Expectations, reference points, and compliance with COVID-19 social distancing measures^☆

3.1. Design

To study how the duration of stay-at-home measures may influence compliance, we collaborated with an established survey firm, SWG, to run three survey rounds on representative samples of the Italian population.9 ^, 10 The first survey round ran on March 18–20, 2020, the second on April 8–10, and the third on April 22–24. Thus, we first surveyed citizens after the lockdown enactment when cases were rising rapidly, then after a first extension announcement when cases were plateauing, and, finally after a second extension announcement during the descending phase of infections.

Each survey round had four main parts. In the first part, respondents answered questions about what actions they were adopting to reduce the likelihood of contagion. Next, we measured respondents’ awareness of the official end date of the stay-at-home measures by asking them to report it. Third, after reminding all respondents of the officially announced end date, to equalize possibly heterogeneous levels of awareness (April 3 in round 1, April 13 in round 2, and May 3 in round 3), we recorded the respondents’ subjective expectations on possible extensions of the measures. In this question, there were four statements and we asked participants to choose the one they believed would be most likely to occur. The options were as follows: (a) “the measures will end on the date indicated by the Government”; (b) “the measures will be extended by a few weeks”; (c) “the measures will be extended by a few months”; and (d) “the measures will be extended indefinitely, until deemed necessary”. Finally, we asked respondents to consider a set of scenarios with different extensions of the stay-at-home measures, and to indicate their intentions to comply with the restrictions under each scenario. There were three cases for them to consider, each with a different duration of the extension: “by a few weeks”, “by a few months”, “indefinitely, until deemed necessary”. In each scenario, participants would select one of five options: “I would significantly increase my isolation”, “I would somewhat increase my isolation”, “I would continue in my current behavior”, “I would somewhat reduce my isolation”, “I might decide to not comply with the rules”. To avoid ordering effects, we presented the scenarios in random sequence.

This design allows us to determine whether the duration of the extension affects intentions to comply with stay-at-home measures. Further, we can assess whether any given extension represented a surprise (positive if the extension was shorter than what the respondent expected, negative otherwise) or matched individuals’ expectations. The survey also included a section with questions about the respondents’ socio-demographic characteristics.

Our survey design does not include exogenous manipulation of individual expectations. In addition to the ethical issues that these manipulations might raise in the context of a pandemic, it is difficult to think about a way to affect expectations that would not also influence compliance in any other way other than through expectations themselves. In the econometric analyses below, we rely on controls for several relevant individual characteristics to address potential explanations for heterogeneities in expectations, and we discuss different interpretations of our findings. The key goal of this study is to document how discrepancies between expectations and formal announcements associate with changes in compliance. Note that this focus on changes in their self-isolation efforts alleviates concerns about unobserved heterogeneity, because individual differences in the assessment of risk (or in other aspects) should affect the level of effort, but not necessarily the change.

Moreover, we measure intentions to comply with social-distancing restrictions, rather than actual behavioral change due to extensions of the measures. As such, social desirability bias could be a concern. For example, people may be inclined to understate the extent to which they are considering reducing their effort. However, respondents’ identities were unknown to the researchers, and the fact that even in the middle of the epidemic more than half of participants admitted not complying with some of the social distancing measures (see below) suggests that, consistent with recent evidence (Jensen, 2020), social desirability may not be a major concern in this context.

3.2. Data and descriptive statistics

The survey reached 894, 891 and 912 individuals in the first, second and third round, respectively. 39.7% of the responses came from participants who reside in one of the five regions where the disease initially spread on a larger scale (the “Red Zone” – see Appendix A).11 About 17% reside in Lombardia, the most populous region that also experienced the largest incidence of COVID-19 cases and deaths. In terms of age profile, 32.8% of participants are older than 60; as such, they are in the sub-population at highest risk of severe effects from COVID-19.12 On average, respondents adopted 4.3 of the six recommended social-isolation behaviors listed in the survey, and 46% of the sample reported adopting all six recommended practices.13

A large majority of respondents (77.2%) reported accurately the date that the Italian Government had announced as the deadline for the stay-at-home measures as of the date of the survey (i.e., April 3, April 13 and May 3 for the first, second and third round, respectively).14 Thus, although a large share of the population was aware of the relevant deadlines, between 25%−30% of the population was either confused or not well informed about the timeline of the measures.

