Expectation management of policy leaders: Evidence from COVID-19.

This paper studies how the communication of political leaders affects the expectation formation of the public. Specifically, we examine the expectation management of the German government regarding COVID-19-related regulatory measures during the early phase of the pandemic. We elicit beliefs about the duration of these restrictions via a high-frequency survey of individuals, accompanied by an additional survey of firms. To quantify the success of policy communication, we use a regression discontinuity design and study how beliefs about the duration of the regulatory measures changed in response to three nationally televised press conferences by former Chancellor Angela Merkel and the Prime Ministers of the German federal states. We find that the announcements of Angela Merkel and her colleagues significantly prolonged the expected duration of restrictions, with effects being strongest for individuals with higher ex-ante optimism.

Introduction

The observation that expectation management is an important component of economic policy has been acknowledged most prominently in the field of monetary policy (e.g. Rivot, 2017) but applies to many markets. It is clear that improving forecasts of demand and supply is important for reducing the uncertainty of economic agents. Correspondingly, analyses of market expectations appear increasingly often and in many guises. A common finding is that the management of expectations is especially important after exogenous shocks or structural breaks in the economy, because agents’ uncertainty about their own forecasts of market fundamentals is at its greatest and the heterogeneity of expectations across agents is correspondingly large.1

This paper studies how policy communication affects expectations of individuals. Specifically, we quantify the effects of statements about COVID-19-related regulatory measures made by former Chancellor Angela Merkel and the prime ministers of the German federal states during the early phase of the pandemic. Our analysis contributes to an emerging literature that studies whether heterogeneous population expectations can be influenced by policymakers.2

The outbreak of the COVID-19 pandemic in early 2020 was a major structural break, taking the world by surprise and creating a world-wide guessing game about the outbreak’s scale and the pandemic’s endurance. We focus on the latter uncertainty, about the pandemic’s duration and the related restrictions, which is of first order for expectation management and hence of major economic importance. Expectations about the pandemic’s duration were highly unrealistic in its early phases: even after the first larger-scale infections in metropolitan centers of different countries, few economic decision-makers anticipated a long-lasting impact. For instance, Bartik et al. (2020) surveyed American businesses between late March and early April 2020, finding that respondents predominantly predicted the crisis to have ended by June 2020.3 What did become clear very quickly is that the crisis, while it lasts, would result in massive changes to economic activity and to private lives. A combination of voluntary measures and imposed restrictions rapidly reduced supply and demand in many markets and affected various other aspects of everyday life. As a consequence, many choices (as well as macroeconomic expectations) of individuals and firms depend on the expected duration of the pandemic. For example, Buchheim et al. (2022) find that German firms that expected the shutdown to last longer were more likely to lay off workers or to cancel or postpone investment projects in 2020. Generally, the scope of economic responses to the duration of the pandemic was seen as large. “How much longer?” was a question that soon reverberated throughout the world’s public media.

In this situation, many political leaders attempted not only to promise a solid policy response with wide-ranging public health measures and economic rescue packages, but also to educate the public about the length of the pandemic. Announcements of policy makers and scientists about the length of the crisis received wide coverage in news and social media – but so did many other public discussions about the pandemic. The empirical question arises to what extent the announcements of policy makers were effective. In a cacophony of media publications and expert views, are policy makers able to move the beliefs of the public? An affirmative answer would be an important prerequisite of successful crisis management. The large majority of the policy response is made up of the regulation of human behaviors – including social distancing, private consumption, and investment decisions – as such, the ability of political leadership to reach the population’s minds is crucial to its effectiveness.

There are potential reasons for and against suspecting a strong policy communication in this context. On the positive side, one may argue that the regulation of social distancing, despite its novelty, is relatively straightforward to describe at least in its basic and undifferentiated form: public events are forbidden or not, schools are closed or not. Consistent with this hypothesis, Goldberg et al. (2020) report evidence that mask wearing increased strongly after a corresponding recommendation by the U.S. Center for Disease Control. Moreover, the main regulatory decisions lie in the hands of a few well-identified policy makers and they are the same individuals who also make the public announcements. Thus, it may be natural to suspect that these announcements are being widely listened to and understood. A further factor may be that the pandemic’s salience and importance makes it plausible that even small modulations in the tone of communication can have significant effects.4 On the negative side, the questions at hand lay in an unchartered territory and politicians are not usually viewed as public health professionals. Even for an experienced politician, the situation was new and highly uncertain. It was therefore not natural for the observer to believe that the policy makers knew what they were doing. Perceived competency is known to be one of the main predictors of a politician’s election success – in a novel context, it is not clear that the public views incumbent politicians as particularly competent.5

In terms of their quantitative importance, these and related arguments may apply more or less strongly to different subgroups of the public. For instance, Lenz and Lawson (2011) document that the less politically educated react more strongly to politicians’ physical appearance in their voting behavior, suggesting that differently educated subgroups of an audience may also react differently to policy announcements about novel and crisis-related policies. More generally, heterogeneity in perspectives of the listeners may have important consequences for the success of policy announcements. A possible reason for heterogeneous reactions is that the politicians’ communications to the public were extraordinarily intrusive for some groups: the politicians told the population, live on TV, whether or not their basic rights are restricted for the foreseeable future. Depending on how strongly the listeners’ lives were affected and depending on whether listeners are offended by the restrictions – which may correlate with observable variables like gender or political affiliation – they may show negative reactions including disbelief (Terkourafi, 2008, Bénabou and Tirole, 2016).

To quantify the effect of expectation management on belief updating, we study the variation in expectations about the duration of restrictions before and after press conferences of German policy makers in the first months of the Corona pandemic’s outbreak. On three occasions during the spring of 2020, leading German politicians, among them former Chancellor Angela Merkel, appeared in widely broadcasted press conferences and made announcements about the state of the pandemic as well as the German regulatory responses. We conduct a large online survey to elicit the beliefs of individuals about the duration of three well-defined restrictions: (1) when will the majority of school children be back in school; (2) when will the premier football league (Bundesliga) return to normal operations with stadium visitors; and (3) when will all current restrictions related to the Corona crisis be fully lifted? The online survey has a fine time structure: the internet panel that we use collects responses on a daily basis within a pre-defined time period. Thus, in the empirical analysis, we can account for time trends in belief formation using a regression discontinuity design.

The empirical analysis in the paper consists of three parts. In the first part, we provide graphical evidence about the evolution of individuals’ expectations over time and, specifically, around the three press conferences. This descriptive exercise provides first suggestive evidence that the press conferences changed individuals’ expectations. In addition, the graphical analysis reveals a clear time trend in expectations. Interestingly, we find a similar time pattern for the expectations of managers of German firms for whom we elicit expectations in a different survey at two points in time. In the second part of the empirical analysis, we control for the overall time trend and isolate the effect of policy communication on expectations. We exploit the variation in expectations shortly before and after each of the three press conferences, using a local polynomial regression discontinuity design. In the main specification, we use a time window of one week before and after each press conference and restrict the time trend to be linear. In additional specification checks, we further show that results are robust to changes in the distance to the cut-off dates, the specification of the time trend and the definition of the outcome variable. Moreover, we conduct placebo tests and estimate the model for all available dates in the survey period. Our results show that policy communication significantly prolonged the expected duration of the restrictions. In particular, we find that the first press conference of Angela Merkel and her colleagues had a sizable impact. In this press conference, Merkel conveyed a strong sense of caution. The expected time until all restrictions would be fully lifted moved by about one month on this day, from mid-October 2020 to mid-November 2020. Likewise, the expected date of school openings moved by about two weeks. In contrast, we do not find significant effects of the second press conference on any measure of restrictions. This is not surprising, as the policy communication in the second press conference was rather vague about the duration of the restrictions. The third press conference also prolonged the expected duration of the general restrictions, but we must caveat that this point estimate is quite sensitive to specification choices and, hence, less robust than the finding for the first press conference. Next, we investigate belief uncertainty and analyze the second moments of individuals’ beliefs before and after the press conferences. We find that while the policy announcements did not significantly affect individuals’ mean beliefs about the duration of sport restrictions, there is some evidence that they reduced the dispersion in these beliefs. Studying the heterogeneity in responses to the policy communication, we find a surprising pattern of consistency – the effects do not vary much with observed respondent characteristics. We do find some differences in response behavior by gender, but no consistent differences with respect to education, age, region, regional exposure to COVID-19 or political preferences. However, in additional quantile regressions, we show that policy communication is most effective for individuals with more optimistic expectations (i.e., individuals who expect a shorter duration of restrictions) as their part of the response distribution is shifted more strongly. In the third and final part of the empirical analysis, we explore some behavioral effects of the policy announcements. Using data on planned consumption expenditure and aggregate mobility indicators, our analyses suggest that behavioral effects of the three press conferences were rather limited, which is consistent with results from previous literature. For example, Coibion et al. (2020b) do not find any effect of the expected duration of the COVID-19 pandemic on individuals’ marginal propensity to consume out of stimulus checks in the United States.

The remainder of this paper is organized as follows. In Section 2, we briefly describe the evolution of the pandemic in Germany in 2020 and explain the content of the three press conferences where policy measures were communicated to the general public. Section 3 presents the data and provides graphical evidence on the evolution of expectations over time. In Section 4, we discuss the econometric approach. Section 5 contains the results on expectations, Section 6 reports additional results on behavioral effects. Section 7 concludes.

Background

In this section, we describe the development of COVID-19 in Germany during the first months of the pandemic and describe the key policy measures implemented in March 2020. We focus on the effectiveness of policy communication at the beginning of the pandemic, studying three main press conferences by former Chancellor Angela Merkel between April and May 2020. In these press conferences, Angela Merkel announced to what extent existing restrictions would be continued or modified. We describe the content of these press conferences below. For a better understanding of how the effectiveness of policy communication may depend on the political context and the popularity of the political leaders, we also provide some background regarding voter support for the German government before and during the pandemic.

