social distancing measures trust the health institutions. We can see similar pattern when we consider the trust in pandemic commission – in this case 71.5 of compliers trust the pandemic commission.
If we consider the subpopulation of non-compliers, we find that those who do not follow the social distancing measures are less likely to trust the government. However, quite interestingly, if we consider the subpopulation of non-compliers with social distancing measures, we find that approximately 71 percent of non-compliers trust the pandemic commission, but they do not follow their recommendations on social distancing.
B. Effect on face covering compliance measures
In Table 4 we present the results of the analysis of compliance with face covering measures, i.e. with wearing face masks.
In contrast to the results presented in the previous section, the estimated relationship between trust in public institutions and compliance with face covering measures is insignificant. This holds when we consider the effect of trust variables independently of each other, as well as when we analyse their joint impact on compliance behaviour. This may be because wearing face coverings was compulsory at the time, when this survey was carried out.
However, we find that some individual characteristics have positive and significant impact on compliance with given measure. Even though we find no statistically significant relationship between being a pensioner and compliance, the estimated results presented in Table 4 suggest that individuals older than 50 years 55 to 57.5 percent more likely to comply with face covering measures. These results are highly statistically significant, i.e. at 1 percent level. This can be due to older people being more vulnerable to the virus and be willing to protect themselves using face covering. However, given that age did not come out as statistically significant, when social distancing measures were considered, we might speculate that this may perhaps be due to lower intensity of social contacts of older people, who may not consider it important for them to comply with social distancing measures. Estimates in Table 4 also suggest that those who fell in danger of COVID-19 are more likely to wear face coverings. In particular, they are approximately 25 to 26 percent more likely to comply with face covering measures. Finally, we find that the number of people in the household (i.e. the size of the household) has a positive statistical impact on compliance with face covering measures. The estimates for all four specifications are significant at 5 percent level. Also, people, who feel in danger of COVID-19 tend to statistically significantly more comply with face covering measures but this is significant at 5, 10 percent significance level only. However, in difference to the analysis in the previous section, the estimate of the number of mobile phones has not a statistically significant impact on the decision to comply with wearing face coverings.
We also looked at the subpopulations of (non)compliers with face covering measures and their trust in public institutions. The results are summarized in Table 5. By looking at the percentages we see that there does not emerge a clear pattern for trust in these institutions among compliers that would distinguish them from non-compliers.
The results suggest that majority of compliers as well as non-compliers with face covering measures trust the health institutions and the pandemic commission.
It is striking that a larger share of non-compliers trusts the health institutions than it is the case in the subpopulation of compliers. This is surprising since health institutions and health experts were highlighting the significance of face covering wearing as an important instrument to slow the spread of the virus from the onset of the pandemic. Thus, even though these people trust these institutions, they do not comply with their face wearing measures.
To evaluate, whether the independent variables reliably predict the dependent variables of our interest, i.e. compliance with social distancing and compliance with face covering measures, we use F-statistics. Thus, we test the significance and joint significance for the specifications used to estimate the causal effect. We find that the reported p-values are smaller than the 5 percent critical value (see Appendix, Table A1), therefore we reject the null hypothesis and conclude that the independent variables are significant and reliably predict the depend variables. Therefore, we conclude that the estimated results presented above are robust.