4.1. Respondents’ expectations about possible extensions of stay-at-home measures

As shown in Fig. 3 , the distribution of expectations about the duration of possible extensions of the stay-at-home measures was different across the three rounds. In the first round, only 2.8% of respondents believed that the lockdown would end on the official date (April 3). Roughly 43% expected the Government to extend the provisions by a few weeks, 20.4% by a few months, and 34.1% for an indefinite time. In the second round, again only a few respondents expected the measures to end at the (new) official date (April 13), but many more believed the extension to last only a few weeks compared to round 1 (70.9%), and fewer expected longer extensions. In the third round, the trend toward expecting a closer end of the lockdown continued, with nearly 40% of respondents anticipating that it would actually end on the newly announced date (May 3), almost half believing it would last only a few more weeks, and a small minority expecting extensions by a few months or an undetermined time.15 The shift in expectations followed at least in part the evolution of the epidemic and the actual policies. Considerable heterogeneity in expectations about how long the measures would be in place remained, however, in each round.

4.2. Respondents’ reactions to possible extensions of stay-at-home measures

We now turn to analyzing how individuals respond to government communication. Our objective is to test how positive or negative surprises (i.e., the duration of self-isolation restrictions is shorter or longer than expected, respectively) affect intentions to comply. We first report some descriptive graphical evidence and then turn to regression analyses.

In Panel A of Fig. 4 , we show the distribution of the respondents’ intentions to comply with stay-at-home restrictions under the three scenarios of potential extensions to the date announced by the Government at a given time: “a few weeks”, “a few months”, or “indefinitely”. Because extreme compliance intentions were infrequent, we grouped intentions to comply into three categories: “maintain”, “reduce” (combining “would reduce self-isolation” and “would consider no longer respecting the rules”), and “increase” (“increase somewhat” and “increase greatly” self-isolation efforts).16

Most respondents (between 63% and 70%) in all scenarios indicated that they intended to “maintain” their current self-isolation efforts, with a decline in case of longer extensions. The proportion of respondents who intended to increase self-isolation is similar across the three scenarios (between 17.2% and 17.7%). Conversely, the share of respondents who would reduce it increases with the length of the expected extensions, going from 12.4% in the scenario where the lockdown is extended by a few weeks to 19.5% in the case of indefinite extension. There are also differences across rounds, with a drop in the proportion of respondents who intend to increase their compliance and a rise in those decreasing it in more recent rounds (see Appendix C for details). Intentions to comply are thus sensitive to the hypothesized duration of lockdown extensions. This could be due to the cost of isolation increasing with its duration. As a longer lockdown is plausibly connected to more severe epidemiological environments, also its benefits could increase with duration, but not necessarily as steeply as the cost.

To investigate the extent to which intentions to comply with the lockdown depended on individual expectations about how long the measures would be in place, we defined a measure of match (or mismatch) between expectations and extension scenarios. A match occurs when the extension scenario coincides with the respondents’ expectations. Mismatch cases can be positive or negative: “shorter“ and “much shorter” indicate instances where the scenario hypothesizes that the measures will end sooner than the respondent expects, whereas “longer” and “much longer” denote cases in which the hypothesized extension dates end later than a respondent's expectation. Table 1 summarizes these different instances.17

Panel B of Fig. 4 displays the distribution of self-isolation intentions by the (mis)match between expectations and hypothesized extension scenario. The sampling-weighted averages show that the proportion of respondents who report a willingness to increase compliance with self-isolation measures decreases as the hypothesized extension of the measures gets longer than the respondents expected. Conversely, the more negative the surprise between scenario and expectations, the higher the proportion of individuals expressing the intention to reduce their level of compliance. For extreme mismatches (where the scenario is much longer than expectations), the share of respondents who indicate they would reduce compliance is higher than the share of those who would increase it. We also observe an asymmetry between positive surprises, which are not associated with changes in compliance intentions, and negative ones, for which we find a reduction in compliance intentions.

On the one hand, one might expect less tolerance in later survey rounds due to people's fatigue with social distancing measures; on the other hand, later in the pandemic people might learn more about the severity of the crisis, and this could lead to greater compliance. Thus, the predicted effect of extensions across waves is ambiguous and worth studying. We find qualitatively similar patterns in each round (Fig. C2 in Appendix C), although the main effects of interest are more pronounced in rounds 1 and, especially, 3. In our econometric analysis, we pooled observations from all three rounds, including round fixed effects in the regressions; the estimates are consistent with the lower willingness to comply being a result of the mismatch between expectations and extension announcements rather than simply being due to survey round effects.