Timeline of COVID-19 and policy responses in Germany

Fig. 1 describes the dynamic development of COVID-19 infections, which started to strongly increase in mid March. At this time, the German government introduced a variety of restrictions that effectively shut down large parts of both economic and private lives. These restrictions included the prohibition of large events, travel restrictions, as well as the closure of stores, schools, and recreational facilities. Citizens were told to stay at home, they could meet only one person from another household, and a minimum distance of 1.5 meters had to be kept whenever contact could not be avoided. These strict contact restrictions were renewed and extended at the end of March, without a fixed expiry date.

Fig. 1

COVID-19 cases and policy measures in Germany in 2020. Notes: Plot shows the evolution of total COVID-19 infections and regulatory measures implemented by the German government in the first months of the pandemic. Data source: RKI COVID19.

The period between March and May, 2020, was characterized by large uncertainty about the spread of infections, the duration of the pandemic, and the appropriate policy response. Individuals received daily information from the media, numerous policy makers, and medical experts. Similar to the evolving scientific knowledge about the COVID-19 pandemic, this information was noisy and often inconsistent.

Three main policy communication events stand out: the press conferences by former Chancellor Angela Merkel between April and May.6 In these press conferences, Angela Merkel announced changes to the restrictions that the federal government and state governments had agreed on. These press conferences had extremely broad media coverage, with the vast majority of Germans following the events live or accessing summaries of the press conferences. For example, on April 15, the day of the first press conference studied in this paper, more than 23 million Germans watched a summary of Angela Merkel’s speech in an evening news show, corresponding to about 30 percent of the German population. This initial media coverage was then multiplied by online and printed press and through social media outlets. Hence, it is credible to assume that most Germans were aware of the content of the press conferences (see Appendix B.1 for more details on media coverage).

We summarize the content of the press conferences below. In the empirical analysis, we then evaluate how these public announcements affected individuals’ expectations about the duration of the restrictions.

Press conferences of Angela Merkel

First press conference (April 15, 2020)7 In the first press conference, former Chancellor Angela Merkel announces that contact restrictions are extended until May 3, specifying that residents can meet at most one person from another household at a time and that minimum distance regulations remain unchanged. Merkel also announces that small steps are being taken to increase the freedom of movement for citizens. Shops up to 800 square meters are allowed to open if they comply with certain hygiene measures. Schools are allowed to open gradually, events with large attendance remain prohibited until August 31. The policy makers also ask the population to refrain from private travels and visits. Overall, rules remain strict and the extent of relaxation measures falls behind expert recommendations issued prior to the press conference.8

Second press conference (April 30, 2020)9 In the second press conference, Angela Merkel announces only minor changes to existing rules. Contact restrictions remain in place. Church services are permitted again, while playgrounds and cultural facilities may reopen if hygienic conditions are met. Economic aid, mostly from federal sources, will be provided to alleviate the negative effects of the crisis. A further evaluation of current policies, and whether further opening measures can take place, is announced for May 6.

Third press conference (May 6, 2020)10 Contact restrictions are modified in that members of two households are allowed to meet. Conceding to pressure from individual state governments, Angela Merkel announces that schools and shops of all sizes are allowed to open under strict conditions. Recreational sport is permitted outside. Further, a de-centralized ‘emergency mechanism’ is imposed according to the regional development of COVID-19 infections: if the cumulative number of new infections per 100,000 inhabitants exceeds a threshold of 50 over a seven-day period in a region, new restrictions will be imposed in that region.

Macroeconomic and political context

The effect of public communication may depend significantly on the context, as well as on the popularity of the political leaders. While Germany was often referred to as “the sick man of Europe” because of low growth and high unemployment before 2005 (Dustmann et al., 2014), this changed in the mid-2000s after a series of labor market and tax reforms. These reforms came into effect when Angela Merkel became German Chancellor in 2005, although they had been initiated by the previous government. These reforms are seen as one reason why neither the Great Recession nor the euro crisis affected the German labor market severely. In contrast to the United States and most other EU countries, Germany experienced almost no increase in unemployment in 2008 and 2009, despite a sharp decline in GDP. Since 2010, the German economy had been growing for 10 consecutive years – the longest period in modern German history. Moreover, labor force participation rates of both women and men increased steadily after 2004 and the unemployment rate fell to 5 percent in 2019.

This stable economic development arguably led to relatively high popularity of former Chancellor Merkel, with approval rates of around 70 percent at the beginning of 2020. Furthermore, Archer and Ron-Levey (2020) report that before the COVID-19 pandemic, 83 percent of the German population said they had a lot or some trust in the government’s medical and health advice, and only 13 percent said they had not much trust or none.11 By and large, the coalition government of Angela Merkel’s center-right CDU/CSU and the center-left social democratic SPD can be described as having worked smoothly in the years preceding the pandemic. The next federal election was scheduled for the fall of 2021, hence, 2020 was not an election year and candidates of all parties were not to be elected before spring 2021. After the outbreak of the COVID-19 pandemic, Angela Merkel’s approval rates - as well as voter support of the federal government - increased to new all time highs (see Fig. A.6 in the Appendix). Angela Merkel had already announced in October 2018 that she would not seek reelection.

Fig. A.6

Popularity of Angela Merkel and the German government over time. Notes: Plot shows the approval ratings for Angela Merkel (solid red line) and the German government (dashed blue line) relative to approval rates in January 2020. Solid vertical lines indicate the three press conferences. Data source: Forschungsgruppe Wahlen, Politbarometer 2019–2020.

Data and graphical evidence

This section describes the data that we collected to study expectations about the duration of the pandemic and shows summary statistics for our sample. We also provide graphical evidence on the evolution of expectations over time around the three press conferences of Angela Merkel. The next section then outlines our empirical approach and quantifies the effect of policy communication on beliefs.

High frequent elicitation of expectations

We elicit expectations about the duration of COVID-19-related restrictions based on daily online surveys conducted by Civey. Civey is a market research and polling institute that provides Germany’s largest open access online panel with over one million active users. Civey collaborates with different online news portals and forums to place short survey modules that can be answered in a multiple-choice set-up (see Fig. A.1 in the Appendix for an example). After participating in a survey, respondents can immediately observe the overall evaluation of all other respondents, which creates an incentive to submit a response (see Fig. A.2 in the Appendix for an example). To obtain results for a balanced sample of the population, surveys are embedded in over 25,000 webpages targeting different audiences. When displaying live results, Civey applies an automated weighting procedure based on self-reported gender, year of birth, postal-code and political party preference. To limit the extent of self-selection into a particular survey, Civey invites survey participants to answer multiple surveys in a row, which are displayed in randomized order – Civey then disregards the answer to the first survey that individuals initially clicked on.

Fig. A.1

Example of an embedded Civey question on online news page.

Fig. A.2

Example of a live display of Civey responses.

We contracted with Civey in March 2020 to survey individuals’ expectations about COVID-19-related restrictions on a daily basis, for a period of two months. Between April 2 and May 27, we obtained a total of 123,840 observations.12 The number of observations varies considerably between the different days and between the different questions, which is partly explained by the display algorithm that makes particular surveys more or less salient on a given day (see Table A.1 in the Appendix for sample statistics).

Table A.1

Observations and sample statistics.

Number observations	Overall	All restrictions	School	Football
Total/ survey period	123,840	82,051	22,693	19,096
Mean/ day	2,211	1,465	405	341
Median/ day	1,043	488	227	187
Min/ day	334	114	84	64
Max/ day	16,291	15,058	2,294	1,846

Note: Civey Online Panel, April 2-May 27, 2020.

We collected answers to the following expectation questions:

When will the current restrictions related to the Corona crisis be fully lifted? (Q1)

When will the majority of school children be back in school? (Q2)

When will the national football league return to normal operations with stadium visitors? (Q3)

Individuals provide answers by stating the number of months they expect it will take until the restrictions are lifted (choosing out of several categorical response options, see Appendix A.2 for details). To take out the mechanical effect of survey time on the choice of categorical response options, we also translate this information into the expected calendar date (see Appendix A.4). In our main analyses, we exclude individuals who responded that the restrictions would never end or that they did not know when they would end, but we use responses to these extreme answer categories to explore uncertainty in beliefs around the press conferences. We further restrict our analyses to individuals with complete information on socio-demographic and geographic covariates.

Owing to its open access nature, the panel is not a representative sample but a convenience sample. In Table 1 we show weighted summary statistics for our estimation sample. Civey provides survey weights separately for each question. In Columns I-III we present the summary statistics based on the weights calculated for three main outcome variables mentioned above. In Column IV we show official statistics for comparison.13 The weighted summary statistics for gender, age, region and political preferences are comparable to the German population. However, the distribution of educational outcomes is very different in the Civey sample. Specifically, more individuals have a university degree than in the official data and only very few individuals have no degree. Thus, a clean analysis of heterogeneous effects by education is not possible. In the main specification of our regression analyses we control for individual characteristics. In addition, we conduct sensitivity analyses and report estimates that use the sample weights provided by Civey in the Appendix.

Table 1

Sample characteristics of Civey respondents.

	Civey Online Panel			Official
	(Q1)	(Q2)	(Q3)
Female	48.1	49.8	47.2	50.7
Age categories
18–39 yrs.	7.3	6.6	6.4	16.3
30–39 yrs.	11.3	12.2	12.0	15.5
40–49 yrs.	16.3	17.7	17.0	14.7
50–64 yrs.	30.3	31.5	32.0	27.5
65+ yrs.	34.9	32.0	32.7	26.0
Region
North/West	33.8	32.5	33.7	37.7
South	41.7	42.6	42.3	42.8
East	24.4	24.9	24.0	19.5
Political party preference
Union/FDP	43.6	44.1	45.7	45.0
Red/Red/Green (RRG)	43.3	39.9	40.3	39.0
AfD	9.4	11.6	10.5	9.0
Other	3.8	4.4	3.6	7.0
Education
University degree	50.0	51.4	51.7	19.6
Vocational degree	47.5	46.0	45.5	59.2
No degree	2.4	2.6	2.8	20.8

Note: Cells contain shares in percent. Official statistics on gender, education, age and region from Federal Statistical Office (Destatis), based on 2019 microcensus and 2019 forward projection of 2011 census. Official statistics on political party preference based on Forsa Sonntagsfrage of May 30th, 2020. Civey samples differ by question: Q1 = All restrictions, Q2 = School closures, Q3 = Bundesliga. Civey means adjusted for population weights.

Expectations of individuals over time

In this section, we show how respondents’ expectations about the duration of the different restrictions in Germany evolved between April and May 2020. Fig. 2 presents the evolution of categorical answers over time. The vertical lines mark the dates of the three press conferences. The figure provides first suggestive evidence that the press conferences affected the expectations of individuals. Specifically, the share of individuals expecting that restrictions will be lifted only in more than nine months increases after the first press conference on April 15th. On the same day, the share of individuals expecting restrictions to be lifted in the next 2–3 months decreases. The picture looks similar when focusing on the restrictions related to schooling. After the first press conference, the share of individuals expecting a re-opening in the next 4 weeks is reduced while the share expecting a longer restriction (2–3 months or 4–5 months) increases. The pattern at the later press conferences and for the restrictions of football events is less pronounced.

Fig. 2

Expected duration of restrictions over time. Notes: Plots show how expectations evolved over time. Solid vertical lines indicate the three press conferences (PC). Data source: Civey Online Panel 2020.

As described above, we translate individuals’ categorical responses into continuous variables which measure the expected duration until restrictions are lifted in days. This allows us to analyze how the mean and the median expected duration evolved over time and in relation to the press conferences. Despite some noise in the daily expectations data with positive and negative outliers, both the mean and the median beliefs show clear shifts around the press conferences in expectations about all restrictions and restrictions related to school closures (Fig. A.3, Fig. A.4 in the Appendix). Consistent with the pattern of the categorical answers, the median of the expected duration for all restrictions and restrictions related to schooling increases after the first press conference, corroborating the suggestive evidence that policy communication can affect expectation formation. In addition, by taking out the mechanical effect of survey time on the choice of categorical response categories, graphical evidence based on the expected calendar date reveals that individuals’ expectations show a sizable time trend over the survey period. For example, individuals surveyed at the beginning of April 2020, on average, expected all restrictions to be fully lifted by November 2020. In contrast, individuals surveyed at the end of May expected an end of all restrictions only in the beginning of 2021. A similar time pattern can be observed for specific restrictions for schools and major sports events (football); however, the expected end date of these restrictions is earlier than for overall restrictions.

Fig. A.3

Mean expectations over time. Notes: Means with 95-% C.I., adjusted for population weights. Vertical lines indicate three major press conferences. Data source: Civey Online Panel 2020.

Fig. A.4

Median expectations over time. Notes: Medians adjusted for population weights. Vertical lines indicate three major press conferences. Civey Online Panel 2020.

The observed time trend in expectations can be explained by various factors. For example, it might be related to the arrival of new scientific information about the pandemic, new media information, or the experience of other countries with longer exposure to Corona. In Appendix E we present additional information about expectations of managers collected in the ifo Manager Survey in two waves in April and May 2020, respectively. We compare the expectations of managers and individuals over the same time periods and find a surprising similarity in the time trends (Fig. A.10). In the following econometric analysis, we control for the overall time trend to isolate the effect of policy communication at the time of the press conferences. In addition, we study heterogeneity by testing whether and how the effect of policy communication differs by observable characteristics and varies along the distribution of expectations.

Fig. A.10

Comparison of household and manager expectations. Notes: Coefficients from multivariate OLS with 95%-C.I.; Data source: Civey Online Panel and ifo Manager Panel 2020.

Model and identification

Empirical model

To identify and quantify the effect of policy communication on expectations, we use the variation in expectations before and after the day of a press conference. We restrict the time window and focus only on changes in expectations one week before and one week after a press conference. In addition, we control for the time trend using a regression discontinuity design14 :where is a measure of the expected end date of the restriction, is the coefficient of interest which captures the effect of the press conference, while and account for the time trend before and after the press conference. The date of the survey, measured in days, is described by and c is the cut-off date. In the main specification, we use a context-based definition of the distance to the press conferences. Specifically, we use a 7-day distance wherever possible but restrict the length if the 7 day default generates overlap with other events.15 We use a local linear polynomial in our preferred specification as recommended by Gelman and Imbens, 2019, Calonico et al., 2014, but also show results based on a standard regression discontinuity design with a global linear time trend. In addition, we control for the effect of further explanatory variables summarized by .16 In the empirical analysis we extend the main specification and analyze the sensitivity of our results to changes in the distance to the cut-off date, the specification of the time trend and the definition of the outcome variable.

Finally, in Table A.3 in C.1 we provide evidence that manipulation around the cut-off dates does not pose a threat to identification in our setting. Importantly, the characteristics of the respondents are very similar in the days before and after the three press conferences. Differences in the observed variables (gender, education, age, children, political party preference, postal-code) before and after each press conference are either not statistically significant or, if the difference is significant, very small in magnitude.

Table A.3

Sample composition before and after the press conferences.

	1st Press Conference			2nd Press Conference			3rd Press Conference
	pre	post	p (Δ)	pre	post	p (Δ)	pre	post	p (Δ)
Female	0.31	0.28	0.00	0.31	0.27	0.00	0.27	0.26	0.18
University	0.51	0.51	0.79	0.51	0.52	0.09	0.52	0.54	0.01
Age above 50	0.79	0.84	0.00	0.82	0.86	0.00	0.85	0.85	0.06
Children in HH	0.20	0.16	0.00	0.17	0.14	0.00	0.14	0.15	0.03
Political pref.: Union/FDP	0.39	0.39	0.59	0.36	0.34	0.03	0.34	0.38	0.00
Political pref.: RRG	0.34	0.31	0.00	0.29	0.30	0.09	0.30	0.31	0.02
Political pref.: AfD	0.19	0.23	0.00	0.26	0.28	0.02	0.28	0.23	0.00
Political pref.: Other	0.08	0.07	0.04	0.09	0.08	0.01	0.08	0.08	0.70
Pop Density: High	0.41	0.40	0.54	0.41	0.41	0.43	0.41	0.42	0.32
Purch. Power: High	0.47	0.49	0.01	0.48	0.48	0.75	0.48	0.50	0.02
Region: North/West	0.33	0.32	0.33	0.33	0.32	0.02	0.32	0.34	0.00
Region: South	0.41	0.42	0.17	0.41	0.42	0.31	0.41	0.42	0.21
Region: East	0.26	0.26	0.62	0.26	0.27	0.19	0.27	0.24	0.00

Note: Civey Online Panel, April 2 - May 27, 2020. Cells contain sample means before (pre) and after (post) the press conferences (PC) and p-values (p) from two sample mean comparison tests on the pre/post mean difference ().

Effects of policy communication on expectations

Graphical evidence

Before we turn to the results of the econometric analysis we present further graphical evidence about the changes in expectations around the three press conferences. Figs. 3 a–i show binned sample means with fitted local linear trends before and after the press conferences. For the expectations about the duration of all restrictions, the graphical evidence points at discontinuities at the first and the third press conference. The same is true for expectations about school closures before and after the first press conference. The evidence is less clear for the second press conference (as expected, given that only minor changes to existing rules were announced) and for the expectations about the restrictions regarding football events.

Fig. 3

Conditional means with local linear fit. Notes: Each observation represents the daily average expected date until restrictions are lifted. The vertical lines denote the press-conference cut-offs. The solid trend lines are based on local linear regressions.

Mean effects

In Table 2 we present our estimates of the impact of policy communication on the expected duration of restrictions. In addition to the main specification - a local linear polynomial regression discontinuity design with controls for observable characteristics - we show results from bivariate OLS without any further control variables, as well as from multivariate OLS with an interacted (global) linear time trend. As documented in the graphical analysis, the time trend has a sizable effect on expectations. Therefore, we focus on the main specification controlling for local linear time trends when discussing the effects of policy communication.

Table 2

Estimation results: Expectation updating in response to new COVID-19 announcements.