4.3. Regression analyses

We investigate the statistical significance of these differences, and their robustness to controls, with multinomial logit regressions. The outcome variable is the choice between increasing, maintaining (the default option) and reducing compliance. The covariates on which we focus are the different extension scenarios, the expectations about extensions that the respondent reported, and the type of mismatch between scenarios and expectations. Additional covariates include indicators for each survey round and the set of socio-demographic characteristics reported in Table A2 in Appendix A.18 Table 2 reports the parameter estimates in relative-risk-ratio transformations. For each specification, the first column displays the estimated coefficients for the option to increase compliance, and the second the estimates for the option to reduce compliance.

The estimated coefficients from specification (1) indicate that respondents, on average, were significantly more likely to express the intention to reduce their compliance when presented with a scenario that implied an extension longer than a few weeks. They were also less likely to report a willingness to increase (and more likely to decrease) compliance in round 3. In specification (2) we included indicators of the respondents’ expectations, setting “the lockdown will end as planned” as the omitted category. Respondents who expected the Government to extend the duration of the lockdown were less likely to reduce their intention to comply compared to those who expected the lockdown to end on the announced date (without any large or statistically significant differences across groups who expected different extensions). In specification (3) we add among the covariates the deviations from the extension scenarios, with “extension matching expectations” as the omitted category. The results confirm the descriptive evidence from Fig. 4: respondents who experienced negative surprises were significantly less likely to report the intention to increase compliance, and (especially) more likely to report the intention to reduce it, compared with the “extension matching expectations” case. Positive surprises imply reactions in the opposite direction, although these effects are weaker and, in most cases, not statistically significant. Wald tests for differences between the estimated coefficients on the positive and negative surprise indicators provide further evidence of the significance of the asymmetric effect of deviations of extension scenarios from expectations.19 Thus, the most significant reaction (statistically and in size) to the discrepancy between extension scenarios and expectations is the intention to reduce compliance when the hypothesized extension is longer than expected, while the reaction to shorter than expected extensions is negligible.

Because expectations about the duration of the lockdown measures vary, we can further investigate whether all types of expectations are equally relevant reference points. Some respondents believed that the isolation restrictions would end at the officially announced dates, and this belief was more frequent in the third round. The official date was plausibly particularly salient, making it especially consequential in order to anticipate behavioral responses. Here, we focus on the positive and negative surprise effects, leaving as the omitted category, again, the “extension matching expectations”. In specification (4), we report estimates from regressions where the covariates include interaction terms between the indicators for the hypothesized extension being longer or shorter than the expectations, and the different expectations. Given the increased detail of the analysis, we grouped the expectation-mismatch indicators for “shorter” and “much shorter”, and for “longer” and “much longer” into only two categories (“shorter” and “longer”, respectively). Given the design of the survey, the length of the restrictive measures in the scenarios could be shorter that the expectations only if the expectations were for the duration to extend for a few months or indefinitely. Therefore, there are only two interaction terms between the indicator for shorter-than-expected extension scenarios and reported expectations. The estimates show that the negative reactions to expectation mismatches are particularly strong for those who expected an end of the lockdown at the announced date (who are more likely to reduce compliance), as well as for those who expected an extension by a few weeks (who are less likely to increase compliance). The non-significant estimates for those expecting the lockdown to be extended by a few months suggest that the difference between an extension by a few months and an indefinite one may not be strong.20

A more fine-grained investigation of individual responses to the three scenarios provides further information on the origin of these results. First, about 10% of respondents reported that they would reduce their compliance in all three extension scenarios. The majority of these participants expected that the restriction would end at the date that the Government announced. Thus, for these respondents all hypothesized extension scenarios were “longer” than what they expected. Second, roughly 15% of individuals indicated the intention to keep or increase compliance for shorter extensions, and the intention to reduce compliance for longer extensions, i.e., in the scenario where individuals were more likely to receive a negative surprise. It is these two subsets of respondents that largely determine the “negative surprise effect” that we detected.

4.4. Alternative explanations and additional analyses

To address the possibility that factors other than expectation mismatches explain our findings, we perform several additional tests. One concern is that positively and negatively surprised individuals, in part by construction, may hold different expectations or have different levels of baseline compliance.21 Those who experience negative surprises, for example, are also those who expect the lockdown to end sooner. In turn, these people might also believe that the pandemic is not very serious, which might explain why they are less likely to maintain or increase their self-isolation efforts. Conversely, those who expect the restrictions to be extended for longer periods are more likely to be positively surprised; they might be more concerned about the pandemic, and thus more likely to state they would maintain or increase their self-isolation efforts. On the one hand, regardless of the underlying determinants of our findings, relevant policy implications remain; in particular, repeated extensions of the end-dates of the isolation measures may have overall negative consequences in terms of compliance with those restrictions. On the other hand, it is useful to understand whether these overall consequences derive from individual heterogeneities or from systematic behavioral responses to the different scenarios, given the expectations.