	1st Press Conference			2nd Press Conference			3rd Press Conference
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
All restrictions	30∗∗∗	25∗∗∗	25∗∗∗	−14∗∗∗	10	8	38∗∗∗	50∗∗∗	30∗∗∗
MDV (pre-event)	14 Oct	14 Oct	14 Oct	28 Nov	28 Nov	28 Nov	16 Nov	16 Nov	16 Nov
S.D. (pre-event)	124	124	124	131	131	131	132	132	132
N	15,560	15,560	15,560	8,675	8,675	8,675	12,708	12,708	12,708
School closures	26∗∗∗	13∗∗	13∗∗	2	0	−6	15∗∗∗	11	4
MDV (pre-event)	05 Jun	05 Jun	05 Jun	29 Jul	29 Jul	29 Jul	01 Aug	01 Aug	01 Aug
S.D. (pre-event)	54	54	54	80	80	80	80	80	80
N	4,215	4,215	4,215	2,913	2,913	2,913	3,450	3,450	3,450
Bundesliga	8	−20	−44∗∗∗	−27∗∗∗	−28	−21	15∗	−28	−41∗
MDV (pre-event)	13 Dec	13 Dec	13 Dec	19 Jan	19 Jan	19 Jan	27 Dec	27 Dec	27 Dec
S.D. (pre-event)	171	171	171	175	175	175	175	175	175
N	3,560	3,560	3,560	2,448	2,448	2,448	3,017	3,017	3,017
Covariates	-	✓	✓	-	✓	✓	-	✓	✓
Time trend	-	global linear	local linear	-	global linear	local linear	-	global linear	local linear

Note: Civey Online Panel, April 2-May 27, 2020. Table shows the change in the expected duration in days associated with each of the three public announcements. Columns 1/4/7 show estimates from bivariate OLS on a binary indicator that takes on 0 if the outcome was measured before the event and 1 if it was measured after the event. Columns 2/5/8 show multivariate OLS estimates with a global linear time trend centered at zero at the event interacted with the before/after indicator. Columns 3/6/9 present multivariate local linear polynomial regression discontinuity estimates. Covariates include gender (male/female), education (university/other), age (below/above 50), children in household (yes/no), region (northwest/south/east), population density (high/low), purchasing power (high/low), political party preference (Union/FDP, Red/Red/Green, AfD, Other) and county-level quartile of COVID-19 new infection rate. MDV = mean dependent variable measured before the event. S.D. = baseline standard deviation in days. Estimation with standard errors clustered at the person-level. ∗, ∗∗, ∗∗∗.

Overall, the results provide evidence that policy communication can have a significant effect on expectations. The first press conference in which Merkel conveyed a strong sense of caution significantly shifted citizens’ expectations about the duration of the pandemic. The expected time until all restrictions are fully lifted moved by almost one month (25 days) after the first press conference. In other words, the press conference shifted beliefs about the end of restrictions from mid-October 2020 to mid-November 2020. This shift amounts to one fifth of the baseline standard deviation and is equivalent to the baseline linear time trend increase over a period of 4 days. We find a similar but smaller effect on the expected duration of school restrictions, which increase by about two weeks (13 days, one quarter of the baseline standard deviation, equivalent to a 4 day baseline linear trend increase). The effect of the announcement on beliefs regarding sports events is negative, but given the sensitivity of this finding to functional form assumptions, we interpret this result with caution.

For the second press conference, we do not find significant effects on any measure of restrictions after controlling for the time trend. This is not surprising, as Angela Merkel announced only minor changes to existing rules during the second press conference and policy communication was rather vague. For the third press conference, we document that expectations about general restrictions shifted by about a month (30 days, one fifth of the baseline standard deviation), postponing the expected end of restrictions from mid-November to mid-December 2020. However, we must caveat that this result is somewhat sensitive to specification choices (see Section 5.3). Likewise, the small positive effect of the third press conference on the expected duration of school restrictions (4 days) is imprecisely estimated and, given the instability of the point estimate regarding sport events, we are also reluctant to put much emphasis on this result. In sum, while we document significant and sizeable effects on citizens’ beliefs of the first press conference and none for the second, the evidence regarding the third one is less clear.

Our results of strong effects for the first press conference while finding none for the second and unclear effects for the third one are consistent with a story of momentum. While at the beginning of the pandemic, there was still a lot of momentum and attention this has diminished over time. The evidence from the declining Tagesschau TV ratings in Appendix B.1 supports this view. A take-away for political communicators who are active during crises may, therefore, be that they should use the momentum at the beginning of a crisis as their effectiveness in communicating might diminish quickly over time.

Specification checks and placebo tests

In the following we describe the specification checks mentioned above and show results from placebo tests that support the central findings of our analysis. Appendix C contains additional robustness checks, such as reweighted estimates and sensitivity to inattentive respondents.

Distance to cut-off We alternatively specify our main model for a fixed 5-day, 6-day, and 7-day pre/post distance around the respective events to analyze the sensitivity of our results to the chosen distances around the cut-off dates (Table A.4 in Appendix C.2). The results for the first and second press conference are robust to the variation in the time window. For the third press conference, we must caveat that the sensitivity analysis based on a narrow 5-day cut-off distance renders insignificant the effect on beliefs about general restrictions and, hence, this result is less stable and should be interpreted with some caution.

Table A.4

Estimation results: Sensitivity to the chosen distance to cut-off.

	1st Press Conference				2nd Press Conference				3rd Press Conference
	c	5d	6d	7d	c	5d	6d	7d	c	5d	6d	7d
All restrictions	25∗∗∗	26∗∗∗	24∗∗∗	22∗∗∗	8	11	6	4	30∗∗∗	15	30∗∗∗	39∗∗∗
	(4)	(4)	(4)	(3)	(6)	(7)	(6)	(6)	(8)	(10)	(8)	(7)
MDV (pre-event)	14 Oct	14 Oct	14 Oct	13 Oct	28 Nov	23 Nov	25 Nov	28 Nov	16 Nov	10 Nov	16 Nov	16 Nov
N	15,560	15,468	15,979	16,501	8,675	7,357	8,351	9,354	12,708	5,735	9,868	13,534
School closures	13∗∗	12∗∗	12∗∗	11∗∗	−6	−7	−10	−9	4	1	4	9
	(4)	(4)	(4)	(4)	(8)	(9)	(8)	(8)	(7)	(8)	(7)	(7)
MDV (pre-event)	05 Jun	06 Jun	05 Jun	06 Jun	29 Jul	01 Aug	01 Aug	29 Jul	01 Aug	30 Jul	01 Aug	02 Aug
N	4,215	4,100	4,401	4,700	2,913	2,652	3,013	3,327	3,450	2,775	3,289	3,589
Bundesliga	−44∗∗∗	−49∗∗∗	−47∗∗∗	−43∗∗∗	−21	−28	−27	−23	−41∗	−43∗	−41∗	−36∗
	(12)	(13)	(12)	(11)	(18)	(23)	(20)	(18)	(16)	(18)	(16)	(15)
MDV (pre-event)	13 Dec	13 Dec	13 Dec	09 Dec	19 Jan	14 Jan	16 Jan	19 Jan	27 Dec	27 Dec	27 Dec	28 Dec
N	3,560	3,481	3,736	3,977	2,448	2,233	2,625	2,894	3,017	2,457	2,884	3,131

Note: Civey Online Panel, April 2-May 27, 2020. Table shows the change in the expected duration in days associated with each of the three public announcements, comparing different distances to the respective event: main specification (c = context-based, 5–7 days avoiding overlap with other events), as well as 5 days, 6 days or 7 days respectively. Coefficient estimates from local polynomial regression discontinuity estimation with local linear time trends. MDV = mean dependent variable measured before the event. Estimates adjusted for gender (male/female), education (university/other), age (below/above 50), children in household (yes/no), region (northwest/south/east), population density (high/low), purchasing power (high/low), political party preference (Union/FDP, Red/Red/Green, AfD, Other) and county-level quartile of COVID-19 new infection rate. Estimation with standard errors clustered at the person-level in parentheses. ∗, ∗∗, ∗∗∗.

Functional form of the time trend Next, we compare our main specification based on a local linear time trend assumption to models with a global linear trend, a local quadratic trend and a global quadratic trend (Table A.5 in Appendix C.2). Our main finding is robust to different specifications of the time polynomial: in all specifications, we find that the first press conference significantly shifted the expectations about all restrictions by about one month. Results for the first press conference are also mostly stable for beliefs related to school closures. For the third press conference, we again document some instability in our finding for beliefs about general restrictions, corroborating our conclusion that this result should be interpreted with some caution.

Table A.5

Estimation results: Sensitivity to the functional form of the time trend.

	1st Press Conference				2nd Press Conference				3rd Press Conference
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)	(11)	(12)
All restrictions	25∗∗∗	25∗∗∗	36∗∗∗	38∗∗∗	8	10	16	34∗∗∗	30∗∗∗	50∗∗∗	−16	25∗
	(4)	(3)	(6)	(7)	(6)	(5)	(10)	(9)	(8)	(7)	(15)	(12)
MDV (pre-event)	14 Oct	14 Oct	14 Oct	14 Oct	28 Nov	28 Nov	28 Nov	28 Nov	16 Nov	16 Nov	16 Nov	16 Nov
N	15,560	15,560	15,560	15,560	8,675	8,675	8,675	8,675	12,708	12,708	12,708	12,708
School closures	13∗∗	13∗∗	13	30∗∗∗	−6	0	−7	17	4	11	−11	2
	(4)	(4)	(7)	(8)	(8)	(7)	(14)	(12)	(7)	(7)	(13)	(11)
MDV (pre-event)	05 Jun	05 Jun	05 Jun	05 Jun	29 Jul	29 Jul	29 Jul	29 Jul	01 Aug	01 Aug	01 Aug	01 Aug
N	4,215	4,215	4,215	4,215	2,913	2,913	2,913	2,913	3,450	3,450	3,450	3,450
Bundesliga	−44∗∗∗	−20	−51∗	−3	−21	−28	−32	−2	−41∗	−28	−46	−50∗
	(12)	(11)	(22)	(21)	(18)	(15)	(33)	(26)	(16)	(14)	(27)	(24)
MDV (pre-event)	13 Dec	13 Dec	13 Dec	13 Dec	19 Jan	19 Jan	19 Jan	19 Jan	27 Dec	27 Dec	27 Dec	27 Dec
N	3,560	3,560	3,560	3,560	2,448	2,448	2,448	2,448	3,017	3,017	3,017	3,017
Time trend	local linear	global linear	local quadratic	global quadratic	local linear	global linear	local quadratic	global quadratic	local linear	global linear	local quadratic	global quadratic

Note: Civey Online Panel, April 2-May 27, 2020. Table shows the change in the expected duration of the restrictions in days associated with each of the three public announcements, for different specifications of the time trend. All estimates adjusted for gender (male/female), education (university/other), age (below/above 50), children in household (yes/no), region (northwest/south/east), population density (high/low), purchasing power (high/low), political party preference (Union/FDP, Red/Red/Green, AfD, Other) and county-level quartile of COVID-19 new infection rate. Estimation with standard errors clustered at the person-level in parentheses ∗, ∗∗, ∗∗∗.

Specification of the dependent variable In Table A.6 we directly estimate the effect on the categorical variables instead of using the constructed variable of expected duration. In more detail, in Panel A we estimate the effect of the press conferences on the probability of choosing the lowest category, in Panel B we focus on the probability of the highest category and in Panel C we use an ordered probit to estimate how the shares in all categories are shifted. Again, we find significant effects of the first press conference. For example, consistent with Figs. A.9a and b, we find that after the first press conference the probability of choosing the lowest category is reduced for all restrictions and school restrictions while the probability for the highest category increases.

Table A.6

Estimation results: Ordered probit and LPM of boundary responses.

	1st Press Conference			2nd Press Conference			3rd Press Conference
	All	School	BL	All	School	BL	All	School	BL
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
A. Lowest category	−0.03∗∗	−0.12∗∗∗	0.03∗	−0.01	0.06	0.03	−0.08∗∗∗	−0.01	0.01
	(0.01)	(0.03)	(0.01)	(0.01)	(0.04)	(0.03)	(0.02)	(0.04)	(0.03)
MDV (pre-event)	0.09	0.42	0.04	0.09	0.21	0.04	0.13	0.23	0.09
B. Highest category	0.06∗∗∗	0.02∗	−0.08∗∗	0.03	0.01	−0.02	0.06∗	−0.01	−0.09∗∗
	(0.01)	(0.01)	(0.03)	(0.02)	(0.02)	(0.04)	(0.03)	(0.02)	(0.04)
MDV (pre-event)	0.24	0.01	0.19	0.37	0.04	0.22	0.31	0.04	0.19
C. Ordered probit	0.28∗∗∗	0.39∗∗∗	−0.04	0.22∗∗∗	−0.00	−0.12	0.29∗∗	0.15	−0.24
	(0.04)	(0.09)	(0.09)	(0.06)	(0.11)	(0.12)	(0.10)	(0.13)	(0.14)
N	15,560	4,215	3,560	8,675	2,913	2,448	12,708	3,450	3,017

Note: Civey Online Panel, April 2-May 27, 2020. Panels A and B show results from linear probability models with the dependent variables indicating if individuals chose the lower bound category (Panel A, “Within the next 4 weeks”) or the upper bound category (Panel B, “More than 9 months” for all restrictions and school closures, “More than 12 months” for football Bundesliga). Panel C shows results from ordered probit regressions of the categorical response variables excluding undecided and extreme responses (“Don’t know”/“Never”). Coefficient estimates from multivariate local polynomial regression discontinuity estimation with local linear time trends. Estimation with standard errors clustered at the person level in parentheses. MDV = mean dependent variable measured before the event. ∗, ∗∗, ∗∗∗.

Fig. A.9

Expected duration of restrictions and of economic recovery. Notes: Means with 95%-C.I.; Data source: ifo Manager Panel 2020.

Placebo tests To provide empirical support for our econometric strategy, we conduct a series of placebo analyses and estimate our model for all available dates in the survey period. We use a two-day distance before and after each true press conference date and each placebo date, excluding only those days for which the two-day distance generates overlap with true treatments or for which there are fewer than four data points available (in the beginning and in the end of the survey period).17 This results in 37 placebo estimates and three treatment estimates for each of the three outcomes (all restrictions, school closures and Bundesliga). We distinguish two types of placebo estimates: (i) ‘true’ placebo days at which no event took place and (ii) days at which events other than the three main press conferences took place. These other events include speeches and meetings of Angela Merkel and the prime ministers that were also related to COVID-19 but which, in contrast to the three events followed by the main press conferences studied, did not change official COVID-19 regulation. Given the narrow two-day time corridor around each date, we cannot estimate a local linear trend and use a linear trend in this set of placebo analyses instead.

In Fig. 4 we show the tests for the question on general restrictions.18 Out of the 37 placebo coefficients, 32 (36) are insignificant at a significance level of 95% (99%). Point estimates are mostly small or even negative. This is also documented in Panel b where we plot the empirical cumulative distribution function of placebo and treatment estimates. The coefficients of the first and the third press conference are above the 90% percentile of all estimates, which underlines the impact of the events in terms of magnitude of the point estimates. However, the effect of the third press conference is estimated much less precisely. Taken together with the evidence from other sensitivity analyses presented above, we are confident in our estimate regarding the first press conference, but more cautious in interpreting the effect of the third press conference.

Fig. 4

Placebo checks and comparison to other events: All restrictions. Notes: Plots show coefficient estimates of the pre/post indicator from multivariate regression with an interacted linear trend and a two-day distance to cut-off. Red squares denote the three press conference treatments (PC), blue triangles show other events: (1,4) Corona Cabinet, (2) Easter address A. Merkel, (3) Leopoldina report, (5) Government statement A. Merkel, (6,8) Meeting A. Merkel with trade associations and unions, (7) Government interrogation Bundestag, (9) Meeting with OECD, IMF and ILO. Data source: Civey Online Panel 2020.

Uncertainty in beliefs

In this section, we investigate if policy communication also affects uncertainty in beliefs. We analyze the occurrence of undecided and extreme responses to the expectations questions and, in addition, analyze the second moments of individuals’ beliefs before and after the press conferences.

Undecided and extreme responses Based on linear probability models with our preferred multivariate specification, we analyze whether the press conferences changed undecided and extreme response behavior (Table A.11 in Appendix D). The share of respondents who said they did not know when restrictions would end moderately increased over time. Moreover, we find that the first press conference significantly increased the share of undecided respondents, indicating a moderate increase in uncertainty. There is also some evidence that the second press conference reduced the probability of choosing an extreme response to the question on general restrictions: the share of respondents who said restrictions would never end decreased by 6 percentage points. Overall, however, extreme response behavior remains relatively stable over time and does not vary systematically with policy communication.

Table A.11

Effect of the press conferences on undecided and extreme responses (LPM).

	1st Press Conference			2nd Press Conference			3rd Press Conference
	All	School	BL	All	School	BL	All	School	BL
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
A. Don’t know	0.02∗∗	0.06∗∗∗	0.08∗∗∗	0.00	−0.02	−0.07∗	0.01	−0.02	0.04
	(0.01)	(0.01)	(0.02)	(0.01)	(0.02)	(0.04)	(0.01)	(0.02)	(0.03)
MDV (pre-event)	0.05	0.05	0.16	0.08	0.08	0.20	0.08	0.08	0.16
N	17,671	4,581	4,560	10,763	3,227	3,274	15,545	3,841	4,024
B. Will never end	0.01	0.01	−0.02	−0.06∗∗∗	−0.00	0.03	−0.01	−0.02	−0.06∗∗
	(0.01)	(0.01)	(0.02)	(0.01)	(0.01)	(0.02)	(0.02)	(0.01)	(0.02)
MDV (pre-event)	0.06	0.01	0.06	0.12	0.02	0.07	0.12	0.02	0.09
N	17,671	4,581	4,560	10,763	3,227	3,274	15,545	3,841	4,024

Note: Civey Online Panel, April 2-May 27, 2020. Linear probability models with the dependent variables indicating if individuals say they do not know when restrictions will end (Panel A) or if they say they will never end (Panel B). Coefficient estimates from multivariate local polynomial regression discontinuity estimation with local linear time trends. Estimation with standard errors clustered at the person-level in parentheses. MDV = mean dependent variable measured before the event. ∗, ∗∗, ∗∗∗.

Second moment analysis Intuitively, one can expect individuals’ expectations to be more responsive to policy communication when prior uncertainty is high. While we do not measure uncertainty directly, we can compare the variability in individuals’ responses before and after the press conferences (Fig. A.8 in Appendix D). Overall, variability in expectations is relatively high. For example, the standard deviation of beliefs before the first press conference is 124 days for general restrictions and 54 days for restrictions related to school closures. Variability is highest in expectations about the duration of sports restrictions, with a baseline standard deviation of 171 days. We run two-sample variance comparison tests to analyze descriptively if standard deviations changed after the press conferences. We find a sizeable and significant decrease by 14 days in the standard deviation of beliefs about the duration of sports restrictions following the first press conference (p = 0.007), and further reductions that do not pass the threshold of statistical significance after the other two events. Hence, although the press conferences did not significantly affect mean beliefs about sport restrictions, a descriptive analysis indicates that policy communication may have reduced the dispersion of these beliefs. The variability in beliefs about general restrictions did not change notably after the first two press conferences but fell slightly by 4 days following the third press conference (p = 0.0237). In contrast, the variability in beliefs about restrictions related to schooling increased significantly after the first (13 days, p = 0.0000) and the third (9 days, p = 0.0000) press conference. This is partly explained by an overall increase in the variability of beliefs about school restrictions over time, and we note that we do not control for a time trend in this analysis. Nevertheless, these descriptive results corroborate our main findings that individuals adjusted their expectations about when children would be back in school in response to policy announcements.

Fig. A.8

Standard deviation pre/post event. Notes: Plots show the standard deviation of beliefs about all restrictions, restrictions related to schooling and sport restrictions before and after the three press conferences. Data source: Civey Online Panel 2020.

Heterogeneous effects

Next, we explore effect heterogeneity and study the extent to which policy communication has different effects on subgroups and varies over the distribution of expectations.