We perform two additional analyses to address this question. First, we rely on additional information, available only for the first round of the survey: the answers to a question about whether respondents were "afraid of getting sick with COVID-19". As shown in Appendix Table C7, respondents who expected longer extensions were indeed more likely to be "quite afraid" or "very afraid" to contract the virus. The regression results displayed in Appendix Table C8, however, indicate that controlling for this variable (a binary indicator taking the value of 1 if people are either “quite afraid” or “very afraid”) does not change meaningfully the estimates of our coefficients of interest.22 Second, we test whether our results hold also when we account for the respondents’ baseline levels of compliance with social-distancing restrictions, by adding to the covariates a binary indicator of whether the respondents reported complying with all six listed social-isolation behaviors (as reported in Section 3, this splits the sample approximately in half). The results, reported in Appendix Table C9, show that the estimated coefficients on baseline compliance are similar in magnitude (about 0.66) and statistically significant for both "increase compliance" and "reduce compliance". This indicates that being fully compliant with recommended social-distancing measures is associated with an increased likelihood of maintaining current levels of compliance as opposed to either reducing it or increasing it. Again, the magnitude and statistical significance of our main estimated coefficients of interest are unchanged.

In addition to the self-reported measure of concerns for the virus in the first survey round, we can also control for the actual health risks based on the respondents’ place of residence, for all three rounds. Throughout the first wave of the pandemic, some areas of the more populous regions of the north of Italy were more affected by the health crisis. In particular, the regions of Emilia Romagna, Lombardia, Marche, Piemonte, and Veneto – often referred to by the media and government officials as the “Red Zone” – accounted for 75% of the total number of COVID-19 cases recorded across the country by May 2020, despite having 42% of the population. As reported in Table C10, our findings hold also after controlling for this factual risk indicator; respondents in the Red Zone were equally significantly likely to report an intention to reduce their self-isolation efforts if the extensions were longer or much longer than they expected, with the latter having a larger estimated coefficient. Similarly, we do not see a reduction on intended compliance in case of positive surprises (i.e., if extensions are shorter or much shorter than respondents’ expected end date). The main difference between respondents living inside and outside of the Red Zone is that, overall, a higher share of Red Zone respondents reports an intention to increase their compliance efforts regardless of whether the scenarios matched their expectations. Despite this, the willingness to increase their self-isolation efforts decreases as the length of hypothesized extensions increases relative to their expectations. In columns 3 and 4 of Table C10, we split the population by age, one of the most important individual factors affecting risk. Younger citizens show a higher propensity to reduce compliance when facing longer than expected extensions, whereas the elderly are less likely to increase compliance.

These additional findings suggest that the association between deviations of lockdown extension scenarios from expectations is not likely due to underlying differences between positively and negatively surprised individuals in terms of their concern about the coronavirus, their relative risk of contagion, or their level of compliance with the social-distancing restrictions.

Moreover, although we can rule out that "ordering effects" may explain the responses to the hypothetical scenarios because the respondents saw the scenarios in random sequence, a remaining concern is that the responses may be the result of a preference for "consistency" by the participants. Limited to the third round of the survey, SWG kept track of what scenario appeared first to each respondent; we therefore perform our main analyses restricting the sample to the responses to the first displayed scenario in the third round.23 In Figs. C3 and C4 and in Table C11 we show that our main findings holds also with this subsample of observations.24

Another possible explanation for the asymmetry between positive and negative surprises is that if individuals are already exerting maximal self-isolation efforts, increasing effort might be difficult to do. To evaluate the plausibility of such a “ceiling effect”, in Fig. 5 , we replicate the analyses from Fig. 4 separately for two groups of respondents, those who reported being compliant with some but not all recommended preventative behaviors (“partially compliant”) and those who reported complying with all six recommended behaviors (“fully compliant”). We observe very similar patterns for the two groups of respondents. In particular, although the “fully compliant” individuals display a higher propensity to maintain their current level of compliance, they have ample room to both reduce or increase their self-isolation efforts. Thus, ceiling effects are unlikely to explain the results.

Thus, overall, our results are consistent with expectations functioning as reference points for individuals, and with discrepancies between actual duration of a policy and expectations asymmetrically affecting intentions. The negative impact of “bad surprises”, in particular, dominates the positive effects of “positive surprises”.