First, we study whether individuals’ responses to the press conferences differ by individual characteristics (Table 3 ).19 We split the sample by observable variables, such as gender, education, age, presence of children in the household, political preferences, region, and regional exposure to Corona20 and run separate local polynomial regression discontinuity estimates using Eq. Appendix 1 with local linear time trends for the different subgroups. Overall, effect heterogeneity is rather low. The point estimates suggest different effects by gender. Specifically, the first and the third press conference shift expectations of women more than those of men. This gender difference is consistent with previous results on gender differences in COVID-19 attitudes and behavior. For example, using data from eight OECD countries including Germany, Galasso et al. (2020) show that women are more likely to perceive the pandemic as a very serious health problem and are also more likely to agree and to comply with restraining measures. Moreover, Coibion et al. (2022) propose that women may respond more strongly to information treatments because of lower ex-ante confidence in their beliefs. While point estimates are different, the gender differences are no longer significant when accounting for multiple hypotheses testing. Therefore we report the gender effect with caution.21 The same holds true for the other subgroups: effects are not significantly different. One important reason for this lack of heterogeneity might be related to the high uncertainty and missing knowledge about COVID-19, which affects all groups alike.

Table 3

Heterogeneity: Expectation updating in response to new COVID-19 announcements.

	1st Press Conference			2nd Press Conference			3rd Press Conference
	All	School	BL	All	School	BL	All	School	BL
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
Full sample	25∗∗∗	13∗∗	−44∗∗∗	8	−6	−21	30∗∗∗	4	−41∗
Demographic variation
Women	30∗∗∗	18	−37	7	−1	−35	45∗∗	18	−51
Men	22∗∗∗	11∗∗	−49∗∗∗	9	−7	−18	23∗	0	−38∗
University	21∗∗∗	16∗∗	−51∗∗	12	3	4	24∗	7	−46∗
No university	29∗∗∗	9	−38∗	4	−19	−46	37∗∗	−1	−37
Age: < 40	35∗∗	−5	−62	13	23	−353∗∗∗	35	−62	−58
Age: 40–64	24∗∗∗	14∗∗∗	−45∗∗∗	7	−7	−13	29∗∗∗	6	−38∗
Age: 65+	27∗∗∗	10∗	−44∗∗	11	−17	13	45∗∗∗	17	−21
Children in HH	25∗∗	25∗	−60	18	−15	−141∗∗	14	−8	−84∗
No children	25∗∗∗	11∗∗	−42∗∗∗	7	−5	−8	33∗∗∗	6	−32
Political pref.: Union/FDP	28∗∗∗	13∗	−50∗	15	−16	−45	35∗∗	−1	−19
Political pref.: RRG	18∗∗	9	−29	−3	−6	3	19	3	−45
Political pref.: AfD	31∗∗∗	14	−87∗∗	12	−4	12	31	9	−82∗
Geographic variation
Region: North/West	19∗∗	6	−57∗∗	20∗	−2	−27	33∗	−1	−35
Region: South	25∗∗∗	19∗∗	−45∗	5	−10	−31	28∗	10	−5
Region: East	32∗∗∗	14	−24	−7	1	−1	28	−2	−98∗∗∗
Pop. Density: High	21∗∗∗	9	−32	9	−0	−31	24	−5	−28
Pop. Density: Low	27∗∗∗	15∗∗	−56∗∗∗	8	−9	−12	35∗∗∗	11	−50∗
Purch. Power: High	21∗∗∗	17∗∗	−50∗∗	4	−8	−22	30∗	−8	−21
Purch. Power: Low	27∗∗∗	9	−39∗	11	−1	−22	29∗∗	17	−60∗∗
By COVID-19 exposure
New cases/ state: < P25	76∗∗∗	0	−51	0	−9	−70	15	−6	−75
New cases/ state: P25-P50	39∗∗∗	27∗∗	−20	−5	−24	14	31	−23	−65
New cases/ state: P50-P75	18∗	2	−49∗	8	−17	−24	42∗	−0	−73∗
New cases/ state: > P75	18∗∗∗	16∗∗	−47∗∗	11	2	−26	24∗	14	−12

Note: Civey Online Panel, April 2-May 27, 2020. Table shows the change in the expected duration in days associated with each of the three public announcements for various subgroups. Coefficient estimates from local polynomial regression discontinuity estimation, adjusted for gender (male/female), education (university/other), age (below/above 50), children in household (yes/no), region (northwest/south/east), population density (high/low), purchasing power (high/low), political party preference (Union/FDP, Red/Red/Green, AfD, Other) and county-level quartile of COVID-19 new infection rate. Estimation with standard errors clustered at the person-level. ∗, ∗∗, ∗∗∗.

In Table 4 , we turn to the effects of policy communication on the distribution of expectations and present results from unconditional quantile regressions. Specifically, we present estimates of Eq. 1 for the median as well as the 25th and the 75th percentiles using the method proposed by Firpo et al. (2009). Policy communication does not just shift mean expectations but it also significantly affects the distribution. The pattern for the expectations regarding all restrictions suggests that more optimistic individuals, who expect a shorter duration of restrictions, respond more strongly to policy communication than more pessimistic individuals with expectations at the 75th percentile. Interestingly, we even find sizeable and significant effects for the second press conference at the 25th percentile and the median, despite the insignificant mean effect. The pattern for the duration of school restrictions and restrictions of sports events is less clear and more sensitive to specification choices (see Table A.7 in the Appendix).

Table 4

Heterogeneity: Quantile regression estimates.

	1st Press Conference				2nd Press Conference				3rd Press Conference
	Mean	Q25	Q50	Q75	Mean	Q25	Q50	Q75	Mean	Q25	Q50	Q75
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)	(11)	(12)
All restrictions	25∗∗∗	18∗∗∗	35∗∗∗	11∗∗∗	8	74∗∗∗	90∗∗∗	8∗	30∗∗∗	45∗∗∗	34	0
	(4)	(3)	(6)	(3)	(6)	(7)	(11)	(4)	(8)	(13)	(19)	(6)
N	15,560	15,560	15,560	15,560	8,675	8,675	8,675	8,675	12,708	12,708	12,708	12,708
School closures	13∗∗	−6∗	−12∗∗	−17∗∗∗	−6	−10	33∗∗∗	57∗∗∗	4	3	3	18
	(4)	(3)	(4)	(3)	(8)	(8)	(6)	(10)	(7)	(9)	(7)	(11)
N	4,215	4,215	4,215	4,215	2,913	2,913	2,913	2,913	3,450	3,450	3,450	3,450
Bundesliga	−44∗∗∗	15	−31	−18	−21	−11	−10	−64∗	−41∗	−21	−52	−36
	(12)	(10)	(18)	(18)	(18)	(15)	(26)	(27)	(16)	(24)	(31)	(33)
N	3,560	3,560	3,560	3,560	2,448	2,448	2,448	2,448	3,017	3,017	3,017	3,017

Note: Civey Online Panel, April 2-May 27, 2020. Table shows the change in the expected duration of restrictions in days associated with each of the three public announcements. Mean effects from local polynomial regression discontinuity estimation. Coefficient estimates at the 25th, 50th and 75th percentile (Q25/Q50/Q75) based on unconditional quantile regression discontinuity estimation with local linear time splines at four knots around the event. Estimates adjusted for gender (male/female), education (university/other), age (below/above 50), children in household (yes/no), region (northwest/south/east), population density (high/low), purchasing power (high/low), political party preference (Union/FDP, Red/Red/Green, AfD, Other) and county-level quartile of COVID-19 new infection rate. Standard errors in parentheses. ∗, ∗∗, ∗∗∗.

Table A.7

Quantile regression estimates: Sensitivity to the functional form of the time trend.

	1st Press Conference				2nd Press Conference				3rd Press Conference
	Mean	Q25	Q50	Q75	Mean	Q25	Q50	Q75	Mean	Q25	Q50	Q75
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)	(11)	(12)
All restrictions
Local trend	25∗∗∗	18∗∗∗	35∗∗∗	11∗∗∗	8	74∗∗∗	90∗∗∗	8∗	30∗∗∗	45∗∗∗	34	0
Global trend	25∗∗∗	22∗∗∗	27∗∗∗	7∗∗	10	62∗∗∗	74∗∗∗	−7∗	50∗∗∗	32∗∗∗	35∗∗	12∗∗∗
N	15,560	15,560	15,560	15,560	8,675	8,675	8,675	8,675	12,708	12,708	12,708	12,708
School closures
Local trend	13∗∗	−6∗	−12∗∗	−17∗∗∗	−6	−10	33∗∗∗	57∗∗∗	4	3	3	18
Global trend	13∗∗	5∗∗	1	−7∗∗	0	−11	35∗∗∗	45∗∗∗	11	−6	−7	3
N	4,215	4,215	4,215	4,215	2,913	2,913	2,913	2,913	3,450	3,450	3,450	3,450
Bundesliga
Local trend	−44∗∗∗	15	−31	−18	−21	−11	−10	−64∗	−41∗	−21	−52	−36
Global trend	−20	15∗	−26∗	2	−28	−23	−56∗∗	−40	−28	−47∗∗	−29	17
N	3,560	3,560	3,560	3,560	2,448	2,448	2,448	2,448	3,017	3,017	3,017	3,017

Note: Civey Online Panel, April 2-May 27, 2020. Table shows the change in the expected duration of restrictions in days associated with each of the three public announcements. Mean effects and unconditional quantile regression estimates at the 25th, 50th and 75th percentile (Q25/Q50/Q75) from multivariate regression discontinuity estimation with local or global time trends. Estimates adjusted for gender (male/female), education (university/other), age (below/above 50), children in household (yes/no), region (northwest/south/east), population density (high/low), purchasing power (high/low), political party preference (Union/FDP, Red/Red/Green, AfD, Other) and county-level quartile of COVID-19 new infection rate. ∗, ∗∗, ∗∗∗.

Behavioral effects: Evidence from planned expenses and mobility indicators

In the final section, we analyze whether the policy communication by Angela Merkel and the German government succeeded in changing the actions and the behavior of German citizens. Data on planned expenses provide some evidence about intended consumption behavior. In addition, we use aggregate mobility data and explore if individuals reduced their mobility in response to the press conferences. We present behavioral effects on planned consumption here and briefly discuss the findings from mobility data, but relegate the details of the mobility analysis to Appendix F.

To measure consumption behavior, we collected information about planned non-routine consumption expenditures between April 2 and May 27, 2020, using an additional Civey survey. The question wording is as follows: Are you planning an unusually high expense within the next 3 months, e.g. for a car, a vacation or a construction measure? Individuals can choose from eight categorical response categories: No, 0–1000 euros, 1001–2500 euros, 2501–5000 euros, 5001–10000 euros, 10001–15000 euros, 15001–20000 euros, more than 20000 euros. Based on the midpoints of the categorical response categories, we construct a measure of planned expenses. We also construct a binary indicator that distinguishes positive and zero amounts to study extensive margin responses.

In contrast to individual expectations about the duration of the pandemic, planned consumption expenses do not show a clear time trend (Fig. 5 ). Moreover, the share of individuals with no planned consumption expenses remains relatively stable over time at about 70 percent. Using regression analysis and the same specification as above, we find no robust extensive margin responses in planned non-routine consumption expenses to any of the three press conferences of Angela Merkel (Table 5 ). Yet, there is some evidence that planned expenses on the intensive margin decreased after the first press conference. The result is even stronger for those individuals with positive planned expenses. However, these intensive-margin effects are not present after the other two press conferences.

Fig. 5

Planned non-routine consumption expenditure over time. Notes: Plots show how responses evolved over time. Solid vertical lines indicate the three press conferences (PC). Data source: Civey Online Panel 2020.

Table 5

Estimation results: Planned non-routine consumption expenses.

	1st Press Conference		2nd Press Conference		3rd Press Conference
	(1)	(2)	(3)	(4)	(5)	(6)
Pr(Planned expenses>0)	−0.06∗	−0.08	−0.01	0.02	−0.09∗	−0.12
	(0.03)	(0.06)	(0.04)	(0.08)	(0.04)	(0.07)
MDV (pre-event)	0.35	0.35	0.30	0.30	0.34	0.34
N	4,441	4,441	3,246	3,246	3,901	3,901
Planned expenses (euros)	−1874∗∗∗	−2567∗∗	0	−556	−438	−745
	(483)	(884)	(595)	(1030)	(538)	(938)
MDV (pre-event)	3296	3296	2657	2657	2935	2935
N	4,441	4,441	3,246	3,246	3,901	3,901
Planned expenses (euros)	−3957∗∗∗	−5846∗∗	24	−2805	534	452
excl. zeros	(1054)	(1944)	(1474)	(2743)	(1160)	(1969)
MDV (pre-event)	9445	9445	8832	8832	8712	8712
N	1,471	1,471	1,045	1,045	1,320	1,320
Covariates	✓	✓	✓	✓	✓	✓
Time trend	local linear	local quadratic	local linear	local quadratic	local linear	local quadratic

Note: Civey Online Panel, April 2-May 27, 2020. Table shows the change in planned non-routine consumption expenditure associated with each of the three public announcements. Local polynomial regression discontinuity estimation with local linear trend or with local quadratic trend, adjusted for gender (male/female), education (university/other), age (below/above 50), children in household (yes/no), region (northwest/south/east), population density (high/low), purchasing power (high/low), political party preference (Union/FDP, Red/Red/Green, AfD, Other) and county-level quartile of COVID-19 new infection rate. Estimation with standard errors clustered at the person-level in parentheses. ∗, ∗∗, ∗∗∗.

Next, we conduct an analysis of mobility behavior before and after the three press conferences. We use aggregate data on the daily mobility of German citizens, relative to pre-year mobility, based on mobile communications data. Mobility plummeted in March 2020, but was beginning to rise again during the time period that we study (Fig. A.11). We then compare how mobility changed after each of the three press conferences, using a similar regression discontinuity design and adjusting for the overall time trend. We find no significant shift in the level of mobility after the first and the third press conference, but some evidence of a trend break in the slope of mobility changes after the latter (Fig. A.12, Table A.13). The second press conference did reduce the mobility of German citizens significantly, but the effect was only short-lived.

Fig. A.11

Mobility behavior in 2020 relative to pre-year. Notes: Plot shows the holiday-adjusted mobility in Germany in 2020 relative to 2019, based on aggregate mobile communications data. Data source: Destatis Mobility Indicators.

Fig. A.12

Conditional means with linear fit. Notes: Plots show the holiday-adjusted mobility in Germany relative to pre-year mobility before and after each press conference (PC). The dashed vertical lines denote the press conference cut-offs. The solid trend lines are based on regressions using binned data. Data source: Destatis Mobility Indicators.

Table A.13

Estimation results: Mobility behavior relative to pre-year.

	1st Press Conference	2nd Press Conference	3rd Press Conference
boldΔMobility/2019	5.95	−19.14∗∗∗	2.19
	(6.81)	(3.25)	(2.60)
N	11	12	13
MDV (pre-event)	−32.33∗∗∗	−21.86∗∗∗	−27.33∗∗∗
	(−13.77)	(−24.71)	(−10.06)

Note: Destatis Mobility Indicators, 2020. Table shows the change in holiday-adjusted mobility relative to pre-year mobility in percentage points associated with each of the three public announcements. Regressions are obtained using binned daily data. Coefficient estimates from bivariate regression discontinuity estimation with global linear trend. Standard errors in parentheses. ∗, ∗∗, ∗∗∗.

Our results on the behavioral effects of policy communication should be interpreted with some caution. First, planned expenses are only an approximation of realized consumption expenses. Second, the intensive-margin effects on consumption behavior obtained in the regression models are not consistent over time. Third, the evidence on mobility patterns is based on aggregate data, which may conceal heterogeneous effects.

Taken together, our results suggest only a limited role of policy communication for consumption behavior and for citizens’ mobility, which is consistent with the results of previous literature. For example, Coibion et al. (2020b) do not find any effect of the expected duration of the COVID-19 pandemic on individuals’ marginal propensity to consume out of stimulus checks in the United States. In their analysis of “unconventional” fiscal and monetary policy measures, D’Acunto et al. (2022) find mixed effects of policy announcements on households’ consumption plans: while the announcement of a VAT change in Germany affected planned spending on durables, the ECB’s forward guidance on inflation does not appear to have such effects. It would be interesting to study the effects of the COVID-19 press conferences further with data on realized consumption and with individual-level data on mobility. In particular, public announcements about the severity of the pandemic might also contain additional information about the state of the economy and might change behavior indirectly (an “information effect” as described by Nakamura and Steinsson (2018)).

Conclusion

In this paper, we provide empirical evidence that the expectation management of policy leaders can affect the expectation formation of the public. For identification, we use variation in expectations about the duration of restrictions before and after press conferences of German policy makers in the first months of the COVID-19 pandemic. We conduct a large online survey to elicit individuals’ beliefs about the duration of three well-defined restrictions: (1) when will the majority of school children be back in school; (2) when will the main football league return to normal operations with stadium visitors; and (3) when will all current restrictions related to the Corona crisis be fully lifted?.

While Coibion et al. (2020a) do not find effects of policy communication in survey experiments in the US, we use real-world press conferences in Germany as natural experiments. Our results show that policy communication indeed did affect expectations in the case at hand. In particular, we find that the first press conference of Angela Merkel and her colleagues had a sizable impact and significantly prolonged the expected duration of the regulatory measures. In this press conference, Merkel conveyed a strong sense of caution. Studying the heterogeneous effects of the policy communication, we document a surprising pattern of consistency. We only find some differences in responses by gender, but no consistent and significant differences by education, age, region, regional exposure to COVID-19, or political preferences. Moreover, our results suggest that policy communication is most effective for individuals with higher ex-ante optimism in expectations (i.e. individuals who expect a shorter duration of restrictions).

Table A.2

Sample characteristics of Civey respondents.

	Civey Online Panel			Official
	(Q1)	(Q2)	(Q3)
Female	29.5	23.0	20.9	50.7
Age categories
18–39 yrs.	1.5	1.1	1.0	16.3
30–39 yrs.	5.7	4.4	4.4	15.5
40–49 yrs.	10.9	8.5	8.4	14.7
50–64 yrs.	38.2	35.8	35.8	27.5
65+ yrs.	43.7	50.2	50.4	26.0
Region
North/West	33.3	33.4	34.9	37.7
South	42.1	41.5	41.3	42.8
East	24.5	25.1	23.7	19.5
Political party preference
Union/FDP	40.9	37.1	39.1	45.0
Red/Red/Green (RRG)	33.8	34.6	34.8	39.0
AfD	20.6	24.6	22.9	9.0
Other	4.7	3.7	3.2	7.0
Education
University degree	52.1	55.7	55.6	19.6
Vocational degree	45.9	42.3	42.4	59.2
No degree	2.0	2.0	2.0	20.8

Note: Cells contain shares in percent. Official statistics on gender, education, age and region from Federal Statistical Office (Destatis), based on 2019 microcensus and 2019 forward projection of 2011 census. Official statistics on political party preference based on Forsa Sonntagsfrage of May 30th, 2020. Civey samples differ by question: Q1 = All restrictions, Q2 = School closures, Q3 = Bundesliga.

Table A.8

Estimation results: Sensitivity to 20% inattentiveness rate.

	1st Press Conference			2nd Press Conference			3rd Press Conference
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
All restrictions	25∗∗∗	22∗∗∗	22∗∗∗	−13∗∗∗	3	3	33∗∗∗	41∗∗∗	26∗∗
MDV (pre-event)	08 Oct	08 Oct	08 Oct	18 Nov	18 Nov	18 Nov	06 Nov	06 Nov	06 Nov
S.D. (pre-event)	123	123	123	130	130	130	129	129	129
N	15,191	15,191	15,191	8,626	8,626	8,626	12,592	12,592	12,592
School closures	24∗∗∗	19∗∗∗	19∗∗∗	3	−5	−8	14∗∗∗	4	1
MDV (pre-event)	21 Jun	21 Jun	21 Jun	08 Aug	08 Aug	08 Aug	12 Aug	12 Aug	12 Aug
S.D. (pre-event)	79	79	79	90	90	90	90	90	90
N	4,077	4,077	4,077	2,821	2,821	2,821	3,346	3,346	3,346
Bundesliga	10	−11	−31∗	−20∗	−30	−27	13	−26	−30
MDV (pre-event)	05 Dec	05 Dec	05 Dec	05 Jan	05 Jan	05 Jan	19 Dec	19 Dec	19 Dec
S.D. (pre-event)	176	176	176	178	178	178	177	177	177
N	3,545	3,545	3,545	2,462	2,462	2,462	3,046	3,046	3,046
Covariates	-	✓	✓	-	✓	✓	-	✓	✓
Time trend	-	global linear	local linear	-	global linear	local linear	-	global linear	local linear

Note: Civey Online Panel, April 2-May 27, 2020. Table shows the change in the expected duration in days associated with each of the three public announcements, assuming 20% of respondents were inattentive and chose answers randomly. Columns 1/4/7 show estimates from bivariate OLS on a binary indicator that takes on 0 if the outcome was measured before the event and 1 if it was measured after the event. Columns 2/5/8 show multivariate OLS estimates with a global linear time trend centered at zero at the event interacted with the before/after indicator. Columns 3/6/9 present multivariate local linear polynomial regression discontinuity estimates. Covariates include gender (male/female), education (university/other), age (below/above 50), children in household (yes/no), region (northwest/south/east), population density (high/low), purchasing power (high/low), political party preference (Union/FDP, Red/Red/Green, AfD, Other) and county-level quartile of COVID-19 new infection rate. MDV = mean dependent variable measured before the event. S.D. = baseline standard deviation in days. Estimation with standard errors clustered at the person-level. ∗, ∗∗, ∗∗∗.

Table A.9

Estimation results: Expectation updating in response to new COVID-19 announcements.

	1st Press Conference			2nd Press Conference			3rd Press Conference
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
All restrictions	32∗∗∗	32∗∗∗	30∗∗∗	−10∗	12	11	39∗∗∗	51∗∗∗	34∗∗
MDV (pre-event)	16 Oct	16 Oct	16 Oct	05 Dec	05 Dec	05 Dec	28 Nov	28 Nov	28 Nov
S.D. (pre-event)	123	123	123	129	129	129	130	130	130
N	15,560	15,560	15,560	8,675	8,675	8,675	12,708	12,708	12,708
School closures	27∗∗∗	8	10	3	22	10	16∗	15	3
MDV (pre-event)	11 Jun	11 Jun	11 Jun	04 Aug	04 Aug	04 Aug	08 Aug	08 Aug	08 Aug
S.D. (pre-event)	64	64	64	80	80	80	83	83	83
N	4,215	4,215	4,215	2,913	2,913	2,913	3,450	3,450	3,450
Bundesliga	12	−11	−37∗	−42∗∗	−32	−35	2	−51	−75∗
MDV (pre-event)	25 Dec	25 Dec	25 Dec	15 Feb	15 Feb	15 Feb	12 Jan	12 Jan	12 Jan
S.D. (pre-event)	171	171	171	176	176	176	183	183	183
N	3,560	3,560	3,560	2,448	2,448	2,448	3,017	3,017	3,017
Covariates	-	✓	✓	-	✓	✓	-	✓	✓
Time trend	-	global linear	local linear	-	global linear	local linear	-	global linear	local linear

Note: Civey Online Panel, April 2-May 27, 2020. Table shows the change in the expected duration in days associated with each of the three public announcements. Columns 1/4/7 show estimates from bivariate OLS on a binary indicator that takes on 0 if the outcome was measured before the event and 1 if it was measured after the event. Columns 2/5/8 show multivariate OLS estimates with a global linear time trend centered at zero at the event interacted with the before/after indicator. Columns 3/6/9 present multivariate local linear polynomial regression discontinuity estimates. Covariates include gender (male/female), education (university/other), age (below/above 50), children in household (yes/no), region (northwest/south/east), population density (high/low), purchasing power (high/low), political party preference (Union/FDP, Red/Red/Green, AfD, Other) and county-level quartile of COVID-19 new infection rate. MDV = mean dependent variable measured before the event. S.D. = baseline standard deviation in days. Estimation adjusted for population weights with standard errors clustered at the person-level. ∗, ∗∗, ∗∗∗.

Table A.10

Heterogeneity: Expectation updating in response to new COVID-19 announcements.

	1st Press Conference			2nd Press Conference			3rd Press Conference
	All	School	BL	All	School	BL	All	School	BL
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
Full sample	30∗∗∗	10	−37∗	11	10	−35	34∗∗	3	−75∗
Demographic variation
Women	32∗∗∗	−0	−35	11	21	7	52∗∗	11	−104∗
Men	27∗∗∗	17∗	−46∗	13	−7	−66∗	14	−8	−47∗
University	27∗∗∗	20∗	−61∗∗	17	21	−13	18	20	−59
No university	31∗∗∗	−2	−18	8	−15	−46	47∗∗	−15	−73
Age: < 40	32	−20	−45	17	30	−291∗∗	51	−86	−121
Age: 40–64	30∗∗∗	18∗∗	−33∗	6	3	−9	30∗∗	13	−53
Age: 65+	33∗∗∗	11	−43∗	−2	−20	−7	55∗∗∗	25∗	−39
Children in HH	34∗∗	30∗	−53	20	18	−99	32	−3	−121∗
No children	28∗∗∗	4	−37∗	9	3	−13	35∗∗	4	−60
Political pref.: Union/FDP	33∗∗∗	5	−46	23∗	−5	−45	49∗∗∗	−3	−39
Political pref.: RRG	25∗∗	2	−28	−5	19	26	16	10	−85
Political pref.: AfD	38∗∗∗	16	−92∗	9	−8	−131∗	15	−21	−136∗∗
Geographic variation
Region: North/West	21∗∗	−3	−45	7	16	−52	36∗	29	9
Region: South	32∗∗∗	9	−51∗	16	3	−53	34∗	−38	−101∗
Region: East	35∗∗∗	29∗	−12	3	6	34	31	30	−138∗∗
Pop. Density: High	24∗∗	−2	−31	16	22	−16	35∗	1	−60
Pop. Density: Low	32∗∗∗	15	−45∗	8	2	−44	34∗	12	−83∗
Purch. Power: High	28∗∗∗	−0	−65∗∗	9	8	−43	41∗∗	−33	−44
Purch. Power: Low	30∗∗∗	16	−15	13	12	−35	25	35	−86∗
By COVID-19 exposure
New cases/ state: < P25	74∗∗∗	21	−137∗	−8	10	−49	−11	13	−44
New cases/ state: P25-P50	50∗∗∗	44∗∗∗	−0	29	−20	34	53∗	−1	−93
New cases/ state: P50-P75	23∗	6	−13	−1	18	−14	31	57	−81
New cases/ state: > P75	21∗∗	1	−52∗	12	6	−63∗	31∗	−16	−63

Note: Civey Online Panel, April 2-May 27, 2020. Table shows the change in the expected duration in days associated with each of the three public announcements for various subgroups. Coefficient estimates from local polynomial regression discontinuity estimation, adjusted for gender (male/female), education (university/other), age (below/above 50), children in household (yes/no), region (northwest/south/east), population density (high/low), purchasing power (high/low), political party preference (Union/FDP, Red/Red/Green, AfD, Other) and county-level quartile of COVID-19 new infection rate. Estimation adjusted for population weights with standard errors clustered at the person-level. ∗, ∗∗, ∗∗∗.

Table A.12

Comparison of households’ and managers’ expectations.

	All restrictions		School closures		Bundesliga
	HH	MP	HH	MP	HH	MP
	(1)	(2)	(3)	(4)	(5)	(6)
MDV (pre)	Aug 20	Jan 21	May 20	May 20	Oct 20	Sep 20
MDV (post)	Dec 20	May 21	Aug 20	Aug 20	Feb 21	Dec 20
Δ(pre/post) BV	111∗∗∗	118∗∗∗	92∗∗∗	96∗∗∗	108∗∗∗	96∗∗∗
	(5)	(13)	(4)	(4)	(9)	(9)
Δ(pre/post) MV	112∗∗∗	131∗∗∗	93∗∗∗	97∗∗∗	106∗∗∗	103∗∗∗
	(5)	(16)	(4)	(5)	(10)	(12)
N	3,294	489	1,459	491	1,236	479

Note: Civey Online Panel 2020 for household (HH) expectations and Manager Panel (MP) 2020 for firm expectations. Table shows mean expectations before (MDV pre = April 6–8) and after (MDV post = May 25–27) the three press conferences, as well as coefficient estimates for the change in expectations over time (in days), using bivariate OLS ( BV) and multivariate OLS ( MV). All estimates with standard errors clustered at the person level in parentheses. ∗, ∗∗, ∗∗∗.